University of Southern California Dissertations and Theses

Real-time channel sounder designs for millimeter-wave and ultra-wideband communications

(USC Thesis Other)

Real-time channel sounder designs for millimeter-wave and ultra-wideband communications

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content REAL-TIME CHANNEL SOUNDER DESIGNS FOR MILLIMETER-WA VE AND ULTRA-WIDEBAND COMMUNICATIONS by Celalettin Umit Bas A dissertation presented to the FACULTY OF THE USC GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In partial fulfillment of the requirements for the degree DOCTOR OF PHILOSOPHY (ELECTRICAL ENGINEERING) August 2018 Copyright 2018 Celalettin Umit Bas Acknowledgements First and foremost, I would like to express my deepest gratitude to my Ph.D. research advisor Dr. Andreas F. Molisch. I am grateful for his unwavering support and guidance since my first day in USC. I would also like to thank my dissertation and qualification exam committee members: Dr. Hossein Hashemi, Dr. Keith Chugg, Dr. Mahta Moghaddam, and Dr. Leana Golubchik. I would like to thank my fellow research group members: Sundar Aditya, Daoud Burghal, Thomas Choi, Jorge Gomez Ponce, Zheda Li, Hao Feng, Vishnu V . Ratnam, Ming-Chun Lee, and Huseein Hammoud. Special thanks to Rui Wang, Seun Sangodoyin, and Vinod Kristem for their contributions to the work presented in this thesis. It has been a great pleasure working with all of you. I am also grateful to our administrative staff: Susan Wiedem, Gerrielyn Ramos, Corine Wong, and Diane Demetras for their help throughout my Ph.D. I would like to thank my friends in Los Angeles: Enes, Hilmi, ˙ Ilknur, Ahsan, Vanessa, Pancham. I am also thankful for my friends from Turkey: Yalc ¸ın, Pelin, Tolga, S ¸ebnem, Burak, Duygu, Umur, Remziye, Berk, Bekir, Merve G., Merve E., ˙ Iskender, S ¸aziye, Nihan, Caner, and Meltem. Despite the distance apart, I have always felt their support and love. I would like to express my deepest gratitude and appreciation to my parents; Ayfer and Nevzat, and my sister (and my first teacher) Kadriye, for their unconditional love and sup- port. Finally, I would like to thank my beloved ˙ Irem whose love, support and patience have 1 encouraged me to pursue my dreams. I am forever grateful to have you in my life. 2 Contents List of Figures 8 List of Tables 16 Abstract 18 1 Introduction 21 1.1 Overview of Channel Sounding . . . . . . . . . . . . . . . . . . . . . . . . . 21 1.2 Thesis Outline and Contributions . . . . . . . . . . . . . . . . . . . . . . . . 27 1.2.1 Real-Time Millimeter-Wave Directional Channel Sounder . . . . . . 27 1.2.2 28 GHz Dynamic Double Directional Propagation Channel Measure- ments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 1.2.3 28 GHz Microcell Propagation Channel Measurements . . . . . . . . 30 1.2.4 28 GHz Foliage Propagation Channel Measurements . . . . . . . . . 30 1.2.5 28 GHz Outdoor to Indoor Propagation Channel Measurements . . . 31 1.2.6 Real-time Ultra-Wideband Channel Sounder Design . . . . . . . . . 32 1.3 Other Publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 2 Real-Time Millimeter-Wave Directional Channel Sounder 35 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 3 2.2 Channel Sounder Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 2.2.1 Sounding Waveform . . . . . . . . . . . . . . . . . . . . . . . . . . 42 2.2.2 Sounder Hardware . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 2.2.3 Sounder Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 2.3 System Verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 2.3.1 Beam-Steering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 2.3.2 Frequency Response . . . . . . . . . . . . . . . . . . . . . . . . . . 54 2.3.3 Phase Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 2.3.4 Path loss Verification . . . . . . . . . . . . . . . . . . . . . . . . . . 58 2.4 Post Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 2.4.1 Power Delay Profile and Power Angular Delay Profile . . . . . . . . 60 2.4.2 Multi-path Extraction . . . . . . . . . . . . . . . . . . . . . . . . . . 61 2.5 Sample Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 2.5.1 Static Directional Measurements . . . . . . . . . . . . . . . . . . . . 64 2.5.2 Dynamic Directional Measurements . . . . . . . . . . . . . . . . . . 65 2.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 3 28 GHz Dynamic Double Directional Propagation Channel Measurements 75 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75 3.2 Measurement Campaign . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 3.3 Measurement Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 3.3.1 Case 1 : Blocking Objects . . . . . . . . . . . . . . . . . . . . . . . 80 3.3.2 Case 2 : Moving Scatterers . . . . . . . . . . . . . . . . . . . . . . . 85 3.3.3 Case 3 : Blocked LOS . . . . . . . . . . . . . . . . . . . . . . . . . 89 3.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 4 4 28 GHz Microcell Propagation Channel Measurements 93 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93 4.2 Measurement Campaign . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95 4.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102 4.3.1 Path Loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102 4.3.2 RMS Delay Spread . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 4.3.3 Extracted Multi-paths . . . . . . . . . . . . . . . . . . . . . . . . . . 107 4.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 5 28 GHz Foliage Propagation Channel Measurements 109 5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 5.2 Measurement Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 5.3 Measurement Environments . . . . . . . . . . . . . . . . . . . . . . . . . . 112 5.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115 5.4.1 Foliage Loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117 5.4.2 Delay Spread . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119 5.4.3 Angular Spread . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122 5.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124 6 28 GHz Outdoor to Indoor Propagation Channel Measurements 125 6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 6.2 Measurement Campaign . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128 6.2.1 Measurement Environments . . . . . . . . . . . . . . . . . . . . . . 128 6.2.2 Measurement Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . 132 6.2.3 Data Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134 6.3 Deterministic Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135 5 6.3.1 Multi-path Components . . . . . . . . . . . . . . . . . . . . . . . . 135 6.3.2 O2I Penetration Loss . . . . . . . . . . . . . . . . . . . . . . . . . . 141 6.4 Statistical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144 6.4.1 O2I Penetration Loss . . . . . . . . . . . . . . . . . . . . . . . . . . 144 6.4.2 Angular Spread . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146 6.4.3 Root Mean Square Delay Spread . . . . . . . . . . . . . . . . . . . . 152 6.5 3GPP-type Model Paremeters for Residential O2I . . . . . . . . . . . . . . . 155 6.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156 7 Real-time Ultra-Wideband Channel Sounder Design 158 7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158 7.2 Channel Sounder Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162 7.2.1 Single Band Measurements . . . . . . . . . . . . . . . . . . . . . . . 164 7.2.2 Multi Band Measurements . . . . . . . . . . . . . . . . . . . . . . . 167 7.3 Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171 7.3.1 Sub-band Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . 171 7.3.2 Stitching Multiple Bands . . . . . . . . . . . . . . . . . . . . . . . . 173 7.4 System Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177 7.4.1 Dynamic Range and Measurable Path Loss . . . . . . . . . . . . . . 177 7.4.2 2-path Coaxial Channel Validation Measurements . . . . . . . . . . . 180 7.4.3 Over The Air (OTA) Validation Measurements . . . . . . . . . . . . 180 7.4.4 Impact of Antenna Pattern and Multiple Input Multiple Output (MIMO) Extension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181 7.5 Measurement Campaign . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184 7.5.1 Noise Averaging and Interference Filtering . . . . . . . . . . . . . . 186 7.5.2 Averaged Power Delay Profile (APDP) computation . . . . . . . . . 186 6 7.5.3 Delay Spread . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189 7.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196 8 Conclusions and Future Directions 197 Bibliography 201 7 List of Figures 1.1 Typical outdoor propagation environment . . . . . . . . . . . . . . . . . . . 22 1.2 Channel sounding basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 1.3 Frequency domain channel sounder . . . . . . . . . . . . . . . . . . . . . . 23 1.4 Time-domain channel sounder . . . . . . . . . . . . . . . . . . . . . . . . . 24 1.5 Multiple transmitter and receiver antennas . . . . . . . . . . . . . . . . . . . 25 1.6 Virtual antenna arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 1.7 Parallel links and antennas . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 1.8 Switched antenna arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 2.1 Rotating horn antenna approach . . . . . . . . . . . . . . . . . . . . . . . . 37 2.2 Beam-switching phased array antenna approach . . . . . . . . . . . . . . . . 38 2.3 In-phase and Quadrature components of the baseband signal, with a sampling rate of 1.25 GSps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 2.4 Spectrum of the baseband multi-tone sounding signal with 801 tones spaced by 500 kHz . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 2.5 Normalized magnitude of the sounding signal . . . . . . . . . . . . . . . . . 41 2.6 TX Block Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43 2.7 RX Block Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43 2.8 TX RFU Block Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 8 2.9 Antenna array . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 2.10 Spectrum of the sounding signal at IF . . . . . . . . . . . . . . . . . . . . . 48 2.11 Beam patterns in Azimuth . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 2.12 Received power vs TX and RX beam pairs when LOS component is at RX anglex and TX angley and (x;y) =f(20; 15); (5; 15); (30; 15)g . . . . . 53 2.13 Received Power Spectrum with different RX beams . . . . . . . . . . . . . . 53 2.14 Phase drift with shared reference, free-running references and GPS-disciplined references . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 2.15 Normalized delay-Doppler function in dB scale for the measurements with the GPS-disciplined references . . . . . . . . . . . . . . . . . . . . . . . . . 56 2.16 Received waveform when switching between 2 beams; Beam #9 and Beam #10 with azimuth directions of5 and 0 , respectively . . . . . . . . . . . 57 2.17 Phase drift for Beam #9 and Beam #10 . . . . . . . . . . . . . . . . . . . . 58 2.18 Line-of-sight (LOS) path loss . . . . . . . . . . . . . . . . . . . . . . . . . . 59 2.19 Significant MPCs (The shown area is 350 m by 80 m) . . . . . . . . . . . . . 61 2.20 Power delay profile for the position given in Fig. 2.19 . . . . . . . . . . . . 63 2.21 DoA Power angular-delay profile (dB) for the location in Fig. 2.19 . . . . . . 65 2.22 DoD Power angular-delay profile (dB) for the location in Fig. 2.19 . . . . . . 66 2.23 Received power (dB) vs TX and RX beams for the location in Fig. 2.19 . . . 66 2.24 DoA and delay for extracted MPCs for the location in Fig. 2.19 . . . . . . . 67 2.25 The environment for the dynamic measurements . . . . . . . . . . . . . . . . 67 2.26 PDP(dB) vs time, 3 moving cars are marked with black arrows and the pedes- trian is marked with red arrow . . . . . . . . . . . . . . . . . . . . . . . . . 69 2.27 Time varying RMS-DS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70 2.28 Mean angle and angular spreads vs time . . . . . . . . . . . . . . . . . . . . 70 9 2.29 Path gains vs time: i)RX and TX tracks the best beam pair, ii) stays on the which best on the average . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71 2.30 Doppler spectrum vs time . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 2.31 Doppler spectrum vs delay att = 7:38s . . . . . . . . . . . . . . . . . . . . 72 3.1 Timeline for the channel sounder operation . . . . . . . . . . . . . . . . . . 77 3.2 Measurement environment . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 3.3 Blocked LOS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 3.4 PDP vs time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 3.5 Power of the MPCs vs DOA/DOD and time . . . . . . . . . . . . . . . . . . 82 3.6 Mean angles and angular spreads vs time . . . . . . . . . . . . . . . . . . . . 83 3.7 Root mean square delay spread vs time . . . . . . . . . . . . . . . . . . . . . 84 3.8 Path gains for best two beam-pairs vs time . . . . . . . . . . . . . . . . . . . 84 3.9 Moving Scatterers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 3.10 Temporal evolution of the multipath components . . . . . . . . . . . . . . . 86 3.11 Doppler spectrum for case 2 . . . . . . . . . . . . . . . . . . . . . . . . . . 87 3.12 Mean angles and angular spreads vs time for Case 2 . . . . . . . . . . . . . . 88 3.13 Path gains for best beam-pairs vs time for Case 2 . . . . . . . . . . . . . . . 88 3.14 Path gain vs time for Case 3 . . . . . . . . . . . . . . . . . . . . . . . . . . 90 3.15 RMS-DS vs time for Case 3 . . . . . . . . . . . . . . . . . . . . . . . . . . 90 3.16 Mean angles and angular spreads vs time for Case 3 . . . . . . . . . . . . . . 91 4.1 Measurement locations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 4.2 RX View on 28th St facing east . . . . . . . . . . . . . . . . . . . . . . . . . 98 4.3 TX View on 28th St facing west . . . . . . . . . . . . . . . . . . . . . . . . 98 4.4 Power-Angular Delay Profile for RX 14 . . . . . . . . . . . . . . . . . . . . 99 10 4.5 Power-Angular Spectrum for RX 14 . . . . . . . . . . . . . . . . . . . . . . 99 4.6 Power Delay Profile for RX 14 . . . . . . . . . . . . . . . . . . . . . . . . . 100 4.7 Path-loss for 28th St. and NLoS locations . . . . . . . . . . . . . . . . . . . 101 4.8 CDF of the fading for 28th St. . . . . . . . . . . . . . . . . . . . . . . . . . 104 4.9 CDF of the fading for NLoS locations . . . . . . . . . . . . . . . . . . . . . 104 4.10 CDF of the logarithm of RMS-DS for 28th St. . . . . . . . . . . . . . . . . . 106 4.11 CDF of the logarithm of RMS-DS for NLoS locations . . . . . . . . . . . . . 106 4.12 Extracted multi-path components . . . . . . . . . . . . . . . . . . . . . . . . 108 5.1 Foliage measurements, location 1 . . . . . . . . . . . . . . . . . . . . . . . . 113 5.2 Foliage measurements, location 3 . . . . . . . . . . . . . . . . . . . . . . . . 114 5.3 Foliage measurements, location 3 . . . . . . . . . . . . . . . . . . . . . . . . 114 5.4 Abstract model for location 3. . . . . . . . . . . . . . . . . . . . . . . . . . 115 5.5 Foliage loss for low and high RX locations with the fitted models . . . . . . . 118 5.6 CDF of the residual values from the proposed model in Table 5.2. . . . . . . 119 5.7 CDFs of the RMS-DS for omnidirectional RX and 90 TX sector for LOS and Foliage shadowed links . . . . . . . . . . . . . . . . . . . . . . . . . . 120 5.8 CDF of the RMS-DS for directional RX and TX with 12 half power beam width for LOS and Foliage shadowed links . . . . . . . . . . . . . . . . . . . 121 5.9 Omnidirectional power delay profiles for two links with LOS and foliage blockage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121 5.10 CDF sof DoA angular spreads for LOS and Foliage shadowed links . . . . . 123 5.11 CDFs of DoD angular spreads for LOS and Foliage shadowed links . . . . . 123 6.1 TX and RX locations for SFU1 . . . . . . . . . . . . . . . . . . . . . . . . . 129 6.2 Layout of RX points for SFU1 . . . . . . . . . . . . . . . . . . . . . . . . . 129 11 6.3 TX and RX locations for SFU2 . . . . . . . . . . . . . . . . . . . . . . . . . 130 6.4 Layout of RX points for SFU2 . . . . . . . . . . . . . . . . . . . . . . . . . 130 6.5 TX and RX locations for BOB . . . . . . . . . . . . . . . . . . . . . . . . . 131 6.6 Layout of RX points for the BOB . . . . . . . . . . . . . . . . . . . . . . . . 131 6.7 Detected MPC components vs DoA - outdoor O1 in BOB, the color and the size of the point indicate the path gain of the MPC . . . . . . . . . . . . . . 136 6.8 Detected MPC components vs DoA - indoor I2 in BOB, the color and the size of the point indicate the path gain of the MPC . . . . . . . . . . . . . . 136 6.9 MPCs vs delay and DoA - indoor I2 in BOB, the color and the size of the point indicate the path gain of the MPC . . . . . . . . . . . . . . . . . . . . 137 6.10 Detected MPCs vs DoA - indoor I7 in SFU1, the color and the size of the point indicate the path gain of the MPC . . . . . . . . . . . . . . . . . . . . 138 6.11 Angular spreads for all RX locations in SFU1, red arrows indicate the mean direction of arrivals, black arrows indicate the RX beam directions for the five highest MPCs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138 6.12 Angular spreads for all RX locations in SFU2 for TX2, red arrows indicate the mean direction of arrivals, black arrows indicate the RX beam directions for the five highest MPCs. . . . . . . . . . . . . . . . . . . . . . . . . . . . 139 6.13 Angular spreads for all RX locations in BOB, red arrows indicate the mean direction of arrivals, black arrows indicate the RX beam directions for the five highest MPCs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139 6.14 Omnidirectional path loss vs the RX locations in SFU1, the indoor locations ordered with respect to the distance from the window . . . . . . . . . . . . . 141 6.15 Omnidirectional path loss vs the RX locations in SFU2-TX1, the indoor lo- cations ordered with respect to the distance from the window . . . . . . . . . 141 12 6.16 Omnidirectional path loss vs the RX locations in SFU2-TX2 the indoor loca- tions ordered with respect to the distance from the window . . . . . . . . . . 142 6.17 Omnidirectional path loss vs the RX locations in BOB, the indoor locations ordered with respect to the distance from the window . . . . . . . . . . . . . 142 6.18 CDF of path loss in the SFU1 for omnidirectional and the directional with the best beam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145 6.19 CDF of path loss in the SFU2 for omnidirectional and the directional with the best beam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145 6.20 CDF of path loss in the BOB for omnidirectional and the directional with the best beam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146 6.21 CDF of DOA angular spread in the SFU1 . . . . . . . . . . . . . . . . . . . 148 6.22 CDF of DOA angular spread in the SFU2 . . . . . . . . . . . . . . . . . . . 148 6.23 CDF of DOA angular spread in the BOB . . . . . . . . . . . . . . . . . . . 148 6.24 SFU1 - Ordered relative power (dB) of other sectors with respect to the sector with the highest power, dashed line indicated where it drops below 10 dB . . 149 6.25 SFU2 TX1 - Ordered relative power (dB) of other sectors with respect to the sector with the highest power, dashed line indicated where it drops below 10 dB . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150 6.26 SFU2 TX2 - Ordered relative power (dB) of other sectors with respect to the sector with the highest power, dashed line indicated where it drops below 10 dB . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150 6.27 BOB - Ordered relative power (dB) of other sectors with respect to the sector with the highest power, dashed line indicated where it drops below 10 dB . . 150 6.28 CDF of RMS-DS in the SFU1 for omnidirectional and the directional with the best beam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152 13 6.29 CDF of RMS-DS in the SFU2 for omnidirectional and the directional with the best beam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153 6.30 CDF of RMS-DS in the BOB for omnidirectional and the directional with the best beam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153 6.31 Sample PDPs for outdoor and indoor RX location for SFU2 . . . . . . . . . . 153 7.1 Transmitter block diagram, the descriptions of the units are given in Table 7.1 162 7.2 Receiver block diagram, the descriptions of the units are given in Table 7.1 . 162 7.3 Measured azimuth patterns of the biconical antenna for frequencies between 3-18 GHz . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166 7.4 Measured elevation patterns of the biconical antenna for frequencies between 3-18 GHz . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166 7.5 Operation of the timing module to trigger AWG (TX) or ADC (RX) and frequency synthesizers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169 7.6 The frequency plan for the multi-band measurements, overlap of the adjacent sub-bands are used to estimate the discontinuities of the frequency and phase responses. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169 7.7 Sideband suppression before and after IQ imbalance correction, the input of the IQ mixer modified to include only the tones in the upper sideband of the LO. The power observed on the left of LO are due to IQ imbalance. . . . . . 174 7.8 Measured sub-band frequency responses (colored lines) and calibrated/patched frequency response (black line) . . . . . . . . . . . . . . . . . . . . . . . . . 175 7.9 Measured sub-band phase responses (colored lines) and calibrated/patched frequency response (black line) . . . . . . . . . . . . . . . . . . . . . . . . . 175 7.10 PDP for the system response with and without phase correction - PDP is shifted on the delay axis for presentation purposes . . . . . . . . . . . . . . . 177 14 7.11 Dynamic range of the RX, black dashed line indicates the RX sensitivity level, magenta line indicate the RX 1 dB compression point . . . . . . . . . . 178 7.12 The analytical and the measured frequency responses for the 2-path test channel179 7.13 The analytical and the measured PDPs for the 2-path test channel, close-in shows the two paths created by using the coaxial components . . . . . . . . 179 7.14 Comparison of the measured frequency responses for same environment with VNA and the proposed setup . . . . . . . . . . . . . . . . . . . . . . . . . . 182 7.15 Comparison of the measured power delay profiles for same environment with VNA and the proposed setup . . . . . . . . . . . . . . . . . . . . . . . . . . 182 7.16 Floor plan for 1st floor (each grid corresponds to 10m) . . . . . . . . . . . . 185 7.17 Floor plan for 2nd floor (each grid corresponds to 10m) . . . . . . . . . . . . 185 7.18 APDP(dB) vs TX-RX distance for the second flor LOS measurements . . . . 188 7.19 Spreading (delay-Doppler) function, power delay profile and Doppler spec- trum for the same measurement snapshot. Stationary TX and RX is moving away from TX with a speed of 0.2 m/s, TX-RX distance: 66.65 m. . . . . . . 190 7.20 RMS-DS for LOS measurements versus TX-RX distance . . . . . . . . . . . 192 7.21 RMS-DS for NLOS measurements versus TX-RX distance, between 26 m and 40 m, 62 m and 70 m, the RMS-DS is significantly higher due to addi- tional paths from outdoor . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192 7.22 1st and 2nd floor NLOS APDP at 35 m, the MPCs with delays more than 500 ns are caused by the reflections from surrounding buildings. . . . . . . . 193 7.23 LOS RMS-DS in logarithmic scale along with means versus the center fre- quency of sub-bands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194 7.24 NLOS RMS-DS in logarithmic scale along with means versus the center fre- quency of sub-bands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194 15 List of Tables 2.1 mm-wave Channel Sounder comparison . . . . . . . . . . . . . . . . . . . . 36 2.2 mm-wave Channel Sounder comparison . . . . . . . . . . . . . . . . . . . . 39 2.3 List of part numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 2.4 Sounder specifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 2.5 Extracted MPC parameters vs Estimated from Google Maps for the location in Fig. 2.19 (only lists a subset of the MPCs) . . . . . . . . . . . . . . . . . 68 3.1 Sounder Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 4.1 Sounder specifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96 4.2 Parameters of the path loss models . . . . . . . . . . . . . . . . . . . . . . . 103 4.3 Parameters of the RMS-DS . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 5.1 Sounder specifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113 5.2 Parameters for the foliage penetration loss model . . . . . . . . . . . . . . . 118 5.3 Delay spread statistics and lognormal fit parameters . . . . . . . . . . . . . 122 5.4 Angular spread statistics and lognormal fit parameters . . . . . . . . . . . . 124 6.1 Sounder specifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133 6.2 Mean path-loss and penetration loss values . . . . . . . . . . . . . . . . . . . 147 6.3 Angular spread statistics and lognormal fit parameters . . . . . . . . . . . . 149 16 6.4 RMS-DS statistics and lognormal fit parameters . . . . . . . . . . . . . . . . 154 7.1 List of the units shown in Figs. 7.1 and 7.2 along with their descriptions and relevant specifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163 7.2 Channel sounder specifications, providing the values used throughout the chapter. Note, however, that the values can be modified on a per-campaign basis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170 7.3 Parameter values for the 2-path test channel . . . . . . . . . . . . . . . . . . 180 7.4 Estimated parameters for the frequency dependent RMS-DS model given in Eq. 7.13 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195 17 Abstract An accurate model for the wireless propagation channel is imperative for designing and test- ing wireless systems. Propagation channel measurements, also known as channel sounding, is the most accurate way to acquire true characteristics of the wireless propagation channel. This dissertation presents two novel channel sounder designs millimeter-wave and ultra-wideband communications. Due to the unique propagation characteristics at those frequencies, accurate channel mod- els are even more important for millimeter-wave bands. It is anticipated that most of the future millimeter-wave systems will utilize beam-forming antenna arrays to overcome the higher path loss that occurs at higher frequencies. Consequently, the angular spectrum and its tem- poral evolution are vital for the efficient design of such systems. Hence, we built a 28 GHz channel sounder which is capable of directionally-resolved measurements in dynamic envi- ronments. Unlike the common practice of using the rotating horn antennas to investigate the directional characteristics of millimeter-wave channels, the developed sounder is capable of performing horizontal and vertical beam steering with the help of phased arrays. With the fast beam-switching capability, the proposed sounder can perform measurements that are di- rectionally resolved both at the transmitter and receiver in 1.44 milliseconds. This not only enables measurement of more transmitter-receiver locations for better statistical inference but also allows to perform directional analysis in dynamic environments. The short measure- 18 ment time combined with the high phase stability limits the phase drift between transmit- ter and receiver, enabling phase-coherent sounding of all beam pairs even when transmitter and receiver have no cabled connection for synchronization. This ensures that the measure- ment data are suitable for high-resolution parameter extraction algorithms and other coherent processing techniques. Furthermore, we present results from measurement campaigns that investigate various aspects at the 28 GHz such as; dynamic environments with moving scat- terers and blocking objects, suburban micro-cellular environments, foliage attenuation are presented. The second channel sounder is an ultra-wideband, real-time channel sounder. It can per- form measurements over the continuous band from 3 GHz to 18 GHz by utilizing a hybrid time/frequency domain approach. The developed sounder sequentially measures overlapping sub-bands of 1 GHz bandwidth to measure up to 15 GHz total bandwidth in 6 ms. This al- lows us to use a waveform generator and a digitizer with relatively lower sample rates and significantly lowers the cost compared to sampling the whole band at once. With the trans- mit power of 40 dBm, the maximum measurable path loss is 132 dB without accounting for possible antenna gains. In this work, we also describe the calibration procedure along with the method to patch sub-bands into a combined channel response. This approach is validated by comparing the results from vector network analyzer measurements performed in the same static environment. The developed channel sounder is real-time and can measure in dynamic environments with maximum Doppler spreads up to 166.7 Hz. At 18 GHz this would corre- spond to a maximum speed of 5 km/h, which is the typical pedestrian speed. In comparison, a vector network analyzer, which is the standard choice of equipment for such large bandwidth measurements, would require several hundreds of milliseconds for a similar measurement, would require a static channel. Additionally, unlike a vector network analyzer, the transmit- ter and the receiver in our setup are physically separated, and they dont require any cable 19 connections since the synchronization is provided by GPS-stabilized rubidium frequency ref- erences. Hence, the sounder can operate in almost any desired measurement environment. The frequency resolution (subcarrier spacing) in our sounder is 500 kHz, corresponding to a maximum measurable excess run-length of multi-path components of 600 m (2 s pseu- dorange), which is sufficient even for most outdoor measurements. Additionally, we present sample results from a measurement campaign performed in an indoor corridor investigating the delay spreads statistics over the range of 3-18 GHz. 20 Chapter 1 Introduction 1.1 Overview of Channel Sounding Wireless propagation channel consists of several paths from the transmitter (TX) to the re- ceiver (RX), see Figure 1.1. Each of these paths may undergo different interactions (e.g., reflections, scattering, diffractions and transmissions) with the interacting objects in the en- vironment. Since TX, RX or the interacting objects might be moving, this whole process is most often is time-varying. The design of any wireless system requires a thorough understanding of the wireless prop- agation channel to perform satisfactorily in thes complex environments. With a well-built model, the designed communication system can compensate or even benefit from the char- acteristics of the corresponding propagation channel. For example, path-loss and shadow- ing characteristics determine distance-dependent outage probability, while delay dispersion determines the spacing of subcarriers (in OFDM) or length of equalizers (in single-carrier systems). Furthermore, all these channel parameters heavily depend on the propagation en- vironment. Hence, measurements of channel characteristics and creation of models derived from them in the environment of interest are vital. 21 Figure 1.1: Typical outdoor propagation environment Figure 1.2: Channel sounding basics 22 Channel sounding is the procedure of exciting the wireless propagation channel with a known signalx(t) and recording received signaly(t) to estimate the channel responseh(t) as seen in Figure 1.2. Channel sounder is the general name for setups specifically designed for performing wireless propagation channel measurements. They can be grouped into two main categories; • Frequency domain channel sounders: These sounders usually based on vector network analyzers (VNA) which are capable of measuring the frequency response of a channel by performing a frequency sweep and calculating S21 at multiple frequency tones. Although VNAs can directly be used for channel sounding [1], one can also extend the VNAs capabilities by adding up/down converters, amplifiers, switched antenna arrays or positioners [2] as shown in Figure 1.3. Figure 1.3: Frequency domain channel sounder • Time domain channel sounders: These sounders consist of separate units for the trans- mitter and the receiver. A pulse generator or an arbitrary waveform generator is used for transmitter while an ADC or a digital sampling scope is used for the receiver. RF units 23 such as mixers and amplifiers are added to these baseband units to up/down-convert the sounding signal to and from the frequency of interest. Additionally, the transmitter and the receiver are synchronized by using either a shared timebase, a shared trigger (Figure 1.4) or highly stable dedicated frequency references. Figure 1.4: Time-domain channel sounder VNA-based sounders are easy to calibrate and don’t require any efforts for synchroniza- tion. However, their applications are fairly limited as the transmitter and receiver antennas can not be separated by large distances. More importantly, the time span during a single fre- quency sweep is usually several hundred milliseconds. Since this is larger than the coherence time of any practical dynamic environment, they are mainly used (or should be used) for mea- surements in static environments [3]. This drawback is even more limiting for multiple input multiple output (MIMO) measurements since the channel sounder needs to perform several sweeps to capture the MIMO channel. Although the development, operation, and calibration of a time domain sounder can be significantly more difficult, they can cover almost any practical scenarios including but not limited to macro/micro cellular environments, highly dynamic vehicle to vehicle communi- cations, multi-level indoor environments, moving transmitter, receiver or scatterers. Both Figures 1.3 and 1.4 depict single input single output (SISO) channel sounders. SISO measurements can only provide insights about non-directional channel characteristics such 24 Figure 1.5: Multiple transmitter and receiver antennas as path-loss, shadowing, root mean delay spread etc. Directional characteristics can only be obtained with a channel sounder which employs real or virtual antenna arrays. A SISO channel sounder can be upgraded toMIMO by using one of the following methods; • Virtual antenna arrays: A virtual antenna array, shown in Figure 1.6 can be created by moving a single antenna with a positioner. Although this is usually the cheapest and the easiest option to implement, the measurement duration for each location can take hours. Consequently, it’s only suitable for static channels. • Parallel links and antennas: Another approach for MIMO measurements is building multiple SISO sounders working in parallel, Figure 1.7. This approach is suitable for dynamic channels. However, its cost is the highest and it requires complicated calibration. • Switched antenna arrays: In this approach, the SISO channel sounder is combined with antenna arrays and multi-throw electrical switches. Then, all TX-RX antenna pairs are measured in order by following a predefined switching pattern. If all desired links are measured within the coherence time of the channel, then the acquired data should be same with parallel measurements. Hence, this approach is also suitable for 25 Figure 1.6: Virtual antenna arrays Figure 1.7: Parallel links and antennas 26 Figure 1.8: Switched antenna arrays dynamic environments, given that the switches are fast enough. In this approach, the biggest challenge is ensuring a tight synchronization between TX and RX switching. Furthermore, usually the high-speed switches can only handle so much power (usually around 1-10 Watt), so the TX power can be limited. • Switched beam arrays: This novel approach is proposed in our recent papers [4,5], see Chapter 2 for details. 1.2 Thesis Outline and Contributions 1.2.1 Real-Time Millimeter-Wave Directional Channel Sounder In this chapter, we present a novel real-time MIMO channel sounder for 28 GHz. Until now, the common practice to investigate the directional characteristics of millimeter-wave chan- 27 nels has been the use of the rotating horn antennas. The sounder presented here is capable of performing horizontal and vertical beam steering with the help of phased arrays. With the fast beam-switching capability, the proposed sounder can perform measurements that are directionally resolved both at TX and RX in 1.44 milliseconds compared to the minutes or even hours required for rotating horn antenna sounders. This not only enables measure- ment of more TX-RX locations for a better statistical inference but also allows to perform directional analysis in dynamic environments. The short measurement time combined with the high phase stability limits the phase drift between TX and RX, enabling phase-coherent sounding of all beam pairs even when TX and RX have no cabled connection for synchro- nization. This ensures that the measurement data are suitable for high-resolution parameter extraction algorithms and other coherent processing techniques. Along with the system de- sign, this work also discusses the measurements performed for verification of the sounder performance. Furthermore, we present sample results from double directional measurements in dynamic environments. Related Publications • C.U. Bas, R. Wang, D. Psychoudakis, T. Henige, R. Monroe, J. Park, J. Zhang and A. F. Molisch; “Real-Time Millimeter-Wave MIMO Channel Sounder for Dynamic Directional Measurements”,submittedforjournalpublication • C. U. Bas, R. Wang, D. Psychoudakis, T. Henige, R. Monroe, J. Park, J. Zhang and A. F. Molisch; “A Real-Time Millimeter-Wave Phased Array MIMO Channel Sounder”, IEEEVTC-FALL2017 28 1.2.2 28 GHz Dynamic Double Directional Propagation Channel Mea- surements This chapter presents results from the (to our knowledge) first dynamic double-directionally resolved measurement campaign at mm-wave frequencies for an outdoor microcellular sce- nario. The measurements are performed with USC’s real-time channel sounder equipped with phased array antennas that can steer beams electrically in microseconds, allowing directional measurements in dynamic environments. Exploiting the phase coherency of the setup, the multi-path components can be tracked over time to investigate the temporal dependencies of the channel characteristics. We present results for time-varying path-loss, delay spread, mean angles and angular spreads observed at the transmitter (TX) and receiver (RX) in the presence of moving vehicles and pedestrians. Additionally, we investigate excess losses observed due to blockage by vehicles and compare the cases when TX and RX are using fixed beams or when they are capable of adjusting beam directions dynamically. Related Publications • C.U. Bas, R. Wang, S. Sangodoyin, S. Hur, K. Whang, J. Park, J. Zhang and A. F. Molisch; “Dynamic Double Directional Propagation Channel Measurements at 28 GHz”,IEEEVTC-SPRING2018(invitedpaper) • C.U. Bas, R. Wang, S. Sangodoyin, S. Hur, K. Whang, J. Park, J. Zhang and A. F. Molisch; “28 GHz Propagation Channel Measurements for 5G Microcellular Environ- ments”,ACES2018(invitedpaper) 29 1.2.3 28 GHz Microcell Propagation Channel Measurements This chapter presents results from the (to our knowledge) first double-directionally resolved measurement campaign at mm-wave frequencies in a suburban microcell. The measurements are performed with a real-time channel sounder equipped with phased antenna arrays that allows electrical beam steering in microseconds, and which can measure path-loss of up to 169 dB. Exploiting the phase coherency of the measurements in the different beams, we obtain both directional and omnidirectional channel power delay profiles without any delay uncertainty. We present statistics of channel characteristics such as path-loss, shadowing and delay spread results for line-of-sight and non-line-of-sight cases, as well as sample results for power angular spectrum and extracted multi-path components. Related Publications • C. U. Bas, R. Wang, S. Sangodoyin, S. Hur, K. Whang, J. Park, J. Zhang and A. F. Molisch; “28 GHz Microcell Measurement Campaign for Residential Environment”, IEEEGlobecom2017 • C.U. Bas, R. Wang, S. Sangodoyin, S. Hur, K. Whang, J. Park, J. Zhang and A. F. Molisch; “28 GHz Propagation Channel Measurements for 5G Microcellular Environ- ments”,ACES2018(invitedpaper) 1.2.4 28 GHz Foliage Propagation Channel Measurements This chapter presents the results from channel sounding campaigns that investigate the im- pact of foliage blockage on the wireless propagation channel characteristics at 28 GHz. The measurements are performed with a real-time channel sounder equipped with phased array antennas that allow beam-forming and electronic beam steering for directionally resolved 30 measurements. Thanks to the short measurement time and the excellent phase stability of the system, we obtain both directional and omnidirectional channel power delay profiles without any delay uncertainty. We investigate and derive models for the distance-dependent foliage attenuation for different receiver heights. Additionally, we compare the angular spread and delay spread for links with and without foliage blockage to understand the effects of foliage on all channel characteristics. Related Publications • C.U. Bas, R. Wang, S. Sangodoyin, S. Hur, K. Whang, J. Park, J. Zhang and A. F. Molisch; “28 GHz Foliage Propagation Channel Measurements”, submitted to IEEE Globecom2018 1.2.5 28 GHz Outdoor to Indoor Propagation Channel Measurements Outdoor to indoor penetration loss is one of the crucial challenges faced at millimeter-wave frequencies. This chapter presents the results from 28 GHz channel sounding campaigns performed to investigate the impact of this phenomenon on the wireless propagation chan- nel characteristics in microcell and fixed wireless access scenarios. The measurements are performed with a real-time channel sounder equipped with phased array antennas that allow beam-forming and electronic beam steering for directionally resolved measurements. Thanks to the short measurement time and the excellent phase stability of the system, we obtain both directional and omnidirectional channel power delay profiles without any delay uncertainty. We compare the measured path loss, delay spreads and angular spreads for indoor and out- door receiver locations for two different types of buildings. We find that the penetration loss strongly depends on the angle of incidence, and that the scatterers on the outside of the build- ing strongly impact how much power is coupled into the building. Based on the results, we 31 provide statistical models for path loss, delay spread, and angular spread. Related Publications • C.U. Bas, R. Wang, S. Sangodoyin, T. Choi, S. Hur, K. Whang, J. Park, J. Zhang and A. F. Molisch; “Outdoor to Indoor Penetration Propagation Channel Measurements at 28 GHz”,submittedforjournalpublication • C.U. Bas, R. Wang, T. Choi, S. Hur, K. Whang, J. Park, J. Zhang and A. F. Molisch; “Outdoor to Indoor Propagation Channel Measurements at 28 GHz for Fixed Wireless Access”,IEEEICC2018 1.2.6 Real-time Ultra-Wideband Channel Sounder Design This chapter presents system design, calibration and example measurements of a novel ultra- wideband, real-time channel sounder. The sounder operates in the frequency range from 3 GHz to 18 GHz thus providing a Fourier delay resolution of 66.7 ps. The developed sounder sequentially measures overlapping sub-bands of 1 GHz bandwidth to cover the whole fre- quency range. This allows us to use a waveform generator and a digitizer with relatively lower sample rates and significantly lowers the cost compared to sampling the whole band at once. With the transmit power of 40 dBm, the maximum measurable path loss is 132 dB without accounting for possible antenna gains. In this work, we also describe the calibration procedure along with the method to patch sub-bands into a combined channel response. This approach is validated by comparing the results from vector network analyzer (VNA) measure- ments performed in the same static environment. In contrast to VNAs, however, our sounder is real-time and can measure in dynamic environments with maximum Doppler spreads up to 166.7 Hz. Additionally, we present sample results from a measurement campaign performed 32 in an indoor corridor investigating the delay spreads statistics over the range of 3-18 GHz. Related Publications • C.U. Bas, V . Kristem, R. Wang, A. F. Molisch; “Real-time Ultra-Wideband Channel Sounder Design for 3-18 GHz”,submittedforjournalpublication • C. U. Bas, V . Kristem, R. Wang, A. F. Molisch; “Real-time Ultra-Wideband Frequency Sweeping Channel Sounder for 3-18 GHz”,IEEEMILCOM2017 1.3 Other Publications This work also enabled following papers: 1. C.U. Bas, V . Kristem, R. Wang, A. F. Molisch; “Indoor Wideband Channel Measure- ments and Modeling in the 3-18 GHz Band”,tobesubmittedforjournalpublication 2. S. Hur, H. Yu, J. Park, W. Roh, C. U. Bas, R. Wang,and A. F. Molisch; “Feasibility of Millimeter-wave Mobility based on Short-term Channel Measurement”, to appear in IEEECommunicationsMagazine 3. V . Kristem, C. U. Bas, R. Wang and A. F. Molisch; “Outdoor Wideband Channel Mea- surements and Modeling in the 3-18 GHz Band”, to appear Transactions on Wireless Communications 4. T. Choi, C. U. Bas, R. Wang, S. Hur, J. Park, J. Zhang and A. F. Molisch; “Measure- ment Based Directional Modeling of Dynamic Human Body Shadowing at 28 GHz”, submittedtoIEEEGlobecom2018 33 5. V . Kristem, C. U. Bas, R. Wang and A. F. Molisch; “Outdoor Macro-Cellular Channel Measurements and Modeling in the 3-18 GHz Band”,IEEEGlobecomm2017 6. R. Wang, C. U. Bas, S. Sangodoyin, S. Hur, K. Whang, J. Park, J. Zhang and A. F. Molisch; “Stationarity Region of Mm-Wave Channel Based on Outdoor Microcellular Measurements at 28 GHz”,IEEEMILCOM2017 34 Chapter 2 Real-Time Millimeter-Wave Directional Channel Sounder 2.1 Introduction The ever-growing need for higher data rates in wireless communications is motivating the use of previously unused spectrum. Consequently, the millimeter wave (mm-wave) band has become a key area of interest for next generation wireless communication systems due to the ample amount of bandwidth available at the frequencies higher than 6 GHz. It is now clear that mm-wave systems will be an essential component of 5th generation cellular networks [6]. The knowledge of true statistical characteristics of the propagation channel is impera- tive for designing and testing wireless systems [7, 8]. An accurate channel model is even more important for mm-wave bands, due to the unique propagation characteristics at those frequencies. It is anticipated that most of the future mm-wave systems will utilize beam- forming antenna arrays to overcome the higher path loss that occurs at higher frequencies. Hence the angular spectrum and its temporal evolution are vital for the efficient design of 35 Table 2.1: mm-wave Channel Sounder comparison Specifications USC Sounder [4] NYU Sounder [27] NIST Sounder [28, 29] Durham Sounder [30, 31] Keysight [32, 33] Center Frequency 27.85 GHz 28,38,60,73 GHz 28.5,60 GHz 30,60,90 GHz Up to 44 GHz Bandwidth 1 GHz 1 GHz 2 GHz 4 GHz 2 GHz TX EIRP 57.1 dBm 54.6 dBm 51.5 dBm 36.7 dBm 23 dBm + Ant. Gain Array Type Switched Beam Rotating Horn Switched Horn Switched TX: Switched RX: Parallel Array Size 1616 NA 168 88 88 RX An- tenna/Beam Gain 19.5 dBi 24.5 dBi 18.5 dBi 20.7 dBi - Direction switch- ing speed 2s >seconds 4s - - TX-RX Synchro- nization GPS-Rubidium Ref. Rubidium Ref. GPS-Rubidium Ref. Rubidium Ref. GPS-Rubidium Ref. such systems [9]. The basic operating principle of a channel sounder is to transmit a known waveform, so that the signal at the receiver can deconvolve the transmitted signal from the received signal to acquire the impulse response of the channel. The sounding waveforms can be pulses [10,11], pn-sequences [12–14], chirp signals [15, 16], or multitone sequences [17, 18] (see also Sec. II.A). Most of the existing directional channel sounders for indoor mm-wave systems are based on vector network analyzers (VNAs), which use slow chirp or frequency stepping, combined with virtual arrays (mechanical movement of a single antenna along a track). Such setups cannot operate in real-time, and need cabled connections between transmitter (TX) and receiver (RX); they are thus mostly used for static indoor channel measurements [19–22]. For outdoor measurements, the prevalent method for directionally resolved measurements is based on mechanically rotating horn antennas [23–27], whose operating principle is sketched in Fig. 1: a (single-input-single-output) channel sounder is connected to horn antennas that are mechanically rotated. For each pair of directions at TX and RX antennas, the sounder measures the impulse response. Note, however, that mechanical rotation requires measure- ment durations on the order of 0.5-5 hours for one measurement location. 36 Figure 2.1: Rotating horn antenna approach In this work, we present a novel real-time double-directional channel sounder setup that is suitable for super-resolution parameter estimation. The channel sounder operates at 28 GHz, although its design principles can be applied to other mm-wave bands as well. It is based on an approach of electronically-switched beams. By using arrays of antennas with phase shifters, we form beams into different directions at both the TX and the RX as shown in Fig. 2.2. With a control interface implemented in field programmable gate array (FPGA), we are capable of switching from one beam to another in less than 2s. Thus, the same effect as in rotating horn antennas (pointing beams in different directions) can be achieved in much shorter time. Since our proposed setup decreases the measurement time for a single MIMO snapshot from hours to milliseconds, it allows collection of tens of thousands of measurement points in a single measurement campaign, i.e., many orders of magnitude faster than with rotating horns. Furthermore, since all TX and RX antenna pairs can be measured within the coherence time of the channel, the developed sounder is suitable for measurement campaigns in dynamic environments. Also with proper repetition of MIMO snapshots, the temporal evolution of multipath components (MPC) can be tracked. The short measurement time together with careful RF design to reduce phase noise limits the relative accumulated phase drift between the sample clocks of the TX and RX within 37 Figure 2.2: Beam-switching phased array antenna approach a single MIMO snapshot even without a cabled connection for synchronizing the clocks. Thanks to this phase coherence, the measurement data are suitable for high resolution param- eter extractions algorithms such as RIMAX [34] or SAGE [35, 36]. These algorithms allow to reveal the true double-directional channels which only depend on the channel of interest and exclude all the effects from the hardware used in the measurements [37]. The other state-of-the-art channel sounders are listed in Table 2.2. Parallel to our work, three other groups have developed real-time capable mm-wave sounders: Ref. [28, 29] re- cently presented a sounder based on an array of horn antennas combined with an electronic switch at the RX and a single TX antenna. This design is capable of faster sounding, and is similar in spirit to our approach. There are two important differences to our setup: (i) it has significantly less equivalent isotropically radiated power (EIRP) even with the horn antennas (ii) the use of mechanically arranged horns limits the flexibility compared to our sounder, which can reconfigure the beam-patterns through software. The sounder described in Ref. [30, 31] is a multi-band and multi-antenna channel sounder operating at carrier fre- quencies of 30 GHz, 60 GHz and 90 GHz with 88, 22 and 22 antenna arrays, respec- tively. The MIMO operation is realized by employing switches at the intermediate frequency 38 Table 2.2: mm-wave Channel Sounder comparison Specifications USC Sounder [4] NYU Sounder [27] NIST Sounder [28, 29] Durham Sounder [30, 31] Keysight [32, 33] Center Frequency 27.85 GHz 28,38,60,73 GHz 28.5,60 GHz 30,60,90 GHz Up to 44 GHz Bandwidth 1 GHz 1 GHz 2 GHz 4 GHz 2 GHz TX EIRP 57.1 dBm 54.6 dBm 51.5 dBm 36.7 dBm 23 dBm + Ant. Gain Array Type Switched Beam Rotating Horn Switched Horn Switched TX: Switched RX: Parallel Array Size 1616 NA 168 88 88 RX An- tenna/Beam Gain 19.5 dBi 24.5 dBi 18.5 dBi 20.7 dBi - Direction switch- ing speed 2s >seconds 4s - - TX-RX Synchro- nization GPS-Rubidium Ref. Rubidium Ref. GPS-Rubidium Ref. Rubidium Ref. GPS-Rubidium Ref. (IF) along with parallel frequency conversions. However, nominal TX output powers are lim- ited to 16 dBm at 30 GHz, 7 dBm at 60 GHz and 4 dBm at 90 GHz. Furthermore, the MIMO order is lower than the 1616 achieved in our setup. Finally, Ref. [33] presents a sounder with 4 TX antennas multiplexed with a switch and 4 RX antennas with 4 down-conversion chains. Similar to [30], the TX power is limited to 24 dBm. Our channel sounder has been used to measure different aspects of the wireless propaga- tion channel at 28 GHz. In [38], we presented path loss and root mean square delay spread for a suburban microcell environment. Ref. [39] investigated the effects of outdoor to in- door penetration on path loss, penetration loss, delay spread and angular spread statistics for two different types of housing. In [40], we presented the first measurement results on the stationarity region of MIMO mm-wave channels, which were based on over 20 million chan- nel impulse responses measured on continuous routes. Finally, Ref. [41] discussed the first measurement campaign for a dynamic double-directionally resolved measurement campaign at mm-wave frequencies for an outdoor microcellular scenario. The real-time measurement capability of the setup enabled these measurements which are not possible with the rotating horn antenna channel sounders. 39 0 0.5 1 1.5 2 t ( s) -1 0 1 I Real part 0 0.5 1 1.5 2 t ( s) -1 0 1 Q Imaginary part Figure 2.3: In-phase and Quadrature components of the baseband signal, with a sampling rate of 1.25 GSps The rest of the chapter is organized as follows. Section 2.2 discusses the proposed chan- nel sounder setup. Section 2.3 explains the measurements that verify system performance. Section 2.4 discusses the post processing methods of measurement data. Section 3.2 presents results from static and dynamic directional measurement campaigns. Finally Section 7.6 summarizes results and suggests directions for future work. 2.2 Channel Sounder Design The general sounder architecture is described in Fig. 2.6 and 2.7. As for any channel sounder, a known waveform is generated in baseband, up-converted to passband and transmitted over the air. The received waveform is down-converted, sampled, and stored, for further post processing. The channel impulse response is extracted from the shape of the received signal. In order to extract the directional channel properties, repetitions of the sounding signal are 40 -800 -600 -400 -200 0 200 400 600 800 Frequency (MHz) -120 -100 -80 -60 -40 -20 0 20 40 |mag| / dB Figure 2.4: Spectrum of the baseband multi-tone sounding signal with 801 tones spaced by 500 kHz 0 0.5 1 1.5 2 t ( s) 0 0.5 1 1.5 |m(t)| - normalized Figure 2.5: Normalized magnitude of the sounding signal 41 sent into (and received from) different directions, where the transmit and receive beams are generated by phased arrays. The following section describes these aspects in more detail. 2.2.1 Sounding Waveform The baseband sounding waveform used throughout this work is a multi-tone waveform. Multi-tone waveforms are a good fit for channel sounding measurements since they can be flat in both in the time and the frequency domain. For mm-wave sounders, they are used e.g., in [17] and in [18]. The sounding waveform can be represented as: m(t) = N X n=N e j(n2ft+n) (2.1) where f is the tone spacing, 2N + 1 is the number of tones and n is the phase of the tone n. Fig. 2.3 and 2.4 show the in-phase and quadrature components, and the spectrum of the complex baseband waveformm(t), respectively. The spectrum of the sounding waveform has a flat-top providing the same signal-to-noise ratio at all frequency tones. As suggested in [42], the values of n can be modified to achieve a low peak to average power ratio (PAPR). Fig. 2.5 shows the normalized amplitude of the waveform which has a PAPR of 0:4 dB, allowing us to transmit with power as close as possible to the 1 dB compression point of the power amplifier without driving it into saturation. Note that while Zadoff-Chu sequences, which are, e.g., used in LTE, provide PAPR=1 under idealized circumstances, this does not hold true for filtered, oversampled sequences, which are relevant here. Our sequences outperform Zadoff-Chu by more than 1 dB. Although all these parameters can be modified on a per campaign basis, Table 7.2 lists the specifications used throughout this chapter. 42 Figure 2.6: TX Block Diagram Figure 2.7: RX Block Diagram 43 Table 2.3: List of part numbers (1) Agilent N8241 (2) Mini-Circuits ZEM-M2TMH+ (3) KL Microwave 11ED50-1900/T500-O/O (4) Pre-Amplifier (5) National Instruments PXIe-6361 (6) National Instruments PXIe-8135 (7) National Instruments PXIe-7961R (8) Phase Matrix FSW-0020 (9) Precision Test Systems GPS10eR (10) Samsung 28 GHz RFU TX (11) Samsung 28 GHz RFU RX (12) Low Pass filter (13) National Instruments PXIe-5160 (14) National Instruments HDD-8265 44 2.2.2 Sounder Hardware The developed channel sounder is a beam-switched multi-carrier setup with 400 MHz instan- taneous bandwidth (can be extended to 1 GHz). Fig. 2.6 shows the block diagram for the TX. A 15-bit, 1.25-GSps arbitrary waveform generator (AWG) generates the baseband sounding signal which has equally spaced 801 tones covering the frequency range from 50 MHz to 450 MHz. After the baseband signal is generated, a mixer up-converts it with a local oscillator (LO) frequency of 1.6 GHz. Since we only utilize the upper sideband of the up-converted signal as the intermediate frequency (IF) input of the beam-former radio frequency unit (BF-RFU), a band-pass filter suppresses the LO leakage of the mixer and the lower sideband. After the band-pass filter and the pre-amplifier, we obtain an IF signal with 400 MHz bandwidth centered at 1.85 GHz as seen in Fig. 2.10. The peak seen at the frequency of 1.6 GHz is due to residual LO leakage. It is at least 10 dB lower than the sounding tones and approxi- mately 40 dB lower than the total power of the multi-tone signal, hence it does not affect the measurements in a significant manner. Finally, the IF signal is fed to the BF-RFU, whose simplified block diagram is shown in Fig. 2.8. Similar to the approach in [43], within the BF-RFU, this IF signal is split into 16 identical signals, which are fed into 16 RF chains. Each chain performs an up-conversion to 27.85 GHz with dedicated amplifiers and phase shifters to perform beam-forming. The antenna array is made of 8 by 2 antenna elements, each element consists of two patches as shown in Fig. 2.9. The subarray spacing is 5:6 mm (0:52 at 28 GHz) horizontal and 12:5 mm (1:16 at 28 GHz) in vertical. The BF-RFU allows to perform 90 horizontal beam steering and 60 vertical beam steering with 5 resolution. The phase shifters can be configured via a control interface, thus allowing an adaptation of the beam-shapes between measurement campaigns, see also Section 2.2.3. The phase shifters used in the phased arrays have a res- 45 olution of 5:625 , the frequency doubler also doubles that to 11:25 . Due to the design, the number of beams between which we can switch can be larger than the number of antennas (subarrays). Another key feature of the RFUs is their low phase noise, see Section 2.3.3. Another key aspect of the sounder is the high EIRP (equivalent isotropically radiated power). By design, the power amplifiers of all RF chains are powered up continuously (ex- cept for breaks during the during the switching between beams). This is a significant differ- ence to sounders where different directions are excited by a circular array of horn antennas, since there one power amplifier (PA) (corresponding to one beam) is active at one time. For the same specifications of the PA, the transmit power in our design is thus higher by 12 dB (factor 16). Since it is desirable to stay in the power amplifiers’ linear regime of operation while maximizing the transmitted power, the output power is set 3.4 dB (PAPR plus 3 dB) less than the 1dB compression point of the amplifiers. The 1 dB compression point of the individual amplifier is 31 dBm per amplifier, however, thanks to 16 power amplifiers used in parallel, the total output power is 39:6 dBm with 3:4 dB back off. After 2 dB feed loss and taking into account the 19:5 dBi antenna gain (12 dBi array gain plus 7:5 dBi gain of 2 element subarrays), we achieve 57:1 dBm EIRP at the output of the TX array. Combined with the RX gain, for 400 MHz bandwidth, we achieve a measurable path loss of 159 dB without any averaging or spreading gain. The state-of-the-art rotating horn antenna channel sounder presented in [27] can measure path loss up to 185 dB. However, this channel sounder’s op- erating principle is based on sliding correlator which requires a much longer measurement duration. The configuration presented in the work achieves 185 dB with an acquisition time of 655.04 ms. To do a fair comparison of the measurable path loss for different types of chan- nel sounders, we have to take this acquisition time into account as well. The channel sounder presented in our work only requires 2s of measurements to achieve 159 dB. If we were to spend the same time for the acquisition as the sounder in [27], we can repeat the same mea- 46 Figure 2.8: TX RFU Block Diagram surement 327500 times which would increase the measurable path loss by 55 dB to a total of 214 dB via RX waveform averaging. Similarly, another state-of-the-art channel sounder presented in [29] has 136 dB measurable path loss excluding the averaging and processing gains. For larger path loss, we can use a smaller bandwidth or employ averaging at the RX, see Section 2.2.3 for further discussion. Similar to the TX RFU, the RX RFU also consists of 16-element beam-forming antenna arrays. After low-noise amplification and combining of the received power, there is an addi- tional step of automatic gain control to utilize the dynamic range of the RX to its limits. The RX gain control has a range of 60 dB with 0:5 dB steps. For each MIMO snapshot, the RX controller estimates the power received from the TX-RX beam pair with the highest power, then adjusts to RX gain accordingly to ensure that the received waveforms always have the same power level prior to the IF to baseband conversion. 47 Figure 2.9: Antenna array Figure 2.10: Spectrum of the sounding signal at IF 48 The gain control is performed prior to the each MIMO snapshot, and the gain levels are recorded along with the measurement data, GPS coordinates, and the coordinated universal time (UTC) from the GPS clock. So far we have used RX BF-RFU providing 90 coverage. Currently, we are working towards upgrading the setup to have 4 BF-RFUs for the RX to achieve complete 360 coverage. In the meantime, we physically rotate the RX to 4 orien- tations to cover 360 in azimuth when required in our sample measurements. At the output of the RX BF-RFU, the received IF signal is filtered, down-converted back to baseband and finally sampled by a 1.25-GSps 10-bit digitizer. The digitizer streams the sampled data to a redundant array of independent disks (RAID) with a rate up to 700 MBps and stores the data for post processing as shown in Fig. 2.7. 2.2.3 Sounder Operation Both the TX and the RX are controlled with LabVIEW scripts running on National Instru- ments PXIe controllers. The beam steering and gain of the variable gain amplifiers at the BF-RFUs are controlled via an FPGA interface with a custom designed control signaling protocol implemented in LabVIEW FPGA. This interface allows us to switch between any beam setting or gain setting in less than 2s. Consequently, the proposed sounder can com- plete a full-sweep a million times faster than a virtual array. All beam pairs can be measured without retriggering the digitizer or the AWG. This avoids triggering jitter which would cre- ate uncertainty in the absolute delay of the paths observed in different beams. TX and RX have no physical connections and they are synchronized with GPS-disciplined Rubidium fre- quency references. These references provide two signals for the timing of the setup; a 10 MHz clock to be used as a timebase for all units and 1 pulse per second (PPS) signals aligned to UTC. Given the measurement period, hardware counters in the NI DAQ Timing modules count 49 the rising edges of the 10 MHz and trigger the rest of the units at given times. These counters are in turn triggered by the 1 PPS. Since the 1 PPS signals in the TX and RX are both aligned to the UTC, they operate synchronously without requiring any physical connections. More importantly, the AWG, the ADC, the frequency synthesizers and the BF-RFUs are disciplined with the 10 MHz signal provided by these frequency references, so that they maintain phase stability during the measurements, which is essential for accurate measurement results [36]. These references also provide GPS locations which are logged along with the measurement data. Finally, Table 7.2 summarizes the hardware and sounding waveform specifications. While this is the configuration used throughout this chapter, the sounding waveform can be modified without any significant changes in the hardware. Equally importantly, the beams can be configured by modifications of the FPGAs in the RFUs. This enables, for example, to get fast scanning in azimuth only for situations where we know that MPCs are mainly incident in the horizontal plane, while a slower sweep through azimuth and elevation can be implemented for other cases. This is a significant advantage compared to setups with switched horn arrays, which cannot be reconfigured without extensive mechanical modifications recabling, and/or switches with larger sizes than the actually used number of antennas. 2.3 System Verification 2.3.1 Beam-Steering Firstly, we show the patterns of the beams formed by the TX BF-RFU, the results are repre- sentative for the RX side as well. Fig. 2.11 shows the beam patterns for all azimuth beams for the TX. In azimuth, 19 beams cover the range -45 to +45 with 5 steps. All beams have approximately 12 3 dB beam-width and the side lobes are -10 dB or less relative to 50 Table 2.4: Sounder specifications Hardware Specifications Center Frequency 27.85 GHz Instantaneous Bandwidth 400 MHz (max 1 GHz) Antenna array size 8 by 2 (for both TX and RX) Horizontal beam steering -45 to 45 Horizontal 3dB beam width 12 Vertical beam steering -30 to 30 Vertical 3dB beam width 22 Horizontal/Vertical steering steps 5 Beam switching speed 2s TX EIRP 57 dBm RX noise figure 5 dB ADC/AWG resolution 10/15-bit Data streaming speed 700 MBps Sounding Waveform Specifications Waveform duration 2s Repetition per beam pair 10 (1 for dynamic) Number of tones 801 Tone spacing 500 kHz PAPR 0.4 dB Total sweep time 1 14.44 ms (400s for dynamic) 51 90 75 60 45 30 15 0 -15 -30 -45 -60 -75 -90 0 10 20 45 40 35 30 25 20 15 10 5 0 -5 -10 -15 -20 -25 -30 -35 -40 -45 Figure 2.11: Beam patterns in Azimuth the main beam. Consequently, in our preliminary results, we investigate the directional chan- nel characteristics by utilizing the beam directionality. However, with the proper (amplitude and phase) calibration of the antenna arrays, in the future, we are planning to apply high- resolution parameter extraction (HRPE) methods such as RIMAX to investigate directional characteristics of the propagation channel. In that case, the actual beam patterns do not af- fect the extracted parameters, as one can completely decouple the system response and the propagation channel [34, 37]. Fig. 2.12 shows the received power for all horizontal TX-RX beam pairs for where the LOS component is at x degree (relative to the 0 degree point of the physical array of the TX, and y degree relative to the RX array). Since the beams are overlapping, significant power is recorded at several beam positions. While this effect naturally limits the angular resolution when directions are determined based on the ”strongest beam” only (similar to horn antennas), the effect is actually desired for HRPE algorithms which usually depend on the relative phase shift of a MPC received by different antennas or beams as in our case. 52 Figure 2.12: Received power vs TX and RX beam pairs when LOS component is at RX angle x and TX angley and (x;y) =f(20; 15); (5; 15); (30; 15)g Figure 2.13: Received Power Spectrum with different RX beams 53 2.3.2 Frequency Response Since the sounding signal has 400 MHz bandwidth, the frequency response of the channel sounder is also relevant. Fig. 2.13 shows the combined channel response of the TX and RX. To compare the response of the different beams, we rotated the RX to align different RX beams towards to the TX. As seen in Fig. 2.13, the frequency responses of different RX beams are within1 dB over the whole frequency band. Furthermore, the ratio of the complex frequency responses of two beams for all frequency tones can be estimated with a complex scalar which has approximately unit gain. This fact fits the narrowband array assumption in RIMAX and greatly facilitates the HRPE, see [34]. Note also that, if so desired, the variations of the received power over frequency could be eliminated by using different transmit power for the different tones. However, this might increase the PAPR and thus require a higher back off. Given the small variations of received power (compared to variations introduced by the channel), we thus retain the sounding sequence design of Section 2.2.1. 2.3.3 Phase Stability The most important features of the proposed sounder are its short measurement time and phase stability. These two features are required for measurements in dynamic environments and HRPE [36]. To investigate the phase stability of the system, we run the sounder continu- ously for 50 ms with fixed beams at the TX and the RX. Fig. 2.14 shows the phase drift of the center tone for i) the best-case scenario: TX and RX use a shared reference, ii) the worst-case scenario: they operate with independent free-running Rubidium frequency references, and iii) the typical measurement configuration where the independent Rubidium references dis- ciplined by GPS receivers. With the shared reference there is no phase drift and the standard deviation of the phase is 5.8 . Even with the free-running references, the total accumulated phase drift observed during a full-sweep of all (19 by 19) TX and RX beam combinations in 54 0 10 20 30 40 50 Time (ms) -180 -160 -140 -120 -100 -80 -60 -40 -20 0 20 Phase drift (degree) Shared Reference Free-running References GPS-disciplined References Figure 2.14: Phase drift with shared reference, free-running references and GPS-disciplined references azimuth is only 4 over 1:444 ms. With the help of GPS-disciplining, this further decreases to less than 1 for the given measurements. During the measurement campaigns, although the performance of the GPS-disciplined case will depend on several factors (i.e. how long the GPS receivers have been connected to GPS satellites, number of active satellites and the quality of the connection), the true phase stability performance will be in between the first two cases. In case of measurements where the GPS might not be available, we also have the option of training one frequency reference with respect to the other prior to measurements to decrease the phase drift for the free-running references. However, this requires a long training duration (12-24 hours), and provide synchronization for only a few hours once the references are separated. Fig. 2.15 shows the delay-Doppler function (;v) of the channel sounder for a short distance LOS measurement in a completely static environment. Thanks to the phase stability of the channel sounder, the delay-Doppler function acquired is a good approximation of the 55 Figure 2.15: Normalized delay-Doppler function in dB scale for the measurements with the GPS-disciplined references ideal case(;v) =()(v), where is the delay,v is the Doppler shift and the() is the Dirac-delta function. The phase stability is important for determining the achievable averaging (or spreading) gain. Averaging achieves noise reduction because the desired signal is added up phase coher- ently, while noise is added up incoherently. Thus, we can achieve benefits only for averaging durations less than the phase coherence time of the sounder. For the admissible measurement time, we have to distinguish between (i) static measurements without HRPE, and (ii) other types of measurements, i.e., measurements in dynamic environments, and/or HRPE. In the former case, the averaging duration within one beam can be the sounder phase coherence time. This means that up to 10,000 repetitions of the training signal can be made, leading to a spreading gain of 40 dB, and thus extending the measurable path loss to more than 200 dB. In the latter case, a complete MIMO snapshot has to be finished within a time that is the shorter of the sounder phase coherence time and the channel coherence time. In this work, 56 Figure 2.16: Received waveform when switching between 2 beams; Beam #9 and Beam #10 with azimuth directions of5 and 0 , respectively we employed 10 repetitions for each beam pair to improve the SNR by averaging resulting a sweep time of 14:44 ms. In this case, the accumulated drift is 40 in the worst case. How- ever, as long as the total drift is guaranteed to be lower than 360 , it can be estimated easily by simply adopting a switching pattern with repeated beam pairs. The estimation is further simplified by the fact that the phase drift is essentially linear even over the fairly long time of 50 ms [44]. Since HRPE algorithms depend on the phase relation between different beam settings, the repeatability of the phase is as important as its stability. To ensure that the calibration measurements are valid, it is crucial that the relative phase between 2 different beams stays constant at all times, even after a complete restart of the sounder. To test this, we recorded the received waveform, while switching back and forth between two beam settings as seen in Fig. 2.16. Fig. 2.17 shows the phase of the center tone for the two beams. The phase offset 57 Figure 2.17: Phase drift for Beam #9 and Beam #10 between two beams does not change significantly over 50 ms. Furthermore, this phase offset does not change even after a complete power cycle of the sounder. Both in the frequency synthesizers and in the BF-RFUs, we have phase locked loops (PLLs) to derive the carrier frequencies from 10 MHz references. Every time they lock, they do so with a random phase, however, the effect of the random phase is the same for all beam pairs and has no effect on the relative phase offset between different beams, which is the relevant quantity for HRPE. Further details about the HRPE implementation and calibration are discussed in [45]. 2.3.4 Path loss Verification Fig. 2.18 shows the measured path loss for line-of-sight measurements conducted in an open area when the TX and the RX were placed on scissor lifts at the height of 5 m. The measure- ments were performed for distances ranging from 30 m to 122 m. The path loss exponents were estimated 1.997 and 2.051 for close-in and alpha-beta-gamma models, respectively. For 58 Figure 2.18: Line-of-sight (LOS) path loss both models, the observed path loss exponents were almost equal to the free-space path loss exponent of 2. The absolute path loss measured in the anechoic chamber at the distance of 6 m is 18:25 dB when the bore-sight of the center beams for the TX and RX are aligned. Af- ter compensating the gains of 12 dB for amplifiers and the 17 dBi for the beamforming (for both the TX and the RX), this corresponds to 76:25 dB over the air path loss showing a good agreement with the free space path loss for the same distance which is 76:9 dB. 2.4 Post Processing In the following, we describe the post processing of the measured data using Fourier-resolution techniques (i.e., no HRPE, which is the subject of future work). 59 2.4.1 Power Delay Profile and Power Angular Delay Profile The directional power delay profile (PDP) for the TX and RX beams with the azimuth angles TX and RX , respectively is estimated as PDP ( TX ; RX ;) = F 1 n W ~ f H TX ; RX ~ f :=H cal ~ f o 2 (2.2) where TX/RX 2 [180; 175],F 1 denotes inverse Fourier transform,H i;j ( ~ f) andH cal ( ~ f) are the frequency responses fori-th TX andj-th RX beam and the calibration response respec- tively;W ( ~ f) is the Hanning Window, ~ f is the vector of frequency tones, and:= is element- wise division. Since all beam pairs are measured without a significant phase drift or trigger jitter, all directional PDPs are already aligned in the delay domain and require no further correction. Consequently, the corresponding PDP can be obtained similarly to the virtual horn mea- surements with a shared reference even though the sounder does not have a shared reference. There are different methods discussed in [25,46] for synthesizing omni-directional PDP from directional measurements. Ultimately, the PDP obtained from HRPE will provide the best representation of the channel. However, we use the approach from [25] to present sample results here. Hence, the omnidirectional PDP is given by PDP () = max TX max RX PDP ( TX ; RX ;) (2.3) Furthermore, the power angular-delay profiles for RX and TX which are calculated as follows. PADP RX ( RX ;) = max TX PDP ( TX ; RX ;) PADP TX ( TX ;) = max RX PDP ( TX ; RX ;) (2.4) 60 Figure 2.19: Significant MPCs (The shown area is 350 m by 80 m) Then the angular power spectrum can be simply calculated as: PAS( TX ; RX ) = X PDP ( TX ; RX ;) (2.5) 2.4.2 Multi-path Extraction For extracting the MPCs in the delay domain, we consider separately each resolvable delay bin. Within each bin, obtaining the MPC directions cannot simply assign an MPC to each beam with significant power; rather we have to take the side-lobes of the beams into account. As seen in Fig. 2.11 and 2.12, in all cases, the main lobe is at least 10 dB stronger than the side-lobes. We first perform a peak search for all peaks in the TX/RX beam domain with power that is higher than a given thresholdN th which is set as average noise power plus 6 dB in this work. Subsequently, we filter out the ghost MPCs due to side-lobes, a ghost MPC must have the same delay with a stronger valid MPC since (due to our Hanning filtering), no significant side-lobes exist in the delay domain. The strongest peak detected at every delay bin is always accepted as a MPC. Since the beam-width is wider than the beam steering steps, a MPC will likely be received in more than a single beam. However, thanks to the concave shape of the beam pattern within the main beam each MPC will correspond to a single peak in the PADP. A peak that has the same delay and a power within 10 dB of the strongest MPC 61 Algorithm 1 Multi-path detection: 1: procedure DETECT MPC(PDP (),N th ) 2: Initialize MPClist 3: Perform 5-D peak search onPDP () 4: Store peaksp(; ), = [ TX ; TX ; RX ; RX ] 5: for all do 6: n 7: ifp( n ; ) is not empty then 8: p max max p( n ; ) 9: max arg max p( n ; ) 10: for allp( n ; ) do 11: ifp( n ; )>p max =10 then 12: Addp( n ; ) to the MPClist 13: else ifp( n ; )>p max =100 then 14: if ( TX,max 6= TX )&( RX,max 6= RX ) then 15: Addp( n ; ) to the MPClist 16: end if 17: end if 18: end for 19: end if 20: end for 21: end procedure 62 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 Delay(us) -175 -170 -165 -160 -155 -150 -145 -140 -135 -130 -125 PDP(dB) Figure 2.20: Power delay profile for the position given in Fig. 2.19 is also accepted as a valid MPC, since it cannot represent a side-lobe. Finally, we accept the peaks that have powers within 20 dB of the strongest MPC and both DoD and DoA are different than the strongest MPC in the same delay bin, since ghost MPCs fulfilling that angle condition would be suppressed by twice the side-lobe suppression (once at TX, once at RX), and thus be suppressed by more than 20 dB. Note that with this extraction algorithm, the dynamic range per delay bin is limited to 20 dB (even for very high SNR), though for different delays, MPCs within a larger dynamic range can be detected. 2.5 Sample Measurements In this section, we will discuss sample results from two measurement campaigns highlighting different operation modes for the channel sounder. The first measurement campaign focusses on static measurements in a suburban residential environment for a micro-cell scenario [38]. The second measurement campaign is a first-of-its-kind directionally resolved dynamic mea- 63 surement campaign performed on an urban street on the University of Southern California University Park Campus in Los Angeles, CA, USA [41]. 2.5.1 Static Directional Measurements During this campaign, we performed measurements in a residential micro-cellular environ- ment for distances up to 400 m. Consequently, we employ 10-times RX waveform averaging to improve the SNR resulting a sweep time of 14:44 ms for 90 sectors at the TX and the RX. We simply repeat the measurements when the RX is rotated to (0; 90; 180; 270) degrees to obtain 360 coverage at the RX side. Although the measurement method used in this cam- paign is not suitable for the dynamic environments with 360 degree coverage, it demonstrates that the channel sounder can provide omni-directional measurements for static environments in much shorter time than the rotating horn antenna channel sounders. Thus it enables mea- suring at more locations to provide statistically significant amount of data. The particular measurement location used as the example here is shown in Fig. 2.19. Fig. 5.9 shows the PDP () for this location. The direct path undergoes an additional attenuation (approximately 20 dB) due to foliage. The second reflected path has a similar power to the direct path since it does not suffer from foliage attenuation. Furthermore, Fig. 2.21 and 2.22 respectively show the power angular-delay profiles for RX and TX. Fig. 2.23 shows the received power for all TX-RX beam combinations. As discussed in section 2.4, some of the peaks observed in the figure are due to the side-lobes of the beams. Hence we employ the Algorithm 1 to detect the MPCs. Fig. 2.24 shows the extracted the MPCs on the DoA and delay plane. These extracted MPCs are compared with the environ- ment by mapping DoA, DoD, and delay to corresponding scatterers. The significant MPCs are marked on Fig. 2.19. Furthermore, Table 2.5 compares the extracted DoD, DoA and delay with their estimated values from Google Maps. Since we are only using the beam 64 Figure 2.21: DoA Power angular-delay profile (dB) for the location in Fig. 2.19 directionality, the angular resolution is 5 . Note that the term MPC is loosely used here. Sim- ilar to the measurements performed with rotating horn antennas, we can actually have more than a single path within the 5 . This ambiguity will be resolved in the future by employing HRPE [45]. 2.5.2 Dynamic Directional Measurements In this section, we will discuss sample results from a first-of-its-kind directionally resolved dynamic measurement campaign. Each SISO measurement (sounding of a TX and a RX beam) takes 4s consisting of 2s sounding waveform and 2s guard time for electronic beam-switching. Although the channel sounder is capable of beam-steering with 5 steps covering45 , to decrease the measurement time we used every other beam resulting in 10 angular resolution. Consequently, both TX and RX perform 90 azimuth sweeps measuring 65 Figure 2.22: DoD Power angular-delay profile (dB) for the location in Fig. 2.19 Figure 2.23: Received power (dB) vs TX and RX beams for the location in Fig. 2.19 66 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 Delay (us) -200 -150 -100 -50 0 50 100 150 200 DOA (degree) -180 -175 -170 -165 -160 -155 -150 -145 -140 -135 -130 -125 Figure 2.24: DoA and delay for extracted MPCs for the location in Fig. 2.19 RX_2 TX_1 Car #3 Car #2 Car #1 Ped Figure 2.25: The environment for the dynamic measurements 67 Table 2.5: Extracted MPC parameters vs Estimated from Google Maps for the location in Fig. 2.19 (only lists a subset of the MPCs) Google Maps Estimated Path # Delay(ns) DoD DoA Delay(ns) DoD DoA 1 412.9 -8 -8 412.5 -10 -5 2 430.8 13 -15 430 15 -15 3 454.7 -13 117 453.5 -15 115 4 475.6 -12 132 473.5 -15 135 5 503.1 -2 -93 502.5 0 -95 6 550.3 -3 -134 550 -5 -135 7 842.1 0 -164 842.5 0 -160 8 844.3 -4 -176 845 -5 -175 9 1056.7 -3 -177 1058.5 -5 -175 10 1110.9 -6 -178 1112.5 -10 -175 11 1376.7 -1 -172 1375 0 -170 12 1436 -7 -179 1435 -10 -180 13 1745.3 -4 -176 1746 -10 -170 68 Figure 2.26: PDP(dB) vs time, 3 moving cars are marked with black arrows and the pedestrian is marked with red arrow 100 beam pairs in only 400s. A single sweep of all possible combinations of TX and RX beams is called a MIMO snapshot. In a burst, 20 MIMO snapshots were measured without any idle time in between. This allows us to estimate Doppler shifts up to1:25 kHz, which corresponds to a maximum relative speed of 48 kph at 28 GHz, which is larger than the maximum permissible speed on the measured street. Furthermore, these 8 ms bursts of MIMO measurements followed by 52 ms of idle time were repeated 200 times with a period of 60 ms to track the evolution of channel parameters as they change due to moving objects in the environment. This configuration provides us a unique capability of performing double- directional measurements under dynamic conditions and investigate the effects of moving objects on the angular channel characteristics and Doppler spectrum. 69 0 2 4 6 8 10 12 Time(s) 40 50 60 70 80 90 100 110 RMS-DS (ns) Figure 2.27: Time varying RMS-DS 0 2 4 6 8 10 12 Time (s) -30 -25 -20 -15 -10 -5 0 5 10 Mean Angle (degree) DoD DoA 0 2 4 6 8 10 12 Time (s) 12 14 16 18 20 22 24 26 28 Angular Spread (degree) DoD DoA Figure 2.28: Mean angle and angular spreads vs time 70 0 2 4 6 8 10 12 Time(s) -104 -102 -100 -98 -96 -94 Path gain(dB) Best beams tracked Fixed beam Figure 2.29: Path gains vs time: i)RX and TX tracks the best beam pair, ii) stays on the which best on the average Fig. 2.25 shows the details about the measurement snapshot presented in this section. During these measurements both TX and RX are placed facing the same direction so that they are out of each other’s visible azimuth range. We can thus observe moving scatterers without a dominant LOS. During 12 s, we observe 4 moving objects; three cars moving towards the TX and RX and a pedestrian walking away. The tracked MPCs corresponding to these four objects are marked in Fig. 2.26. Fig. 2.27 shows the root mean square delay spread (RMS- DS) varying within range from 50 ns to 100 ns through 12 s of measurements. Thanks to the capability of performing directional measurements in dynamic environ- ments, we can also study the evolution of the angular channel statistics. Due to fast moving objects in the environment, and their interactions with each other, the angular spectra for the TX and the RX change quickly. As seen in Fig. 2.28, the mean DoA angle varies from15 to 10 and the angular spread takes values in the range of 18 to 26 . Similarly, the DoD varies between27 and10 with angular spreads from 14 to 23 . While designing a mm-wave system with adaptive beam-forming it is crucial to under- stand the requirements on the beam-switching, i.e., how frequently we have to search for the best beam and when the beam should be switched. Hence, the measurements which can cap- 71 Figure 2.30: Doppler spectrum vs time Figure 2.31: Doppler spectrum vs delay att = 7:38s 72 ture the path gain simultaneously for different TX and RX beam directions are needed. Fig. 3.8 shows the path gains for the fixed beam with highest average power and if we track the best beam at all times. The instantaneous best beam surpasses the fixed beam by more than 8 dB. The phase stability of the system allows us to estimate the Doppler spectrum and its temporal evolution. The 20 fast repetitions within each burst are used to estimate Doppler for each time instance. Fig. 2.30 shows the Doppler spectrum over time. Since all three cars have relatively similar speeds, they create a similar Doppler shift of 1kHz-1.25 kHz. However, if we plot the spreading function with respect to delay att = 7:38s as in Fig. 2.31, we see four MPCs with non-zero Doppler. The three MPCs with high positive Doppler shifts are caused by the cars while the MPC with the negative smaller Doppler is due the pedestrian walking away. 2.6 Conclusion In this chapter, we presented a novel mm-wave channel sounder that can perform double- directional measurements in dynamic environments. By using a beam-forming array, we decreased the measurement time from minutes to milliseconds compared to rotating horn antenna approach. Furthermore, thanks to the beam-forming gain, the measurable path loss we achieved for a 400 MHz bandwidth is 159 dB without waveform averaging and FFT pro- cessing gain. We also validated the sounders phase stability, which is the paramount feature of the proposed design. Compared to the rotating horn antenna setups which can only mea- sure either directional or dynamic channel properties at a given time, the channel sounder presented in this work can simultaneously estimate DoD, DoA, delay, and Doppler in a dy- namic channel. Furthermore, it can provide the temporal variations of the angular spectrum, which is a crucial input for designing beam-forming algorithms. Possible directions of fu- 73 ture research are; upgrading the RX to cover 360 in azimuth, performing extensive channel sounding campaigns and processing the acquired data with high-resolution parameter extrac- tion algorithms. 74 Chapter 3 28 GHz Dynamic Double Directional Propagation Channel Measurements 3.1 Introduction CISCO estimated the global mobile data traffic in 2016 as 7 exabytes (EB, 10 18 bytes) per month and forecast it to reach 49 EB per month by 2021 [47]. One commonly anticipated way to meet this ever-growing demand is to utilize the idle spectrum available at millimeter- wave (mm-wave) frequencies. Especially, 28 GHz band attracts a lot of interest thanks to comparatively lower hardware and implementation costs due to relatively lower carrier fre- quency. An accurate propagation channel model is essential for designing and validating a wire- less communication system. Due to shorter wavelengths at mm-wave frequencies, the nature of the propagation channels in that regime can be significantly different from those sub 6 GHz bands. Consequently, there has been a growing interest in wireless propagation channel mea- surements at mm-wave frequencies [8]. It is commonly accepted that mm-wave systems will 75 have to use beam-forming arrays to overcome the higher pathloss experienced at higher car- rier frequencies [9], [8]. Hence an accurate model for the angular statistics is crucial to assess system performance. Moreover, in case of dynamic channels, an accurate model of temporal characteristics of these angular statistics is imperative for system design. In this work, we use a 28 GHz real-time, double-directional mm-wave channel sounder allowing us to perform directional measurements in dynamic channels [4, 5]. Instead of a rotating horn, our channel sounder uses phased array antennas to form beams that can be steered electronically. By using measurements with this channel sounder, we investigate time-varying PDP, path-loss, delay spread, mean angles and angular spreads observed at the transmitter (TX) and receiver (RX) in the presence of moving vehicles and pedestrians. Ad- ditionally, we investigate excess losses observed due to blockage by vehicles and compare the cases when TX and RX are using fixed beams or when they are capable of adjusting di- rections of beams dynamically. To the best of our knowledge, no such measurements exist in the literature up to now. Most of the directional measurements of sounders used in mm- wave channels are based on rotating horn antennas which are not suitable for the intended measurements here [25, 27, 46]. Conversely, most of the papers in the literature investigating the dynamic channel char- acteristics for the mm-wave focus on the shadowing effects of moving vehicles or humans when the TX and RX antennas are fixed (they may be directional, but cannot adapt their di- rections). In [48], the authors performed 30 GHz indoor measurements with omnidirectional antennas to investigate the human-induced variations. Ref. [49] proposes a model for the fading due to human blockage at 73:5 GHz with Markov models by using measurements per- formed with directional horn antennas with different beam-widths. Ref. [50] measured the impact of people walking through/near the significant paths between TX and RX at 60 GHx. They found that strongest outages (additional attenuation 15 dB or more) occurred when both 76 Figure 3.1: Timeline for the channel sounder operation TX and RX are at/below height of the walking people but smaller when one of the antennas was 2m or higher. Ref. [51] presents the blockage characteristics at 28 GHz due to a station- ary car. They compare path loss, delay spread and angular spread with and without vehicular blockage for static measurements. The references [52] and [53] investigate shadowing char- acteristics due to vehicles and humans with omnidirectional antennas at 28 GHz. Both [54] at 76 GHz and [55] at 36 GHz observed 20 dB attenuation due to blockage by a truck. However, none of those previous works is capable of providing directional information in dynamic channels. Hence they can’t provide insights on the temporal dependencies of the angular statistics, which are crucial for a mobile system operating with beam-forming arrays. 77 Table 3.1: Sounder Parameters Hardware Specifications Center Frequency 27.85 GHz Instantaneous Bandwidth 400 MHz Antenna array size 8 by 2 (for both TX and RX) Horizontal beam steering -45 to 45 degree Horizontal 3dB beam width 12 degrees Vertical 3dB beam width 22 degrees Horizontal steering steps 10 degrees Beam switching speed 2s TX EIRP 36 dBm (max 57 dBm) RX noise figure 5 dB ADC/AWG resolution 10/15-bit Data streaming speed 700 MBps Sounding Waveform Specifications Waveform duration 2s Repetition per beam pair 1 Number of tones 801 Tone spacing 500 kHz PAPR 0.4 dB Total sweep time 400s MIMO repetition per burst 20 Burst repetition 16.66 Hz 78 3.2 Measurement Campaign The measurements were performed on an urban street in the University of Southern California University Park Campus in Los Angeles, CA, USA. TX and RX were always placed across the street from each other as shown in Figure 3.2. The RX height is 1:8 m while the TX height was changed to 2:5 m, 3:5 m and 4:5 m. From the TX and RX orientations shown in the Figure 3.2, the following TX-RX pairs were measured: • Case 1: TX 1 to RX 1 • Case 2: TX 1 to RX 2 • Case 3: TX 2 to RX 3 In Cases 1 and 3, a LOS exists when the environment is idle but it may or may not be blocked by passing cars, buses or trucks. For Case 2, although TX and RX have a visual LOS, they are placed so that TX and RX are out of each others visible azimuth range, i.e., their beams do not point towards each other. During this measurement campaign we only utilize a single elevation angle 0 with 10 azimuth angles both for the TX and the RX. Figure 3.1 shows the time-line for the sounder operation for this measurement campaign. Each SISO measurement takes 4s consisting of 2s sounding waveform and 2s guard time for electronical beam-switching. Although the channel sounder is capable of beam-steering with 5 steps covering45 , to decrease the measurement time we used every other beam resulting in 10 angular resolution. Con- sequently, both TX and RX perform 90 azimuth sweeps measuring 100 beam pairs in only 400s. A single sweep of all possible combinations of TX and RX beams is called a MIMO snapshot. In a burst, 20 MIMO snapshots were measured without any idle time in between. This allows us to estimate Doppler shifts up to1:25 kHz which corresponds to a maximum 79 RX_1 RX_2 RX_3 TX_1 TX_2 Figure 3.2: Measurement environment relative speed of 48 kph at 28 GHz, which is larger than the maximum permissible speed on the measured street.. Furthermore, these bursts of MIMO measurements were repeated 200 times with a period of 60 ms to track evolution of channel parameters as they change due to moving objects in the environment. This configuration provides us a unique capability of performing double-directional measurements under dynamic conditions and investigate the effects of moving objects on the angular channel characteristics and Doppler spectrum. Table 7.2 details the configuration used in this measurement campaign. 3.3 Measurement Results 3.3.1 Case 1 : Blocking Objects In this scenario, we observe two main MPCs while the channel is idle; the LOS path and a reflection as shown in Figure 3.3. The actions during this measurement can be listed as; • The measurements start with an idle channel, 80 RX TX Bus Dominant paths Figure 3.3: Blocked LOS 50 60 70 80 90 Delay(m) 0 2 4 6 8 10 12 Time(s) -125 -120 -115 -110 -105 -100 Figure 3.4: PDP vs time 81 0 2 4 6 8 10 12 Time(s) -45 -35 -25 -15 -5 5 15 25 35 45 DOA (degree) -120 -110 -100 0 2 4 6 8 10 12 Time(s) -45 -35 -25 -15 -5 5 15 25 35 45 DOD (degree) -120 -110 -100 Figure 3.5: Power of the MPCs vs DOA/DOD and time 82 0 2 4 6 8 10 12 Time (s) -40 -30 -20 -10 0 10 20 30 40 Mean Angle (degree) DoD DoA 0 2 4 6 8 10 12 Time (s) 0 5 10 15 20 25 Angular Spread (degree) DoD DoA Figure 3.6: Mean angles and angular spreads vs time • a bus enters to the street moving right to left, • the bus blocks the LOS att = 5 s, • the bus blocks both the LOS and the reflection att = 8:5 s, • the bus stops in front of the RX while still blocking the two main paths. Figure 3.4 shows the PDP versus time, att = 0 s the two dominant paths, LOS and the reflection can be seen at the delays of 51 m and 57:75 m respectively. The corresponding angular spectra for TX and RX are shown in Figure 3.5. The LOS path is at [ TX ; RX ] = [15;25] and reflection, is at [ TX ; RX ] = [35;35]. The effects of a bus blocking the RX on the delay and angular dispersion are visible in the figures as well. As seen in Figure 3.6, due to the two dominant paths with similar DoAs and different DoDs, initially the angular spread for DoA is as small as 5 while it is 20 for DoD. Once the bus blocks the RX sector almost completely, the only strong paths 83 0 2 4 6 8 10 12 Time (s) 10 20 30 40 50 RMS-DS (ns) Figure 3.7: Root mean square delay spread vs time 0 2 4 6 8 10 12 Time(s) -130 -120 -110 -100 -90 Path gain (dB) LOS - [ TX , RX ]=[-15,-25] Ref. - [ TX , RX ]=[35,-35] Figure 3.8: Path gains for best two beam-pairs vs time 84 are through the bus or diffracted around the bus. Consequently, The angular spread for the direction of departure (DoD) is limited compared to the direction of arrival (DoA). Once both dominant paths are blocked the root mean square delay spread (RMS-DS) increases from 12 ns to approximately 40 ns, Figure 3.7. Figure 3.8 shows the temporal variations of the path gains for the two dominant RX-TX beam pairs. Betweent = 5 s andt = 8:5 s, if both TX and RX can select the best beam then the excess loss is 9 dB compared to the 20 dB if we use fixed beams. This shows that beam direction adaptation is essential not only for mm-wave systems in mobile scenarios, but also with fixed TX and RX if significant blockage can occur. 3.3.2 Case 2 : Moving Scatterers In a second measurement, both TX and RX are placed facing the same direction so that they are out of each other’s visible azimuth range. We can thus observe moving scatterers without a dominant LOS. Figure 3.9 shows the details about the measurement snapshot presented in this section. During 12 s, we observe 4 moving objects; Car #1 and Pedestrian #1 are coming towards the TX and RX starting fromt = 0 s. The tracked MPCs corresponding these two objects are marked as Car #1 and Ped #1 in Figure 3.10. Furthermore, these objects are also clearly visible in the Doppler spectrum with positive Doppler shifts as expected, see Figure 3.11. After t = 6 s, Car #2 and Ped #2 enter the picture and move away from TX and RX. Between time t = 9 s and t = 11 s, Ped #1 blocks the path from Car #2 to RX. Their effect on the Doppler spectrum is also marked in Figure 3.11. The speeds of all four objects estimated speeds from the Doppler spectrum matches with the video recordings captured simultaneously with the measurements. Due to fast moving objects in the environment, and their interactions with each other 85 Figure 3.9: Moving Scatterers 20 40 60 80 100 120 Delay(m) 0 2 4 6 8 10 12 Time(s) -150 -145 -140 -135 Figure 3.10: Temporal evolution of the multipath components 86 Figure 3.11: Doppler spectrum for case 2 87 0 2 4 6 8 10 12 Time (s) -40 -30 -20 -10 0 10 20 Mean Angle (degree) DoD DoA 0 2 4 6 8 10 12 Time (s) 10 15 20 25 30 35 Angular Spread (degree) DoD DoA Figure 3.12: Mean angles and angular spreads vs time for Case 2 0 2 4 6 8 10 12 Time(s) -130 -125 -120 -115 -110 -105 -100 Path gain (dB) [ TX , RX ]=[-5,-5] [ TX , RX ]=[15,5] [ TX , RX ]=[-5,25] [ TX , RX ]=[-45,45] Figure 3.13: Path gains for best beam-pairs vs time for Case 2 88 (e.g., Ped #1 shadowing Car #2 ) the angular spectra for the TX and the RX change quickly. As seen in Figure 3.12, between t = 5 s and t = 8 s, the mean DoA varies in between 10 and 10 continuously. During the same time period, angular spread for DoD varies from 12 to 31 . Figure 3.13 shows the evolution of the path gains for four different beam pairs over time. The beam pair #1 at [ TX ; RX ] = [5; 5] point towards a static reflector and has a roughly constant gain of115 dB throughout this snapshot. Although at times due to moving reflectors, the other three beams surpass the beam pair #1 by as much as 10 dB, this improvement is temporary and never lasts more than 1 s. Depending on the overhead of finding the best beam and the required time to update both RX and TX beams, it might not be optimum to use an approach which constantly looks for the best beam. Other approaches such as remaining at a beam pair which is “good” enough or implementing a trigger hysteresis for beam update should be considered as well. 3.3.3 Case 3 : Blocked LOS These measurements were performed with TX 2 and RX 3 in Figure 3.2. In this case, a strong LOS is present with no other significant reflections. During the measurements, the LOS was blocked by a moving bus in betweent = 6 s andt = 8 s. Figure 3.14 shows the time-variant path gain. Due to the metallic parts of the bus in the front and the back, we observe excess losses up to 24 dB. In this particular case, it takes 120 ms for the excess loss to reach to 24 dB and it stays in the shadowing dip for 240 ms. As the bus continues to move, the type of material blocking the direct path changes between metal and glass, hence the excess loss varies between 5 dB and 15 dB for 1:14 s. As the back end of the bus blocks the direct path, we observe another fading deep lasting 270 ms. Afterwards, the excess loss returns to 0 dB in 300 ms as the bus clears the area. Figure 3.15 shows the time-varying RMS-DS for the same duration. While the channel 89 3 4 5 6 7 8 9 10 Time(s) -100 -95 -90 -85 -80 -75 Path Gain (dB) Figure 3.14: Path gain vs time for Case 3 3 4 5 6 7 8 9 10 Time(s) 0 10 20 30 40 50 60 RMS-DS (ns) Figure 3.15: RMS-DS vs time for Case 3 90 4 6 8 10 Time (s) -10 -5 0 5 10 15 Mean Angle (degree) DoD DoA 4 6 8 10 Time (s) 10 12 14 16 18 20 22 24 Angular Spread (degree) DoD DoA Figure 3.16: Mean angles and angular spreads vs time for Case 3 is idle the RMS-DS is around 6:5 ns. The RMS-DS is increasing as high as 50 ns during the aforementioned fading dips observed as the bus enters and exits the area between the TX and RX. During the rest of the blockage, the RMS-DS varies 6 ns to 15 ns. Figure 3.16 shows the angular statistics. Similar to the RMS-DS, the effect of the moving bus on the angular spectra is prominent while the bus enters or exits the visibility of the TX and the RX. Additionally, as the bus is in view (i.e. betweent = 6 s andt = 8 s) both the mean angles and the angular spread vary continuously. 3.4 Conclusions In this chapter, we presented results from a double-directionally resolved measurement cam- paign at mm-wave frequencies for an outdoor microcellular scenario. For three different scenarios, we investigated the angular spectrum, the delay spread, and Doppler spectrum. The unique capabilities of the channel sounder allowed phase-coherent measurements en- 91 abling investigations into the temporal channel characteristics. Hence, we provided results into the evolutions of the channel parameters in a dynamic channel in the presence of moving vehicles. The variation of the angular spectrum is especially important for systems utilizing di- rectional beams. The algorithms, which intend to find the best beam and update the beam direction, would benefit from a better understanding of these temporal variations in the an- gular spectrum. Depending on the overhead of beam discovery and update, it might not be optimum to use an approach which constantly looks for the best beam. Other approaches such as remaining at a beam pair which is “good” enough or implementing a trigger hysteresis for beam update should be considered as well. 92 Chapter 4 28 GHz Microcell Propagation Channel Measurements 4.1 Introduction Due to the ever-increasing demand for wireless data, current networks are becoming overbur- dened. While a variety of different techniques will be used to alleviate this congestion and enable future growth [56] [57], making new spectrum available is among the most promis- ing approaches. For this reason, there is great interest in developing wireless communica- tions systems in the frequency spectrum beyond 6 GHz, which up to now has been mostly fallow [58]. In a recent ruling, the frequency regulator in the USA, the Federal Communi- cations Committee, has allowed usage of more than 10 GHz of bandwidth of that frequency range for new services - considerably more than currently used in all wireless services taken together. Other countries are expected to follow suit, and frequency bands such as 28 GHz, 38 GHz, 60 GHz and 75 GHz are being considered for fifth generation (5G) cellular net- works [59]. For outdoor applications the 28 GHz currently enjoys the greatest interest, since 93 the comparatively low frequency (compared to other mm-wave bands) allows a lower-cost implementation of many components. The design and deployment planning of any wireless system requires a thorough under- standing of the wireless propagation channel. For example, path-loss and shadowing charac- teristics determine distance-dependent outage probability, while delay dispersion determines spacing of subcarriers (in OFDM) or length of equalizers (in single-carrier systems). At the same time, it must be remembered that the channel characteristics strongly depend on the propagation environment. Thus, measurements of channel characteristics and creation of models derived from them in the environment of interest are vital. While there have been sev- eral measurement campaigns for channel characteristics in urban microcellular environments (see below), to the best of our knowledge no such measurements exist in suburban micro- cell environments. This work aims to close this gap. In particular, we will present results from a measurement campaign in a microcellular, suburban environment with a directionally resolving, wide-band channel sounder, and extract some key channel characteristics. Existing work: As mentioned above, a number of directionally resolved measurements have been performed in urban microcellular environments 1 . The works in [24, 60, 61] in downtown New York City. Refs. [62] and [63] provide results in various cities in Korea, while [64] reports channel measurements conducted at 32 GHz on a University campus in Beijing, China. All these environments are densely built up, with high-rise buildings ( 5 floors and/or contiguous facades). All of the measurements use mechanically rotating horn antennas to extract the directional characteristics and make use of the high antenna gain to improve the link budget; the drawback of this approach is that it is very time-intensive, often requiring hours to scan a single measurement location, and thus naturally limiting the number of locations underlying the measurements. 1 Due to space restrictions, we do not review non-directionally resolved measurement campaigns here 94 The suburban measurements reported in the literature have been mainly focused on Fixed Wireless Access (FWA, also known as LMDS) systems [65–67], i.e both transmit and receive antennas are above rooftops, and thus different from the scenario considered here. Similarly, measurements in suburban environments at carrier frequencies below 6 GHz in suburban microcells exist, but due to the different frequency range cannot provide any information about mm-wave propagation. Contributions: In this work, we present the results from the first 28 GHz channel sound- ing campaign in a residential suburban cellular scenario with directionally resolvable results for both TX and RX. The measurements are performed with a real-time channel sounder equipped with phased antenna arrays [4]. The phased arrays form beams at the different TX and RX angles and switch between these beams in microseconds, which allows measurement at a large number of locations within reasonable time, and ensures minimal variation in the environment during the performing of the measurements. We provide key channel charac- teristics, such as path-loss, shadowing, and delay spread, for both line-of-sight (LoS) and non-line-of-sight (NLoS) situations and present sample results for power angular spectrum and extracted multi-path components. 4.2 Measurement Campaign During this measurement campaign we only utilize a single elevation angle 0 with 19 az- imuth angles both for the TX and the RX. With an averaging factor of 10, the total sweep time is 14:44 ms(without averaging it can be as low as 1:444 ms) for 361 total beam pairs. Since phased arrays cover 90 sectors, we rotated the RX tof0 ; 90 ; 180 ; 270 g to cover 360 while using a single orientation at the TX. Consequently, for each measurement location, we obtain a frequency response matrix of size 19 by 72 by 801. The measurements were performed in a typical US suburban residential area at/near 28th 95 Table 4.1: Sounder specifications Hardware Specifications Center Frequency 27.85 GHz Instantaneous Bandwidth 400 MHz Antenna array size 8 by 2 (for both TX and RX) Horizontal beam steering -45 to 45 degree Horizontal 3dB beam width 12 degrees Vertical beam steering -30 to 30 degree Vertical 3dB beam width 22 degrees Horizontal/Vertical steering steps 5 degrees Beam switching speed 2s TX EIRP 57 dBm RX noise figure 5 dB ADC/AWG resolution 10/15-bit Data streaming speed 700 MBps Sounding Waveform Specifications Waveform duration 2s Repetition per beam pair 10 Number of tones 801 Tone spacing 500 kHz PAPR 0.4 dB Total sweep time 14.44 ms 96 Figure 4.1: Measurement locations Street in Los Angeles, CA, USA 2 populated with 2 to 3 story houses along a street that contains trees and other foliage, see Figure 4.1. Consequently, the measurements mimic a real-life cellular deployment scenario including the effect of foliage penetration loss. To imitate a microcell scenario, the TX is placed on a scissor lift at the height of 7:5 m while the RX is on a cart, and the RX antenna height is 1:8 m. The bore-sight of the 90 TX sector is parallel to the 28th St and faced towards to RX. The RX locations are chosen for 2 different scenarios. In the first one, the RX is placed on the same street (28th St) with the TX. Since in some cases the direct path is blocked by foliage or other surrounding objects, throughout the chapter we call this data 28th St. instead of LOS. In the second scenario, the RX is located on the two crossing streets to create NLOS links. All RX locations are either on the sidewalks or in the front yards of the surrounding houses. The range of TX-RX separation vary from 36 m to 400 m for the 28th St and from 130 m to 273 m for NLOS. 97 Figure 4.2: RX View on 28th St facing east Figure 4.3: TX View on 28th St facing west 98 Figure 4.4: Power-Angular Delay Profile for RX 14 Figure 4.5: Power-Angular Spectrum for RX 14 99 Figure 4.6: Power Delay Profile for RX 14 100 Figure 4.7: Path-loss for 28th St. and NLoS locations 101 4.3 Results 4.3.1 Path Loss There are two approaches commonly used for path loss modeling in mm-wave; alpha-beta- gamma (ABG) and close-in (CI) models [68] [8]. For a single frequency band, both can be simplified into: PL(d) = 10nlog 10 (d=1m) +P 0 + (4.1) However, they differ in the estimation of model parametersn andP 0 . For the CI method,P 0 is given by 20log 10 (4f=c) which is the free-space path loss for 1m TX-RX separation at the frequencyf wherec is the speed of light. Once theP 0 is fixed, the path-loss exponent (PLE) n is estimated from measurement data via minimum mean square error estimation. In ABG method, both P 0 and n are estimated together from measurement data with least-squares regression. For both models is a Gaussian random variable with 0 mean and standard deviation of in dB [68]. We used both ABG and CI models to characterize the path-loss for both omnidirectional and directional cases. Figure 4.7 shows the path-loss values for the 28th St and NLoS mea- surement points along with the ABG and CI fits for the omnidirectional RX, and the theo- retical free space path-loss (FSPL). The path-loss model parameters for both directional and omnidirectional RX are given in Table 6.2. For the 28th St, the parameters for the ABG and the CI models are very similar while they differ significantly for NLoS data. Note, however, that while the parameters are different, the resulting line fits in the range of interest, i.e., the range over which measurements have been made and thus the model is applicable, are quite 2 note that despite the location in Los Angeles, the building height and density is suburban, not metropolitan, as can also be seen from Figures 4.1,4.2 and 4.3 102 Table 4.2: Parameters of the path loss models Data n P 0 P-value omni 28th St - ABG Model 2.82 63.47 6.44 0.975 28th St - CI Model 2.92 61.34 6.45 0.978 NLoS - ABG Model 4.97 29.53 2.58 0.745 NLoS - CI Model 3.58 61.34 3.06 0.706 directional 28th St - ABG Model 3.17 58.01 7.75 0.840 28th St - CI Model 3.15 61.34 7.76 0.928 NLoS - ABG Model 5.85 18.12 4.53 0.856 NLoS - CI Model 3.96 61.34 5.06 0.958 similar. Table 6.2 also summarizes the path-loss models for the directional PDP. In the case of directional PDP, the path-loss components are slightly higher than the omnidirectional case. For 28th St, this is expected, since as the RX moves away from the probability of having an optical LoS decreases, resulting relatively higher attenuation at large distances. For both omnidirectional scenarios, the cumulative distribution functions (CDF) of shadow fading are given in Figures 4.8 and 4.9. Both CI and ABG models follow zero-mean Gaussian distributions with the standard deviations listed in Table 6.2. In 28th St, we observe a high shadow fading variance due to the foliage penetration loss and other objects along the street. Furthermore, Table 6.2 also shows the P-values of the fits, acquired via Kolmogorov-Smirnov (KS) test which uses the measure of maximum difference between the CDF of the empirical and the hypothetical distributions [69]. In all cases KS-test do not reject the hypothetical Gaussian distribution with a P-value larger than 0.7. 103 Figure 4.8: CDF of the fading for 28th St. Figure 4.9: CDF of the fading for NLoS lo- cations 104 4.3.2 RMS Delay Spread As is common in the literature, we characterize the delay dispersion by the root-mean-square delay spread (RMS-DS), i.e., the second central moment of the power delay profile. S = v u u u t P ^ PDP (^ )^ 2 P RX 0 @ P ^ PDP (^ )^ P RX 1 A 2 (4.2) wheref^ =jPDP ()> 2 2 noise g and 2 noise is the noise power for omnidirectional PDP. Prior to the RMS-DS calculation, we apply noise filtering to avoid any contribution of the noise floor, which can significantly distort delay spread calculations by creating nonphysical contributions at large delays. Due to the automatic gain control implemented at the RX, the noise level might vary between directional PDPs for different beam pairs. Hence, we first obtain noise-filtered directional PDPs by: P ( TX ; RX ;) = 8 > > < > > : P ( TX ; RX ;) ifP ( TX ; RX ;)> 4 2 ( TX ; RX ) 0 otherwise (4.3) where 2 ( TX ; RX ) is the noise power for TX beam TX and RX beam RX . Figure 4.4 shows the noise-filteredPADP RX for RX 14. Then the omnidirectional PDP is calculated by using Equation 6.2. A sample omnidirectional PDP and the samples used for delay spread calculation are shown in Figure 4.6. Figure 4.10 and 4.11 show the cumulative distribution functions of the Log(RMS-DS) along with the corresponding Gaussian fits for 28th St and NLOS, respec- tively. As listed in Table 4.3, the median RMS-DS are 25:63 ns for 28th St. and 67:18 ns for NLoS. The mean of the Gaussian fits are -7.58 for 28th St and -7.2 for NLOS. In [70], for urban-micro scenario, they modeled the of the Log(RMS-DS) as =0:2Log(1+f)7:2 105 Figure 4.10: CDF of the logarithm of RMS- DS for 28th St. Figure 4.11: CDF of the logarithm of RMS- DS for NLoS locations and =0:21Log(1 +f) 6:88 for LOS and NLOS respectively. At 27:85 GHz, the cor- responding values are -7.49 and -7.19 which are well-aligned with our results. We also investigate the delay spread values for the directional case, i.e., for the TX/RX beam combi- nation that provides that highest receive power. In case of LOS 86% of the links have RMS-DS within 5 ns to 10 ns, see Figure 4.10, and the median is 8:7 ns. For NLOS, the directional RMS-DS vary from 8 ns to 70 ns as seen in Figure 4.11. We thus see that the ratio of omni-directional to directional delay spread is on the order of 3, a result that is comparable to the results in [24] for urban environments. 106 Table 4.3: Parameters of the RMS-DS Median(ns) P-value 28th St - omni 25.63 -7.58 0.263 0.889 28th St - directional 8.19 -8.10 0.101 0.254 NLoS - omni 67.18 -7.20 0.156 0.664 NLoS - directional 28.71 -7.55 0.271 0.991 4.3.3 Extracted Multi-paths By performing 3-dimensional peak detection in the P ( TX ; RX ;) we extract the multi- path components (MPC) with the information of; direction of departure (DOD), direction of arrival (DOA), delay and path gain. To avoid the ghost paths due to sidelobes of the beams, for every delay bin, we filter out any MPCs with 10 dB or less path gain compared to the highest power MPC in the same delay bin. The extracted MPCs for the RX 14 are shown in Figure 4.12. 4.4 Conclusion In this work, we presented results from a channel sounding campaign in a residential suburban environment at 28 GHz. The novel design of the channel sounder allowed phase-coherent measurements of all TX and RX angles. For path-loss, we provided parameters for both ABG and CI models. Although the environment is not urban, we saw that the mean RMS-DS results are inline with the Urban micro-cellular model provided in [70]. We showed that the channel sounder used in this campaign is capable of angular investigations for both TX and RX. In the future, we will provide statistics for angular spreads, perform more measurement campaigns to investigate the outdoor-to-indoor penetration loss and foliage effects. 107 Figure 4.12: Extracted multi-path components 108 Chapter 5 28 GHz Foliage Propagation Channel Measurements 5.1 Introduction There has been a growing interest into the millimeter-wave (mm-wave) channel sounding campaigns to understand the propagation channel characteristics at these frequencies, see [8, 38, 40, 65, 71, 72] and references therein. These measurements allow to understand the relative advantages and drawbacks of mm-wave systems compared to the legacy systems operating at microwave (sub 6 GHz) frequencies. Hence, they provide valuable insights for system design, optimization and validation. Foliage attenuation is one of the anticipated challenges facing mm-wave cellular or fixed wireless access networks. At mm-wave frequencies, the foliage penetration losses have been shown to be higher and more sensitive to factors such as foliage depth and density. Fur- thermore, outdoor mm-wave systems will be dependent on highly directional antennas or arrays to overcome the higher path-loss at higher frequencies [9]. Consequently, the angular 109 spectrum and its statistics are required inputs to the system design. Although foliage attenuation is recognized as a major challenge for mm-wave communi- cations, there has been a limited amount of measurements to investigate its effects at 28 GHz. Furthermore, most of the measurements only focused on the foliage penetration loss and did not consider the effects of foliage blockage on the angular spread and delay spread. As one of the earlier efforts to quantify the vegetation loss at mm-wave frequencies, [73] presents measurement results at 9.6 GHz, 28.8 GHz, and 57.6 GHz in an orchard. It was observed that the excess loss is almost linear with the foliage depth with a slope of 1-2 dB/m in the first 30 m, however, beyond that, the slope is only about 0.05 dB/m. In [74], the authors per- formed 28 GHz channel measurements in the presence of vegetation with a directional chan- nel sounder. They proposed an attenuation model based on the ITU-R terrestrial model [75] and the FITU-R-like frequency-dependent model [76] and observed that the excess loss due to foliage saturates after a certain distance. The frequency and the foliage depth dependent model proposed in [76] covers a frequency range of 10 GHz to 40 GHz. Additionally, [77] presents measurements through vegetation at 24 GHz, with the transmitter (TX) antenna de- ployed above average tree level and receiver (RX) antenna located close to ground level for a single environment. In [78], the authors investigate the excess loss due to dense foliage of a single tree in a tropical environment over the frequency bands of 2 to 18 GHz and 26.5 to 40 GHz. The observed excess losses as high as 9.69 dB/m are considerably higher than other existing measurements which are mostly performed in moderate climates. In this work, we present the results from 28 GHz channel sounding campaigns performed to investigate the effects of foliage blockage on the wireless propagation channel character- istics. A real-time channel sounder equipped with phased array antennas was used for the measurements [4, 5]. The phased arrays form beams at the different TX and RX angles and switch between these beams in microseconds, enabling directionally resolved results while 110 ensuring minimal variation in the environment during the measurements. Furthermore, with the help of beam-forming gain, the channel sounder provides a measurable path loss of up to 169 dB. We report a foliage attenuation model dependent on the foliage depth. Addition- ally, we compare the angular spread and delay spread for the links with and without foliage blockage. The rest of the chapter is organized as follows. Section 5.2 describes the measurement equipment and the configuration for these channel sounding campaigns. Section 6.2 provides details about the measurement scenarios under investigation. Section 5.4 presents results for penetration loss, and angular spreads and delay spread statistics in the presence of foliage blockage. Finally, Section 6.6 summarizes results and concludes the discussion. 5.2 Measurement Setup In this campaign, we used a switched-beam, real-time, wide-band mm-wave sounder with 400 MHz bandwidth [4, 5]. The sounding signal is a multi-tone signal which consists of 801 equally spaced tones covering 400 MHz. A low peak to average power ratio (PAPR) of 0:4 dB is achieved by adjusting the phases of individual tones as suggested in [42]. This allows transmitting with power as close as possible to the 1 dB compression point of the power amplifiers without driving them into saturation. Both the TX and the RX are equipped with 2 by 8 rectangular phased array antennas capable of forming beams that can be electronically steered with 5 resolution in the range of [45 ; 45 ] in azimuth and [30 ; 30 ] in elevation. During this measurement campaign, we utilize a single elevation angle 0 with 19 azimuth angles for the TX, and the RX. With an averaging factor of 10, the total sweep time is 14:44 ms for 361 total beam pairs. Since phased arrays cover 90 sectors, we rotated the RX tof0 ; 90 ; 180 ; 270 g to cover 360 while using a single orientation at the TX. Consequently, for each measurement location, we 111 obtain a frequency response matrix of size 72 by 19 by 801. Moreover, thanks to the beam-forming gain, the TX EIRP is 57 dBm, and the measur- able path loss is 159 dB without considering any averaging or spreading gain. Including the averaging ratio used in this campaign, the measurable path-loss is 169 dB. By using GPS- disciplined Rubidium frequency references, we were able to achieve both short-time and long-time phase stability. Combined with the short measurement time, this limits the phase drift between TX and RX, enabling phase-coherent sounding of all beam pairs even when TX and RX are physically separated and have no cabled connection for synchronization. Con- sequently, the directional power delay profiles (PDP) can be combined easily to acquire the omnidirectional PDP. Table 7.2 summarizes the detailed specification of the sounder and the sounding waveform. References [4] and [5] discuss further details of the sounder setup, the validation measurements, and the data processing. 5.3 Measurement Environments The measurements were performed on or near the University of Southern California, Uni- versity Park Campus in Los Angeles, CA, USA. For all measurements, the TX was placed on a scissor lift at the height of 5 m similar to an urban micro-cell base station. RX points were grouped as high and low which have 5 m and 2 m (actual heights vary from 1.8 m to 2.5 m due to terrain etc.) heights, respectively. During all measurements, we performed 90- degree sweeps for both TX and RX, to avoid any angular mismatch issues due to bore-sight alignment. The three measurement environments are shown in Figures 5.1, 5.2 and 5.3. To estimate the foliage depth for a given path, we build an abstract model of all environments including the RX points and the relevant foliage. Each tree in the path from TX to any RX points is modeled by an elevated cylinder and the corresponding parameters such as the height of the 112 Table 5.1: Sounder specifications Hardware Specifications Center Frequency 27.85 GHz Instantaneous Bandwidth 400 MHz Antenna array size 8 by 2 Horizontal beam steering (RX/TX) 360 / 90 Horizontal 3dB beam width 12 Vertical 3dB beam width 22 Beam steering steps 5 Beam switching speed 2s TX EIRP 57 dBm RX noise figure 5 dB ADC/AWG resolution 10/15-bit Data streaming speed 700 MBps Sounding Waveform Specifications Waveform duration 2s Repetition per beam pair 10 Number of tones 801 Tone spacing 500 kHz PAPR 0.4 dB Total sweep time 14.44 ms (each 90 Sector) Figure 5.1: Foliage measurements, location 1 113 RX Locations TX Figure 5.2: Foliage measurements, location 3 RX Locations TX Figure 5.3: Foliage measurements, location 3 114 Figure 5.4: Abstract model for location 3. foliage from the ground, the top height and the radius of the tree were estimated from an environment survey. Figure 5.4 shows an example of the abstract model for the location 3. By using this model, we then estimated the foliage depth for each TX-RX pair. The obtained foliage depths vary from 7 m to 52 m. 5.4 Results In the following, we describe the post processing of the measured data using Fourier-resolution techniques. The directional power delay profile (PADP) for the TX and RX beams with the 115 azimuth angles TX , RX , respectively is estimated as PADP ( TX ; RX ;) = F 1 n W ~ f H TX ; RX ~ f :=H cal ~ f o 2 (5.1) where RX 2 [180; 175], TX 2 [45; 45],F 1 denotes inverse Fourier transform,H TX ; RX is the frequency response for the TX beam direction TX and the RX beam direction RX and H cal ( ~ f) is the calibration response; W ( ~ f) is the Hanning Window, ~ f is the vector of frequency tones, and:= is element-wise division. Since all beam pairs are measured without a significant phase drift or trigger jitter, all directional PDPs are already aligned in the delay domain and require no further correction. To extract MPCs, we perform peak detection in 3-D space inPADP ( TX ; RX ;). Then, for each delay bin we filter the ghost MPCs caused by side-lobes in the beam pattern. The resulting delay resolution is 2:5 ns and the angular resolution is 5 . Similar to the approach in [25], the omnidirectional PDP is calculated as follows. PDP () = max TX max RX PADP ( TX ; RX ;) (5.2) Note that omnidirectional PDP throughout the chapter implies 360 azimuth coverage for the RX and 90 sector for the TX. Finally, the angular power spectrum is given by; APS( TX ; RX ) = X PADP ( TX ; RX ;) (5.3) Further details of the post-processing are described in [4, 5]. 116 5.4.1 Foliage Loss The excess loss due to foliage attenuation is calculated with respect to the theoretical free space path loss (FSPL). The excess loss due to foliage attenuation A ev (d v ) with a foliage depth ofd v is given by; A ev (d v ) =PL R (d;d v )FSPL(d) (5.4) where theFSPL(d) is the free space path loss at distanced (taking into account the antenna gain at TX and RX) and thePL R (d;d v ) is the observed path loss at the RX at the distanced and a foliage depthd v . The beam-forming gain and the response of the channel sounder are calibrated as discussed in [5]. As suggested in the ITU-R model [75], the excess loss due to foliage attenuation,A ev is modelled by: A ev (d v ) =A m 1exp A m d v (5.5) where, is the specified attenuation per distance (dB/m), andA m is the maximum attenuation for a given type of vegetation. Another model used in Ref. [74] isA ev (d v ) = af b d c v where a;b;c are the model parameters, f is the carrier frequency in MHz. However, this model did not show a good agreement with the measured data. Even though the estimated values were similar to the model presented in Ref. [74] and they provided comparable root mean square error (RMSE) with the previous model, the confidence intervals for the parameters were several magnitude of order larger than the values of the estimated parameters. The scatter plots in Figure 5.5 show the A ev values obtained from measurements with respect to the estimated foliage depths for low and high RX points. The excess loss due to foliage varies from 10 dB to 40 dB for both cases. For both data sets, the corresponding models obtained from the Equation 5.5 are also shown in Figure 5.5. Table 5.2 summarizes 117 0 10 20 30 40 50 60 Foliage depth d v (m) 0 10 20 30 40 Excess Loss(dB) Low RX High RX Low RX - fit High RX - fit Figure 5.5: Foliage loss for low and high RX locations with the fitted models Table 5.2: Parameters for the foliage penetration loss model RX A m RMSE() Value 95% Conf. Value 95% Conf. (dB) Low 2.85 (2.14, 3.56) 33.34 (30.08, 36.59) 4.65 High 4.07 (3.35, 4.80) 33.57 (32.22, 34.93) 2.85 estimated parameters for the model along with the 95% confidence internals and the RMSE of the estimation. We observed that the losses are 3 to 5 dB higher for the high RX locations when the foliage depth is relatively small. Note that the figure plots the loss as a function of the foliage depth, not the geometrical distance; for the same geometrical distance, the attenuation for a high RX is smaller than for a low RX. However, the maximum losses are similar in both cases. Consequently, the values for the estimated parameter A m are also similar. Figure 5.6 shows the residual deviation from the model proposed in Table 5.2. For both cases, the cumulative distribution function (CDF) of the residual values can be well approximated with zero-mean Normal distribution with a standard deviation of 4.65 and 2.85 for Low and High RX, respectively. Hence, to include the shadowing variation into the foliage model, we add a zero-mean Normal random variable with the given standard deviation (i.e N(0;)) to the model in Equation (5.5) resulting in: 118 -15 -10 -5 0 5 10 15 Residual values (dB) 0 0.2 0.4 0.6 0.8 1 CDF(x) Low RX High RX Low RX- Normal Fit High RX- Normal Fit Figure 5.6: CDF of the residual values from the proposed model in Table 5.2. A ev (d v ) =A m 1exp A m d v +N(0;): (5.6) 5.4.2 Delay Spread In this section, we compare the delay spreads of the links that are shadowed by foliage with the line-of-sight (LOS) measurements taken in the environment shown in Figure 5.1. As is common in the literature, we characterize the delay dispersion by the root-mean-square delay spread (RMS-DS), i.e., the second central moment of the power delay profile [79] S = v u u u t P ^ PDP (^ )^ 2 P ^ PDP (^ ) 0 @ P ^ PDP (^ )^ P ^ PDP (^ ) 1 A 2 (5.7) wheref^ =jPDP ()> 4 2 noise g and 2 noise is the noise power estimated from the first 30 samples of the PDP. We both consider the omnidirectional PDP given in the Equation (6.2) and the directional PDP which is defined as the PDP acquired from the TX-RX beam pair 119 0 20 40 60 80 100 RMS-DS(ns) 0 0.2 0.4 0.6 0.8 1 CDF(x) LOS Foliage Figure 5.7: CDFs of the RMS-DS for omnidirectional RX and 90 TX sector for LOS and Foliage shadowed links with the highest received power. The noise thresholding is used for both cases. Figure 5.7 shows the CDFs of the omnidirectional RMS-DS for the LOS links and links blocked by foliage. The mean RMS-DS for LOS is 9:3 ns, while it is 62:7 ns for the links blocked by the foliage. The mean directional RMS-DS is significantly smaller than the mean omnidirectional RMS-DS for both LOS and foliage links, namely 3:5 ns for LOS and 12:5 ns for foliage links. To understand the increase the RMS-DS, we further investigate the omnidi- rectional PDPs for a LOS link (LOS 2 in Figure 5.1) and a link shadowed by foliage, shown in Figure 5.9. Both links have approximately 90 m TX-RX distance. Although the foliage penetration does not introduce additional echoes, it decreases the power in the direct path. Since some of the other paths with larger delays are reflections from the surrounding struc- tures and do not go through the foliage to reach the RX, they do not suffer foliage attenuation. Consequently, the foliage shadowed links have significantly larger RMS-DS. Finally, Table 7.4 summarizes the RMS-DS statistics along with the lognormal fits to the data. Note that the fit for the LOS case is only provided for comparison, and are not statistically reliable due to the limited number of data points. 120 0 10 20 30 40 50 60 RMS-DS(ns) 0 0.2 0.4 0.6 0.8 1 CDF(x) LOS Foliage Figure 5.8: CDF of the RMS-DS for directional RX and TX with 12 half power beam width for LOS and Foliage shadowed links 300 400 500 600 700 800 900 1000 Delay (ns) -170 -160 -150 -140 -130 -120 -110 -100 PDP (dB) LOS Foliage Figure 5.9: Omnidirectional power delay profiles for two links with LOS and foliage block- age 121 Table 5.3: Delay spread statistics and lognormal fit parameters Data Mean Median Lognormal fit (ns) (ns) Directional - LOS 3.5 2.8 -8.47 0.15 Directional - Foliage 12.5 8.8 -8.02 0.30 Omni - LOS 9.3 9.2 -8.07 0.22 Omni - Foliage 62.7 60.4 -7.22 0.14 5.4.3 Angular Spread In this section, we compare the angular spreads for the direction of arrival (DoA) and the direction of departure (DoD) of the links which are shadowed by foliage with the LOS mea- surements taken in the same environment shown in Figure 5.1. The angular spreadsS are calculated as follows [80] S = v u u u u t P jexp(j) j 2 APS() P APS() (5.8) and = P exp(j)APS() P APS() (5.9) where2 [;) for DoA and2 [=4;=4] for DoD. Figure 5.10 shows the CDFs of the DoA angular spreads for LOS and foliage shadowed links. The mean angular spread for the shadowed links is 36:4 which is significantly higher than the LOS case. Figure 5.11 depicts the CDFs of the DoD angular spreads for LOS and foliage shadowed links. The DoD angular spread is obtained from a 90 sector in the azimuth, however, this actually resembles a typical sectoral deployment of a micro-cell in a realistic scenario. As there are no objects nearby the TX and the LOS path is quite dominant, DoD 122 0 10 20 30 40 50 60 DoA Angular Spread (degree) 0 0.2 0.4 0.6 0.8 1 CDF(x) LOS Foliage Figure 5.10: CDF sof DoA angular spreads for LOS and Foliage shadowed links 0 10 20 30 40 50 60 DoD Angular Spread (degree) 0 0.2 0.4 0.6 0.8 1 CDF(x) LOS Foliage Figure 5.11: CDFs of DoD angular spreads for LOS and Foliage shadowed links angular spread is limited to only 2:7 for LOS. Once the LOS is blocked by foliage, the angular spread increases by almost 500%. Finally, Table 6.3 summarizes the angular spread statistics along with the lognormal fits to the data. Note that the fit for the LOS cases is only provided for comparison, and are not statistically reliable due to the limited number of data points. 123 Table 5.4: Angular spread statistics and lognormal fit parameters Data Mean Median Lognormal fit DoA - LOS 13.0 13.1 1.11 0.04 DoA - Foliage 36.4 36.5 1.55 0.12 DoD - LOS 2.7 2.6 0.41 0.18 DoD - Foliage 15.9 15.3 1.17 0.17 5.5 Conclusions In this chapter, we presented measurements at 28 GHz investigating effects of foliage block- age on the RX power, delay spread and angular spreads. We observed that the foliage loss follows the functional form of the ITU-R model, with parameters extracted from our mea- surements. The foliage penetration loss is observed to be higher for higher RX locations for relatively small foliage depths. However, for both RX heights, the loss saturates around 33 dB. Furthermore, we observed that the foliage shadowed links have significantly higher RMS-DS and DoA/DoD angular spreads compared to the LOS links in the same environ- ments. 124 Chapter 6 28 GHz Outdoor to Indoor Propagation Channel Measurements 6.1 Introduction The number of connected devices and their data requirements have been increasing expo- nentially. Especially with the introduction of new applications such as augmented reality, virtual reality, and Ultra-HD video streaming, monthly global IP traffic is expected to reach 278 exabytes by 2021 [47]. While a variety of different techniques will be employed in 5G to enable this growth [57], [58], utilizing the fallow spectrum beyond 6 GHz is among the most promising approaches [56]. Gbps broadband connections to multiple users can be realized by exploiting the large bandwidths available at mm-wave frequencies. [81, 82]. An accurate channel model is crucial for any efficient wireless system design. The prospect of mm-wave wireless systems has thus fueled interest in mm-wave propagation channel measurements, e.g., [65, 71]. Since many of the future applications will demand connectivity between outdoor base stations and indoor users, Outdoor to indoor (O2I) pene- 125 tration is one of the most important factors affecting the performance of mm-wave systems. Not only does the penetration loss take on higher values than at cm-wave frequencies, but it also is highly sensitive to the several factors such as density or the construction material types. Furthermore, at mm-wave frequencies, the angular characteristics of the propagation channel are of the utmost importance, since these systems are highly dependent on the beam-forming gain to overcome the higher path loss occurring at higher frequencies [9]. Most of the current literature at 28 GHz focus on outdoor to outdoor or indoor to indoor measurements, e.g., [8, 25, 26, 41, 83, 84] and references therein. As an exception from this trend, [85] reported O2I measurements at 28 GHz by using a rotating horn antenna channel sounder which can provide an accurate absolute delay as well as angular information. The paper observed a larger number of clusters, larger excess delays and larger angular spreads in- door compared to the corresponding outdoor locations. However, there was a limited number of indoor RX locations and a single type of building was investigated. Ref. [86] performed narrowband measurements at 28 GHz where the RX was in an office building, and reported O2I penetration loss values varying from 3 dB to 60 dB depending on the RX location and the construction material types. However, the measurements were performed for highly direc- tional receivers and do not consider the effects of scattering-rich indoor environments on the angular spectrum. Furthermore due to the limited bandwidth, the delay spread statistics were not considered. Similarly, [87] presents narrowband building penetration measurements in a suburban residential neighborhood. It proposes a common-slope cross-comparison method and estimates building penetration losses varying from 9 dB to 17 dB. Ref. [88] investigates the penetration loss for external and internal walls at different carrier frequencies ranging from 0:8 GHz to 28 GHz with narrowband signals, and proposes a linear model for frequency dependency of the penetration loss. In [89], the authors present results for penetration loss and reflection coefficient measurements for different types of building materials. For exam- 126 ple; the penetration loss for clear glass is measured as 3:9 dB while it is 40 dB for tinted glass. However, the reported O2I measurement are resembling a device-to-device use case rather than a cellular deployment. A summary of penetration loss results, and a frequency- dependent model, is given in [72] and also used in 3GPP. None of references [72, 86–89] provide any delay or angular statistics. There are other studies investigating the O2I propagation channel at different mm-wave frequencies. Ref. [90] provides delay and angular statistics at 20 GHz for one office building on different floors. For a urban micro-cellular environment, [91] presents the penetration loss at 26 GHz and 37 GHz, and compares it with microwave frequencies. It also reports the effect of incident angle on penetration loss and compare the measurement results with the 3GPP and IMT-Advanced models, however, it does not discuss delay spread or angular spread. [92] investigates the O2I penetration loss and delay spreads for the O2I channels at 60 GHz. However, the distance from the base-station to the building under test was only 10 m which is not a typical scenario for either mobile networks or fixed wireless access. In this work, we present the results from 28 GHz channel sounding campaigns performed to investigate the effects of O2I penetration on the wireless propagation channel character- istics. A real-time channel sounder equipped with phased array antennas was used for the measurements [4, 5]. The phased arrays form beams at the different transmitter (TX) and receiver (RX) and switch between these beams in microseconds, enabling directionally re- solved results while ensuring minimal variation in the environment during the measurements. Furthermore, with the help of beam-forming gain, the channel sounder provides a measur- able path loss of 169 dB. We compare the outdoor to outdoor and outdoor to indoor channel characteristics and report the effects of O2I penetration on the received power, the angular spread of the direction of arrival (DoA) and the delay spread. Furthermore, we compare these channel characteristics for omnidirectional and directional RX cases. Finally, we investigate 127 the available beam diversity and the expected additional loss due to a blocked sector. The rest of the chapter is organized as follows. Section 6.2 describes the measurement equipment and the details about the measurement scenarios under investigation. Section 6.3 presents sample results that are related to the site-specific propagation physics. Section 6.4 presents statistical results and models for O2I penetration loss, delay and angular spread for indoor and outdoor RX locations. Section 6.5 compares the channel parameters with the most relevant 3GPP model. Finally, Section 6.6 summarizes results and presents conclusions for system design.. 6.2 Measurement Campaign 6.2.1 Measurement Environments The measurements were performed at three different locations on or near the University of Southern California campus. For all cases, the RX was on the first floor, while the TX height was on a scissor lift at the height of 5 m imitating an urban micro-cell (UMi) scenario. To investigate the more challenging cases, we chose buildings surrounded by foliage and made sure the angle of the direct path is quasi-grazing with respect to the front facade of the target building. 6.2.1.1 Single-Family Units We performed O2I measurements in two different single-family units (SFU). Figure 6.1 shows TX and RX locations on the campus for the first SFU which is named as SFU1. The building is a two-story, detached, single-family unit built using wood frames and drywalls, as is typical in California. There is also a covered, first-floor patio wrapping around the build- ing. Measurement points are marked in Figure 6.2. There are 5 outdoor points placed right 128 TX RX Figure 6.1: TX and RX locations for SFU1 I6 O1 O2 O3 O4 O5 0 270 180 90 I1 I2 I3 I4 I5 I7 I8 I9 I10 I11 I12 Patio Indoor 2m 1.5m 2.5m 3m 3m 1m TX direction 10 degrees 185 m Receiver Orientations Figure 6.2: Layout of RX points for SFU1 in front of the windows on the patio. 12 indoor measurement locations are placed throughout the room as the furniture allows. All RX points are at 1.8 m height above the first floor. The TX is placed on the same street with the house at a distance of 185 m at an angle of 10 according to the given directions in Figure 6.1. The direct path from the TX to the house is blocked by foliage from trees. Additionally, points O1 and O2 are also shadowed by the building across the street. The second SFU (SFU2) has a similar structure as SFU1 except the fact that SFU1 was on the corner of an intersection while SFU2 is not, as seen in Figure 6.3. In this case, we 129 TX1 RX TX2 Figure 6.3: TX and RX locations for SFU2 Indoor Patio 1m 1m 1m O1 I5 I4 I3 I2 I6 0 270 180 90 Receiver orientations I1 3.5m 6.5m 1.75m 2.5m 3.65m 2m 1.3m 1.2m O2 O3 Direction from TX 2 12 degrees 110m Direction from TX 1 25 degrees 50m Figure 6.4: Layout of RX points for SFU2 130 TX RX Figure 6.5: TX and RX locations for BOB Indoor O1 I1 0 270 180 90 Receiver orientations I2 I3 I4 I5 I6 I7 I8 I9 O2 O3 TX direction 16 degrees 114 m 4.1m 2.4m 4.7m 1.6m 2.58m 4.64m 4.62m 0.94m 3.7m 3.15m 3.15m 0.94m 1m 0.4m 1m 1m Figure 6.6: Layout of RX points for the BOB repeated the measurements for two transmitter points; TX1 and TX2. TX1 is located 50 m away from the house and at the angle of 25 , Figure 6.4. TX2 is a more challenging location as it is placed 110 m away from the house at the angle of 12 . Furthermore, the side window (the window facing 0 ) is visible from TX1, while it is blocked by the neighboring house in case of TX2. For both TX locations, the TX height is 5 m. The RX heights for all indoor points and O1 are 1.8 m above the first floor while the points O2 and O3 are 1.8 m with respect to the street level. 131 6.2.1.2 Brick Office Building The second measurement scenario is a multi-story, brick building (BOB) surrounded by heavy foliage as shown in Figure 6.5. The TX-RX distance is 114 m, and the angle of the direct path is 16 . Three outdoor measurement locations are just outside of the three front- facing windows. For each window, there are three indoor locations placed on a line along with the corresponding outdoor point. The distances of the indoor measurement points from the window are 0:4 m, 1:4 m and 2:4 m, see Figure 6.6. We finally notice that none of the windows is of the energy-saving type. This has important consequences for propagation, since energy-saving windows. which are typically covered with a thin metal film, have much higher attenuation than regular ones. 6.2.2 Measurement Setup In this campaign, we used a switched-beam, real-time, wide-band mm-wave sounder with 400 MHz bandwidth [4, 5]. The sounding signal is a multi-tone signal which consists of equally spaced 801 tones covering 400 MHz. A low peak to average power ratio (PAPR) of 0:4 dB is achieved by adjusting the phases of individual tones as suggested in [42]. This allows to transmit with power as close as possible to the 1 dB compression point of the power amplifiers without driving them into saturation. Both the TX and the RX are equipped with 2 by 8 rectangular phased array antennas capable of forming beams that can be electronically steered with 5 resolution in the range of [45 ; 45 ] in azimuth and [30 ; 30 ] in elevation. During this measurement campaign we utilize a single elevation angle, 0 , with 19 azimuth angles for the TX, and 7 elevation angles along with 19 azimuth angles for the RX. With an averaging factor of 10, the total sweep time is 101:08 ms for 2527 total beam pairs. Since phased arrays cover 90 sectors, we rotated the RX tof0 ; 90 ; 180 ; 270 g to cover 360 while using a single orientation at the 132 Table 6.1: Sounder specifications Hardware Specifications Center Frequency 27.85 GHz Instantaneous Bandwidth 400 MHz Antenna array size 8 by 2 Horizontal beam steering (RX/TX) 360 / 90 Horizontal 3dB beam width 12 Vertical beam steering (RX/TX) 30 to 30 / 0 Vertical 3dB beam width 22 Horizontal/Vertical steering steps 5 / 10 Beam switching speed 2s TX EIRP 57 dBm RX noise figure 5 dB ADC/AWG resolution 10/15-bit Data streaming speed 700 MBps Sounding Waveform Specifications Waveform duration 2s Repetition per beam pair 10 Number of tones 801 Tone spacing 500 kHz PAPR 0.4 dB Total sweep time 101.08 ms (each 90 sector) 133 TX. Consequently, for each measurement location, we obtain a frequency response matrix of size 7 by 72 by 19 by 801. Moreover, thanks to the beam-forming gain, the TX equivalent isotropically radiated power (EIRP) is 57 dBm, and the measurable path loss is 159 dB without considering any averaging or spreading gain. Including the averaging ratio used in this campaign, the mea- surable path-loss is 169 dB. By using GPS-disciplined Rubidium frequency references, we were able to achieve both short-time and long-time phase stability. Combined with the short measurement time this limits the phase drift between TX and RX, enabling phase-coherent sounding of all beam pairs even when TX and RX are physically separated and have no cabled connection for synchronization. Consequently, the directional power delay profiles (PDP) can be combined easily to acquire the omnidirectional PDP. Table 7.2 summarizes the detailed specification of the sounder and the sounding waveform. References [4] and [5] discuss further details of the sounder setup, the validation measurements, and the data pro- cessing. 6.2.3 Data Processing In the following, we describe the post processing of the measured data using Fourier-resolution techniques. The directional power delay profile (PADP) for the TX and RX beams with the azimuth angles TX , RX , respectively and RX elevation angle RX is estimated as PADP ( TX ; RX ; RX ;) = F 1 n W ~ f H TX; RX; RX ~ f :=H cal ~ f o 2 (6.1) where RX 2 [180; 175], TX 2 [45; 45], RX 2 [30; 30],F 1 denotes inverse Fourier transform, H TX ; RX ; RX is the frequency response for the TX beam direction TX and the RX beam direction ( RX ; RX ) andH cal ( ~ f) is the calibration response;W ( ~ f) is the Hanning 134 Window, ~ f is the vector of frequency tones, and:= is element-wise division. Since all beam pairs are measured without a significant phase drift or trigger jitter, all directional PDPs are already aligned in the delay domain and require no further correction. To extract multi-path components (MPCs), we perform peak detection in 4-D space inPADP ( TX ; RX ; RX ;). Then, for each delay bin we filter the ghost MPCs caused by side-lobes in the beam pattern. Similar to the approach in [25], the power angular-delay profiles (PADP) for RX and the omnidirectional PDP are calculated as follows. PADP RX ( RX ; RX ;) = max TX PADP ( TX ; RX ; RX ;) PDP () = max RX max RX PADP RX ( RX ; RX ;) (6.2) The resulting delay resolution is 2:5 ns and the angular resolutions are 5 and 10 for azimuth and elevation, respectively. The RX elevation resolution of 10 is not sufficient to study angular statistics in elevation. However, the elevation steering ensures estimating true power reflected from nearby objects at different heights including the window frames and furniture. Further details of the post-processing are described in [4, 5]. 6.3 Deterministic Results 6.3.1 Multi-path Components In this section, we discuss deterministic results to understand the effects of different scenarios on the propagation channel. We study the distribution of the extracted MPCs and compare them with the photos taken from the RX point of view 1 and the floor plans. First, we compare the indoor and outdoor locations for the BOB. Figure 6.7 shows the 1 Due to height difference between camera and actual antenna, and the fish-eye lens effect the observed elevation angles in the photos are distorted 135 -150 -100 -50 0 50 100 150 Azimuth (degree) -80 -60 -40 -20 0 20 40 Elevation (degree) -150 -140 -130 -120 Figure 6.7: Detected MPC components vs DoA - outdoor O1 in BOB, the color and the size of the point indicate the path gain of the MPC -150 -100 -50 0 50 100 150 Azimuth (degree) -80 -60 -40 -20 0 20 40 Elevation (degree) -170 -165 -160 -155 -150 -145 Figure 6.8: Detected MPC components vs DoA - indoor I2 in BOB, the color and the size of the point indicate the path gain of the MPC MPCs observed at O1 outside BOB. The strongest MPC is at 15 and it is the line-of-sight (LOS) path propagating through the foliage. Additionally, there is a cluster of MPCs with RX 2 [50 ;5 ] due to reflections from the building under investigation. The MPCs that have RX 2 [20 ; 180 ] are caused by reflections some further-away objects (e.g., buildings and poles), they have larger delays and less path gain due to the larger propagation distance. For the indoor location in Figure 6.8, we see that almost all of the MPCs are entering the building via the windows, as the brick walls do not allow any significant penetration. Figure 6.9 shows the PADP for the same environment. The MPCs with the smallest delay also have the same DoA of 15 as the direct path outdoors, indicating that this is the path through the brick wall. However, this path is attenuated by more than 45 dB. After the first 136 -150 -100 -50 0 50 100 150 Azimuth (degree) 350 400 450 500 550 Delay (ns) -170 -165 -160 -155 -150 -145 Figure 6.9: MPCs vs delay and DoA - indoor I2 in BOB, the color and the size of the point indicate the path gain of the MPC MPCs, there is the first cluster of MPCs with delays less than 400 ns, which are caused by the interactions of the direct path with the window frames. When we compare the path gains of these MPCs with the direct path observed outdoor, we see that the excess loss is more than 30 dB, caused by the large incident angle. However, the other MPCs, which are reflected by surrounding structures, have more favourable angles of incidence. Consequently, their path gain only decreases by 5 dB to 15 dB compared to the outside locations. This indicates the effects of O2I penetration loss does not only depend on the structure under investigation but it also depends on the surrounding buildings. Furthermore, this weighting of MPCs increases the indoor root mean square delay spread (RMS-DS). The effects of these phenomena on the O2I penetration loss, RMS-DS and angular spread will be further discussed in Sections 6.4.2 and 6.4.3. Figure 6.10 shows the DoAs of the MPCs seen at RX location I7 in SFU1. SFU1 is a corner house with large windows on two sides, which are both illuminated by the TX (compare Figure 6.2). In addition to the MPCs through windows, there are several additional MPCs entering the house through the walls, which were built by wooden frames and drywalls. 137 -150 -100 -50 0 50 100 150 Azimuth (degree) -80 -60 -40 -20 0 20 40 Elevation (degree) -170 -165 -160 -155 -150 Figure 6.10: Detected MPCs vs DoA - indoor I7 in SFU1, the color and the size of the point indicate the path gain of the MPC 25 30 35 40 45 50 Angular Spread (degree) Figure 6.11: Angular spreads for all RX locations in SFU1, red arrows indicate the mean direction of arrivals, black arrows indicate the RX beam directions for the five highest MPCs. 138 25 30 35 40 45 50 Angular Spread (degree) Figure 6.12: Angular spreads for all RX locations in SFU2 for TX2, red arrows indicate the mean direction of arrivals, black arrows indicate the RX beam directions for the five highest MPCs. 32 34 36 38 40 42 44 46 48 Angular Spread (degree) Figure 6.13: Angular spreads for all RX locations in BOB, red arrows indicate the mean direction of arrivals, black arrows indicate the RX beam directions for the five highest MPCs. 139 Consequently, we see a larger range of azimuth angles indoors, compared to the BOB. As seen in Figure 6.11, for every RX point, the observed set of paths varies depending on the exact location within the house. As a result of this, the angular spread values and the mean angles change significantly depending on the RX location within the room (note that Figure 6.11 shows only the mean angle and the strongest 5 paths for representation purposes; as can be observed from Figure 6.10, the number of actual MPCs is much larger). Although SFU2’s building materials are similar to the SFU1, as seen in Figure 6.12, the observed MPCs are mostly arriving from the front side of the building since the neighboring structures block the side wall. Hence, most of the dominant paths enter the buildings via the front side. The resulting angular distribution of the MPCs in SFU2 is thus more similar to that of the BOB shown in Figure 6.13. Furthermore, the mean angles for indoor RX locations are always towards the windows. In summary, in a realistic environment, there will be multiple paths arriving from the BS to the building under consideration. Since these paths might have a varying range of azimuth angles, they will experience different levels of O2I attenuation depending on their incident angles. Hence the O2I penetration will filter out some of the MPCs and favour others. The distribution of the angles incident on the building depends on the outside environment, while the filtering is a function of the building material and window layout. Consequently, observed propagation channel statistics and their differences for outdoor and indoor location will depend on site-specific details. Thus, it is crucial to design measurement campaigns which can consider the composite effect of these variables. It also becomes clear that a model that simply attenuates all MPCs by a “bulk” attenuation is not reflecting physical reality. 140 All Os I1,I4,I6,I7,I9,I12 I2,I5,I7,I10 I3,I8,I11 RX Locations 110 120 130 140 150 Path loss (dB) Outdoor Indoor Mean PL Figure 6.14: Omnidirectional path loss vs the RX locations in SFU1, the indoor locations ordered with respect to the distance from the window O3 O2 O1 I1,I2 I3 I4 I5 RX Locations 95 100 105 110 115 120 Path loss (dB) Outdoor Indoor Mean PL Figure 6.15: Omnidirectional path loss vs the RX locations in SFU2-TX1, the indoor loca- tions ordered with respect to the distance from the window 6.3.2 O2I Penetration Loss In this section, we discuss O2I penetration loss for omnidirectional and directional RX an- tennas. The omnidirectional path loss is calculated as described in Section 6.2.3, while the directional RX simply chooses the best available beam in terms of RX power. As discussed above, this pathloss is meaningful when assessing coverage, but should not be interpreted as constant loss for each MPC. Figures 6.14, 6.15, 6.16 and 6.17 show the path loss for all RX points in all measurement locations. The points are ordered with respect to their distance from the windows (i.e. the outdoor points on the left are further away from the window and indoor points on the right are further away from the windows). In case of SFU1, the free space path loss for 185 m TX-RX distance at 28 GHz is 106.7 141 O3 O2 O1 I1,I2,I6 I3 I4 I5 RX Locations 100 110 120 130 Path loss (dB) Outdoor Indoor Mean PL Figure 6.16: Omnidirectional path loss vs the RX locations in SFU2-TX2 the indoor locations ordered with respect to the distance from the window O1,O2,O3 I1,I4,I7 I2,I5,I8 I3,I6,I9 RX Locations 110 120 130 140 150 Path loss (dB) Outdoor Indoor Mean PL Figure 6.17: Omnidirectional path loss vs the RX locations in BOB, the indoor locations ordered with respect to the distance from the window 142 dB. However, due to shadowing from foliage, the path-loss for the outdoor RX locations varies between 117 dB to 135 dB with a mean of 127.8 dB. According to the path-loss model in [38], based on measurements in a similar environment, the anticipated path loss is 127.4 dB for the distance of 185 m which shows a good agreement with our results. In case of SFU2, the two outdoor RX locations;O2 andO3 have almost line-of-sight LOS channels, hence the observed path loss values are similar to the LOS path loss for both TX1 and TX2. However, O1’s view to the TX is blocked by the neighboring structure and foliage resulting in a rela- tively higher path-loss. Especially in case of TX2, O1 has 13 dB higher path loss than other outdoor points, although they all have similar TX-RX distances. The outdoor path loss values observed in the BOB vary from 111 dB to 122 dB with a mean of 117 dB, which indicates a loss due to foliage around 15 dB. We first consider the impact of building material on the penetration loss by comparing points directly outside of the house and directly inside for three different scenarios SFU1, SFU2-TX2, and BOB that have similar incident angles. For SFU1 and SFU2-TX2 the excess losses are 10.0 dB and 8.1 dB, respectively. Care must be taken in the interpretation of the penetration loss in SFU2, in particular when using location O1 as a reference. The patio structure obstructs the propagation of quasi-line-of sight components, so that the outdoor attenuation is higher. Thus, outdoor locations O2 and O3 provide better estimates of the overall outdoor arriving power. If we take the other outdoor points into account as well, the excess loss for SFU2-TX2 is as high as 16.6 dB which is significantly higher than SFU1; this is due to the fact that only one facade has windows, as discussed in 6.3.1. It is noteworthy that the BOB structure has a considerably higher pathloss than the SFU locations, namely more than 22 dB. As discussed earlier, in this location, the only viable paths from TX to RX are through the windows, which decreases the indoor RX power significantly, see Figure 6.8 and Section 6.3.1. 143 As seen in Figures 6.14, 6.15, 6.16 and 6.17, we did not observe any significant increase in the path loss at the RX locations far away from the windows. Except at SFU1 (where the existence of the side windows leads to significant variations of the path loss within the room), all other measurements observed standard deviations of path loss within the room of only 1-2 dB without a significant dependency on the distance from the window. For BOB and SFU2, in parallel to the discussions in Section 6.3.1, we see that for the indoor locations the MPCs with the highest power are either reflections from the window frames or the MPCs that are reflected/diffracted from outdoor objects near the receiver and propagate almost perpen- dicularly into the building. Neither of these two propagation paths are significantly affected by the additional distance for the RX points deeper in the rooms. 6.4 Statistical Results In this section, we discuss the statistical distributions of the estimated O2I penetration loss, angular spread and RMS-DS. 6.4.1 O2I Penetration Loss Figures 6.18, 6.19 and 6.20 show the cumulative distribution functions (CDFs) of the path loss values for all measurement environments. The corresponding means and standard de- viations of path loss values for indoor and outdoor RX points are summarized in Table 6.2. To investigate the distribution of the O2I penetration loss, we subtract the mean outdoor path loss from indoor path loss values and consider these values as realizations of an ensemble 2 . For each case, the normality of the distribution of the penetration losses in logarithmic scale is tested with Kolmogorov-Smirnov test [69] and it is not rejected with a 5% significance 2 Note the observed distribution is affected by both the varying O2I penetration losses and the indoor propa- gation channel. 144 115 120 125 130 135 140 145 150 Path loss (dB) 0 0.2 0.4 0.6 0.8 1 CDF(x) Outdoor-omni Indoor-omni Outdoor-bb Indoor-bb Figure 6.18: CDF of path loss in the SFU1 for omnidirectional and the directional with the best beam 90 100 110 120 130 140 150 Path loss (dB) 0 0.2 0.4 0.6 0.8 1 CDF(x) Outdoor-omni Indoor-omni Outdoor-bb Indoor-bb Figure 6.19: CDF of path loss in the SFU2 for omnidirectional and the directional with the best beam level. The corresponding and values are given in Table 6.2 as well. We first discuss the O2I penetration loss for omnidirectional RX. In case of SFU1, the O2I penetration loss has a mean of 10.6 dB and a standard deviation of 5.2 dB. Even though the two buildings are similar in terms of construction, the penetration loss for SFU2 is higher than SFU1. In SFU2, the mean penetration losses are 14.4 dB and 18.2 dB for TX1 and TX2, respectively. In case of SFU2-TX2 which has a similar incident angle with SFU1, the penetration loss is 8 dB higher than SFU1. As described in Section 6.3.1, a likely cause is that the SFU1 is a corner house and has less blockage from the surrounding buildings. Furthermore, the standard deviations of the penetration losses are lower than SFU1, namely 1.8 dB and 2.3 dB for TX1 and TX2, respectively. When we compare TX1 and TX2 results 145 110 115 120 125 130 135 140 145 Path loss (dB) 0 0.2 0.4 0.6 0.8 1 CDF(x) Outdoor-omni Indoor-omni Outdoor-bb Indoor-bb Figure 6.20: CDF of path loss in the BOB for omnidirectional and the directional with the best beam for SFU2, we also see that the penetration loss is 4 dB higher for TX2, as the more grazing angle of incidence presents a more challenging situation for radio-waves’ penetration. Similar to SFU2 for all BOB indoor locations, MPCs undergo similar propagation paths, the observed path losses do not vary significantly. The mean path losses for outdoor and indoor locations are 117 dB and 139.7 dB, respectively. Hence the mean excess loss due to O2I penetration is 22.7 dB and the standard deviation is 1.2 dB. For indoor RX points in SFUs the directional path loss is approximately 2.5 dB higher than the omnidirectional path loss. However, the differences between directional and omni- directional path losses for outdoor RX points are around 1 dB. Consequently, the directional path loss is approximately 1.5 dB more than the omnidirectional case for all SFU measure- ments. The difference is more prominent in the case of BOB and the directional penetration losses are 2.5 dB more than the omnidirectional case. Note that this excludes the antenna gain, which needs to be considered for a realistic link budget. 6.4.2 Angular Spread In this section, we study the DoA angular spread of the MPCs shown in Section 6.3.1 and Figures 6.11, 6.12 and 6.13. 146 Table 6.2: Mean path-loss and penetration loss values Location Outdoor Indoor Penetration Loss omnidirectional SFU1 127.8 dB 7.4 dB 138.4 dB 5.2 dB 10.6 dB 5.2 dB SFU2 - TX1 102.0 dB 4.4 dB 116.4 dB 1.8 dB 14.4 dB 1.8 dB SFU2 - TX2 108.3 dB 7.4 dB 126.5 dB 2.3 dB 18.2 dB 2.3 dB BOB 117.0 dB 4.9 dB 139.7 dB 1.2 dB 22.7 dB 1.2 dB directional SFU1 128.6 dB 6.4 dB 140.9 dB 7.5 dB 12.3 dB 7.5 dB SFU2 - TX1 102.8 dB 5.5 dB 119.0 dB 1.8 dB 16.1 dB 1.8 dB SFU2 - TX2 109.9 dB 9.7 dB 128.8 dB 3.0 dB 19.0 dB 3.0 dB BOB 117.9 dB 5.6 dB 143.1 dB 1.3 dB 25.2 dB 1.3 dB In parallel to our earlier observations, for the indoor RX points SFU1, the observed subset of paths varies depends on the exact location within the house. As seen in Figure 6.21, the angular spread values vary from 21 to 55 although all outdoor points have angular spreads in the range of 38.5 to 47 . This variation within the room has important implications for system design. Deployment of antennas indoors might require higher adaptivity of the antenna pattern, and the angular spread could change drastically when relocating the antenna within a room. The wide range of mean angle of arrivals shown in Figure 6.11 supports this observation as well. Additionally, the mean angular spread for indoor is 43.5 , which is quite similar to the outdoor angular spread of 41 . For both SFU2 and BOB, the main path of propagation is via the front side of the house, due to blockage by the neighboring structure for SFU2 and lack of side windows for BOB. The observed mean angular spreads are slightly smaller for the indoor compared to the out- door RX points for both cases, see Table 6.3. However, in case of BOB, due to the limited number of outdoor locations, it is not certain that the results can be generalized. Further- more, the observed mean and sigma for indoor are quite similar for these two structures even 147 20 25 30 35 40 45 50 55 DOA angular spread (degree) 0 0.2 0.4 0.6 0.8 1 CDF(x) Outdoor Indoor Figure 6.21: CDF of DOA angular spread in the SFU1 20 25 30 35 40 45 50 55 DOA angular spread (degree) 0 0.2 0.4 0.6 0.8 1 CDF(x) Outdoor Indoor Figure 6.22: CDF of DOA angular spread in the SFU2 30 35 40 45 50 DOA angular spread (degree) 0 0.2 0.4 0.6 0.8 1 CDF(x) Outdoor Indoor Figure 6.23: CDF of DOA angular spread in the BOB 148 Table 6.3: Angular spread statistics and lognormal fit parameters Location Mean Median Lognormal fit SFU1-Indoor 43.51 46.31 1.62 0.13 SFU1-Outdoor 41.00 39.77 1.61 0.04 SFU2-Indoor 36.46 34.49 1.56 0.07 SFU2-Outdoor 38.12 37.25 1.54 0.18 BOB-Indoor 37.53 37.60 1.57 0.06 BOB-Outdoor 39.21 36.32 1.59 - O1 O2 O3 O4 O5 I1 I2 I3 I4 I5 I6 I7 I8 I9 I10 I11 I12 RX Locations 7 6 5 4 3 2 1 Sector order -30 -25 -20 -15 -10 -5 0 Figure 6.24: SFU1 - Ordered relative power (dB) of other sectors with respect to the sector with the highest power, dashed line indicated where it drops below 10 dB though they are built with significantly different types of materials. This indicates that the effects of the floor plan (i.e. number and dimensions of windows) are prominent compared to the effects of the building material. Finally, Table 6.3 summarizes the statistics of the angular spreads for all SFU and BOB angular spread values along with the parameters of the corresponding Lognormal fits. The proposed models are tested with Kolmogorov-Smirnov test [69] and they are not rejected with a 5% significance level. We do not propose a model for the BOB outdoor case due to the limited number of data points. 149 O1 O2 O3 I1 I2 I3 I4 I5 RX Locations 7 6 5 4 3 2 1 Sector order -30 -25 -20 -15 -10 -5 0 Figure 6.25: SFU2 TX1 - Ordered relative power (dB) of other sectors with respect to the sector with the highest power, dashed line indicated where it drops below 10 dB O1 O2 O3 I1 I2 I3 I4 I5 I6 RX Locations 7 6 5 4 3 2 1 Sector order -30 -25 -20 -15 -10 -5 0 Figure 6.26: SFU2 TX2 - Ordered relative power (dB) of other sectors with respect to the sector with the highest power, dashed line indicated where it drops below 10 dB O1 O2 O3 I1 I2 I3 I4 I5 I6 I7 I8 I9 RX Locations 7 6 5 4 3 2 1 Sector order -30 -25 -20 -15 -10 -5 0 Figure 6.27: BOB - Ordered relative power (dB) of other sectors with respect to the sector with the highest power, dashed line indicated where it drops below 10 dB 150 We next analyze the available beam diversity in the various measured environments. This is important in case that the strongest MPC is blocked, e.g., by a human person walking through the propagation path. We define the available diversity as follows: we divide the 360 azimuth into 8 sectors (of 45 each - we assume that a human can block one such sector - an assumption that of course depends on the distance between the human and the receiver). The sector centers are chosen as [0 , 45 , 90 ,..., 315 ]. We then determine the total arriving power in each sector and consider the number of sectors in which the arriving power is within 10 dB of the power in the strongest sector. We name these sectors as selectable sectors. This measure thus indicates whether switching to a different direction can maintain the connection. Figures 6.24, 6.25, 6.25 and 6.27 show the relative received power in each sector (in descending order) with respect to best sector for that RX location for all measurement envi- ronments. The dashed line indicates where the received power drops 10 dB below the best sector. In line with the large range of angular spread values for the SFU1 indoor RX loca- tions, the number of selectable sectors also varies from 1 to 6 with a mean of 3.9, see Figure 6.24. The mean received power differences between the best sector and the second best sec- tor are 4 dB and 3.2 dB for indoor and outdoor, respectively. As seen in Figures 6.25 and 6.26, O1, which is the point right outside of the windows, suffers shadowing from foliage and neighboring structures. In addition, it receives several reflections from the house itself, con- sequently, unlike other outdoor RX locations, 6 out of 8 sectors are within 10 dB of the best sector for both TX locations. The other outdoor RX locations have 0 to 3 available sectors. In case of indoors, the mean number of selectable sectors are 3.2 and 2.3 for TX1 and TX2, respectively. Furthermore, the expected loss in the received power when the best sector is blocked are 5.1 dB for outdoor and 3.5 dB for indoor. Another interesting observation is that the second-best sector is a neighboring sector to the best one for 10 out of 11 indoor RX lo- cations for the SFU2, while this ratio is only 6 out of 12 for the SFU1. Hence the availability 151 0 10 20 30 40 RMS-DS(ns) 0 0.2 0.4 0.6 0.8 1 CDF(x) Outdoor-omni Indoor-omni Outdoor-bb Indoor-bb Figure 6.28: CDF of RMS-DS in the SFU1 for omnidirectional and the directional with the best beam of a side window affects the correlation of the received power for neighboring sectors. In case of BOB, the mean number of selectable sectors are 2.7 and all indoor RX locations have a number of selections between 2 to 4. The mean received power differences between the best sector and the second best sector are 6.3 dB for outdoor and 4.3 dB for indoor which is slightly higher than in the other measurement locations. Similar to SFU2, for most of the cases (7 out of 9), the second best sector is a neighboring sector of the best one. This is anticipated as both structures only allow MPCs from a limited azimuth range. In summary, we observed that for most the indoor locations, there is at least one more sector within 10 dB of the main sector. Additionally, the mean RX power difference between the best and the second-best section varies from 3 to 5 dB. Considering the excess losses due to body blockage at 28 GHz are shown to be as high as 15 dB [93, 94], it is clearly beneficial to employ a structure with adaptive beam-forming for the indoor environments. 6.4.3 Root Mean Square Delay Spread Figures 6.28, 6.29 and 6.30 show the CDF of the RMS-DS values. Similar to the penetration loss, we investigate the cases where the RX has an omnidirectional view or if it selects the RX beam with the highest power (i.e. the directional RX). 152 0 20 40 60 80 RMS-DS(ns) 0 0.2 0.4 0.6 0.8 1 CDF(x) Outdoor-omni Indoor-omni Outdoor-bb Indoor-bb Figure 6.29: CDF of RMS-DS in the SFU2 for omnidirectional and the directional with the best beam 0 20 40 60 80 RMS-DS(ns) 0 0.2 0.4 0.6 0.8 1 CDF(x) Outdoor-omni Indoor-omni Outdoor-bb Indoor-bb Figure 6.30: CDF of RMS-DS in the BOB for omnidirectional and the directional with the best beam 300 400 500 600 700 800 900 1000 Delay (ns) -180 -160 -140 -120 PDP (dB) Outdoor Indoor Figure 6.31: Sample PDPs for outdoor and indoor RX location for SFU2 153 Table 6.4: RMS-DS statistics and lognormal fit parameters Location Mean Median Lognormal fit (ns) (ns) omnidirectional SFU1-Indoor 26.15 27.36 -7.61 0.19 SFU1-Outdoor 23.89 23.32 -7.66 0.23 SFU2-Indoor 36.74 25.73 -7.51 0.28 SFU2-Outdoor 18.26 14.18 -7.79 0.24 BOB-Indoor 51.99 47.86 -7.32 0.06 BOB-Outdoor 37.88 28.01 -7.46 - directional SFU1-Indoor 18.09 16.11 -7.79 0.21 SFU1-Outdoor 9.66 12.37 -8.09 0.33 SFU2-Indoor 26.80 17.64 -7.70 0.38 SFU2-Outdoor 7.55 6.51 -8.19 0.29 BOB-Indoor 46.18 43.18 -7.41 0.34 BOB-Outdoor 14.37 14.26 -7.84 - Figure 6.31 shows a comparison between outdoor and indoor PDPs from SFU2-TX2. In this case, the outdoor and indoor RMS-DS are 35 ns and 75 ns RMS-DS, respectively. Although the O2I propagation does not introduce new MPCs with large excess delays, the varying amount of penetration loss affecting MPCs causes significant changes in RMS-DS. In line with the discussion of Section 6.3, we see that early MPCs, which have grazing incidence, have stronger attenuation than later arriving components; this explains the increase in the delay spread. Following our earlier discussions in Section 6.3.1, O2I penetration forces all paths into a relatively limited angular range. Especially, in BOB and SFU2 this process is more promi- nent, and results in directional RMS-DS values similar to the omnidirectional ones for indoor locations even though the outdoor directional RMS-DS values are significantly less than the omnidirectional ones for these locations. 154 The mean and the median RMS-DS values and the corresponding Lognormal distribution parameters are listed in Table 7.4. Similar to the angular spread, the proposed models are tested with the Kolmogorov-Smirnov test [69] and they are not rejected with a 5% signifi- cance level. A model is not proposed for BOB-outdoor due to the limited number of data points. 6.5 3GPP-type Model Paremeters for Residential O2I In this section, we compare our findings with the O2I channel model parameters suggested in Table 7.5-6 in the most recent 3GPP standard for mm-wave frequency channels [95]. In terms of base-station height and the environment type, the urban micro (UMi) scenario presented in the report is the closest one to our measurements. However, the UMi channel model parameters are only given for a street canyon scenario. As discussed in Section 6.3.1, the relative locations of the surrounding buildings not only have an effect on the absolute values of the channel parameters, but they also affect how the channel parameters change during the O2I penetration. For 28 GHz UMi, the 3GPP standard reports the means of the logarithm of the RMS-DS DS as -7.49 for LOS, -7.18 for NLOS and -6.62 for O2I. Since these values are given for a street canyon scenario, as one would expect that they are significantly higher than our obser- vations. Especially in the O2I case the prescribed DS corresponds to 240 ns which is almost an order of magnitude larger than our measurements. Furthermore the difference between outdoor and indoor RMS-DS is much higher than our observation. Consequently, these val- ues are not applicable to the residential micro-cellular environments. From our observations, for a given house location if the DS,out of the outdoor RMS-DS is known, the indoor can be estimated by DS,in = DS,out + 0:2 for an omnidirectional RX, and by DS,in = DS,out + 0:45 for a directional RX, when the building under study has a single illuminated wall. If the 155 structure and environment allow MPCs to penetrate into the building from more than one side, then the indoor RMS-DS only increases slightly i.e. DS,in = DS,out + 0:05 for the omnidirectional RX and DS,in = DS,out + 0:3 for directional RX. The 3GPP standard also reports the mean of the logarithms of the angular spread of az- imuth DoA for LOS, NLOS, and O2I as 1.69, 1.61 and 1.76, respectively. Hence, the average indoor angular spreads are approximately 20% and 40% higher than the outdoor NLOS and LOS points. We observed this increase in only around 5% for the SFU cases. Furthermore, in case of BOB, which was a brick building (i.e. penetration is only possible via windows), the mean indoor angular spread was 5% smaller than the outdoor values. 6.6 Conclusions In this chapter, we presented measurements at 28 GHz investigating effects of outdoor to indoor penetration. We observed that the O2I penetration losses vary from 10 dB to 18 dB for a single family unit depending on the site-specific details such as place of the house on the street or the incident angle from the base station to the house. In the case of a brick building, the penetration loss is as high as 23 dB. Furthermore, the penetration loss is 2-3 dB higher when the RX is forced to choose a single directional beam with 12 half power beam width for all cases. For buildings with a single visible front from the TX that would allow O2I penetration, we have not observed any significant change in the indoor RX power with on the indoor location or the distance from the window. However, the RX power fluctuates more than 10 dB in the presence of a second side offering an alternative path of propagation. Perhaps most importantly, we find that the common approach of adding a “bulk” penetration loss to an outdoor model is not a viable way to model outdoor-to-indoor mm-wave chan- nels. Neither is a concatenation of an outdoor channel with an indoor channel; rather the environment-specific interaction has to be accounted for. Thus, effects of O2I propagation 156 on the DoA and delay spread statistics are more affected by the floor plan and the relative location of the building under investigation with respect to the surrounding structures rather than its building materials. Further investigations into the available beam-diversity showed that an adaptive beam- forming could improve the mean RX SNR by 5 to 7 dB in case the best sector is blocked by a human body. Hence, even in the case of fixed TX and RX, due to presence of other moving objects, the angular spectrum can change significantly over time. Consequently, further measurements investigating the temporal characteristics of O2I channels are required for more accurate propagation models. Finally, we compare our findings about RMS-DS and angular spread with the most recent 3GPP standard and find that the type of environment we measured in is not represented by an existing model. Furthermore, the estimated channel parameters significantly differ from the other defined environments in the standard. We therefore provide model parameters for a “suburban O2I” environment, which will be important for 5G deployment. 157 Chapter 7 Real-time Ultra-Wideband Channel Sounder Design 7.1 Introduction As the number of applications and their bandwidth requirements for wireless communica- tions increase, the need for frequency spectrum has also grown [96]. Since the frequencies lower than 6 GHz are mostly occupied, this need can only be met by utilizing the ample spec- trum at the higher frequencies that is currently lying fallow or under-utilized. In particular, the emergence of fifth generation cellular (5G) communications has significantly increased interest in millimeter-wave (mm-wave) communications and motivated propagation channel measurement at higher frequencies [71, 97, 98]. Although there has been increasing interest in the band between 6 GHz and 20 GHz [99], most of the measurement campaigns performed so far focus on bands below 6 GHz [44, 100–103] or above 20 GHz [4, 13, 22, 27, 28, 31, 104]. It is commonly accepted that the new mm-wave systems will have to coexist with the legacy networks (4G, LTE, WiFi 158 etc.) operating mostly below 6 GHz. Consequently, a host of other papers [105–109] com- pared the characteristics of below-6 GHz channels with mm-wave propagation and discuss the frequency dependency of the channel parameters. However, all of these work measure disjoint sub-bands rather than a continuous frequency band. Furthermore, except [106, 107], the measurements at the different carrier frequencies were not performed with the same setup (i.e., either separate up and down-conversion chains for each frequency or completely iso- lated channel sounders). Consequently, comparisons of parameters such as delay spread are difficult due to varying dynamic range of the receivers. Another type of wideband system that has drawn interest is ultra-wideband (UWB) com- munications which operate from 3.1 GHz to 10.6 GHz. UWB channels have been well investigated for indoor, outdoor or vehicular scenarios, e.g., [1, 110, 111]. Most of these measurements were performed with vector network analyzers (VNA), which can only oper- ate in static environments, or with time domain channel sounders requiring costly pieces of equipment such as high-speed (more than 10 GSps) arbitrary waveform generators and dig- itizers [112]. Hence it is neither practical nor cost-efficient to expand these setups beyond 10 GHz. To fill this gap, we designed a channel sounder that can perform measurements over the continuous band from 3 GHz to 18 GHz by utilizing a hybrid time/frequency domain ap- proach. Even though the instantaneous bandwidth of the setup is only 1 GHz (thus allowing the use of relatively low-cost components), by utilizing a sweeping sub-band approach, it can measure up to 15 GHz total bandwidth within 6 ms. The short measurement time allows to measure in dynamic environments as long as the coherence time of the channel is more than 6 ms. At 18 GHz this would correspond to a maximum speed of 5 km/h, which is the typical pedestrian speed [7]. Since the measurement time for our setup is directly proportional to the total bandwidth, measurements in more mobile environments can be performed with smaller 159 bandwidth. In comparison, a VNA, which is the standard choice of equipment for such large bandwidth measurements, would require a much longer time for similar measurements. For example, for a Keysight programmable VNA (PNA) 5224B, for the same frequency range (3-18 GHz) and with the same number of frequency points (30001) and 600 kHz intermedi- ate frequency bandwidth (similar to our frequency spacing of 500 kHz), the minimum sweep time is 72 ms compared to the 6 ms for our channel sounder. The frequency resolution (sub- carrier spacing) in our channel sounder is 500 kHz, corresponding to a maximum measurable excess run-length of multi-path components of 600 m (2s pseudorange), which is sufficient even for most outdoor measurements. Additionally, unlike a VNA, the transmitter (TX) and the receiver (RX) in our setup are physically separated, and they do not require any cable con- nections since the synchronization is provided by global positioning system (GPS)-stabilized rubidium frequency references. Hence, the channel sounder can operate in almost any desired measurement environment with a coherence time more than 6 ms. Consequently, the channel sounder is used for outdoor cellular measurements [113, 114] and the indoor measurement campaign discussed in this chapter. There has been limited work utilizing a sub-band or multi-band measurement approach. In [115], a similar approach is used to perform measurements at 60 GHz. However the total measurement time in the proposed setup is 10 s for 5 GHz of total bandwidth so that (contrary to our setup) the channel sounder in [115] can only be used in static scenarios. Furthermore, the frequency resolution achieved was 8 MHz limiting the maximum measurable delay spread to 37.5 m. Similarly, the channel sounder proposed in [116] uses a narrow band system to perform wide band measurements with a total 1 GHz bandwidth. Another channel sounder with the multi-band approach was presented in [117]. The channel sounder in [117] utilizes software defined radios and measures 10 sub-bands of 20 MHz providing a total bandwidth of 200 MHz around 5.6 GHz in 3 ms. While this enables measurements in dynamic environ- 160 ments, the bandwidth is almost two orders of magnitude lower than in our setup. Furthermore in both of the previous works, the TX and the RX units require a cable connection for sharing reference signals and control signals. Consequently, neither of the setups are suitable for long range measurements. In addition, the enormous bandwidth of our designed channel sounder introduces additional challenges such as frequency dependent IQ imbalance and system gain. The frequency dependencies necessitate additional steps to be taken during the operation of the channel sounder and the post processing of the data, see Sections 7.2 and 7.3. The contribution of this work thus include: • Designing a low-cost hybrid time/frequency domain channel sounder setup with a total bandwidth of 15 GHz in 6 ms measurement time, • Describing the method for channel sounder calibration to overcome hardware imper- fections in setups, • Proposing a method to concatenate multiple sub-bands into a wide-band frequency response, • Presenting validation measurements and investigating channel sounder performance, • Showing sample results for the frequency dependency of the root mean square delay spreads (RMS-DS) in an indoor channel sounding campaign with the proposed setup. This chapter is organized as follows. Section 7.2 describes the details of the channel sounder setup and its operation principles. Section 7.3 discusses the channel sounder calibra- tion steps and Section 7.4 demonstrates performance evaluations for the setup. Section 7.5 describes the measurement environment for the channel sounding campaign along with the detailed the post-processing steps and the results. Finally, Section 7.6 concludes the chapter with a summary and discussion of future work. 161 Figure 7.1: Transmitter block diagram, the descriptions of the units are given in Table 7.1 Figure 7.2: Receiver block diagram, the descriptions of the units are given in Table 7.1 7.2 Channel Sounder Design The proposed setup is a real-time, frequency-hopped multi-band channel sounder with direct up/down conversion. The TX and the RX were built as physically separate structures and they do not require any cable connection, allowing arbitrary placement of the TX and the RX. Figs. 7.1 and 7.2 show the block diagrams for the TX and the RX respectively. Furthermore, Table 7.1 lists the part numbers and descriptions of all the units used in the setup. 162 Table 7.1: List of the units shown in Figs. 7.1 and 7.2 along with their descriptions and relevant specifications Unit Number Unit Name Unit Description (1) Agilent N8241A 2-channel AWG, 15-bit resolution, 1.25-GSps sample rate (2) MLIQ-0218I IQ Mixer, 8.5 dB conversion loss (3) RHPF23G03G18 High pass filter, 2.1 dBa insertion loss, 3-18 GHz pass band (4) GT1000B Power amplifier, 40 dBm output power (5) SAS-547 Biconical antenna, 1-18 GHz frequency range (6) PAM-118A Low noise amplifier, 3 dB noise figure, 40 dB gain (7) NI PXIe-5160 2-channel ADC, 10-bit resolution, 1.25- GSps sample rate (8) NI HDD-8265 Raid array, 6 TB capacity, 700 MBps read/write speed (9) Phase Matrix FSW-0020 Frequncy synthesizer, 0.2-20 GHz fre- quency range (10) Ptsyst GPS10eR GPS-disciplined Rubidium Reference, Allan Deviation(1s)< 1.5e-12 (11) NI PXIe-8135 2.3 GHz Quad-Core PXI Controller (12) NI PXIe-6361 Multifunction I/O Module, 32-bit coun- ters at 100 MHz clock (13) NI PXIe-1082 PXIe Chassis 163 7.2.1 Single Band Measurements The TX operation for a single band measurement can be summarized as follows. A 15-bit, 1.25-GSps arbitrary waveform generator (AWG) generates the complex baseband sounding signal. In this measurement campaign, the sounding signal is a multi-tone waveform repre- sented as: m(t) = N X n=N e j(n2ft+n) (7.1) where f is the tone spacing, 2N + 1 is the number of tones and n is the phase of the tone n. In order to find the phases of a multi-tone signal that minimizes the peak to average power ratio (PAPR), we implemented an iterative algorithm that attempts to reduce the high peaks in the time-domain waveformm(t) [42]. The algorithm switches between the time and the frequency domain representation of the waveforms. In each iteration step, the error signal is computed by cropping the parts ofm(t) whose amplitudes are larger than a certain threshold. The removal of these high peaks is performed in the frequency domain by subtracting the frequency coefficients of the error signal from the current frequency coefficients. The new frequency coefficients are normalized to maintain the flat-top multi-tone waveform (i.e., the magnitude of each frequency coefficient is 1). Then, the updatedm(t) is calculated from new frequency coefficients. The algorithm iterates until the PAPR drops below a predetermined threshold. Note that the waveform can be precomputed off-line, and then is just stored in the AWG. In our case the PAPR was 0:4 dB, allowing us to transmit with power as close as possible to the 1 dB compression point of the power amplifier without driving it into saturation. The in-phase and quadrature (I and Q) components of the signal are the real and imaginary parts of m(t) respectively. The 500 MHz I and Q signals are generated by 2 channels of the AWG and transmitted to an IQ mixer via phase-matched cables. The IQ 164 mixer directly up-converts the baseband I and Q signals to a 1 GHz radio frequency (RF) signal around the carrier frequency provided by the frequency synthesizer. Finally, the RF signal is amplified and transmitted by a biconical antenna with a frequency range of 1 GHz to 18 GHz. Figs. 7.3 and 7.4 show the azimuth and elevation patterns of the antenna measured in an anechoic chamber. The RX operates in a similar manner to the TX. The received RF signal is down-converted to baseband I/Q signals and sampled with a 10-bit, 2-channel analog to digital converter (ADC). The acquired data is streamed into a 6 terabyte (TB) redundant array of independent disks (RAID), which is equipped with a PCIe x4 connection allowing 700 MBps sustained data write speed. With the fast streaming capability and the 6 TB of data storage, the chan- nel sounder can perform measurements for more than 2 hours with a 40% duty cycle (i.e 40% of the time, the channel sounder performs measurements and in parallel, continuously streaming the acquired data to the RAID array). With the current configuration the 10-bit ADC generates 2-Bytes per sample. The duty cycle can be doubled by limiting the ADC to 8-bit per sample, if necessary. We also note that this is an important difference to the use of a sampling oscilloscope at the RX. While a suitable scope could easily receive the whole 15 GHz bandwidth of interest, it could not provide sustained reading/writing required, e.g., for measuring the channel evolution as the mobile station moves on a trajectory. Both the TX and the RX are controlled with Labview scripts running on National Instru- ments (NI) real-time controllers. GPS-stabilized Rubidium frequency references provide two signals for the timing of the setup; a 10 MHz clock to be used as a timebase for all units and 1 pulse per second (PPS) signal aligned to coordinated universal time (UTC). Since the 1 PPS signals in the TX and the RX are both aligned to the UTC, they operate synchronously without requiring any physical connections. More importantly, the AWG, the ADC, and the frequency synthesizers were disciplined with the 10 MHz signal provided by these frequency 165 0 30 60 90 120 150 180 210 240 270 300 330 -10 -5 0 5 Azimuth pattern (dB) 3 GHz 6 GHz 9 GHz 12 GHz 15 GHz 18 GHz Figure 7.3: Measured azimuth patterns of the biconical antenna for frequencies between 3- 18 GHz 0 30 60 90 120 150 180 210 240 270 300 330 -20 -10 0 Elevation pattern (dB) 3 GHz 6 GHz 9 GHz 12 GHz 15 GHz 18 GHz Figure 7.4: Measured elevation patterns of the biconical antenna for frequencies between 3-18 GHz 166 references, thereby maintaining phase stability during the measurements, which is essential for the accurate measurement results [118]. The frequency references also provide GPS lo- cations, which are logged along with the measurement data. 7.2.2 Multi Band Measurements All RF units mentioned in Section 7.2.1 can operate from 3 GHz to 18 GHz. Additionally, the frequency synthesizers can switch between two arbitrary frequencies within this range in less than 100 s. Hence they can change the carrier frequency every 100 s, which is the main feature which allows the construction of the real-time channel sounder with a frequency sweep approach. One can think of the proposed channel sounder as a hybrid design lying be- tween a VNA and a time-domain channel sounder setup. Similar to a VNA, it sweeps through different frequency tones to obtain the frequency response of the channel under investigation. However, unlike a VNA using a single tone at a time, it uses a wide-band signal that itself consists of 2001 simultaneously transmitted tones. The first algorithm describes the operation for the multi-band measurements; the same procedure runs in the TX and the RX in parallel. In summary, we perform 30 channel mea- surements each with 1 GHz bandwidth and center frequencies of 3.25 GHz to 17.75 GHz with 500 MHz spacing. Given the measurement period and the duration to measure each sub-band, hardware counters in the NI DAQ timing modules count the rising edges of the 10 MHz clock and generate the trigger waveforms for the frequency synthesizer (to switch to the next carrier frequency) as well as the AWG at the TX and the ADC at the RX. Fig. 7.5 depicts the counter operation for the counter values corresponding to the 200s per band measurement duration and 6 ms total sweep time. To allow the frequency synthesizers to be stabilized, both the AWG and ADC wait for 120s once they received the trigger and then operate for 80s, which consists of 20 repetitions of the sounding waveform. Furthermore, 167 Algorithm 2 Multi-band Operation:N m ,N bands andN sweeps are the measurement duration in seconds, number of sub-bands to be measured, and repetition of multi-band measurements per second respectively. 1: procedure SWEEP SOUNDER(N m ,N bands ,N sweeps ) 2: i 0 3: whilei<N m do 4: while 1 PPS trigger not received do 5: Wait 6: end while 7: Start counter for sub-band trigger 8: for k = 0,k++,k<N bands N sweeps do 9: l kmodN bands 10: s kdivN bands 11: while sub-band trigger not received do 12: Wait 13: end while 14: Wait for 120s 15: Channel sounding for sub-bandl in sweeps 16: end for 17: Store the data 18: Log GPS location and time 19: Stop counter 20: i i + 1 21: end while 22: end procedure 168 C o u n te r Start trigger Signal to count 10 MHz 1 PPS Sweep trigger Synthesizer, ADC, AWG triggers N1: 60000 N2: 2000 1/N 1 1/N 2 Figure 7.5: Operation of the timing module to trigger AWG (TX) or ADC (RX) and frequency synthesizers Figure 7.6: The frequency plan for the multi-band measurements, overlap of the adjacent sub-bands are used to estimate the discontinuities of the frequency and phase responses. since both the TX and the RX counters start with the 1 PPS signals provided by the frequency references, the triggers at the TX and the RX are well-aligned. The NI DAQ timing module has an internal base clock of 100 MHz, which limits the maximum offset between the TX and the RX to less than 10 ns. This process is repeated for all sub-bands and the acquired data and the metadata (including GPS location, time, ADC gain etc) are transferred from the field programmable gate array (FPGA) of the RX into the permanent storage in 1-second chunks. Since the frequency synthesizer is basically a phase locked loop (PLL), every time it switches into a new carrier frequency it introduces a random phase offset relative to the previous carrier. Moreover, the triggering uncertainties of the AWG and ADC add additional phase offsets between the TX and the RX. The phase offsets can be estimated and corrected if two adjacent sub-bands have overlapping frequency tones. Thus, the frequency plan for the 169 Table 7.2: Channel sounder specifications, providing the values used throughout the chapter. Note, however, that the values can be modified on a per-campaign basis. Hardware Specifications Frequency range 3-18 GHz Instantaneous bandwidth 1 GHz Carrier spacing 500 MHz Frequency switching speed 100s TX power 40 dBm RX noise figure 5 dB ADC/AWG resolution 10/15-bit Data streaming speed 700MBps Sounding Waveform Specifications Waveform duration 2s Repetition per band 20 Number of tones per band 2001 Tone spacing 500 kHz PAPR 0.4 dB Total sweep time 6 ms Total number of tones 30001 Delay resolution 66.667 ps (2 cm) Max delay spread 2s multi-band measurements is designed accordingly as shown in Fig. 7.6. More details about phase correction are given in Section 7.3.2. One of the main motivation of this setup was to investigate the frequency dependency of the channel statistics. Hence for a fair comparison, we ensured a similar dynamic range through-out the whole frequency range. To achieve this, the transmitting signal was pre-distorted to compensate the gain of the system response per band basis. The final dynamic range only varies1:5 dB due to the variations of the RX noise figure caused by the LNA noise figure. Table 7.2 summarizes the configuration of the channel sounder for the measurements presented in this work. The full sweep consists of 30001 tones with 500 kHz spacing over 170 15 GHz total bandwidth. This configuration provides a time resolution of 66.67 ps with a maximum measurable delay spread of 2 s. Hence the channel sounder is capable of distinguishing two multi-path components (MPC), if their run-lengths differ by more than 2 cm, and it can measure up to 600 m of maximum run-length for MPCs. However, thanks to the flexible design, almost all the parameters given in the table can be modified according to the goal of the particular channel sounding campaign. For example, the channel sounder can operate as low as 2 GHz. However, due to interference from WLAN and cellular networks, and the presence of licensed spectrum bands we limited our measurements to 3-18 GHz during this campaign. Note that with the addition of band-pass filters, the channel sounder can also operate like a generic time-domain channel sounder setup with 1 GHz bandwidth and a carrier frequency anywhere between 2 to 18 GHz without any modifications. 7.3 Calibration 7.3.1 Sub-band Estimation The frequency response of each sub-band k is estimated with a least squares approach as follows: ^ H k (f t ) = (H m;k (f t )=H cal;k (f t )) E fH ant;k (f t ;)g (7.2) whereH m;k (f) andH cal;k (f)are the measured channel response and system response for the k-th sub-band, respectively. H ant;k (f;) is the antenna response for thek-th sub-band at the azimuth angle 2 [0; 2). E fg is the mean taken over to average out the variations in the azimuth pattern. The system response,H cal;k (f), is measured with a thru connection (i.e., by connecting TX and RX RF ports directly) prior to the each to the each measurement 171 campaign. The least squares estimation is only performed on the tone frequencies, i.e.,f t 2 [f LO 500 MHz;f LO +500 MHz]. Since eachf t in the given frequency tone is occupied with a sounding tone and the calibration responseH cal;k (f) and the antenna responseH ant;k (f;) are measured with signal to noise ratios (SNR) better than 40 dB, the noise enhancement in the least squares estimation is not an issue. Even though the biconical antennas have a nominally omnidirectional pattern, there are some deviations from that ideal behavior, see Fig. 7.3. Thus, the measured frequency responses of antennas are averaged over all azimuth angles for single-antenna measurements, such as the ones presented here (for directional measurements, obtained, e.g., with an array, the actual directional patterns would be taken into account in the calibration and evaluation). Since the proposed setup utilizes zero-IF up and down conversion, there are two main imperfections of the IQ mixers that need to be dealt with; local oscillator (LO) leakage and IQ phase/amplitude imbalance. For the down-conversion mixer (RX side), the LO leakage is filtered out by the low pass filter of the ADC and does not affect the measurements. Con- versely in the up-conversion (TX side), if LO power is not properly managed, LO leakage might drive the power amplifier into saturation since the leaked LO will be in the same fre- quency range as the up-converted RF signal. Consequently, LO power and the power of the baseband sounding signal are adjusted on a per-band basis to ensure that the up-converted RF signal has always more power than the LO leakage and the total power is at least 2 dB less than the input 1 dB compression point of the power amplifier. At the RX end, the tones located in [f LO 2f;f LO + 2f] are simply ignored to avoid any misinterpretations due to LO leakage of the IQ mixers. The imbalances between I and Q channels are measured separately for the TX and the RX with the method suggested in [119] prior to the measurement campaigns. For both mixers, the deviation of the relative phase between I and Q channels from 90 degrees, and the mismatch 172 in the amplitudes, are estimated on a per sub-band base. Then for all measurements, the TX sounding waveform is digitally pre-distorted according to the obtained phase and amplitude correction values for each sub-band. The RX IQ imbalance correction is performed off-line as described in [120] before further processing. Prior to each measurement campaign, the accuracy of the IQ imbalance correction is validated by checking the sideband suppression ratio with IQ inputs to form a single-sideband modulated signal. Hence for this test, to include only the tones in the upper sideband of the LO, the sounding signal is modified as: m USSB (t) = N X n=1 e j(n2ft+n) (7.3) where f is the tone spacing, N is the number of tones and n is the phase of the tone n. Again the I and Q signals are real and imaginary parts of m USSB (t). Fig. 7.7 shows the spectrum of the received signal with and without IQ imbalance correction for a thru connection. The sideband suppression ratio, which is the ratio of the power in the desired sideband over to the suppressed sideband (in this case ratio of the upper sideband over lower sideband) is only 9 dB before correction, for a carrier frequency of 11.75 GHz (the band with the worst IQ imbalance initially). The applied IQ imbalance correction increases the side-band suppression ratio to more than 25 dB. 7.3.2 Stitching Multiple Bands In this section we investigate the measurement results for a thru connection to verify the calibration and stitching approach. To achieve a comparable SNR in different sub-bands, the power of the sounding signal and the LO is adjusted per sub-band. The same power levels are used in the calibration measurements as well. Fig. 7.8 shows the measured frequency responses (i.e.,H m;k ) for all sub-bands. The stitching of the multiple bands is based on the fact that the channel does not change during the time it takes to measure two adjacent bands 173 10.15 10.35 10.55 11.75 11.95 12.15 12.35 Frequency (GHz) -80 -70 -60 -50 -40 -30 -20 -10 0 10 20 Power Spectrum (dB) Before IQ imbalance correction After IQ imbalance correction Figure 7.7: Sideband suppression before and after IQ imbalance correction, the input of the IQ mixer modified to include only the tones in the upper sideband of the LO. The power observed on the left of LO are due to IQ imbalance. (in the calibration setup, using a cable connection as the channel, there is no change at all, but even for the later measurements of the wireless channel, phase changes due to the Doppler effect occur on a larger time scale than the frequency switching time). Thus, any differences between the transfer function at the same subcarrier frequency, when measured with two different (overlapping) bands, has to be due to the channel sounder response, which has to be compensated as described in more detail below. Fig. 7.8 also shows the stitched frequency response after calibration and IQ imbalance correction. Even though the overlapping tones in the adjacent sub-bands may look different in the initial measurements, once calibrated they are well aligned within 1 dB, and there are no significant discontinuities in the final frequency response. Hence there is no need for additional correction for the amplitude of the frequency response while stitching adjacent bands. In case of the phase response; due to the nature of the PLL in the frequency synthesizer, every time the synthesizer switches to a new carrier frequency, it does so with a random phase. Consequently, there is a random phase offset between consecutive sub-band measurements as 174 2 4 6 8 10 12 14 16 18 Frequency (GHz) -45 -40 -35 -30 -25 -20 -15 -10 -5 0 5 Frequency Response(dB) Figure 7.8: Measured sub-band frequency responses (colored lines) and calibrated/patched frequency response (black line) 2 4 6 8 10 12 14 16 18 20 Frequency (GHz) -200 -150 -100 -50 0 50 100 150 200 Phase Response(degrees) Figure 7.9: Measured sub-band phase responses (colored lines) and calibrated/patched fre- quency response (black line) 175 seen Fig. 7.9. The phase offsets change for every run of a full sweep, hence they need to be estimated and corrected to acquire the true combined phase response for 3 GHz to 18 GHz. The random phase offset from sub-bandk 1 to sub-bandk is calculated with a maximum likelihood estimator which is formulated as follows: k = 8 > > > < > > > : 0; for k = 1 \ fs X f=0 ^ H k1 (f) ^ H k (ff s ) ! ; for 2kN bands (7.4) where\ denotes the phase of a complex number andN bands is the number sub-bands. Finally, the stitched complex frequency response for all sub-bands is calculated by; H(f k +f) = ^ H k (f k +f)exp i k X n=1 n ! Where: 1kN bands ; f s =2<ff s =2; f s is the step size for the carrier frequencies; f k is thek-th carrier frequency: (7.5) Fig. 7.9 shows the phase response for the uncalibrated sub-band measurements along with the phase response of the full-sweep after performing calibration, IQ imbalance and phase corrections. Additionally Fig. 7.10 shows the power delay profile (PDP) for a thru connection with and without phase correction. The phase correction removes the undesired ripples caused by phase jumps in the frequency response. To test the temporal stability of the calibration, we repeated the measurement with the thru connection 100 times; the maximum deviation in power was 0.2 dB. Phase and gain offset corrections are implemented as part of the post-processing scripts and performed automatically for every single measurement point. 176 75 80 85 90 95 -70 -60 -50 -40 -30 -20 -10 0 10 Figure 7.10: PDP for the system response with and without phase correction - PDP is shifted on the delay axis for presentation purposes 7.4 System Performance 7.4.1 Dynamic Range and Measurable Path Loss Given the values in Table 7.1, the estimated noise figure for the RX is 5 dB. At any given time, the RX measures a sub-band with 1 GHz bandwidth. The resulting RX sensitivity is: RX sensitivity =174 dBm=Hz + 5 dB + 10log 10 (1e9) Hz =79dBm (7.6) With the 40 dBm transmitting power, the instantaneous measurable path loss for the channel sounder is 119 dB. Since we employ a RX waveform averaging factor of 20 (i.e., sounding 177 -100 -90 -80 -70 -60 -50 -40 -30 -20 -10 P in (dBm) -100 -90 -80 -70 -60 -50 -40 -30 -20 -10 P out (dBm) Ideal Curve Measured levels RX sensitivity RX P1dB Figure 7.11: Dynamic range of the RX, black dashed line indicates the RX sensitivity level, magenta line indicate the RX 1 dB compression point waveform is transmitted 20 times for each sub-band), we gain additional 13 dB resulting in a measurable path loss of 132 dB. Fig. 7.11 shows the dynamic range of the RX. P in is the input power to the RX RF connector and theP out is the estimated power from the recorded waveform averaged over the whole band. In parallel to the estimated RX sensitivity, noise power distorts the estimated power if the input power is below -79 dBm. For -80 dBm input power, the estimated power is -78.2 dBm. Since the 1 dB compression input power for the RX LNA is -26 dBm, the dynamic range of the RX is 53 dB. Note that during the measurement campaigns for the points with RX power more than -30 dBm, a variable attenuator 1 is placed before the RX LNA to ensure the linearity of the RX. For the points within the dynamic range of the RX, the root mean square error (RMSE) of the estimated power levels with respect to the input power is 0.4 dB. 1 A manual variable attenuator was used for this purpose, since the implementation of an automatic gain control amplifier which can operate over the 15 GHz band is challenging. 178 6 6.5 7 7.5 8 8.5 9 Frequency (GHz) -14 -12 -10 -8 -6 -4 -2 0 2 Frequency Response (dB) Measured Analytical Figure 7.12: The analytical and the measured frequency responses for the 2-path test channel 20 40 60 80 100 120 140 Delay (ns) -70 -60 -50 -40 -30 -20 -10 0 PDP (dB) Measured Analytical 20 30 40 -15 -10 -5 Figure 7.13: The analytical and the measured PDPs for the 2-path test channel, close-in shows the two paths created by using the coaxial components 179 7.4.2 2-path Coaxial Channel Validation Measurements In this section, we use a deterministic channel to test the channel sounder. A coaxial 2-path channel was created by using a power divider, delay line and a power combiner. The impulse response of the corresponding analytical channel ish() = 1 e j 1 ( 1 )+ 2 e j 2 ( 2 ), where() is the Dirac delta function and k , k and k are respectively the amplitude, phase and delay of thek’th path. The values of these parameters are given in Table 7.3. Path numberk Amplitude k Phase k Delay k 1 0.31 -2.7 radians 25.2 ns 2 0.23 1.2 radians 36.9 ns Table 7.3: Parameter values for the 2-path test channel Fig. 7.12 compares the frequency response from the analytical channel with the frequency response with the channel sounder. The RMSE for the measured frequency response with respect to the analytical channel is 0.89 dB. The estimated errors in the powers of the paths are 0.01 dB for the first path and 0.19 dB for the second path. Furthermore, in the measured PDP, there are additional peaks due to the reflections between the components in the test channel. However, the power of these reflections are more than 30 dB below the paths under investigation. 7.4.3 Over The Air (OTA) Validation Measurements As the final verification we compared the results from the channel sounder setup with those from a VNA i.e., measurements of the same wireless channel were taken with the channel sounder setup and the VNA. For a sample location, Fig. 7.14 shows the frequency responses obtained from VNA and the calibrated and stitched frequency response acquired with the proposed channel sounder. If we calculate the RMSE in dB-scale similar to the Section 7.4.2, the RMSE for the frequency response is 4.8 dB. However, this error is mainly dominated by 180 the larger errors caused by not-aligned fading dips (i.e., although measurements are taken in the same environment, slight changes in the TX-RX locations affect the frequency of a fading dip). Additionally, Fig. 7.15 presents the same comparison for power delay profiles. The same specular components can be observed in both responses, and the power of the line-of- sight (LOS) path differs only by 0.1 dB between two cases. The total powers estimated for the two measurement differ by 0.4 dB between two measurements. Consequently, the mea- surement acquired via the two methods are in good agreement, which validates the proposed channel sounder and calibration procedure. 7.4.4 Impact of Antenna Pattern and Multiple Input Multiple Output (MIMO) Extension As a single input single output (SISO) setup, the channel sounder is only capable of mea- suring the “radio channel”, i.e., the concatenation of the propagation channel with the an- tennas [37]. The frequency characteristics of that radio channel are thus impacted by the frequency dependence of the antennas as well as that of the propagation. Since we aim to characterize the propagation, the antenna patterns should be as independent of frequency as possible. We stress that a frequency dependence of the antenna gain is not problematic as long as it is the same for all directions, since it can be calibrated out from the overall mea- surements (as indeed done in our setup). However, if the gain emphasizes some directions at one frequency and others at a different frequency, the results become specific to this particu- lar antenna/channel combination, and are thus less useful. For this reason, we used bi-conical antennas whose pattern shape is fairly constant over frequency as can be see in Figures 7.3 and 7.4, though some residual impact remains, see also the examples in Ref. [114]. A complete elimination of the impact of the antennas would require measurements with a MIMO channel sounder (also known as double-directional), as this allows to determine 181 Figure 7.14: Comparison of the measured frequency responses for same environment with VNA and the proposed setup Figure 7.15: Comparison of the measured power delay profiles for same environment with VNA and the proposed setup 182 the directions of the MPCs at TX and RX. With the proper calibration of the antenna ar- ray responses and post-processing, it becomes possible to completely decouple the response of the propagation channel from the response of the antenna arrays [37] by using high- resolution parameter extraction algorithms such as space-alternating generalized expectation- maximization (SAGE) [35] or joint maximum likelihood estimation (RIMAX) [34]. Conse- quently, the resulting channel characteristics would solely depend on the characteristics of the channel. While the standard RIMAX implementation assumes frequency independence of the antenna patterns (i.e., only applicable to narrowband measurements in which the an- tenna pattern is not dependent on the frequency), a version of the algorithm extending it to the frequency-dependent (UWB) case is given in Ref. [121] for a single input multiple output (SIMO) case, and generalization to MIMO is relatively straightforward. Although the channel sounder presented here is a single input single output (SISO) setup, the same principle can be used to build a MIMO channel sounder by employing antenna arrays at the TX and the RX. Extension to a MIMO channel sounder could be achieved by connecting the output of the TX amplifier to an electronic switch that connects the signal sequentially to the different elements of the transmit array (and similarly at the receiver) [122]. However, hardware implementation of such a structure is beyond the scope of this work. Even with the MIMO extension, the designed channel sounder will still be able to operate in dynamic channels. Since most of the measurement duration is due to the frequency switching delay, the best way to scan all antenna pairs at all frequencies would be measuring all MIMO channels before switching to next carrier frequency. The resulting time spent for recording a MIMO snapshot would be: SweepTime = ((N T N R T SISO ) +T switch )N bands (7.7) whereN T andN R are the number of antennas at the TX and the RX, respectively. T SISO is 183 the time spent for sounding one TX-RX antenna pair,T switch is the buffer time for frequency switching and theN bands is the number of bands to be measure. In our current configurations; we setT SISO = 4s, T switch = 100s andN bands = 30. If we assume that we extend our current setup to a typical arrangement of such arrays with 8 antenna elements at the TX and the RX, the total time is approximately 10 ms which is five orders of magnitude faster than the state of the art method of combining a VNA with a virtual array [123], and an order of magnitude smaller than what a VNA would require even to measure a SISO channel. With uniform switching, the maximum measurable Doppler would be 83.33 Hz. How- ever, one can improve the measurement Doppler by using the irregular switching approach described in [124, 125]. Note that the IQ imbalance, phase discontinuity between adjacent frequency bands and the amplitude pre-distortion would be same for all antenna pairs in the same sub-band, so the post-processing steps would still follow Section 7.3. 7.5 Measurement Campaign In this section, we describe the test measurements we performed with the channel sounder to demonstrate both the usefulness of the channel sounder and the validity of operation in a real environment. The measurements were performed in the Grace Ford Salvatori Hall (GFS) on the University of Southern California campus. The measurements were repeated on two different floors, with slightly different floor plans as shown in Figs. 7.16 and 7.17. For each floor, there are one LOS and one non-line-of-sight (NLOS) TX locations. For both cases, the TX is stationary while the RX continuously moves on the same line along the corridor, at a speed of 0:2 m=s. Measurements were taken every 1 s and in each measurement, our setup records five snapshots of the channel impulse response(i.e five frequency sweeps covering 3 GHz to 18 GHz), with a time gap of 10 ms between successive snapshots. Both the TX and the RX antennas are placed at 1:5 m height, similar to a typical device 184 NLOS TX LOS TX RX Route Figure 7.16: Floor plan for 1st floor (each grid corresponds to 10m) NLOS TX LOS TX RX Route Figure 7.17: Floor plan for 2nd floor (each grid corresponds to 10m) 185 to device (D2D) scenario. In case of LOS, the TX-RX distances range from 5 m to 53 m for the first floor and from 5 m to 70 m for the second floor. Similarly, in case of NLOS, the distances range from 9 m to 55 m for the first floor and from 9 m to 73 m for the second floor. Since the GPS-based location accuracy deteriorates indoor, we also marked checkpoints on the measurement route and time the RX movement to ensure a more accurate TX-RX distance determination. 7.5.1 Noise Averaging and Interference Filtering The multiple snapshots acquired within the same burst experience similar small-scale fading, since the maximum distance covered within 40 ms (i.e., maximum time gap between the first and the last snapshot) is less than half a wavelength for all frequencies. Hence these snapshots can be used for noise averaging. The multiple snapshots can also be used to detect and suppress the bursty 5 GHz WiFi interference, which was observed occasionally. Since the interference might be present only in a subset of the snapshots, using a pair wise correlation of the snapshot channel impulse responses, followed by median filtering, we can discard the snapshots corrupted by interfer- ence [44, 114]. The remaining snapshots are used for noise averaging. 7.5.2 Averaged Power Delay Profile (APDP) computation LetfH T (f k );k = 1N F g be the wideband channel frequency response measured at time T seconds (time is measured relative to the first measurement in that route), where f 1 = 3 GHz, f = f 2 f 1 = 0:5 MHz andN F = 30001. Then the wideband impulse response h T () is calculated by taking an inverse fast Fourier Transform (IFFT) with a Hann win- dow to suppress the side lobes. The magnitude squared of the impulse response gives the instantaneous power delay profile,PDP T (). 186 Along the route, consecutive measurements were taken 0:2 m apart, which corresponds to 2 spacing at 3 GHz carrier frequency and 12 at 18 GHz carrier frequency. Consequently, the successive measurements experience independent small-scale fading. Hence, the effects of the small-scale fading can be removed by averaging successive N PDP measurements, whereN is the number of consecutive measurements in which the MPC have similar powers with independent phase realizations.N is characterized by using the correlation between the instantaneous PDPs and the variation in the overall received power. Finally, the APDP is given by APDP () = 1 N k+N1 X T=k PDP T () (7.8) whereN , min (N 1 ;N 2 ). N 1 denotes the number of consecutive measurements over which the PDP’s are correlated (i.e., correlation is larger than 0.5 [126–128]) and N 2 denotes the number of consecutive measurements whose received power does not vary by more than 3 dB. N 1 = minfn :Corr (PDP k ();PDP k+n ())< 0:5g (7.9) N 2 = minfn :jP k P k+n j> 3dBg (7.10) where Corr(:;:) denotes the correlation coefficient and P T = 10 log 10 P N F k=1 jH T (f k )j 2 denotes the power in the channel frequency response. More details about the post-processing can be found in Ref. [114]. Fig. 7.18 shows all LOS APDPs measured on the the second floor. The first two MPCs with initial delays of 17 ns and 80 ns correspond to the LOS path and the reflection from wall behind the TX. As the RX moves away from the TX, the delays of these MPCs increase linearly with a slope of 3e8 meters per second as expected. Two other dominant MPCs with initial delays of 405 ns and 470 ns are caused by the reflections from the walls at the end of the corridor. Another MPC with initial delay of 535 ns is caused by the double reflection 187 Figure 7.18: APDP(dB) vs TX-RX distance for the second flor LOS measurements 188 from the wall behind the TX and the wall at the end of the corridor. Hence the delays of these three MPCs decrease linearly as the RX moves down the corridor. Next, we demonstrate the channel sounders capability of operating in dynamic environ- ments by investigating the spreading function observed for a time-varying measurement. Fig. 7.19 shows the spreading function along with the power delay profile and Doppler spectrum for the same measurement snapshot which corresponds to the TX-RX distance of 66 m in Fig. 7.18.. During this measurement, the TX is stationary while the RX is moving away from TX with a speed of 0.2 m/s. The dominant clusters are marked on the PDP in Fig. 7.19; clusters #1 and #3 are LOS path and the reflection from the wall behind the TX, hence they have positive Doppler. Due to the wide measurement bandwidth, the observed Doppler shifts vary from 2 Hz at 3 GHz to 12 Hz at 18 GHz. All other clusters (i.e #2, #4, #5 and #6) are reflections (single or double) from the end-wall and have the same amount of negative Doppler. Consequently, the Doppler spectrum shown on the left hand-side in Fig. 7.19 has two peaks at approximately8 Hz. Due to slow movement of the RX, the observed Doppler shifts are relatively small. However, with 15 GHz bandwidth, the channel sounder is capable of operating channels with Doppler shifts up to83.33 Hz. Larger Doppler (i.e., more dy- namic channels) can be measured by sacrificing the bandwidth, maximum measurable excess delay or measurement SNR. 7.5.3 Delay Spread The RMS-DS RMS is computed as the second central moment of the APDP [7]. RMS = s R 1 0 ( ) 2 APDP ()d R 1 0 APDP ()d (7.11) 189 Figure 7.19: Spreading (delay-Doppler) function, power delay profile and Doppler spectrum for the same measurement snapshot. Stationary TX and RX is moving away from TX with a speed of 0.2 m/s, TX-RX distance: 66.65 m. 190 where is the mean delay, which is given by = R 1 0 APDP ()d R 1 0 APDP ()d (7.12) To reduce the effects of noise, we remove the noise power from the APDP, by a noise- thresholding filter, in which the APDP samples whose magnitude are below a threshold are set to zero. The threshold is set to be 6 dB above the noise floor. The noise floor is computed from the noise-only region of the APDP (samples before the first MPC). Fig. 7.20 shows the LOS RMS-DS with respect to the TX-RX distance. The values we observe vary between 15 ns and 55 ns. For both floors, the RMS-DS increase with distance until 30 m, after which it decreases. After further inspection of PDPs, we justify this trend with the following observation that are related to the particular geometry of the environment. Initially, the direct path is much stronger than any other paths, resulting in a relatively smaller delay spread, see Fig. 7.18. As we move along the corridor, the other paths with larger excess delays are getting relatively stronger. Consequently, RMS delay spread slightly increases. Especially, due to its large excess delay, the emergence of the path reflected by the wall at the end of the route is the main contributor to the increase in the RMS DS values. The travel distance for the path reflected by the wall at the end decreases as we move towards that wall. Hence, this path is also getting absolutely stronger as the TX-RX distance increases and causes RMS-DS to increase as well. As we move even closer to this wall the excess delay of the reflected path relative to direct path gets smaller, hence the RMS delay spread gradually decreases again after 30m. For the NLOS case, the RMS-DS values are within 25 ns to 60 ns except two segments [25 m, 40 m] and [62 m, 67 m] on the second floor as shown in Fig. 7.21. Within these segments, the RMS-DS takes on values as high as 250 ns. When the RX is in one of these segments, both the NLOS TX and the RX have clear views to windows, allowing reflections 191 0 10 20 30 40 50 60 70 Distance(m) 15 20 25 30 35 40 45 50 55 60 RMS-DS(ns) 1st floor 2nd floor Figure 7.20: RMS-DS for LOS measurements versus TX-RX distance 0 10 20 30 40 50 60 70 80 Distance(m) 0 50 100 150 200 250 300 RMS-DS(ns) 1st floor 2nd floor Figure 7.21: RMS-DS for NLOS measurements versus TX-RX distance, between 26 m and 40 m, 62 m and 70 m, the RMS-DS is significantly higher due to additional paths from outdoor 192 200 400 600 800 1000 1200 Delay (ns) -160 -140 -120 PDP (dB) NLOS - 1st Floor 200 400 600 800 1000 1200 Delay (ns) -160 -140 -120 PDP (dB) NLOS - 2nd Floor Figure 7.22: 1st and 2nd floor NLOS APDP at 35 m, the MPCs with delays more than 500 ns are caused by the reflections from surrounding buildings. from surrounding buildings to arrive at the RX. An example APDP for this case in comparison to the first floor is shown in Fig. 7.22. The MPCs with an excess delay more than 500 ns on the second floor are due to these external MPCs. Note that the excess delays for these paths are consistent through-out the whole segment, and they are observed in all individual sub- bands from 3 GHz to 18 GHz, hence they are indeed MPCs and can not be caused by some kind of interference. Ref. [129] presents RMS-DS values observed in 13-17 GHz frequency band for two indoor environments. The non-parametric results (i.e., delay spread is calculated from the power delay profile and not from the extracted MPCs which are the output of SAGE algorithm) presented in the paper indicate RMS-DS values ranging from 8 ns to 35 ns. Since the measured TX-RX distances and the measurement environments are relatively smaller than in our case, the observed RMS-DS are also less than what we observed. Next, we investigate the frequency dependency of the RMS-DS by dividing the measured frequency responses into 1 GHz sub-bands and calculating the RMS-DS for each sub-band. Figs. 7.23 and 7.24 show the logarithm of the RMS-DS in seconds for all measurement points 193 4 6 8 10 12 14 16 18 Frequency(GHz) -8 -7.5 -7 -6.5 log 10 (DS/s) LOS - 1st Floor Mean 4 6 8 10 12 14 16 18 Frequency(GHz) -8 -7.5 -7 -6.5 log 10 (DS/s) LOS - 2nd Floor Mean Figure 7.23: LOS RMS-DS in logarithmic scale along with means versus the center frequency of sub-bands 4 6 8 10 12 14 16 18 Frequency(GHz) -8 -7 -6 log 10 (DS/s) LOS - 1st Floor Mean 4 6 8 10 12 14 16 18 Frequency(GHz) -8 -7 -6 log 10 (DS/s) LOS - 2nd Floor Mean Figure 7.24: NLOS RMS-DS in logarithmic scale along with means versus the center fre- quency of sub-bands 194 Table 7.4: Estimated parameters for the frequency dependent RMS-DS model given in Eq. 7.13 Location (%95 confidence interval) LOS-1st Floor -0.53 (-0.67, -0.39) -6.82 LOS-2nd Floor -0.04 (-0.17, 0.09) -7.41 NLOS-1st Floor -0.29 (-0.43, -0.16) -7.09 NLOS-2nd Floor -0.12 (-0.19, -0.04) -7.17 versus the center frequency of the sub-band. Following the the proposed model of the 3rd Generation Partnership Project (3GPP) [95], we model the frequency-dependent RMS-DS in logarithmic scales as follows; log(DS) (f c ) =log(1 +f c ) + (7.13) where log(DS) (f c ) is the mean of the logarithm of the RMS-DS (in seconds) at the center frequency off c in GHz. and are the model parameters to be estimated from the measured data. As seen in Table 7.4, for all scenarios the log(DS) decreases the frequency (i.e., is negative). However, for LOS-2nd Floor data the 95% confidence interval includes positive values for, indicating the decreasing trend can not be confirmed with 95% confidence level. In [108], similar models for the frequency dependency of the RMS-DS in several indoor environments were proposed. The most relevant campaign to our work is the EAB Office measurements performed at 2.4 GHz, 5.8 GHz, 14.8 GHz and 58.7 GHz center frequencies. The estimated’s are -0.07 and -0.01 for LOS and NLOS, respectively. As described in Section 7.2.2, the pre-distortion of the transmitting waveform provides a comparable dynamic range for all frequency sub-bands and thus allows a fair comparison unlike the measurements results provided by setups operating with different dynamic ranges at different carrier frequencies. 195 7.6 Conclusions In this chapter, we presented an UWB real-time channel sounder operating from 3 GHz to 18 GHz. Using a frequency-hopped sub-band approach, the channel sounder can measure the whole band in 6 ms without requiring high-speed digitizers and waveform generators. We discussed calibration and post-processing steps to combine sub-bands into a single 15 GHz band. The achieved RX sensitivity and the dynamic range of the channel sounder are -79 dBm and 53 dB, respectively. The channel sounder operation is tested with a deterministic coaxial channel, the observed RMSE of the estimated frequency response was 0.89 dB. The con- structed channel sounder and the post-processing techniques are also validated by comparing the measured impulse response with a VNA measurement. Since the TX and the RX do not require any physical connection, the setup is suitable for both indoor and outdoor channel sounding campaigns. Initial measurement results for indoor environment showed that the wideband RMS-DS varies from 15 ns to 55 ns for LOS, and 30 ns to 250 ns for NLOS mea- surements. Additionally, using 1 GHz sub-bands of the frequency response, we investigated the frequency dependency of the RMS-DS. We observed that the RMS-DS decreases with the frequency in all scenarios, and the trend is statistically significant for three out of four measurement scenarios. In the future, the channel sounder will be used to investigate the dependencies of the path loss exponents, shadow fading, RMS-DS, Ricean factor and coher- ence bandwidth on the frequency in the 3-18 GHz band for other environments. We also discussed the fundamental possibilities of extending the measurement principle to MIMO channel sounders. 196 Chapter 8 Conclusions and Future Directions We presented a novel mm-wave channel sounder that can perform double-directional mea- surements in dynamic environments. By employing a phased array, the channel sounder decreases the measurement time from minutes to milliseconds compared to rotating horn antenna approach for 90-degree sectors. Within these sectors, it can provide the temporal variations of the angular spectrum. Compared to the rotating horn antenna setups which can only measure either directional or dynamic channel properties at a given time, the channel sounder presented in this work can simultaneously estimate DoD, DoA, delay, and Doppler in a dynamic channel [4, 5]. With the constructed mm-wave channel sounder, we performed measurements investigating various aspects of the propagation channel at 28 GHz band. First, with the help of the unique capabilities of the channel sounder, we performed first double-directionally resolved measurement campaign at mm-wave frequencies for an out- door microcellular scenario [41]. We showed that even for stationary TX-RX scenarios (i.e. fixed wireless access), the adaptive beam-forming could increase the RX power by more than 10 dB. Furthermore we also showed that the delay spread and angular spread statistics change significantly in the existence of moving vehicles in the environment. This work establishes a framework which will enable developing spatio-temporal channel models describing time- 197 varying channel under the influence of human or vehicle movement in the channel. A RX upgrade to cover 360 in azimuth will allow continuous measurements with moving RX with full 360 view and enable measurements in highly mobile TX and/or RX such as vehicle to vehicle and vehicle to infrastructure scenarios. The findings of these studies and developed channel models, that jointly describe the directional and temporal channel characteristics, will be essential to evaluate beam-forming and tracking algorithms. Further measurements in the effects of human body blockage are already being performed by using our mm-wave channel sounder [94]. These measurements will provide further insights into effects of user own body and the surrounding other bodies on the time-varying propagation channel at 28 GHz. Second group of channel sounding campaigns were concerned with various aspects of the residential micro-cellular environment such as outdoor to outdoor channels, the foliage attenuation and outdoor to indoor channels. We first modelled the path-loss for outdoor to outdoor measurements [38]. Although the environment is not urban, we saw that the mean RMS-DS results are inline with the Urban micro-cellular model provided in [70]. We also investigated the effects of foliage blockage on the RX power, delay spread and angular spreads [130]. We observed that the foliage loss at 28 GHz follows the functional form of the ITU-R model, with parameters extracted from our measurements. Furthermore, we observed that the foliage shadowed links have significantly higher RMS-DS and DoA/DoD angular spreads compared to the LOS links in the same environments. As the penetration loss assumed to be one of the most challenging issues at higher fre- quency, we studied the outdoor to indoor penetration at 28 GHz [39, 131]. We observed that the O2I penetration losses vary from 10 dB to 23 dB depending not only on the building ma- terial but also on the surrounding environment. Furthermore, the penetration loss is 2-3 dB higher when the RX is forced to choose a single directional beam with 12 half power beam width for all cases. Perhaps most importantly, we find that the common approach of adding 198 a “bulk” penetration loss to an outdoor model is not a viable way to model outdoor-to-indoor mm-wave channels. Furthermore, the effects of O2I propagation on the DoA and delay spread statistics are more affected by the floor plan and the relative location of the building under investigation with respect to the surrounding structures rather than its building materials. Fur- ther investigations into the available beam-diversity showed that an adaptive beam-forming could improve the mean RX SNR by 5 to 7 dB in case the best sector is blocked by a human body. Finally, we provide model parameters for a suburban outdoor to indoor micro-cellular propagation channel which were not proposed by the current 3GPP standard [70]. The second setup presented in this thesis is a UWB real-time channel sounder which operates from 3 GHz to 18 GHz [132]. Using a frequency-hopped sub-band approach, the sounder can measure the whole band in less than 6 ms. The achieved RX sensitivity and the dynamic range of the channel sounder are -79 dBm and 53 dB, respectively. The channel sounder operation is tested with a deterministic coaxial channel, the observed RMSE of the estimated frequency response was 0.89 dB. The constructed channel sounder and the post- processing techniques are also validated by comparing the measured impulse response with a VNA measurement. In addition to the hardware design, we discussed calibration steps which allow compara- ble dynamic range for all subbands and post-processing method which combine sub-bands into a single frequency response with 15 GHz bandwidth. This allows a fair comparison of propagation channel parameters over the whole frequency range. Since TX and RX don’t require any physical connection, the channel sounder can operate in almost any desired mea- surement environment. Hence, we performed first-of-its-kind ultra-wideband macro/micro- cell propagation channel measurements with this channel sounder and investigated frequency dependency of the channel statistics [113,114]. Furthermore through our initial investigations of the indoor measurements, we observed that the RMS-DS decreases with the frequency in 199 all scenarios explored during our measurement campaign. Similar to the outdoor results, further investigations into frequency dependency of the channel parameters (e.g., path loss, shadow fading, delay spread, ricean factor and coherence bandwidth) for indoor environ- ments will be the next step. 200 Bibliography [1] C. U. Bas and S. C. Ergen, “Ultra-wideband channel model for intra-vehicular wireless sensor networks beneath the chassis: From statistical model to simulations,” IEEE Transactions on Vehicular Technology, vol. 62, no. 1, pp. 14–25, January 2013. [2] S. Ranvier, M. Kyro, K. Haneda, T. Mustonen, C. Icheln, and P. Vainikainen, “VNA- based wideband 60 GHz MIMO channel sounder with 3-D arrays,” in 2009 IEEE Radio and Wireless Symposium, Jan 2009, pp. 308–311. [3] C. Garcia-Pardo, J. M. Molina-Garcia-Pardo, J. V . Rodriguez, and L. Juan-Llacer, “Comparison between time and frequency domain mimo channel sounders,” in 2010 IEEE 72nd Vehicular Technology Conference - Fall, September 2010, pp. 1–5. [4] C. U. Bas, R. Wang, D. Psychoudakis, T. Henige, R. Monroe, J. Park, J. Zhang, and A. F. Molisch, “A real-time millimeter-wave phased array mimo channel sounder,” in 2017 IEEE 86th Vehicular Technology Conference (VTC-Fall), September 2017, pp. 1–6. [5] C. U. Bas, R. Wang, S. Sangodoyin, D. Psychoudakis, T. Henige, R. Monroe, J. Park, J. Zhang, and A. F. Molisch, “Real-time millimeter-wave MIMO channel sounder for dynamic directional measurements,” submitted. 201 [6] S. Rangan, T. S. Rappaport, and E. Erkip, “Millimeter-wave cellular wireless net- works: Potentials and challenges,” Proceedings of the IEEE, vol. 102, no. 3, pp. 366– 385, March 2014. [7] A. F. Molisch, Wireless communications, 2nd ed. IEEE Press - Wiley, 2010. [8] A. F. Molisch, A. Karttunen, R. Wang, C. U. Bas, S. Hur, J. Park, and J. Zhang, “Millimeter-wave channels in urban environments,” in 2016 10th European Conference on Antennas and Propagation (EuCAP), April 2016, pp. 1–5. [9] W. Roh, J. Y . Seol, J. Park, B. Lee, J. Lee, Y . Kim, J. Cho, K. Cheun, and F. Aryanfar, “Millimeter-wave beamforming as an enabling technology for 5G cellular communica- tions: theoretical feasibility and prototype results,” IEEE Communications Magazine, vol. 52, no. 2, pp. 106–113, February 2014. [10] N. Deparis, A. Bendjabballah, A. Boe, M. Fryziel, C. Loyez, L. Clavier, N. Rolland, and P. A. Rolland, “Transposition of a baseband uwb signal at 60 ghz for high data rate indoor wlan,” IEEE Microwave and Wireless Components Letters, vol. 15, no. 10, pp. 609–611, Oct 2005. [11] U. Demir and S. C. Ergen, “Arima-based time variation model for beneath the chassis UWB channel,” EURASIP J. Wireless Comm. and Networking, vol. 2016, p. 178, February 2016. [12] D. Cassioli, “60 ghz uwb channel measurement and model,” in 2012 IEEE International Conference on Ultra-Wideband, Sept 2012, pp. 145–149. [13] S. Ranvier, J. Kivinen, and P. Vainikainen, “Millimeter-wave mimo radio channel sounder,” IEEE Transactions on Instrumentation and Measurement, vol. 56, no. 3, pp. 1018–1024, June 2007. 202 [14] S. Piersanti, L. A. Annoni, and D. Cassioli, “Millimeter waves channel measurements and path loss models,” in 2012 IEEE International Conference on Communications (ICC), June 2012, pp. 4552–4556. [15] J. E. Hakegard, H. Rustad, I. Tardy, T. A. Myrvoll, and V . Ringset, “mm-wave channel sounding using a fully programmable soc,” in 2016 IEEE 27th Annual International Symposium on Personal, Indoor, and Mobile Radio Communications (PIMRC), Sept 2016, pp. 1–6. [16] A. Prokes, J. Vychodil, M. Pospisil, J. Blumenstein, T. Mikulasek, and A. Chandra, “Time-domain nonstationary intra-car channel measurement in 60 GHz band,” in 2016 International Conference on Advanced Technologies for Communications (ATC), Oct 2016, pp. 1–6. [17] J. M. Conrat, P. Pajusco, and J. Y . Thiriet, “A multibands wideband propagation channel sounder from 2 to 60 ghz,” in 2006 IEEE Instrumentation and Measurement Technology Conference Proceedings, April 2006, pp. 590–595. [18] M. Peter, W. Keusgen, and R. Felbecker, “Measurement and ray-tracing simulation of the 60 ghz indoor broadband channel: Model accuracy and parameterization,” in The Second European Conference on Antennas and Propagation, EuCAP 2007, Nov 2007, pp. 1–8. [19] P. F. M. Smulders and A. G. Wagemans, “Wide-band measurements of MM-wave in- door radio channels,” in [1992 Proceedings] The Third IEEE International Symposium on Personal, Indoor and Mobile Radio Communications, Oct 1992, pp. 329–333. 203 [20] C. Gustafson, F. Tufvesson, S. Wyne, K. Haneda, and A. F. Molisch, “Directional Analysis of Measured 60 GHz Indoor Radio Channels Using SAGE,” in 2011 IEEE 73rd Vehicular Technology Conference (VTC Spring), May 2011, pp. 1–5. [21] C. Gustafson and F. Tufvesson, “Characterization of 60 GHz shadowing by human bodies and simple phantoms,” in Antennas and Propagation (EUCAP), 2012 6th European Conference on, March 2012, pp. 473–477. [22] W. Fu, J. Hu, and S. Zhang, “Frequency-domain measurement of 60 GHz indoor channels: a measurement setup, literature data, and analysis,” IEEE Instrumentation Measurement Magazine, vol. 16, no. 2, pp. 34–40, April 2013. [23] T. S. Rappaport, F. Gutierrez, E. Ben-Dor, J. N. Murdock, Y . Qiao, and J. I. Tamir, “Broadband millimeter-wave propagation measurements and models using adaptive-beam antennas for outdoor urban cellular communications,” Antennas and Propagation, IEEE Transactions on, vol. 61, no. 4, pp. 1850–1859, April 2013. [24] T. S. Rappaport, G. R. Maccartney, M. K. Samimi, and S. Sun, “Wideband millimeter- wave propagation measurements and channel models for future wireless communi- cation system design,” Communications, IEEE Transactions on, vol. 63, no. 9, pp. 3029–3056, September 2015. [25] S. Hur, Y . J. Cho, J. Lee, N.-G. Kang, J. Park, and H. Benn, “Synchronous channel sounder using horn antenna and indoor measurements on 28 GHz,” in 2014 IEEE International Black Sea Conference on Communications and Networking (BlackSeaCom), May 2014, pp. 83–87. [26] M.-D. Kim, J. Liang, Y . K. Yoon, and J. H. Kim, “28 GHz path loss measurements in urban environments using wideband channel sounder,” in Antennas and Propagation 204 USNC/URSI National Radio Science Meeting, 2015 IEEE International Symposium on, July 2015, pp. 1798–1799. [27] G. R. MacCartney and T. S. Rappaport, “A flexible millimeter-wave channel sounder with absolute timing,” IEEE Journal on Selected Areas in Communications, vol. PP, no. 99, pp. 1–1, 2017. [28] P. B. Papazian, C. Gentile, K. A. Remley, J. Senic, and N. Golmie, “A radio chan- nel sounder for mobile millimeter-wave communications: System implementation and measurement assessment,” IEEE Transactions on Microwave Theory and Techniques, vol. 64, no. 9, pp. 2924–2932, Sept 2016. [29] R. Sun, P. B. Papazian, J. Senic, Y . Lo, J.-K. Choi, K. A. Remley, and C. Gentile, “Design and calibration of a double-directional 60 ghz channel sounder for multipath component tracking,” in Antennas and Propagation (EUCAP), 2017 11th European Conference on. IEEE, 2017, pp. 3336–3340. [30] S. Salous, “Multi-band multi-antenna chirp channel sounder for frequencies above 6 GHz,” in 2016 10th European Conference on Antennas and Propagation (EuCAP), April 2016, pp. 1–4. [31] S. Salous, S. M. Feeney, X. Raimundo, and A. A. Cheema, “Wideband MIMO Chan- nel Sounder for Radio Measurements in the 60 GHz Band,” IEEE Transactions on Wireless Communications, vol. 15, no. 4, pp. 2825–2832, April 2016. [32] 5G Channel Sounding, Reference Solution. [33] Z. Wen, H. Kong, Q. Wang, S. Li, X. Zhao, M. Wang, and S. Sun, “mmWave channel sounder based on COTS instruments for 5G and indoor channel measurement,” in 2016 205 IEEE Wireless Communications and Networking Conference Workshops (WCNCW), April 2016, pp. 37–43. [34] A. Richter, “Estimation of radio channel parameters: Models and algorithms,” Ph.D. dissertation, Techn. Univ. Ilmenau, Ilmenau, Germany, May 2005. [Online]. Available: http://www.db-thueringen.de. [35] B. H. Fleury, M. Tschudin, R. Heddergott, D. Dahlhaus, and K. I. Pedersen, “Channel parameter estimation in mobile radio environments using the SAGE algorithm,” IEEE Journal on selected areas in communications, vol. 17, no. 3, pp. 434–450, March 1999. [36] P. Almers, S. Wyne, F. Tufvesson, and A. F. Molisch, “Effect of random walk phase noise on MIMO measurements,” in Vehicular Technology Conference, 2005. VTC 2005-Spring. 2005 IEEE 61st, vol. 1. IEEE, May 2005, pp. 141–145. [37] M. Steinbauer, A. F. Molisch, and E. Bonek, “The double-directional radio channel,” Antennas and Propagation Magazine, IEEE, vol. 43, no. 4, pp. 51–63, Aug 2001. [38] C. U. Bas, R. Wang, S. Sangodoyin, S. Hur, K. Whang, J. Park, J. Zhang, and A. F. Molisch, “28 GHz Microcell Measurement Campaign for Residential Environment,” in GLOBECOM 2017 - 2017 IEEE Global Communications Conference, December 2017, pp. 1–6. [39] C. U. Bas, R. Wang, T. Choi, S. Hur, K. Whang, J. Park, J. Zhang, and A. F. Molisch, “Outdoor to Indoor Penetration Loss at 28 GHz for Fixed Wireless Access,” in Communications (ICC), 2018 IEEE International Conference on, 2018. [40] R. Wang, C. U. Bas, S. Sangodoyin, S. Hur, J. Park, J. Zhang, and A. F. Molisch, “Sta- tionarity region of mm-wave channel based on outdoor microcellular measurements 206 at 28 ghz,” in MILCOM 2017 - 2017 IEEE Military Communications Conference (MILCOM), October 2017, pp. 782–787. [41] C. U. Bas, R. Wang, S. Sangodoyin, S. Hur, K. Whang, J. Park, J. Zhang, and A. F. Molisch, “Dynamic Double Directional Propagation Channel Measurements at 28 GHz,” in 2018 IEEE Vehicular Technology Conference (VTC-Spring). [42] M. Friese, “Multitone signals with low crest factor,” IEEE Transactions on Communications, vol. 45, no. 10, pp. 1338–1344, October 1997. [43] D. Psychoudakis, H. Zhou, B. Biglarbegian, T. Henige, and F. Aryanfar, “Mobile sta- tion radio frequency unit for 5G communications at 28 GHz,” in 2016 IEEE MTT-S International Microwave Symposium (IMS), May 2016, pp. 1–3. [44] V . Kristem, S. Sangodoyin, C. U. Bas, M. Kske, J. Lee, C. Schneider, G. Sommerkorn, C. J. Zhang, R. S. Thom, and A. F. Molisch, “3D MIMO Outdoor-to-Indoor Propaga- tion Channel Measurement,” IEEE Transactions on Wireless Communications, vol. 16, no. 7, pp. 4600–4613, July 2017. [45] R. Wang and et.al., “Enabling super-resolution parameter estimation algorithm on mm- Wave MIMO channel sounder ,” to be submitted. [46] K. Haneda, S. L. H. Nguyen, J. JŁrvelŁinen, and J. Putkonen, “Estimating the omni-directional pathloss from directional channel sounding,” in 2016 10th European Conference on Antennas and Propagation (EuCAP), April 2016, pp. 1–5. [47] C. V . Forecast, “Cisco visual networking index: Global mobile data traffic forecast update, 2016–2021 white paper,” Cisco Public Information, 2017. 207 [48] P. Marinier, G. Y . Delisle, and C. L. Despins, “Temporal variations of the indoor wireless millimeter-wave channel,” IEEE Transactions on Antennas and Propagation, vol. 46, no. 6, pp. 928–934, Jun 1998. [49] G. R. MacCartney, T. S. Rappaport, and S. Rangan, “Rapid Fading Due to Human Blockage in Pedestrian Crowds at 5G Millimeter-Wave Frequencies,” in GLOBECOM 2017 - 2017 IEEE Global Communications Conference, Dec 2017, pp. 1–7. [50] S. Collonge, G. Zaharia, and G. E. Zein, “Influence of the human activity on wide-band characteristics of the 60 ghz indoor radio channel,” IEEE Transactions on Wireless Communications, vol. 3, no. 6, pp. 2396–2406, 2004. [51] J.-J. Park, J. Lee, J. Liang, K.-W. Kim, K.-C. Lee, and M.-D. Kim, “Millimeter Wave Vehicular Blockage Characteristics Based on 28 GHz Measurements,” in Vehicular Technology Conference, 2017. VTC 2017-Fall. IEEE, September 2017. [52] R. J. Weiler, M. Peter, W. Keusgen, A. Maltsev, I. Karls, A. Pudeyev, I. Bolotin, I. Siaud, and A.-M. Ulmer-Moll, “Quasi-deterministic millimeter-wave channel mod- els in miweba,” EURASIP Journal on Wireless Communications and Networking, vol. 2016, no. 1, p. 84, 2016. [53] R. J. Weiler, M. Peter, W. Keusgen, K. Sakaguchi, and F. Undi, “Environment induced shadowing of urban millimeter-wave access links,” IEEE Wireless Communications Letters, vol. 5, no. 4, pp. 440–443, Aug 2016. [54] V . Semkin, U. Virk, A. Karttunen, K. Haneda, and A. V . RŁisŁnen, “E-band prop- agation channel measurements in an urban street canyon,” in 2015 9th European Conference on Antennas and Propagation (EuCAP), May 2015, pp. 1–4. 208 [55] K. Sato, M. Fujise, R. Tachita, E. Hase, and T. Nose, “Propagation in rof road- vehicle communication system using millimeter wave,” in IVEC2001. Proceedings of the IEEE International Vehicle Electronics Conference 2001. IVEC 2001 (Cat. No.01EX522), 2001, pp. 131–135. [56] N. Al-Falahy and O. Y . Alani, “Technologies for 5G Networks: Challenges and Op- portunities,” IT Professional, vol. 19, no. 1, pp. 12–20, January 2017. [57] J. G. Andrews, S. Buzzi, W. Choi, S. V . Hanly, A. Lozano, A. C. Soong, and J. C. Zhang, “What will 5G be?” Selected Areas in Communications, IEEE Journal on, vol. 32, no. 6, pp. 1065–1082, 2014. [58] F. Boccardi, R. W. Heath, A. Lozano, T. L. Marzetta, and P. Popovski, “Five disruptive technology directions for 5G,” IEEE Communications Magazine, vol. 52, no. 2, pp. 74–80, February 2014. [59] F. C. Commission et al., “In the matter of use of spectrum bands above 24 GHz for mobile radio services,” Notice of Proposed Rulemaking in FCC, pp. 15–138, 2015. [60] G. R. MacCartney, M. K. Samimi, and T. S. Rappaport, “Omnidirectional path loss models in New York City at 28 GHz and 73 GHz,” in Personal, Indoor, and Mobile Radio Communication (PIMRC), 2014 IEEE 25th Annual International Symposium on. IEEE, 2014, pp. 227–231. [61] A. I. Sulyman, A. T. Nassar, M. K. Samimi, G. R. Maccartney, T. S. Rappaport, and A. Alsanie, “Radio propagation path loss models for 5G cellular networks in the 28 GHz and 38 GHz millimeter-wave bands,” IEEE Communications Magazine, vol. 52, no. 9, pp. 78–86, 2014. 209 [62] J.-J. Park, J. Liang, J. Lee, H.-K. Kwon, M.-D. Kim, and B. Park, “Millimeter- wave channel model parameters for urban microcellular environment based on 28 and 38 GHz measurements,” in Personal, Indoor, and Mobile Radio Communications (PIMRC), 2016 IEEE 27th Annual International Symposium on. IEEE, 2016, pp. 1–5. [63] S. Hur, S. Baek, B. Kim, J. Park, A. Molisch, K. Haneda, and M. Peter, “28 GHz channel modeling using 3D ray-tracing in urban environments,” in Antennas and Propagation (EuCAP), 2015 9th European Conference on, April 2015, pp. 1–5. [64] X. Zhao, S. Li, Q. Wang, M. Wang, S. Sun, and W. Hong, “Channel measurements, modeling, simulation and validation at 32 GHz in outdoor microcells for 5G radio systems,” IEEE Access, vol. 5, pp. 1062–1072, 2017. [65] P. B. Papazian, G. A. Hufford, R. J. Achatz, and R. Hoffman, “Study of the local multipoint distribution service radio channel,” IEEE Transactions on Broadcasting, vol. 43, no. 2, pp. 175–184, 1997. [66] Z. Muhi-Eldeen, L. P. Ivrissimtzis, and M. Al-Nuaimi, “Modelling and measure- ments of millimetre wavelength propagation in urban environments,” IET Microwaves, Antennas Propagation, vol. 4, no. 9, pp. 1300–1309, September 2010. [67] H. Xu, T. S. Rappaport, R. J. Boyle, and J. H. Schaffner, “Measurements and mod- els for 38-GHz point-to-multipoint radiowave propagation,” IEEE Journal on Selected Areas in Communications, vol. 18, no. 3, pp. 310–321, March 2000. [68] T. S. Rappaport, G. R. MacCartney, M. K. Samimi, and S. Sun, “Wideband Millimeter- Wave Propagation Measurements and Channel Models for Future Wireless Commu- 210 nication System Design,” IEEE Transactions on Communications, vol. 63, no. 9, pp. 3029–3056, September 2015. [69] F. J. Massey Jr, “The kolmogorov-smirnov test for goodness of fit,” Journal of the American statistical Association, vol. 46, no. 253, pp. 68–78, 1951. [70] “5G Channel Model for bands up to 100 GHz,” Aalto University, AT&T, BUPT, CMCC Ericsson Huawei, INTEL, KT Corporation, Nokia, NTT DOCOMO, New York University, Qualcomm, Samsung, University of Bristol, University of Southern Cali- fornia, Tech. Rep., 04 2016. [71] T. Rappaport, S. Sun, R. Mayzus, H. Zhao, Y . Azar, K. Wang, G. N. Wong, J. K. Schulz, M. Samimi, and F. Gutierrez, “Millimeter wave mobile communications for 5g cellular: It will work!” Access, IEEE, vol. 1, pp. 335–349, 2013. [72] K. Haneda, J. Zhang, L. Tan, G. Liu, Y . Zheng, H. Asplund, J. Li, Y . Wang, D. Steer, C. Li et al., “5g 3gpp-like channel models for outdoor urban microcellular and macro- cellular environments,” in Vehicular Technology Conference (VTC Spring), 2016 IEEE 83rd. IEEE, 2016, pp. 1–7. [73] F. K. Schwering, E. J. Violette, and R. H. Espeland, “Millimeter-wave propagation in vegetation: experiments and theory,” IEEE Transactions on Geoscience and Remote Sensing, vol. 26, no. 3, pp. 355–367, May 1988. [74] J. Ko, Y . S. Noh, Y . C. Kim, S. Hur, S. r. Yoon, D. Park, K. Whang, D. J. Park, and D. H. Cho, “28 GHz millimeter-wave measurements and models for signal attenuation in vegetated areas,” in 2017 11th European Conference on Antennas and Propagation (EUCAP), March 2017, pp. 1808–1812. [75] Rec. ITU-R P.833-8, “Attenuation in vegetation,” September 2013. 211 [76] M. O. Al-Nuaimi and R. B. L. Stephens, “Measurements and prediction model optimi- sation for signal attenuation in vegetation media at centimetre wave frequencies,” IEE Proceedings - Microwaves, Antennas and Propagation, vol. 145, no. 3, pp. 201–206, June 1998. [77] I. Rodriguez, R. Abreu, E. P. L. Almeida, M. Lauridsen, A. Loureiro, and P. Mogensen, “24 GHz cmwave radio propagation through vegetation: Suburban tree clutter atten- uation,” in 2016 10th European Conference on Antennas and Propagation (EuCAP), April 2016, pp. 1–5. [78] H. M. Rahim, C. Y . Leow, T. A. Rahman, A. Arsad, and M. A. Malek, “Foliage at- tenuation measurement at millimeter wave frequencies in tropical vegetation,” in 2017 IEEE 13th Malaysia International Conference on Communications (MICC), Nov 2017, pp. 241–246. [79] A. F. Molisch, Wireless communications. John Wiley & Sons, 2012, vol. 34. [80] B. H. Fleury, “First- and second-order characterization of direction dispersion and space selectivity in the radio channel,” IEEE Transactions on Information Theory, vol. 46, no. 6, pp. 2027–2044, September 2000. [81] Z. Pi, J. Choi, and R. Heath, “Millimeter-wave gigabit broadband evolution toward 5g: fixed access and backhaul,” IEEE Communications Magazine, vol. 54, no. 4, pp. 138–144, 2016. [82] J. Wells, “Faster than fiber: The future of multi-g/s wireless,” IEEE microwave magazine, vol. 10, no. 3, 2009. 212 [83] S. Hur, Y .-J. Cho, T. Kim, J. Park, A. Molisch, K. Haneda, and M. Peter, “Wideband spatial channel model in an urban cellular environments at 28 GHz,” in Antennas and Propagation (EuCAP), 2015 9th European Conference on, April 2015, pp. 1–5. [84] K. Haneda, J. J¨ arvel¨ ainen, A. Karttunen, M. Kyr¨ o, and J. Putkonen, “A statistical spatio-temporal radio channel model for large indoor environments at 60 and 70 GHz,” EEE Transactions on Antennas and Propagation, vol. 63, pp. 2694–2704, 2015. [85] J. Ko, K. Lee, Y . J. Cho, S. Oh, S. Hur, N. G. Kang, J. Park, D. J. Park, and D. H. Cho, “Feasibility study and spatial-temporal characteristics analysis for 28 GHz outdoor wireless channel modelling,” IET Communications, vol. 10, no. 17, pp. 2352–2362, 2016. [86] C. Larsson, F. Harrysson, B. E. Olsson, and J. E. Berg, “An outdoor-to-indoor prop- agation scenario at 28 GHz,” in The 8th European Conference on Antennas and Propagation (EuCAP 2014), April 2014, pp. 3301–3304. [87] J. Du, D. Chizhik, R. Feick, G. Castro, M. Rodrguez, and R. A. Valenzuela, “Suburban residential building penetration loss at 28 ghz for fixed wireless access,” 2018. [88] I. Rodriguez, H. C. Nguyen, I. Z. Kovcs, T. B. Srensen, and P. Mogensen, “An Empir- ical Outdoor-to-Indoor Path Loss Model From Below 6 GHz to cm-Wave Frequency Bands,” IEEE Antennas and Wireless Propagation Letters, vol. 16, pp. 1329–1332, 2017. [89] H. Zhao, R. Mayzus, S. Sun, M. Samimi, J. K. Schulz, Y . Azar, K. Wang, G. N. Wong, F. Gutierrez, and T. S. Rappaport, “28 GHz millimeter wave cellular communication measurements for reflection and penetration loss in and around buildings in New York 213 city,” in 2013 IEEE International Conference on Communications (ICC), June 2013, pp. 5163–5167. [90] N. Tran, T. Imai, and Y . Okumura, “Outdoor-to-indoor channel characteristics at 20 GHz,” in 2016 International Symposium on Antennas and Propagation (ISAP), Oct 2016, pp. 612–613. [91] T. Imai, K. Kitao, N. Tran, N. Omaki, Y . Okumura, and K. Nishimori, “Outdoor-to- Indoor path loss modeling for 0.8 to 37 GHz band,” in 2016 10th European Conference on Antennas and Propagation (EuCAP), April 2016, pp. 1–4. [92] C. A. L. Diakhate, J. M. Conrat, J. C. Cousin, and A. Sibille, “Millimeter-wave outdoor-to-indoor channel measurements at 3, 10, 17 and 60 GHz,” in 2017 11th European Conference on Antennas and Propagation (EUCAP), March 2017, pp. 1798– 1802. [93] X. Chen, L. Tian, P. Tang, and J. Zhang, “Modelling of human body shadowing based on 28 ghz indoor measurement results,” in 2016 IEEE 84th Vehicular Technology Conference (VTC-Fall), September 2016, pp. 1–5. [94] T. Choi, C. U. Bas, R. Wang, S. Hur, K. Whang, J. Park, J. Zhang, and A. F. Molisch, “Measurement Based Directional Modeling of Dynamic Human Body Shadowing at 28 GHz,” in submitted. [95] 3GPP, “5G; Study on channel model for frequencies from 0.5 to 100 GHz,” 3GPP TR 38.901 version 14.3.0 Release 14, January 2018. [96] J. Gozalvez, “Mobile traffic expected to grow more than 30-fold [mobile radio],” IEEE Vehicular Technology Magazine, vol. 6, no. 3, pp. 9–15, September 2011. 214 [97] A. F. Molisch, A. Karttunen, R. Wang, C. U. Bas, S. H. Hur, J. Park, and J. Zhang, “Millimeter-wave channels in urban environments,” to appear in EuCAP, 2016. [98] K. Haneda, “Channel models and beamforming at millimeter-wave frequency bands,” IEICE Transactions on Communications, vol. 98, no. 5, pp. 755–772, 2015. [99] J. Gozalvez, “Prestandard 5G Developments [Mobile Radio],” IEEE Vehicular Technology Magazine, vol. 9, no. 4, pp. 14–28, December 2014. [100] D. Cox, “910 mhz urban mobile radio propagation: Multipath characteristics in new york city,” IEEE Transactions on Communications, vol. 21, no. 11, pp. 1188–1194, November 1973. [101] V . Erceg, L. J. Greenstein, S. Y . Tjandra, S. R. Parkoff, A. Gupta, B. Kulic, A. A. Julius, and R. Bianchi, “An empirically based path loss model for wireless channels in suburban environments,” IEEE Journal on Selected Areas in Communications, vol. 17, no. 7, pp. 1205–1211, July 1999. [102] M. Toeltsch, J. Laurila, K. Kalliola, A. F. Molisch, P. Vainikainen, and E. Bonek, “Statistical characterization of urban spatial radio channels,” IEEE Journal on Selected Areas in Communications, vol. 20, no. 3, pp. 539–549, April 2002. [103] Z. Sun and I. F. Akyildiz, “Channel modeling and analysis for wireless networks in underground mines and road tunnels,” IEEE Transactions on Communications, vol. 58, no. 6, pp. 1758–1768, June 2010. [104] T. Rappaport, G. Maccartney, M. Samimi, and S. Sun, “Wideband millimeter-wave propagation measurements and channel models for future wireless communication system design,” Communications, IEEE Transactions on, vol. 63, no. 9, pp. 3029– 3056, September 2015. 215 [105] R. Mller, S. Hfner, D. Dupleich, R. S. Thom, G. Steinbck, J. Luo, E. Schulz, X. Lu, and G. Wang, “Simultaneous multi-band channel sounding at mm-Wave frequencies,” in 2016 10th European Conference on Antennas and Propagation (EuCAP), April 2016, pp. 1–5. [106] W. Fan, I. Carton, and G. F. Pedersen, “Comparative study of centimetric and millimet- ric propagation channels in indoor environments,” in 2016 10th European Conference on Antennas and Propagation (EuCAP), April 2016, pp. 1–5. [107] W. Fan, I. Llorente, J. Nielsen, K. Olesen, and G. Pedersen, “Measured wideband characteristics of indoor channels at centimetric and millimetric bands,” EURASIP Journal on Wireless Communications and Networking, vol. 2016, no. 58, 2016, special issue on Radio Channel models for higher frequency bands. [108] S. L. Nguyen, J. Medbo, M. Peter, A. Karttunen, K. Haneda, A. Bamba, R. D’Errico, N. Iqbal, C. Diakhate, and J.-M. Conrat, “On the Frequency Dependency of Radio Channel’s Delay Spread: Analyses and Findings From mmMAGIC Multi-frequency Channel Sounding,” arXiv preprint arXiv:1712.09435, 2017. [109] A. Al-Samman, T. Rahman, M. Azmi, M. Hindia, I. Khan, and E. Hanafi, “Statistical modelling and characterization of experimental mm-wave indoor channels for future 5G wireless communication networks,” PloS one, vol. 11, no. 9, p. e0163034, 2016. [110] A. F. Molisch, “Ultra-Wide-Band Propagation Channels,” Proceedings of the IEEE, vol. 97, no. 2, pp. 353–371, Feb 2009. [111] P. Pagani, F. T. Talom, P. Pajusco, and B. Uguen, Ultra Wide Band Radio Propagation Channel. John Wiley & Sons, 2013. 216 [112] A. Dezfooliyan and A. M. Weiner, “Evaluation of Time Domain Propagation Mea- surements of UWB Systems Using Spread Spectrum Channel Sounding,” IEEE Transactions on Antennas and Propagation, vol. 60, no. 10, pp. 4855–4865, October 2012. [113] V . Kristem, C. U. Bas, R. Wang, and A. F. Molisch, “Outdoor Macro-Cellular Chan- nel Measurements and Modeling in the 3-18 GHz Band,” in 2017 IEEE Globecom Workshops (GC Wkshps), December 2017, pp. 1–7. [114] ——, “Outdoor Wideband Channel Measurements and Modeling in the 3-18 GHz Band,” IEEE Transactions on Wireless Communications, pp. 1–1, 2018. [115] T. Zwick, T. J. Beukema, and H. Nam, “Wideband channel sounder with measurements and model for the 60 GHz indoor radio channel,” IEEE Transactions on Vehicular Technology, vol. 54, no. 4, pp. 1266–1277, July 2005. [116] M.-S. Choi, G. Grosskopf, and D. Rohde, “A Novel Wideband Space-Time Channel Measurement Method at 60 GHz,” in 2005 IEEE 16th International Symposium on Personal, Indoor and Mobile Radio Communications, vol. 1, September 2005, pp. 584–588. [117] J. li, Y . Zhao, C. Tao, and B. Ai, “System Design and Calibration for Wideband Chan- nel Sounding With Multiple Frequency Bands,” IEEE Access, vol. 5, pp. 781–793, 2017. [118] P. Almers, S. Wyne, F. Tufvesson, and A. F. Molisch, “Effect of random walk phase noise on mimo measurements,” in 2005 IEEE 61st Vehicular Technology Conference, vol. 1, May 2005, pp. 141–145 V ol. 1. 217 [119] Tektronix, Application note: Baseband Response Characterization of I-Q Modulators, February 2014, no. 37W-29553-1. [120] R. Zetik, M. Kmec, J. Sachs, and R. S. Thom¨ a, “Real-time MIMO channel sounder for emulation of distributed ultrawideband systems,” International Journal of Antennas and Propagation, September 2014. [121] J. Salmi and A. F. Molisch, “Propagation Parameter Estimation, Modeling and Mea- surements for Ultrawideband MIMO Radar,” IEEE Transactions on Antennas and Propagation, vol. 59, no. 11, pp. 4257–4267, November 2011. [122] R. S. Thoma, D. Hampicke, A. Richter, G. Sommerkorn, A. Schneider, U. Trautwein, and W. Wirnitzer, “Identification of time-variant directional mobile radio channels,” IEEE Transactions on Instrumentation and Measurement, vol. 49, no. 2, pp. 357–364, April 2000. [123] S. Sangodoyin, V . Kristem, A. F. Molisch, R. He, F. Tufvesson, and H. M. Behairy, “Statistical modeling of ultrawideband mimo propagation channel in a warehouse en- vironment,” IEEE Transactions on Antennas and Propagation, vol. 64, no. 9, pp. 4049– 4063, September 2016. [124] R. Wang, O. Renaudin, C. U. Bas, S. Sangodoyin, and A. F. Molisch, “Antenna switching sequence design for channel sounding in a fast time-varying channel,” in International Conference on Communications. IEEE, 2018 (accepted). [125] ——, “On channel sounding with switched arrays in fast time-varying channels,” arXiv preprint arXiv:1805.08069, 2018. [126] K. Guan, B. Ai, Z. Zhong, C. F. Lpez, L. Zhang, C. Briso-Rodrguez, A. Hrovat, B. Zhang, R. He, and T. Tang, “Measurements and analysis of large-scale fading char- 218 acteristics in curved subway tunnels at 920 mhz, 2400 mhz, and 5705 mhz,” IEEE Transactions on Intelligent Transportation Systems, vol. 16, no. 5, pp. 2393–2405, Oct 2015. [127] 3GPP, “Spatial Channel Model for Multiple Input Multiple Output (MIMO) Simula- tions,,” 3GPP TR 25.996 version 6.1.0 Release 15, September 2003. [128] G. Senarath, “Multi-hop relay system evaluation methodology (channel model and performance metric),” http://ieee802. org/16/relay/docs/80216j-06 013r3. pdf. [129] C. Ling, X. Yin, H. Wang, and R. S. Thom, “Experimental Characterization and Multipath Cluster Modeling for 13-17 GHz Indoor Propagation Channels,” IEEE Transactions on Antennas and Propagation, vol. 65, no. 12, pp. 6549–6561, December 2017. [130] C. U. Bas, R. Wang, S. Sangodoyin, S. Hur, K. Whang, J. Park, J. Zhang, and A. F. Molisch, “28 GHz Foliage Propagation Channel Measurements,” in submitted. [131] C. U. Bas, R. Wang, T. Choi, S. Hur, K. Whang, J. Park, J. Zhang, and A. F. Molisch, “Outdoor to Indoor Penetration Propagation Channel Measurements at 28 GHz,” in submitted. [132] C. U. Bas, V . Kristem, R. Wang, and A. F. Molisch, “Real-time ultra-wideband fre- quency sweeping channel sounder for 3-18 GHz,” in MILCOM 2017 - 2017 IEEE Military Communications Conference (MILCOM), October 2017, pp. 775–781. 219

Abstract (if available)

Abstract An accurate model for the wireless propagation channel is imperative for designing and testing wireless systems. Propagation channel measurements, also known as channel sounding, is the most accurate way to acquire true characteristics of the wireless propagation channel. This dissertation presents two novel channel sounder designs millimeter-wave and ultra-wideband communications. Due to the unique propagation characteristics at those frequencies, accurate channel models are even more important for millimeter-wave bands. It is anticipated that most of the future millimeter-wave systems will utilize beam-forming antenna arrays to overcome the higher path loss that occurs at higher frequencies. Consequently, the angular spectrum and its temporal evolution are vital for the efficient design of such systems. Hence, we built a 28 GHz channel sounder which is capable of directionally-resolved measurements in dynamic environments. Unlike the common practice of using the rotating horn antennas to investigate the directional characteristics of millimeter-wave channels, the developed sounder is capable of performing horizontal and vertical beam steering with the help of phased arrays. With the fast beam-switching capability, the proposed sounder can perform measurements that are directionally resolved both at the transmitter and receiver in 1.44 milliseconds. This not only enables measurement of more transmitter-receiver locations for better statistical inference but also allows to perform directional analysis in dynamic environments. The short measurement time combined with the high phase stability limits the phase drift between transmitter and receiver, enabling phase-coherent sounding of all beam pairs even when transmitter and receiver have no cabled connection for synchronization. This ensures that the measurement data are suitable for high-resolution parameter extraction algorithms and other coherent processing techniques. Furthermore, we present results from measurement campaigns that investigate various aspects at the 28 GHz such as

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Signal processing for channel sounding: parameter estimation and calibration

PDF

Double-directional channel sounding for next generation wireless communications

PDF

mmWave dynamic channel measurements for localization and communications

PDF

Propagation channel characterization and interference mitigation strategies for ultrawideband systems

PDF

Multidimensional characterization of propagation channels for next-generation wireless and localization systems

PDF

Channel sounding for next-generation wireless communication systems

PDF

Hybrid beamforming for massive MIMO

PDF

Elements of next-generation wireless video systems: millimeter-wave and device-to-device algorithms

PDF

Silicon-based wideband & mm-wave power amplifier architectures and implementations

PDF

Wideband low phase-noise RF and mm-wave frequency generation

PDF

Frequency channelized receiver for ultra-wideband and serial-link systems

PDF

RF and mm-wave blocker-tolerant reconfigurable receiver front-ends

PDF

Silicon-based RF/mm-wave power amplifiers and transmitters for future energy efficient communication systems

PDF

Design and implementation of frequency channelized ultra-wide-band (UWB) transceivers

PDF

High power, highly efficient millimeter-wave switching power amplifiers for watt-level high-speed silicon transmitters

PDF

UWB polarization measurements in multipath channels

PDF

Design and analysis of large scale antenna systems

PDF

Channel state information feedback, prediction and scheduling for the downlink of MIMO-OFDM wireless systems

PDF

optical communication systems and space-time wave packet generation using frequency combs and spatial modes

PDF

Optical and mm-wave orbital-angular-momentum beam generation and its applications in communications

Asset Metadata

Creator Bas, Celalettin Umit (author)

Core Title Real-time channel sounder designs for millimeter-wave and ultra-wideband communications

School Viterbi School of Engineering

Degree Doctor of Philosophy

Degree Program Electrical Engineering

Publication Date 08/07/2018

Defense Date 06/08/2018

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag channel sounder,channel sounding,millimeter-wave channels,OAI-PMH Harvest,ultra-wideband channels,wireless propagation channel

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Molisch, Andreas F. (committee chair), Golubchik, Leana (committee member), Moghaddam, Mahta (committee member)

Creator Email cbas@usc.edu,cumitbas@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-56181

Unique identifier UC11672484

Identifier etd-BasCelalet-6677.pdf (filename),usctheses-c89-56181 (legacy record id)

Legacy Identifier etd-BasCelalet-6677.pdf

Dmrecord 56181

Document Type Dissertation

Format application/pdf (imt)

Rights Bas, Celalettin Umit

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA