University of Southern California Dissertations and Theses

mmWave dynamic channel measurements for localization and communications

(USC Thesis Other)

mmWave dynamic channel measurements for localization and communications

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content MM-WAVE DYNAMIC CHANNEL MEASUREMENTS FOR LOCALIZATION AND
COMMUNICATIONS
by
Guillermo Andres Castro Carranza
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)
December 2024
Copyright 2024 Guillermo Andres Castro Carranza

Dedication
A los locos, a los alienados, y a los reprimidos del mundo.
- Luis Alberto Spinetta, 1984
ii

Acknowledgements
In short, I need to thank everyone who has the supportive network of people that have been part of my
journey through the last five years of my life. If any merit is mine, then it also belongs to them. No man
is an island.
Specifically, I want to start by giving thanks to Prof. Andreas F. Molisch for his great scientific advice.
My gratitude also extends to the people participating as my Qualification Exam and Defense Committee:
Prof. Constantine Sideris, Prof. Ramesh Govindan, Prof. Alan Willner, and Prof. Mahta Moghaddam.
I’m also grateful for all the talks had with Dr. Hussein Hammoud, Dr. Naveed Abbasi, and Dr. Jorge
Gomez Ponce during my stay at WiDeS lab. Special thanks to Kelvin Arana and Yuning Zhang, for helping
me carry out endless nights of measurements and direct the measurement squad.
Additional thanks to the team at Extreme Waves Inc. in San Diego. The phased array prototypes they
built were fundamental to this work, also their never-stopping support.
Thanks to Prof. Rodolfo Feick and Dr. Mauricio Rodriguez from Chile who made me think a PhD was
possible.
To my parents Marta G. Carranza M. and Guillermo E. Castro G. for doing the right thing when my
younger self started the obsessive question for how everything around worked. The screwdrivers and
hammers, trips to Santiago Chile for parts, excessive confidence in my childhood electrical expertise (the
house didn’t burn!) and their eternal support. Finally, thanks to my partner Ana for having bigger-thaninfinite patience and supporting me through this chaotic time.
iii

Table of Contents
Dedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi
Chapter 1: Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Background and Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Propagation Mechanisms of Wireless Channels . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2.1 Free-space Attenuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2.2 Reflection and Transmission . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.3 Diffraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.4 Scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.5 Other Propagation Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Existing Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Chapter 2: mmWave Channel Sounder Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.1 System Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.1 Sounding Waveform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1.2 Phased-Array Antennas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1.3 Transmission and reception ends . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.1.4 Timing, Control, and Synchronization . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.1.4.1 Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.1.4.2 Timing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.1.4.3 Acquired data and pre-processing . . . . . . . . . . . . . . . . . . . . . . 27
2.2 Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.2.1 System Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.2.2 Array Tapering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.3 Verification Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.3.1 Data Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.3.2 Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.3.2.1 T1 - Flat Reflector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.3.2.2 T2 - Blocker . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
iv

Chapter 3: Fundamental Operational and Mechanical Challenges . . . . . . . . . . . . . . . . . . . 44
3.1 System Timing Accuracy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.1.1 Digitizer Temporal Drift and Jumps . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.1.1.1 Characterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.1.1.2 Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.1.2 Arbitrary Waveform Generator Jitter . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.1.2.1 Characterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.1.2.2 Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.2 Elevation Scanning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.2.1 Characterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.2.2 Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.3 Self Propelled Sounder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.4 Movement Tracking for Ground Truth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.4.1 Location Tracking Systems for Channel Sounding . . . . . . . . . . . . . . . . . . . 56
3.4.2 Location Tracking Solution for mmWave . . . . . . . . . . . . . . . . . . . . . . . . 58
3.4.3 Additional Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.5 Calibration Positioner . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.6 Control Signal Integrity and Overhead Reduction . . . . . . . . . . . . . . . . . . . . . . . 61
3.6.1 Signal Integrity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.6.2 Control Signal Overhead . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Chapter 4: mmWave Localization Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.1 Measurement Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.1.1 Scenario 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.1.2 Scenario 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.1.3 Scenario 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.2 Data and Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.2.1 Scenario 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.2.2 Scenario 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.2.3 Scenario 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
Chapter 5: mmWave Microcell Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.1 Measurement Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.1.1 Sounder Modifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.1.2 Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.2 Data and Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.2.1 PDP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.2.2 Path Gain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.2.3 APS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
Chapter 6: Conclusions and Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
v

List of Tables
1.1 millimeter-wave (mmWave) Channel Sounders on Existing Literature . . . . . . . . . . . . 7
2.1 Channel Sounder Equipment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2 Timing Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
vi

List of Figures
1.1 Representation of sphere over which energy distributes. [30] . . . . . . . . . . . . . . . . . 3
1.2 Reflection and transmission events. [30]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 (a) Diffraction event. [30]. (b) Scattering event. [30]. . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Atmospheric absorption rates versus frequency. [41]. . . . . . . . . . . . . . . . . . . . . . 6
2.1 Sounder diagram for spatially compact measurements. The spatially distributed version is
based on this same basic concept, with modifications described in Sections 2.1.3 and 2.1.4. 12
2.2 Example sounder deployment scenario showing the RX side cart in the foreground and
TX side cart in the background. Array boxes mounted on top of each cart. Extra single
panel TXs not visible. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3 Single polarization panels used in the TX unit. . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.4 Dual polarization panels used in the RX unit and as additional distributed TXs. . . . . . . . 17
2.5 Rasied cosine window with factors 6 and 12 dB. . . . . . . . . . . . . . . . . . . . . . . . . 18
2.6 Custom upconverter. Input signal is centered around 700 MHz, with 1 GHz bandwidth.
Output is centered around 3.5 GHz. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.7 Custom downconverter. Input signal is centered around 3.5 GHz, with 1 GHz bandwidth.
Output is centered around 700 MHz. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.8 Distributed anchor RF chain and RFoF distribution network. . . . . . . . . . . . . . . . . . 20
2.9 TCS structure and important timing signals. Note that the structure varies for the
distributed case. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.10 TCS synchronization diagram showing all relevant timing signals. . . . . . . . . . . . . . . 23
vii

2.11 MRK1 signal cycle for a single Transmitter (TX) beam. This cycle repeats once per TX
beam to form an Receiver (RX) beam cycle. “W" and “L" represent wait and loading
time-slots, respectively. Numbers in green represent the index of the active beam. The
MRK1 signal from Fig. 2.10 is shown to correlate time events. . . . . . . . . . . . . . . . . 24
2.12 Distributed sounder TCS TX side structure. . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.13 TCS synchronization diagram showing all relevant timing signals for the distributed case. 28
2.14 Sounder as a linear system during calibration. . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.15 Delay response for all beam combinations of TX panel A2 and RX panel 1. . . . . . . . . . 32
2.16 Tapered phase array radiation patterns for TX unit example panel. Single frequency
response for 28 GHz shown. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.17 TX Array example panel beamwidths and sidelobe performance. . . . . . . . . . . . . . . . 34
2.18 Measurement scenario Layout, top view. A3 and RX are mounted on self-propelled carts
(green-black rectangles). A4 and A2 are mounted on tripods (smaller black rectangles).
This scenario corresponds to the one shown in Fig. 2.2 . . . . . . . . . . . . . . . . . . . . 37
2.19 P DPomni(τ, t) and P DPomni(τ, t) - P DP g omni(τ ). (a), (b) and (c) show P DPomni(τ, t)
for TXs A2, A3 and A4, respectively. (d), (e) and (f) shown P DPomni(τ, t) - P DP g omni(τ )
for each TX. Trajectory is T1: reflector. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.20 APS for the reflector case, T1. All static MPCs are removed from the data and then only
MPCs associated to movement are kept. In this case we show only movement MPCs with
powers up to 5 dB lower than the maximum. Color values represent the power deviation
with respect to the static channel for the specific AoD/AoA combination. . . . . . . . . . . 40
2.21 P DPomni(τ, t) and Temporal evolution of power for diferent angular combinations. (a),
(b) and (c) show P DPomni(τ, t) for TXs A2, A3 and A4, respectively. (d), (e) and (f) shown
the received power evolution for selected directions. Trajectory is T2: blocker. . . . . . . . 41
2.22 APS for the blockage case, T2. All static MPCs are removed from the data and then only
MPCs associated to movement are kept. In this case we show only movement MPCs with
powers up to 7.5 dB greater than the deepest fade. Color values represent the power
deviation with respect to the static channel for the specific AoD/AoA combination. . . . . 43
3.1 Digitizer Jump Pattern. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.2 Temporal PDP showing delay bin movements due to AWG jitter. . . . . . . . . . . . . . . . 48
3.3 AWG Jitter as shown by static measurement. (a) shows received time domain waveform
for a single MIMO snapshot slot. (b) zoom in time to show apparent wave overlap. (c)
shows overlap discrepancy of 8 samples. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
viii

3.4 Sounding sequence repetition start location as a relative count of number of samples.(a)
shows results for Arbitrary Waveform Generator (AWG) trigger threshold set to 1.2 V. (b)
shows the same for trigger threshold level 1.4 V. . . . . . . . . . . . . . . . . . . . . . . . . 50
3.5 Measured beam pattern for scanning angle −10◦
in elevation and 0
◦
in azimuth. . . . . . . 53
3.6 Measured phase for each row of the RX array when the scanning angle is -10◦
. (a) Phases
leading to erroneous beam shape. (b) Corrected phases. . . . . . . . . . . . . . . . . . . . . 54
3.7 Beam shape after fixing phase intersection . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.8 3D rendering of cart design. Two units were built: one for the TX and one for RX. . . . . . 56
3.9 Intel RealSense T265 Camera. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.10 Custom Positioning Arm for Calibration. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.11 (a) Unbuffered signals for data and clock lines of SPI array control. Low slew rate leads
to high distortion and array does not work. (b) Buffered signals for data and clock lines
of SPI array control. Slew rate is high enough to enable 50 MHz SPI communications.
Enabled by custom PCB with careful design contraints. . . . . . . . . . . . . . . . . . . . . 62
4.1 Scenario 1 trajectory (in blue) and most significant MPCs. . . . . . . . . . . . . . . . . . . . 66
4.2 Scenario 2 trajectory (in blue) and most significant MPCs. . . . . . . . . . . . . . . . . . . . 67
4.3 Scenario 3 trajectory (in blue) and most significant MPCs. . . . . . . . . . . . . . . . . . . . 68
4.4 Time-variant PDPs for Scenario 1. The horizontal axis is time in seconds, the vertical
axis is delay in meters. Path gain intensity is shown as a color scale. (a) shows the
time-variant PDP for anchor A2. (b) shows the time-variant PDP for anchor A3. (c) shows
the time-variant PDP for anchor A4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.5 APS snapshots for Scenario 1. On each subplot, the horizontal axis corresponds to AoA
while the vertical axis is AoD. For a particular AoA/AoD combination, the path gain
is shown as a color-mapped intensity. The multiple plots correspond to multiple time
instants: (a) t = 0 s. (b) t = 8 s. (c) t = 16 s. (d) t = 24 s. (e) t = 32 s. (f) t = 40 s. (g)
t = 48 s. (h) t = 56 s. (i) t = 64 s. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.6 Time-variant PDPs for Scenario 2. The horizontal axis is time in seconds, the vertical
axis is delay in meters. Path gain intensity is shown as a colorscale. (a) shows the
time-variant PDP for anchor A2. (b) shows the time-variant PDP for anchor A3. (c) shows
the time-variant PDP for anchor A4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.7 Time-variant PDPs for Scenario 3. The horizontal axis is time in seconds, the vertical
axis is delay in meters. Path gain intensity is shown as a color scale. (a) shows the
time-variant PDP for anchor A2. (b) shows the time-variant PDP for anchor A3. (c) shows
the time-variant PDP for anchor A4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
ix

5.1 Marker cycle for microcell measurements. . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
5.2 Trajectories for the microcell measurement campaign. . . . . . . . . . . . . . . . . . . . . . 78
5.3 Example PDP and synthetic PDP omni. (a) is a PDP. (b) is a synthetic omnidirectional PDP. 79
5.4 Measured path gain for all routes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.5 APS snapshots for microcell measurement route 1. On each subplot, the horizontal
axis corresponds to azimuth AoA while the vertical axis is azimuth AoD. Rows of
subplots correspond to a fixed elevation on the transmit side. For a particular AoA/AoD
combination, the path gain is shown as a color-mapped intensity. The multiple plots
correspond to fixed RX elevations: (a) -30◦
. (b) -20◦
. (c) -10◦
s. (d) 0◦
. (e) 10◦
. (f) 20◦
. (g) 30◦
. 81
x

Abstract
The mmWave channel offers significant advantages for both communication and localization applications
of wireless systems. For this reason, the mmWave band is a key component of fifth generation(5G) and future sixth generation (6G) communication systems. The large bandwidth available at mmWave frequencies
provides the potential for ultra-high data rate communications; it also results in higher delay resolution
that greatly improves the accuracy of localization when compared against sub-6GHz systems. As for all
wireless communication and localization systems, also those operating in the mmWave band require deep
knowledge of the physical characteristics of the channel to ensure good performance. The channel characteristics create significant channel-sounding challenges in the case of mmWave, since the channel exhibits
fast fluctuations and losses can be higher than in sub-6GHz bands if antenna gain is not exploited. More
generally, there is a fundamental technological trade-off for existing sounders in the gain-speed-bandwidth
triangle. In this work, we developed a state-of-the-art double-directional phased array channel sounder
that has high EIRP, is real-time and can capture mmWave channel dynamics, and has a bandwidth of 1
GHz. With this, we have captured statistically significant amounts of channel data that can address the
problems of mmWave communications modelling for single link and cooperative multipoint systems, as
well as localization with and without multiple anchor nodes.
xi

Chapter 1
Introduction
1.1 Background and Motivation
Operation in the mmWave band offers significant advantages for both communication and localization
applications of wireless systems. For this reason, the mmWave band is a key component of Fifth generation
(5G) cellular systems and WiFi (802.11ad/ay), and will be part of Sixth generation (6G) communication
systems [38]. Specifically, the large bandwidth available at mmWave frequencies provides the potential
for ultra-high data rate communications; it also results in higher delay resolution that greatly improves the
accuracy of localization [49, 40] when compared against sub-6GHz systems. The directional nature of the
channels also opens up the possibility for communications and localization while reducing detectability
by unwanted third parties.
Actually realizing these potential advantages requires the use of antennas with large effective area to
compensate for higher isotopic pathloss. Given the unpredictable and time-varying locations of the User
Equipments (UEs), and thus fast-changing mmWave channels, these antennas need to be directional and
adaptive. This opens the door also to spatial multiplexing, which can further increase the sum communication rate over a service area. For localization applications, multiple anchor nodes with adaptive antennas
need to exchange information with the agent (UE) to be localized [30].
1

In all these situations, the wireless system design needs a complete empirical characterization of the
mmWave channels; in particular, since adaptive antennas are used, we require double-directional (i.e., directionally resolved at both link ends) channel characterization. Early measurements of double-directional
mmWave channels used mechanically rotated horn antennas. While valuable, this approach is only applicable if both scatterers as well as TX and RX are static. Measuring double-directional fast fading characteristics of channels where either one or both communication ends are moving, or where there are moving
scatterers or blockers, requires fast sounding techniques [30]. Furthermore, while traditional measurements with vector network analyzers (VNAs) can sweep sequentially through a large overall bandwidth
with small instantaneous bandwidth at each time, real-time measurements require large instantaneous
measurement bandwidth, and thus fast Analog-to-Digital (AD) and Digital-to-Analog (DA) conversion.
Finally, the fact that localization generally requires connection to multiple anchors simultaneously leads
to a requirement for distributed channel sounding.
1.2 Propagation Mechanisms of Wireless Channels
1.2.1 Free-space Attenuation
As waves travel through space the total radiated energy spreads over a increasingly bigger sphere. The
receiving antenna then captures some of that energy, a fraction of the total sphere surface (see Fig. 1.1).
The relationship between antenna gains, transmitted power, received power and distance is formulated in
Friis’ equation for transmission:
PRX = PT X + GT X + GRX + 20 log
λ
4πd
, (1.1)
2

Figure 1.1: Representation of sphere over which energy distributes. [30]
where PRX represents received power in dBm, PT X is the transmitted power in dBm, GT X is the transmitting antenna gain in dBi, GRX is the receiving antenna gain, λ is the wavelength and d is the distance
separating the transmitter from the receiver.
It is important to note that Friis’ equation is only valid in the far-field, for planar wavefronts. Far
field occurs at distances much greater than the Fraunhoffer distance dF =
2L2
a
λ
, with La being the largest
dimension of the antenna.
For the mmWave frequency bands the relationship between antenna effective area and antenna gain
has to be kept in mind. The gain of an antenna can be written as GRX =
4π
λ2 ARX, where ARX is the
effective antenna area. This is important because it tells us that the increased losses can be counteracted
with larger, directional antennas.
1.2.2 Reflection and Transmission
Reflection and Transmission are two phenomena that are modelled by Snell’s law. When electromagnetic
waves propagate through an impedance discontinuity boundary, part of the energy will travel through the
boundary (transmission) and part of it will generate a wave that goes back to the medium from which the
wave comes from (reflection). This is illustrated in Fig. 1.2.
3

Figure 1.2: Reflection and transmission events. [30].
Snell’s law establishes that the incidence and reflection angles are identical, while the transmission
angle is determined by the ratio of the squared roots of the dielectric constants of each medium. Reflection
are important for mmWave bands since the small wavelengths results in a lot of common objects to work
as significant mirrors. On the other hand, transmission losses are high for high frequency bands.
1.2.3 Diffraction
Diffraction occurs when the waves impinge on structures with finite dimensions, and are particularly affected by the edges of objects. Huygen’s principle shows that edges do not create sharp shadows, but
rather energy seems to "bend" around corners. The amount of bending will depend on the Fresnel parameter νF = −y
q
2
λx , where x and y are distances from the sharp edge. Fig. 1.3 (a) shows a diffraction
event.
1.2.4 Scattering
Rough surfaces will create an ensemble of reflection interactions that will spread the energy in all directions, not necessarily just those indicated by Snell’s law. Surface roughness is a critical parameter that is
not only hard to model, but also dependent on the frequency. The shorter the wavelength the bigger the
4

(a) (b)
Figure 1.3: (a) Diffraction event. [30]. (b) Scattering event. [30].
impact of these irregularities, since the key quantity is the amount of surface variation in units of wavelength. Fig. 1.3 (b) illustrates this mechanism. mmWave bands experience increased scattering at high
frequencies.
1.2.5 Other Propagation Mechanisms
Two other important propagation mechanisms important for mmWave channels in urban environments
are waveguiding and atmospheric absorption. Waveguiding is the phenomenon by which the power seems
to be "carried" by an enclosed geometrical structure. It is a well studied mechanism for cylindrical waveguides with closed boundaries, but it shows differences in typical urban environments. Such environments
usually exhibit waveguides made of lossy materials in addition to wall discontinuities. Surface roughness
and the presence of interacting objects inside the waveguides make the use of equations intractable. This
phenomenon has been reported at lower frequency bands in street canyons, corridors and mines, among
others.
Atmospheric absorption refers to the effect of airborne particles absorbing part of the wave energy.
This results in an increased rate of power loss versus traveled distance. mmWave frequencies are subject
to increased rates of atmospheric absorption.
5

Figure 1.4: Atmospheric absorption rates versus frequency. [41].
Fig. 1.4 shows the gaseous specific attenuation as a function of frequency. Note the two spikes occurring at mmWave band frequencies 28 GHz and 60 GHz.
1.3 Existing Literature
The mechanisms shown in the previous section will create multiple paths for the signal to go from the TX
ti the RX. The RX can receive waves coming directly from the TX (Line of Sight (LOS)) or from reflection,
diffraction and others. This defines the fundamental nature of wireless channel: multi-path propagation.
Moreover, different paths can have different lengths, arrive from different direction or have departed from
the TX in a particular direction. In order to completely describe the channel, the double-directional impulse
response needs to be measured.
h(t, τ, Ω, Φ) = X
L
l=1
ρlδ(Φ − Φl)δ(Ω − Ωl)δ(τ − τl) (1.2)
,where l is the path index, ρl
, τl are the complex gain and delay of the l
th path, Ωl
is the direction of
departure Ω = [ϕ, θ], where ϕ, θ are azimuth and elevation angles, and Φ is the direction of arrival Φ =
[ϕ, θ]
The process of measuring this response is referred to as double-directional channel sounding.
6

Table 1.1: mmWave Channel Sounders on Existing Literature
Parameter Our
Sounder
USC [5] NIST [7] AT&T [9,
10, 21]
NYU [42,
43]
NIST [8,
34]
Center Frequency (GHz) 28 27.85 28.5 28 60 60
Bandwidth (GHz) 1 0.4 2 1 1 1
TX EIRP (dBm) 51 57 30 30 N/A 36
Array Size (N panels) 8x8 (10) 8x2 (2) 8x8 (6) 16x16/8x8
(2/4)
12 elements
(2)
8x32 (6)
Channel Sweep Time (ms) 2.4 1.4 1.3 6.25 (1.16) 1 4.2 (0.26)
TX-RX Azimuth FoV (◦
) 360-360 90-90 180-360 100-360 90-90 180-360
TX-RX Elevation FoV (◦
) 60-60 60-60 50-50 60-60 - -
Timing Standard GPSRubidium
GPSRubidium
Rubidium GPSRubidium
GPSRubidium
Rubidium
Traditional double-directional sounding methods for mmWave channels are based on mechanically
steered horns [28, 37, 23, 26, 22, 13, 20] and virtual arrays [36, 29, 48, 31, 45, 24]. These sounding techniques
are limited only to static channels: in both methods a mechanical movement (rotational or translational)
needs to be achieved, e.g., via stepper motors. The achievable speed with such motors is severely limited,
so that measuring in different directions/locations at both link ends can take between tens of minutes and
hours, during which the channel should not change. Some faster horn-based channel sounders have been
developed for mmWave dynamic channel characterization, but they sacrifice bandwidth to improve their
measurement speed and only allow single-directional measurements [12, 33]. A novel-approach 60 GHz
channel sounder is presented in [17], where fast spinning reflectors are used to measure double-directional
channel responses. However, its MIMO snapshot time of 28 ms is still slower than the channel’s coherence
time and only fixed elevation measurements are possible. Switched arrays use an electronic switch to
select a subset of the elements of an actual physical array [44, 6, 7]. They can provide fast operation when
compared to mechanical rotation sounders, but differ from our solution as explained next.
A more recent approach, enabled by developments in mmWave antenna arrays is the use of switchedbeam channel sounders. These channel sounders rely on phased-array antennas to perform full (doubledirectional) angular channel sweeps at speeds comparable to the mmWave channel’s coherence time. It
7

is important to highlight that phased and switched array sounders are related. Switched sounders are a
popular alternative for channel sounding at mmWave frequencies since they can be faster than horn-based
sounders and provide high Equivalent Isotropic Radiated Power (EIRP) (just like the phase-array sounder
at the center of this thesis). However, they exhibit some differences in the Signal-to-Noise Ratio (SNR)
and/or power management aspects when compared to array-based sounders. Generally, the power gains
for the phased arrays arise from two factors: (i) beamforming gain, and (ii) time-continuous use of multiple
Power Amplifiers (PA). Beamforming gain arises from the constructive adding up of the radiation to/from
the different antenna elements. Phased-array and switched-array sounders achieve the same beamforming
gain. The gain due to multiple PAs depends on both the configuration, and the assumptions about the
output power. In our phased array, there is a separate PA for each antenna element. In the interest of
fairness, we thus compare with a switched sounder in which there is also one PA per antenna element.
The key difference is now that in the phased-array sounder, each PA is on continuously, resulting in a total
conducted instantaneous power that is N (with N being the number of antenna elements and thus PAs)
times larger than that in a switched sounder, under the assumption that the maximum (peak) power of
the PAs is the same in both cases. This is a meaningful assumption if the peak output power is the key
limitation. For the case that the total conducted power is the limitation (e.g., because of thermal reasons,
or due to regulatory constraints), then the two sounder principles would result in the same SNR, though
in this case the phased array can achieve the same SNR with lower gain and maximum output power (and
thus lower cost) PAs.
Phased and switched array based sounders can achieve measurement acquisition times much faster
than mechanically rotating horn sounders. However, they are both affected by tradeoff between the acquisition time versus number of angular samples: the higher the number of angular directions being measured,
the higher the measurement time will be. When comparing alternatives, it is important to keep in mind
the number of measured angular directions.
8

Table 1.1 summarizes the available sounders for mmWave based on the switched-beam approach that
has been developed by various groups in the world. The WiDeS group at USC developed the first phasedarray sounder for the 28 GHz band [4, 5] channels. This sounder only covered a 90◦
sector on TX/RX sides
with limited resolution in elevation and has a bandwidth of 400 MHz. The authors in [7] presented a 28
GHz switched array sounder with a EIRP of 41 dBm covering 360◦
in the RX side, but only 180◦
at the
TX side. More recently, NYU and AT&T developed a 28 GHz phase array channel sounder for dynamic
measurements. This sounder has 30 dBm EIRP and completes a full Multiple Input Multiple Output (MIMO)
snapshot in 6.25 ms (can be reduced to 1.16 ms by reducing the number of beams). The TX side can cover
a 100◦
sector while the RX side can cover the entire 360◦
azimuth range.
Sounders have also been developed for the 60 GHz mmWave band. In [42, 43] a 60 GHz sounder is
presented. This sounder is based on an off-the-shelf array, the elevation angle is fixed and 90◦ of azimuth
are covered at the TX and RX sides. It can complete a beam sweep in less than 1 ms, but with an overall
period of 3 ms. NIST also developed a 60 GHz sounder [8, 34] with fixed elevation. This sounder has a
MIMO snapshot period of 4.2 ms (can be reduced to 262 µs by reducing the number of active array elements
and beams) and a TX that can cover 180◦
.
Finally, to the best of our knowledge, no phased-array sounders for simultaneous measurement of
multiple links (as relevant for both localization and cooperative multipoint (CoMP) communications) currently exists. The only multi-link measurement we are aware of is [27]); however, this is not real-time as
it is done with a rotating-horn sounder and furthermore measurements of the different links are done at
different times, moving the Base Station (BS) from one location to another.
9

Chapter 2
mmWave Channel Sounder Design
The fundamental contribution of this work consists of the development of a 28 GHz phased-array channel
sounder for dynamic double-directional channel measurements that can work in both spatially compact
and distributed setups, and provides a unique combination of Field-of-View (FoV), speed, power, and bandwidth. This sounder enables channel data acquisition for a wide range of mmWave use cases.
• The compact regime allows measuring with a single TX and RX, covering an azimuth FoV of 360◦
on both link ends. The elevation FoV is 60◦ on both ends. The MIMO snapshot can allocate up to
784 TX-RX beam combinations.
• Due to careful array panel switching and fast phased-array antenna control, the sounder can perform a full MIMO snapshot sweep in 2.4 ms. This allows tracking channel dynamics, in particular
movement of scatterers or moving link ends with speeds under 8 km/h if traditional uniform scanning is used. This limit could be further increased by carefully organizing the beam sequence (see
[47]).
• We design and implement an elaborate and effective timing and control circuitry that enables the fast
switching. Importantly, it only uses inexpensive (non-proprietary) off-the-shelf components. The
main building blocks of this circuitry are Microcontroller Unit (MCU)s and small Field Programmable
10

Gate Array (FPGA)s. Furthermore, the sounder acquires data without any temporal gaps enabling
complete analysis of time-domain phenomena.
• In the distributed regime, the sounder can have 3 TX locations distributed in space to enable triangulation of the RX for localization and the study of CoMP communication schemes. Up to 588 beam
combinations can be measured.
• The sounder bandwidth is 1 GHz, exceeding the maximum bandwidth foreseen in 5G for the 28 GHz
band. The EIRP is 50 dBm, which is significantly higher than that of most other existing sounders.
• The sounder is mounted on self-propelling carts with location tracking capabilities. It allows simultaneous movement of the RX and 1 TX end.
This chapter is organized as follows: Section 2.1 describes the system architecture in detail, Section
2.2 explains how the sounder operation was verified and calibrated, and Section 2.3 shows some example
measurements performed with the sounder.
2.1 System Architecture
Fig. 3.1 shows the basic structure of the channel sounder, more specifically the spatially compact sounder
setup. Its main components are described in Table 2.1. Periodic repetitions of a 1 GHz multi-tone sounding
waveform are generated by a large-bandwidth AWG, then upconverted to a 28 GHz center frequency using
two conversion stages at the TX, and transmitted via the phased arrays. At the RX end, the signal is received
with phased arrays, downconverted (again using two conversion stages), and the resulting waveform is
captured by a large-bandwidth digitizer without any gaps for a current maximum) of 8 s. More details will
be given in the following sections.
11

AWG
UPC IF
LO IF LO RF
DWC IF
DIGITIZER
TX ARRAY BOX
GPSDO
LO IFLO RF
RX ARRAY BOX
DISTRIBUTION AMP DISTRIBUTION AMP
Figure 2.1: Sounder diagram for spatially compact measurements. The spatially distributed version is based
on this same basic concept, with modifications described in Sections 2.1.3 and 2.1.4.
Table 2.1: Channel Sounder Equipment
Unit Name Description
AWG Tektronix
AWG7051
Arbitrary Waveform Generator, 5GSa/s, 1.5 GHz, 8 bit
LO IF Mercury
Frequency
Synthesizer
Frequency Synthesizer, +13 dBm, 0.1 - 20 GHz, low phase noise
LO RF NI Frequency
Synth
Frequency Synthesizer
UPC Custom
Upconverter
1 GHz bandwidth upconverter, output center frequency 3.5 GHz
DWC Custom
Downconverter
1 GHz bandwidth upconverter, output frequency 0.2 - 1.2 GHz
DIGITIZER Guzik
ADC6131
Digitizer, 13 GHz, 40 GSa/s, 8 bit
GPSDO Jackson Labs
LN Rubidium
GPS disciplined oscillator, low phase noise
DISTRIBUTION AMP SRS FS735 Distribution Amplifier, 10 MHz and CMOS amplifiers
12

2.1.1 Sounding Waveform
A baseband multi-tone waveform (MTW) is used as the sounding waveform. It can be mathematically
represented as:
m(t) = X
N
n=−N
e
j(n2π∆f t+θn)
(2.1)
where t is time, ∆f is the tone spacing in frequency, 2N + 1 is the number of tones and θn is the
phase of the n-th tone. MTW have been previously used in mmWave sounders (see, e.g., [11, 35]). The
specific MTW used here is a variant of the Zadoff-Chu waveforms used in LTE/NR and other sounders,
but it has the tone phases θn chosen to achieve a low Peak-to-Average Power Ratio (PAPR) even in the
presence of oversampling [14]. This means that the sounder’s amplifiers can be driven as close to their
1 dB compression point as possible. The sounding sequence is similar to the one used in [5]. Note that
optimizing the tone phases preserves the nice autocorrelation properties of the base Zadoff-Chu sequence,
which is important for our sounding application.
We chose ∆f to be 500 kHz for this sounder, thus N = 2000 (and thus unambiguous multipath range
of 600 m), and duration of 2 µs with a center frequency of 700 MHz, such that only a purely real (low-IF
bandpass signal) needs to be generated by the AWG. Our AWG allows flexibility in the sounding sequence
selection as long as it stays within the 1 GHz bandwidth. For example, the subcarrier spacing could be
reduced to the range of a few tens of kHz, to match those used in LTE/NR.
2.1.2 Phased-Array Antennas
The transmitting antennas are phased arrays with 8x8 vertically-polarized elements, as shown in Fig. 2.3.
A description of the specific single-polarization arrays used here is not available in the literature, but they
are identical to the ones described in [50] except for the number of elements. A similar array architecture
13

is used at the RX: two 4x8 arrays on a single board, one operating in vertical (V) polarization and a second array for horizontal (H) polarization, see Fig. 2.4. For further details of the dual-polarization panels
see [25]. For channel sounding, it is important to note that the V and H polarized arrays are about 3.5
wavelengths apart. This is may be farther than the (vertical) coherence distance for the mmWave channels we intend to deploy the sounder on. This means that the V and H panels will not experience the
same realization of the channel, i.e. they will not "see" the same small scale fading phenomena. However,
High-Resolution Parameter Extraction (HRPE) is not impacted by this offset, which would be simply incorporated in the data model. Each of the arrays could act as both TX and RX, with the single-polarized
arrays having 51 dBm EIRP, while the dual-polarization arrays have 43 dBm EIRP. However, our sounder
is operating always with the single-polarized arrays transmitting and the dual-polarized receiving. A full
polarimetric description of the channel requires measuring the channel’s vertical-to-vertical (V-V), verticalto-horizontal (V-H), horizontal-to-horizontal (H-H), and horizontal-to-vertical (H-V) responses. However,
from a statistical point of view, the H-V/H-H ratio is similar to the V-H/V-V ratio. This means that we can
get a good estimate of the co- and cross-polarized responses by measuring V-V and V-H. The choice of
the polarization arrangement of our sounder was motivated by: (i) it allows getting data to get estimates
for co- and cross-polarized channel responses (though possible differences between V-H and H-V strength
(or between V-V and H-H) cannot be detected), (ii) using dual-polarization only on one link end allows
faster completion of a full MIMO sweep, (iii) due to manufacturing reasons, the total number of antenna
elements was constrained, requiring a trade-off between dual-polarization capability and number of antenna elements per polarization. Thus, the single-polarized arrays provide higher TX power and (due to
the larger number of antennas for the given polarization) better elevation resolution, which is especially
useful for measurements at an emulated BS location.
The 3 dB bandwidth for both array types is 5 GHz, covering 24-29 GHz. The arrays have an azimuth
FoV covering ± 60◦
, and an elevation FoV covering ± 40◦
. Four arrays are on the RX unit, arranged in a
14

square formation so as to cover the full 360◦
azimuth. For the TX end, two configurations are employed:
in the concentrated mode, 4 panels are used, similar to the RX geometry, and thus obtaining full azimuth
coverage. In the distributed mode, only one TX panel is used at each TX location, resulting in 90◦
azimuth FoV in the TX side. The front of the TX unit is shown in Fig. 2.2. The TX and RX units have one
Intermediate-Frequency (IF) chain per array panel inside. This IF chain includes an upconverter on the TX
side and a downconverter on the RX side. The IF range is 3-9 GHz.
Two additional array panels, identical to the dual-polarization panels used in the RX side, are available
as additional TX units for the distributed sounder setup. These panels do not include an internal upconversion stage, so a custom upconversion stage was developed, as explained in Section 2.1.3. The decision
to use the dual-polarized panels as TX for the remote panels in the distributed setup was governed by the
availability of components. Additionally, distributing the RX panels would result in the TX unit becoming
the unit moved on the trajectory, which might expose the sounder operator (who has to move the mobile
cart) to potentially elevated levels of radiation. Note that the measurements for both the concentrated and
the distributed operation mode in Sec. 2.3 are done for V-V polarization only.
All array elements have an independent Radio-Frequency (RF) chain that includes Variable Gain Amplifier (VGA)s and a 6-bit phase shifter. These elements allow for sidelobe control through tapering and
beamforming. The phase shifter resolution is 5.625◦
. The TX unit has a maximum gain of 84 dB. The RX
unit has a maximum gain of 54.5 dB. The single dual-polarization panels used in the distributed setup have
41 dB maximum gain.
All phased array panels can be controlled through a serial interface. The single polarization panels
have a Quadruple Serial Peripheral Interface (QSPI) interface that runs at 45 MHz. The QSPI interface
can transmit 4 bits per communication clock cycle. A beam update requires 3 bytes per antenna element.
With 64 antennas this equals to 1536 bits, thus the beam update period is 8.5 µs. The dual-polarization
arrays have a Serial Peripheral Interface (SPI) interface, i.e. single data rate. This interface runs at 50 MHz.
15

Figure 2.2: Example sounder deployment scenario showing the RX side cart in the foreground and TX side
cart in the background. Array boxes mounted on top of each cart. Extra single panel TXs not visible.
A beam update event requires transmission of 3 bytes to each of the 32 elements, totalling 768 bits per
polarization, and so the beam update period for the dual polarization arrays is around 15.5 µs. Note that
this timing performance does not include intrinsic control interface overheads. These overheads will be
discussed in Section 2.1.4.
All array panels were tapered with a raised cosine window shown in Fig. 2.5 [19] to improve their
sidelobe level [3]. An αdB dB cosine window is defined by its weights wn:
wn = αsc + (1 − αsc) cos
πn
N − 1
−
π
2

(2.2)
16

Figure 2.3: Single polarization panels
used in the TX unit.
Figure 2.4: Dual polarization panels used in the RX unit
and as additional distributed TXs.
where αsc = 10−
αdB
20 is the nominal value of the cosine window, n ∈ [0, N − 1] is the weight index, and
N is the window size. Details about tapering and calibration will be discussed in Section 2.2.
2.1.3 Transmission and reception ends
A Tektronix AWG7051 AWG generates the previously described sounding sequence. The maximum AWG
bandwidth is 1.5 GHz, and its sampling rate is 5 GSa/s. When the waveform marker outputs are used, like
in this sounder, the sample depth is 8-bit. The total waveform memory is 64 MB. The AWG’s single ended
output is used, the output SNR is at least 40 dB. The AWG memory contains a pre-loaded sequence of
sounding waveform repetitions that is triggered once per full scan of all 28 beams covering all faces. This
triggering signal is managed by the Timing and Control and Synchronization (TCS) as will be described
below.
17

Figure 2.5: Rasied cosine window with factors 6 and 12 dB.
The output signal from the AWG is then upconverted using a custom upconversion stage shown in
Fig. 2.6. This stage takes the sounding waveform from 700 MHz to 3.5 GHz center frequency. The upconverter uses a single mixing stage followed by a bandpass filter for image and Local Oscillator (LO) leakage
rejection. The upconverter can handle up to 0 dBm, which is higher than the input power to the phased
arrays that leads to the maximum EIRP. A low-phase noise (-102 dBc/Hz at 100 Hz) frequency synthesizer
is used for this stage.
The IF at the receiving array unit is centered at 3.5 GHz. This waveform is further downconverted by
a custom downconversion stage shown in Fig. 2.7. This conversion stage centers the 1 GHz bandwidth
spectrum around 700 MHz before digitizing. The downconverter uses a single mixing stage that is bandpass
filtered before being fed into an amplifying stage. This is done to remove the image and limit the LO
leakage, as well as adapting the signal level to the digitizing stage. The OP1dB of the RX array box is
-2 dBm and the downconverter stage has a loss of 7 dB. Thus the maximum output before filtering is -9
dBm. Two baseband amplifiers are used to match the digitizer’s amplitude range. They take the received
sounding signal to a maximum of 14 dBm when the RX unit’s downconverter is saturated (limiting factor
for the received power).
18

RF
LO IF
Figure 2.6: Custom upconverter. Input signal is centered around 700 MHz, with 1 GHz bandwidth. Output
is centered around 3.5 GHz.
12 dB 12 dB
RF LO
IF
Figure 2.7: Custom downconverter. Input signal is centered around 3.5 GHz, with 1 GHz bandwidth.
Output is centered around 700 MHz.
19

3.5 GHz IF
LO
RoF link
Splitter
To other TXs SP3T
RoF link
x2
To other TXs
Figure 2.8: Distributed anchor RF chain and RFoF distribution network.
A Guzik ADC6131 digitizer is used to capture the downconverted measurement data without any gaps
at a sampling rate of 5 GSa/s. The digitizer’s sample depth is 8 bits, and its bandwidth is set to 2.5 GHz to
limit noise power (maximum bandwidth is 40 GHz). The digitizer acquires the downconverted spectrum
centered at 700 MHz and is triggered once per run by the 1 PPS signal from a GPS-Disciplined Oscillator (GPSDO) and then ignores this trigger while the acquisition is active to prevent phase jumps at the
triggering times. The digitizer captures data for a maximum of 8 seconds before needing to offload the
40 GB of data to long-term storage. The captured MIMO snapshots are then split in post-processing to
evaluate channel parameters.
Two additional dual polarization arrays are used in the sounder for distributed measurements. These
panels are used in TX mode to function as extra anchors (for localization) or BSs (for communication). To
minimize losses, the LO and sounding waveforms are transmitted through RF Over Fiber (RFoF) links from
the main UPC and LOs located in the TX unit cart. The LO signal is passed through a 3-way splitter to
generate the necessary copies. The sounding waveform is distributed using a fast RF switch to maximize
SNR and limit interference between TXs. IF and LO signals are distributed through same-length fibers
20

Init
Steer
10 MHz
TX
FPGA AWG TX ARRAY 10 MHz
10 MHz
1 PPS
AWG TRIG MRK1
SWITCH
FPGA SWITCH
10 MHz
1 PPS
10 MHz
1 PPS
MCU
SWITCH
FPGA SWITCH RX ARRAY
MCU
Init
Steer
TIME BASE
RX
BEAM TRIG
RST
BEAM TRIG
RST
SPI
QSPI
Figure 2.9: TCS structure and important timing signals. Note that the structure varies for the distributed
case.
to ensure timing consistency between all anchors. LO recovery, conditioning and the final upconversion
stage are performed at the end of the RFoF link, just before the array so cable losses are minimal. Fig.
2.8 shows this upconversion stage that takes the 3.5 GHz signal to 28 GHz and adjusts signal power to
maximize the transmitted power.
2.1.4 Timing, Control, and Synchronization
2.1.4.1 Structure
A custom-built TCS unit was developed for this channel sounder. The TCS is based on a distributed, multidevice structure that is disciplined by single GPSDO in order to maintain acceptable system jitters and
delays. Fig. 2.9 shows the general view of the TCS architecture for the TX and RX ends. Due to their
accurate and reproducible timing behavior, FPGAs are used to control timing and synchronization while
being supplied with 1 Pulse Per Second (1 PPS) and 10 MHz signals from a Rubidium clock that is disciplined
by GPS. A custom MCU+FPGA (“satellite") board is attached to every array for array control. The MCUs
21

have firmware implementations of the phase array register structure for QSPI/SPI control. MCUs exhibit
large jitters that make them incompatible with the design of a sounder of this complexity. Furthermore,
the jitter behavior is dependent on the number of tasks the MCU processing units is assigned with which
makes predicting their temporal accuracy very difficult. Due to this high temporal uncertainty, the MCUs
are triggered by their mating FPGA. This ensures strict timing for the sounding system, and reduces the
processing load on the MCUs (which lowers their jitters). MCUs were custom-coded at the register level
in order to remove overheads and approach nominal beam loading times for arrays. They achieve 25%
overhead and their jitter is around 20% the nominal control time. This MCU jitter has no impact on system
performance since the jitters are absorbed by dead times while the system performs other tasks.
The 10 MHz signal is distributed to each device and then converted into a TTL level squared signal
that is used as a reference clock by the FPGAs. This allows to align sounder triggers to a 100 ns grid with
jitters on the order of 100 ps. These jitters are primarily because of the variations in the 10 MHz signal and
its signal conditioning circuits. FPGAs have jitters on the order of a few 10s of ps, much lower than every
other device in the system. If finer triggering resolution is needed, high accuracy PLLs on board of FPGA
boards can be used to align the system to a 10 ns grid.
The MCUs implement an array initialization routine where all amplifiers and phase shifters are set
in order to operate the array for measurements. This routine also includes tapering if applicable, with
antenna element weights stored in MCU memory. After the initialization sequence, the MCU waits for a
trigger signal from its corresponding FPGA to cycle through the beams for each array. The beam sequence
is pre-loaded to the MCU memory before a measurement. After a triggering event happens, a full set of
phase shifter weights is transferred over the SPI/QSPI bus to the antenna panel(s).
The AWG waveform/marker triggering cycle covers the time for all TX beams and panels and all RX
panels during one fixed RX beam cycle. This means that for each AWG trigger, the AWG will transmit
16×7 waveform repetitions to cover the 16 TX RX combinations for 7 TX beams while generating 1 marker
22

A1 A2 A3 A4
110 000 100 010
WVFRM WVFRM WVFRM WVFRM
12-14 μs
A1 A2 A3 A4
110 000 100 010
WVFRM WVFRM WVFRM WVFRM
A1 A2 A3 A4
110 000 100 010
WVFRM WVFRM WVFRM WVFRM
12-14 μs 12-14 μs
A1 A2 A3 A4
110 000 100 010
WVFRM WVFRM WVFRM WVFRM
x4 x4 x4 x4 x4 x4 x4 x4 x4 x4 x4 x4 x4 x4 x4 x4 2.7
μ
s
TX
RX
1 PPS
AWG TRIG
MRK1
SWITCH
BEAM TRIG
QSPI
AWG WAVFRM
SWITCH
BEAM TRIG
SPI
10.8
μ
s
Figure 2.10: TCS synchronization diagram showing all relevant timing signals.
pulse signal every 4 waveform repetitions to synchronize the TX beam loading and panel switching. This
is shown by signals MRK1, SWITCH, AWG WVFRM on the TX side of Fig. 2.10. The MRK1 signal triggers
a secondary FPGA that is attached to the satellite board next to the antenna array. This secondary FPGA
is disciplined by its GPS-derived 10 MHz square clock. The FPGA triggers the MCU beam-steering and the
array box switches to select a specific array panel in the TX.
On the receiving end, the RX FPGA handles all timing events while being disciplined by 1 PPS and the
GPS-derived 10 MHz clock signal. This FPGA cycles the array box faces by means of a switch and sends
beam switching triggers to the RX MCU. This is shown by the SWITCH and BEAM_TRIG signals on RX
side of Fig. 2.10. The digitizer on reception is triggered by the 1 PPS signal and set to capture continuously
for the total length of the measurement up a maximum of 8 seconds. Careful timing verification was done
to ensure consistency of the captured MIMO snapshots.
23

Table 2.2: Timing Signals
Signal Description Purpose
1 PPS 1 pulse per second signal from GPSDO Provide accurate time base for sounder
AWG TRIG FPGA generated pulse Trigger AWG cycle
MRK1 Synchronization pulse generated by AWG FPGA and MCU disciplining for beam
switching
SWITCH TX and RX panel switch control 3 bits to select an array panel for transmission and reception
BEAM TRIG FPGA generated pulse Trigger MCU to transmit new beam information to array phase shifters
SPI Array control bus Used to set array register for RF parameters and beamforming phase shifters
AWG WAVEFRM Sounding waveform Waveform described in Section 2.1.1
Figure 2.11: MRK1 signal cycle for a single TX beam. This cycle repeats once per TX beam to form an RX
beam cycle. “W" and “L" represent wait and loading time-slots, respectively. Numbers in green represent
the index of the active beam. The MRK1 signal from Fig. 2.10 is shown to correlate time events.
24

2.1.4.2 Timing
The system allows to steer beams and panels on the TX and RX sides. In our current implementation, up
to 7 beams per panel for a maximum of 4 panels on TX and 4 panels on RX side can be steered, though
different numbers can be achieved with different MCU programs and associated FPGA setups. The two
fundamental switching ideas are:
• A beam can be loaded into any array box panel while a different panel is actively transmitting.
• TX side switching is faster than RX side switching. TX units are switched more often for this reason.
A description of all important timing signals is shown in Table 2.2. Fig. 2.11 shows the fundamental
repetition cycle for a single TX RX beam combination. This cycle is triggered on every MRK1 pulse. Rows
correspond to states of a single array panel, while columns represent one waveform repetition slot. Green
indicates that the panel is active, red means that it is being loaded with a beam, and yellow means that it
is idle. The cell number indicates the current beam loaded into a panel. The MRK1 cycle starts with TX
panel A1 being active during 4 waveform slots. During this time, RX cycles through all 4 faces and at the
same time TX panels A3 and A4 are being loaded with beam number 1. Note that TX panels A1 and A2
were loaded with beam one on the previous MRK1 cycle. Once all 4 RX panels have been active, TX panel
A2 becomes active and the RX cycles through all 4 panels again. At this point, 8 waveform slots have
been completed and TX panels A3 and A4 are ready to become active. The MRK1 cycle continues with TX
panel A3 becoming active for 4 waveform slots and RX cycling through all of its 4 panels. After this, TX
panel A4 becomes the active panel for 4 slots and the corresponding RX cycle happens. During the time
where TX panels A3 and A4 are active, TX panels A1 and A2 are being loaded with beam number 2 for
the next MRK1 cycle. The MRK1 cycle happens once per TX beam. Once all TX beams have been cycled
through, the RX is loaded with the next beam and the sequence of MRK1 cycles repeats again, until the
25

MRK1 sequence has been repeated once per RX beam. This completes one MIMO snapshot, covering all
TX RX beam/panel combination.
Following this sequence, the theoretical MIMO snapshot period is described by (2.3):
TMIMO = (Nb,TXNp,TXNp,RXTw + Tb,RX)Nb,RX , (2.3)
where TMIMO is the theoretical MIMO snapshot period; Nb,TX, Nb,RX are the number of beams per panel
on the TX side and RX side, respectively; Np,TX, Np,RX are the number of panels on the TX side and
RX side, respectively; Tb,RX is the beam loading time for the RX; and Tw is the duration of the sounding
waveform.
However, the switching of panels and beams includes guard times of 300 ns for the switches at the
beginning of each waveform slot, and 400 ns at the end.
Tw is thus 2.7 µs. There’s also guard times for TX side panel switching and SPI jitter. Fig. 2.10 shows
the real timing signals and their alignment. This diagram has been color-coded in agreement with Fig. 2.9.
This means that the real TMIMO is described by (2.4), which is based on (2.3):
TMIMO = (Nb,TXNp,TXNp,RX(Tw + Tg) + Tb,RX + Tj)Nb,RX , (2.4)
where the new terms Tg and Ts are the total guard time per waveform slot and total SPI jitter guard-time
per RX beam cycle.
Tw +Tg is 2.7 µs on the real system. Tb,RX is 37 µs. Tj
is 2.1 µs. The time per RX beam is thus 341.5 µs
and TMIMO = 2.3905 ms. The system is set to take 411 MIMO snapshots per second. This means that
a maximum Doppler shift of 411 Hz can be detected, which translates to 8 km/h speed of the cart when
operating at 28 GHz.
26

SW CTL
1 PPS
AWG
UPC IF
SW
UPC RF
10 MHz
RF Cable
Fiber
DB9 Cable
Samtec Cable
LO
A2, A3, A4
RoF link
A3
A4
A2
UPC RF
RoF link
RoF link
Splitter
Figure 2.12: Distributed sounder TCS TX side structure.
Two additional TCS satellite boards were designed for the additional anchors based on the 32 element
TX panels. The general structure of the distributed TCS, TX side, is shown in Fig. 2.12. The RX side TCS
stays the same as in the spatially compact case. The new timing signal alignment is shown in Fig. 2.13.
The additional anchors replace TX panels A2 and A4 from the non-distributed case. All timing signals are
distributed through same length cables/fibers to account for propagation delays. The additional anchors
have single SPI control interfaces. This makes them slower to steer, so their triggering happens while other
panels are active. This is shown by signals A2 SPI, A4 SPI on Fig. 2.13.
2.1.4.3 Acquired data and pre-processing
The recorded signal is a continuous data stream at 5GSa/s with a bandwidth of 2.5 GHz that lasts 8 seconds. The digitized signal shows a pattern of timing jumps that causes the acquired signal to shift in time
27

A1 A2 A3 A4
110 000 100 010
WVFRM WVFRM WVFRM WVFRM
A1 A2 A3 A4
110 000 100 010
WVFRM WVFRM WVFRM WVFRM
A1 A2 A3 A4
110 000 100 010
WVFRM WVFRM WVFRM WVFRM
A1 A2 A3 A4
110 000 100 010
WVFRM WVFRM WVFRM WVFRM
x4 x4 x4 x4 x4 x4 x4 x4 x4 x4 x4 x4 x4 x4 x4 x4 2.7
μ
s
TX
RX
1 PPS
AWG TRIG
MRK1
SWITCH
BEAM TRIG
A4 SPI
AWG WAVFRM
SWITCH
BEAM TRIG
SPI
10.8
μ
s
A3 QSPI
A2 SPI
Figure 2.13: TCS synchronization diagram showing all relevant timing signals for the distributed case.
approximately every 1 s. This pattern is stable over time; it is characterized before each measurement, and
the acquired data are corrected according to it in postprocessing.
A total of 3288 MIMO snapshots are recorded during the 8 s capture. After correcting the jump pattern,
the MIMO snapshots are parsed according to the switching scheme shown before to obtain up to 784 beam
combinations for each run. In the case of the distributed measurements only three TX panels were used,
which means that 588 beam combinations were measured.
The downconverted received signal of each beam combination is centered around 700 MHz, with 1 GHz
bandwidth. The FFT of this time-domain signal is frequency-filtered to isolate the bandwidth of interest,
from 200 MHz to 1.2 GHz. This data is impacted by the frequency response of both channel and sounder;
the latter is removed, for each beam combination, using the calibration that will be discussed in 2.2. This
finally provides the channel’s frequency response that is used for further analysis.
28

2.2 Calibration
The observed signal at the RX is the multiplication (in the frequency/angle domain) of the sounding signal
with the concatenation of the sounder (device) response and the propagation channel response. Thus, the
analysis of the channel requires to determine, and compensate for, the device response. Determination of
the device response is also known as calibration, and proceeds in several steps. The first step calibrates
the transfer function of the whole system, for transmission and reception at broadside and using a beam
that points into the direction of the broadside. In the second step, we measure the directional patterns of
the phased arrays separately at Tx and Rx, which can be assessed relative to the pattern value at broadside. The combination of the two calibrations then provides a full calibration of the whole system for all
directions. The details of this approach, which considerably simplifies overall calibration, are described in
the following.
2.2.1 System Response
To facilitate the calibration of the TX and RX boxes, our array panel design included a directional coupler
connected to each panel’s internal RF chain; the single panel TXs allow direct connection to the array panel
RF ports for this same purpose. These couplers allow a separate characterization of the RF chains and the
panels themselves, easing the challenges that arise when only IF ports are accessible for calibration (see
[46] for approaches to handle that situation). The single array panels and the array boxes are small enough
to allow anechoic chamber calibration. To obtain the full system response, the complete sounder (including
up/downconversion chains) needed to be measured for one fixed orientation, as discussed further below.
This characterization was done outdoors for two reasons: (i) the complete sounder is too large to fit onto
the anechoic chamber without damage to the absorbers there, and (ii) the distance is too short to allow
transmission at full power without damage to the RX boards.
29

HT1 HT2 HR2 HR1
HTC HRC
HC
Figure 2.14: Sounder as a linear system during calibration.
In the context of a measurement, the sounder and channel can be modelled as a linear system shown
in Fig. 2.14. For simplicity, we only show a single instance of this linear system (where the system is
using a single TX-RX panel-beam combination). Note that this frequency responses will change for each
TX and RX panel-beam combination, needing a total of Nb,TXNp,TXNp,RXNb,RX characterizations. In
this diagram HT1 includes all components from AWG to the input ports of the TX array panel, which is
identical to the output of the upconverter to 28 GHz (in the case of the TX box, this includes its internal
RF chains), HT2 includes all array panel components operating on the 28 GHz signal, namely beam former
and antenna elements, HTC is the TX coupler response, HC is the channel response, HR2 includes all
devices from array panel to the input of the 28 GHz downconverter, HR1 includes all devices from that
input to the digitizer, and HRC is the RX coupler response. The signal captured by the sounder RX, divided
by the sounding signal, provides a frequency response Hsounder = HT1HT2HCHR2HR1. The objective of
calibration is to find a way to isolate HC, the channel’s frequency response.
There are several beams per TX and RX panel, but due to the complexity of the outdoor calibration
measurements only one beam per panel (beam pointing at broadside) was measured, and only at a single
orientation of the panel, i.e., broadside. In other words, all panels were set to their broadside beams and
then one outdoor calibration measurement was made for all TX-RX panel combinations. To complete the
calibrations, the beams for each panel were measured in the anechoic chamber, as discussed below. The
anechoic chamber measurements include the broadside beam, in order to use it as a reference beam and
obtain the complete response of each panel-beam combination.
30

The impact of the RF chains can be computed by measuring twice in the same (greenfield) location.
The first measurement is with a Vector Network Analyzer (VNA) that is connected to the arrays’ coupler
ports, providing HVNA = HTCHT2HCHR2HRC. The second is with the “normal" sounder setup, providing Hsounder = HT1HT2HCHR2HR1 as discussed above. The ratio Hbroadside = Hsounder/HVNA thus
becomes HT1HR1
HTCHRC
for arrays when TX and RX boxes are used, and HT1HR1
HRC
for the case of the dual polarization panels used as TX, while the RX box is the same as above. Delay-gating [2] was used to ensure
that no reflections from objects in the environment (which violate the greenfield assumption) impacted
the calibration measurements.
All of the above serves to measure the complete system response when measuring at broadside (more
explicitly, the combination of every TX and every RX panel is measured at broadside). We next describe
the second step of the overall procedure, namely the measurement of the directional patterns for all beams.
The beam patterns were measured with a VNA, in an anechoic chamber using an automated antenna
pattern measurement system. Since the 28 GHz ports are directly accessible in those panels, we measured
Hdirectional = HTCHT2 for the TX arrays in the box, Hdirectional = HR2HRC for the RX arrays in the box,
and Hdirectional = HT2 for the single panels. In all of these cases, the to-be-measured panel was at one link
end, while a standard horn antenna with known gain was at the other. The product of the responses with
the outdoor calibration measurement results in Hcal = HbroadsideHdirectional = HT1HT2HR2HR1 for the
arrays in the boxes and also for the single panel cases. Note that dividing any Hsounder measurement by
Hcal results in HC, which is the frequency response of the channel.
The array panels’ longest dimension is 8.6 cm, thus their Fraunhofer distance is about 1.4 m. The
reference antenna was placed 3.5 m away from every panel in order to ensure far-field measurements. A
static standard gain horn was used as a reference antenna acting as a TX or RX and its frequency response
removed from the measured patterns. The data was also corrected for the free space channel response
inside the chamber. The phased array under test was rotated around its phase center for azimuth and
31

Figure 2.15: Delay response for all beam combinations of TX panel A2 and RX panel 1.
elevation cuts. The patterns were measured with an angular resolution of 1◦
. TX and RX boxes were
measured one panel at a time. Single panel TX were also measured in the same way. Careful alignment
(to establish the true broadside angle) of the antennas was done by using laser alignment tools.
We measured the patterns of 7 beams per panel. Specifically, we chose azimuth angles -45◦
,-30◦
, -15◦
,
0
◦
, 15◦
, 30◦
, 45◦
. A set of phase shifter weights to create a single beam was loaded onto the panel under test
and then the calibration system performed one full mechanical rotation. Then the next beam was loaded
and the cycle repeated. This was done for each beam on each panel.
Thus, a total of 70 beam patterns were measured in the 27-29 GHz frequency range with 1.25 MHz
carrier spacing. The VNA’s IF bandwidth was set to 1 kHz; the resulting SNR is 40 dB.
As an example, the IFFT of Hcal, i.e. the delay response of the sounder for all 49 beam combinations
of A2-RX1 are shown in Fig. 2.15.
32

Figure 2.16: Tapered phase array radiation patterns for TX unit example panel. Single frequency response
for 28 GHz shown.
2.2.2 Array Tapering
While sidelobes of the beampatterns are not necessarily detrimental when HRPE is used, they do make it
more difficult to interpret “raw" Fourier-based measurements. Thus we use tapering to reduce the sidelobe
level at the price of slightly increased main lobe width [3]. The TX unit panels were tapered using a 12
dB cosine window, while the RX and the dual polarization TX panels were tapered using a 6 dB window.
Multiple cosine windows were tested for each array to achieve good sidelobe level suppression without
broadening the main lobe beyond 18◦ on the steering range. As an example, the resulting tapered phased
array radiation patterns for a TX box example panel are shown in Fig. 2.16.
Beamwidths were analyzed for every beam. The left axis of Fig. 2.17 (dashed line) shows the 3-dB
beamwidths after tapering for each beam of a single panel. The widest beamwidth is 18 degrees at the
±45◦ beams. The beamwidths are less than 15◦
for angles in ±30◦
range. This reflects the fact that for a
planar array with uniformly space elements, the beamwidth increases approximately as sec (β) where β
is the beam pointing angle [3].
33

Figure 2.17: TX Array example panel beamwidths and sidelobe performance.
The sidelobe levels were also studied for every panel. The right axis of Fig. 2.17 (solid line) shows the
maximum sidelobe level for each beam of a single panel. Outer angles ±45◦
show at least 10 dB sidelobe
level. For angles in ±30◦
the sidelobe suppression is at least 13 dB, with a maximum of 17 dB. The bigger
sidelobes for the ±45◦
case is due to the array factor sidelobes starting to coincide with the mainlobe of
the element radiation pattern. This does not happen for cases that are closer to the array’s broadside.
2.3 Verification Measurements
This section shows sample measurements with the sounder performed for two purposes: (i) verifying
the viability and accuracy of the sounder, and (ii) demonstrating the capabilities, in particular the multilink capabilities, of the sounder. As mentioned before, the sounder is capable of operating as a multinode or single-node on the transmit side and single node with 360◦ FoV on reception. The sounder can
continuously acquire MIMO snapshots without gaps, allowing the measurement of non-stationarities that
conventional sounders with acquisition duty-cycling cannot. These characteristics will be shown below.
34

2.3.1 Data Processing
In a first step, the time-variant Double-Directional Delay Power Spectrum (DDDPS) [30] can be then calculated as:
DDDP S(ϕT X, ϕRX, τ, t) =

F
−1
f
{W(f) · HϕTX,ϕRX,t(f)}

2
, (2.5)
where f denotes frequency in the 0.2 to 1.2 GHz range, τ represents delay between 0 and 600 m, and t
is time. In this article, and knowing that the propagation speed is the speed of light, we use τ in [m] to
make comparisons against geometrical markers easier. ϕT X/RX = −180◦
, −165◦
, ..., 165◦
, F
−1
f
denotes
the inverse Fourier transform with respect to frequency, HϕTX,ϕRX,t(f) is the measured frequency response (after accounting for calibration) for the combination of pointing angles ϕT X, ϕRX, and W(f) is a
windowing function. Note that the only time-variant element on the right side is the measured frequency
response. A Hann window was used as a good compromise between mainlobe width and sidelobes. This
impacts precursor power as well as interference between MPCs on plots.
Noise thresholding [16, 2] is applied to the measured DDDP S(ϕT X, ϕRX, τ, t) data. Then, the max
method to synthesize P DPomni(τ, t) [5, 2] is used:
P DPomni(τ, t) = max
ϕTX
max
ϕRX
DDDP S(ϕT X, ϕRX, τ, t) , (2.6)
where NT X is the number of beams on the TX side, and NRX is the number of beams on the RX side.
Similarly, the Angular Power Spectrum (APS) is obtained by integrating the DDDPS over delay:
AP S(ϕT X, ϕRX, t) = X
τ
DDDP S(ϕT X, ϕRX, τ, t) . (2.7)
Note that all these quantities still depend on time/location.
35

The temporal evolution of the APS is useful to show the directional characteristics of the channel and
also single/cluster MPCs.
2.3.2 Measurements
Measurements took place on University Park Campus of University of Southern California (USC) in Los
Angeles, CA, USA. The sounder was deployed in its spatially distributed mode, and placed in an 1400 m2
open area surrounded by buildings and high-canopy trees. There are several notable metal objects in the
measured environment: An emergency phone booth (label Phone), a 0.5 m2 metal sign on a wall (label
Sign), a pole (label Pole), a wire fence around a storage area (label Fence), a self-powered utility cart next
to it (label Cart 1), a second utility cart behind the location of TX A3 (label Cart 2); their locations were
carefully recorded. The sounder was operating in distributed mode, with TX units placed around the RX
according to Fig. 2.18. Distances between TX and RX were 11.6 m, 11.9 m and 12.1 m for A4, A2, and
A3 respectively. The measurements focused on the impact of moving objects, while the sounder carts
were kept at a fixed location; laser tools were used to align antennas and measurement wheels to measure
distances.
Two different object trajectories were recorded on this scenario: T1 is a linear trajectory over which a
flat aluminum reflector screen was moved, and T2 is circular trajectory around the RX over which a flat 1
m2 blocking screen was moved. The reflector and blocker were kept at antenna height (1.6 m) at all times
during their movement.
2.3.2.1 T1 - Flat Reflector
Fig. 2.19 (a), (b), (c) shows the P DPomni(τ, t) of the T1 trajectory for the three distributed TX nodes. Each
plot consists of 3288 PDPs taken at the different snapshot times. The plot for each of the TX panels shows
the PDP summed over 196 beam combinations, i.e. 7 TX beams covering ±45◦ with 28 RX beams covering
36

A2
A4
A3
RX
T1
T2
Lamp
Phone
Fence
Cart 1
Sign
Pole
Cart 2
Bins
36 m
39 m
-45o 45o
-45o 45o
-45o 45o
0o t = 0
t = 8
t = 0
t = 8
+ -
Figure 2.18: Measurement scenario Layout, top view. A3 and RX are mounted on self-propelled carts
(green-black rectangles). A4 and A2 are mounted on tripods (smaller black rectangles). This scenario
corresponds to the one shown in Fig. 2.2
37

A2
2 4 6 8
Time [s]
10
20
30
40
D
ela
y [m]
-130
-120
-110
-100
-90
P
a
t
h
G
ain [d
B]
Lamp/Phone LOS
T1
Fence/Cart
Sign
(a)
A3
2 4 6 8
Time [s]
10
20
30
40
D
ela
y [m]
-130
-120
-110
-100
-90
P
a
t
h
G
ain [d
B]
LOS
Fence/Pole
T1
Phone/Lamp
Cart 1
(b)
A4
2 4 6 8
Time [s]
10
20
30
40
D
ela
y [m]
-130
-120
-110
-100
-90
P
a
t
h
G
ain [d
B]
LOS
T1
Pole/Fence
Phone
Lamp
(c)
2 4 6 8
Time [s]
10
20
30
40
D
ela
y [m]
5
10
15
P
o
w
e
r [d
B]
T1
Digitizer Jumps
(d)
2 4 6 8
Time [s]
10
20
30
40
D
ela
y [m]
5
10
15
P
o
w
e
r [d
B]
T1
(e)
2 4 6 8
Time [s]
10
20
30
40
D
ela
y [m]
5
10
15
P
o
w
e
r [d
B]
T1
(f)
Figure 2.19: P DPomni(τ, t) and P DPomni(τ, t)- P DP ^omni(τ ). (a), (b) and (c) show P DPomni(τ, t)for TXs
A2, A3 and A4, respectively. (d), (e) and (f) shown P DPomni(τ, t) - P DP ^omni(τ ) for each TX. Trajectory
is T1: reflector.
360◦
around the RX. Note that values under the noise threshold are represented with the same color as the
minimum value of the colorscale.
The notable scatterers are labeled in the figures. The calculated path length of wavefronts between TX
units and RX and metal reflectors can be matched to peaks in the Power-Delay Profile (PDP) with an error
of ±0.5m of which 0.3 m can be attributed to the delay resolution of the sounder. The remaining error
might be attributed to the distance measurement tool’s intrinsic and handling error. It can be noted that
tall and slim edge-like reflectors give rise to Multi-Path Components (MPCs) with significant power: for
TX A4, Fence and Pole are within 7 dB of the LOS power; for TX A2, Lamp and Sign are within 11 dB from
LOS; for TX A3, Fence and Pole are within 18 dB from LOS. The aluminium reflector can generate MPCs
which maximum power is 14 dB under the LOS power. These results are not accounting for the increased
free-space loss due the extra path length.
38

Next, for each run, the median (with respect to t) PDP P DP ^omni(τ ) is calculated and subtracted from
each TX case. The P DP ^omni(τ ) describes the static scatterers, since all movements are brief perturbations
compared to the long term capture. Removing P DP ^omni(τ ) allows to better isolate the effects of the
moving objects in the channel. In the case of T1 this corresponds to the moving flat reflector screen, which
can be seen in Fig. 2.19 (d), (e), (f). Notice that imperfect cancellation of the digitzer jump pattern (see Sec.
II) lead to residual apparent power changes of strong MPCs, which appear in the figures. They have been
marked in Fig. 2.19 (d), but they are also observable in Fig. 2.19 (e) and (f).
Each MIMO snapshot is separated into 3 APS sets, one for each TX. Using our carefully calibrated
system response we apply the MUSIC algorithm [39] to each APS set to obtain a high-resolution version
of the APS that we refer to as MUSIC APS.
Also for the MUSIC APS, we compute the median MUSIC APS and subtract it from the time-variant
MUSIC APS, so that only variations with respect to the static case are shown in the following figures.
Furthermore, only variations within 5 dB from the maximum are shown to de-clutter the plots. The result
is shown in Fig. 2.20, where all the 3288 APSs have been overlapped to show the scatterer movement
effect in the angular domain. Angle of Arrival (AoA) 0◦
is as marked on the RX in Fig. 2.18 (looking south)
and its range is -165◦
to 180 Positive AoAs are measured clockwise and negative AoAs counterclockwise.
Ground truth reference trajectories are plotted as red dashed lines. Each of the two almost-parallel lines
corresponds to the trajectory of one of the horizontal ends (left and right) of the reflector used.
The flat reflector’s trajectory is clearly seen in the angular domain. As Fig. 2.18 shows, T1 causes MPCs
with AoA mostly in the -60◦
to -140◦
range which corresponds to T1 angles as seen from the RX. At t = 0,
the MPC generated by A2 is visible at AoA -60◦
. At this time the MPCs generated by A3 are still to weak
to be visible, but they become visible when the reflector is in the -110◦
to -135◦ AoA range. A few weaker
MPCs can be seen for A2 and A4 for high Angle of Departure (AoD) values. According to our 360 video
recordings from the measurement runs, they correspond to reflections on P ole and Cart.
39

Figure 2.20: APS for the reflector case, T1. All static MPCs are removed from the data and then only MPCs
associated to movement are kept. In this case we show only movement MPCs with powers up to 5 dB
lower than the maximum. Color values represent the power deviation with respect to the static channel
for the specific AoD/AoA combination.
2.3.2.2 T2 - Blocker
Fig. 2.21 shows the P DPomni(τ, t) of the T2 trajectory for the three distributed TX antennas. At t = 0, the
blocker is placed between TX A2 and the RX. The blocker is moved in clockwise direction (ref Fig. 2.18)
around the RX at walking speed (around 1 m/s). At t = 8, the blocker ends in between the LOSs A4-RX
and A3-RX. As time progresses, the scatterer blocks the RX first from receiving the LOS from A2 and then
that from A4. As in Fig. 2.19, values under the noise threshold are represented with the same color as the
minimum value of the colorscale.
Fig. 2.21 (a) shows the blockage of the LOS between TX A2 and the RX from t = 0 s to t = 2 s. As
the blocker moves, it blocks the scatterer labeled Sign from t = 2.5 s to t = 4.5 s. At some point during
this time, the blocker also interrupts the wavefront that departs from TX A2 to the scatterer labeled Fence,
which after reflection towards the RX generates the MPC that appears at 26.5 m in the P DPomni(τ, t).
40

A2
2 4 6 8
Time [s]
10
20
30
40
D
ela
y [m]
-130
-120
-110
-100
-90
P
a
t
h
G
ain [d
B]
Lamp/Phone LOS
Sign Fence/Cart
(a)
A3
2 4 6 8
Time [s]
10
20
30
40
D
ela
y [m]
-130
-120
-110
-100
-90
P
a
t
h
G
ain [d
B]
LOS
Fence/Pole
Phone/Lamp
Cart 1
(b)
A4
2 4 6 8
Time [s]
10
20
30
40
D
ela
y [m]
-130
-120
-110
-100
-90
P
a
t
h
G
ain [d
B]
LOS
Phone
Pole/Fence
Lamp
(c)
0 2 4 6 8
Time [s]
-80
-70
-60
-50
R
e
c
eiv
e
d
P
o
w
e
r [d
B
m]
A2
LOS
Pmax;TX
Pmax;All
(d)
0 2 4 6 8
Time [s]
-80
-70
-60
-50
R
e
c
eiv
e
d
P
o
w
e
r [d
B
m]
A3
LOS
Pmax;TX
Pmax;All
(e)
0 2 4 6 8
Time [s]
-80
-70
-60
-50
R
e
c
eiv
e
d
P
o
w
e
r [d
B
m]
A4
LOS
Pmax;TX
Pmax;All
(f)
Figure 2.21: P DPomni(τ, t) and Temporal evolution of power for diferent angular combinations. (a), (b)
and (c) show P DPomni(τ, t) for TXs A2, A3 and A4, respectively. (d), (e) and (f) shown the received power
evolution for selected directions. Trajectory is T2: blocker.
41

The following blockage events happen for TX A4, when the blocker is around its LOS with the RX.
From t = 4 s to t = 7, the LOS between TX A4 and the RX unit is blocked. The path from the scatterers
around Fence are blocked then as well, until the end of the run at t = 8 s.
The LOS between TX A3 and the RX is not interrupted during this run, but there interruptions of scatterers around Fence that can be seen at the same time intervals as they happen in the other TX nodes. This
is because they are interruptions of the reflected wavefront, not the wavefront incident on the scatterer.
Thanks to the angular scanning capabilities of the sounder, the blockage events can also be simultaneously characterized for all angular combinations. For each TX, the direction combination that represents
the LOS component is selected and its power shown in Fig. 2.22 (d), (e), (f) (blue-continuous line). At each
time instant, Pmax,TX, the maximum power that can be obtained across all direction combinations for each
TX is calculated and shown as well (red discontinuous line). Similarly, Pmax,All, the maximum obtainable
power across all TX and RX direction combinations, across all TX units, is shown for comparison (black
dash-dot line). Fig. 2.22 (a) shows a significant fade from the blocking event on the LOS between A2 and
the RX (solid line). However, the maximum obtainable power when switching within A2 during the fade
is at least 11 dB higher. The maximum the RX can see, from any TX is 18 dB higher. Fig 2.22 (e) shows that
A3 doesn’t suffer fades, but its power is lower than the maximum the RX can see, from any TX. Finally,
Fig. 2.22 (f) shows that, similar to A2, the blockage-induced fade can be mitigated by looking at different
directions. In this case, the RX would see an improvement of at least 2 dB and at most 16 dB if it could
scan azimutally. This shows the advantages that a system with beam-scanning capabilities would have in
this example case.
Fig. 2.22 shows the temporal evolution of the MUSIC APS for the T2 case. To prevent clutter, only
negative variations larger within 7 dB from the deepest fade are shown, highlighting the blockage events
from this run. These events match the geometry and location of scatterers and have been checked against
42

Figure 2.22: APS for the blockage case, T2. All static MPCs are removed from the data and then only MPCs
associated to movement are kept. In this case we show only movement MPCs with powers up to 7.5 dB
greater than the deepest fade. Color values represent the power deviation with respect to the static channel
for the specific AoD/AoA combination.
our 360◦
camera videos. We have added rectangular regions with the angular locations of the main scatterers for comparison. Two new scatterers, not shown in Fig. 2.18, are included in the MUSIC APS: T ree1
and T ree2. These correspond to hanging tree branches located between Sign and P ole.
Fig. 2.22 shows the temporal evolution of the APSs for the T2 case. Only negative variations larger
than -5 dB than are plotted in this case, highlighting the blockage events from this run. These events match
the events shown for the PDP case.
43

Chapter 3
Fundamental Operational and Mechanical Challenges
Real devices often show behaviors that are not well defined or analyzed by manufacturers. These are
qualified here as unexpected challenges, since they can’t be predicted in the design stage nor in the verification testing stages. Low probability events and other specific use case performance items are part of
this chapter. These challenges meet two criteria: they have a negative impact when using the sounder for
high-precision measurements, and they are cumbersome to solve (often significant time-consumers). For
the sounder described here, the main impacted sounder performance indicator was system timing. This
indicator has multiple effects on measurement quality, but one of the most important effect is distortions
in wave arrival times (crucial for localization). Additional challenges were found in the array scanning
subsystem. The challenges, its characterization, and solutions are described below.
3.1 System Timing Accuracy
Phase stability is a crucial system parameter for any sounder, regardless of their operation in the time or
frequency domain. Since frequency domain sounders are usually based on VNAs, the phase stability is
usually guaranteed and tested by the manufacturer of the VNA. This is a much more complex topic in the
case of channel sounders with spatially separated transmitter and receiver, and it is usually a design aspect
that is present at every stage from foundations to the high-level operation.
44

As discussed in Chapter 2, the measurement of channel phase is based on the synchronized transmission (by an AWG) and acquisition (by a digitizer) of a waveform. The receiver knows the exact moment
the waveform was transmitted, and uses this knowledge to determine how much delay the channel under
measurement is responsible for. The important assumption here is that, for practical effects, the known
transmission-acquisition delay is a time-invariant quantity. This assumption can be violated by any component of the timing system. In this case, both digitizer and AWG broke it in the ways that are detailed
below. Solutions for this issues are also presented.
3.1.1 Digitizer Temporal Drift and Jumps
The digitizer used in this sounder has a sampling rate of 5 GSa/s and enough memory throughput to record
its input signal without any gaps for 8 seconds. Its internal sample clock can be disciplined by an external
10 MHz reference to achieve a clock/phase stability that closely follows that of the external reference.
However, while testing the direct connection from AWG to digitizer (static test) a drift and a jump pattern
was detected.
3.1.1.1 Characterization
The drift pattern resulted in a uniformly increasing delay for the received waveform. Since the AWG
generating the waveform is locked to the shared 10 MHz external reference (tested with independent
instrument), it was determined that the digitizer wasn’t following the 10 MHz signal. After talking to the
manufacturer, it was determined that a firmware upgrade was necessary. This succesfully eliminated the
drift.
The digitizer also exhibits a timing jump pattern, even when fully locked to the 10 MHz external reference. Moreover, this jump pattern appears only for certain timing configurations applied to the digitizer.
Specifically, the jump was not present for captures below 2 seconds. Whenever this was the configured
45

Figure 3.1: Digitizer Jump Pattern.
acquisition time, the digitizer timing would work perfectly. However, the 8 second acquisition time case
showed a timing jump pattern. The first jump happens at around t = 0.43 s and then it repeats with
periodicity 1.71 s. On every jump event, the digitizer waveform is delayed an extra 384 Sa.
3.1.1.2 Solution
This issue was discussed multiple times with the equipment manufacturer and no solution was provided.
Initially, an 8 s capture was recorded for a totally static scenario at every measurement deployment. At
this point the jump pattern was not tested across multiple on/off cycles. The initial static jump pattern was
then used to correct the real channel measurement data. In order to perform this correction, the waveform
is first rearranged in an M × N matrix, where M is the number of sounding sequence repetition in the
waveform (3288 for 8 s), and N is the number of samples per sounding sequence repetition (13500 at 5
GSa/s). Each row of this matrix is then shifted an amount equal to the additive inverse of the recorded
jump pattern.
46

The shift was applied according to the following equation:
sc[i, j] =



su[i, j + p[i]] i ≥ p[i]
su[i, j + (M + p[i])] i < p[i] ,
(3.1)
where sc[i, j] is the jump-corrected signal, su is the uncorrected signal, p[i] is the amount of jump for the
i-th sounding sequence repetition, M as described earlier, i = 1, ..., M indexes rows and j = 1, ..., N
indexes columns (time samples). This operation is equivalent to the circshift function available in most
scripting languages.
After many tests in different conditions of temperature it was concluded that this jump pattern is
invariant through on/off cycles. This means we need to measure it only once and then we can correct all
jumps for the same digitizer configuration with the same data. This was the solution implemented in the
end, successfully solving the digitizer timing jumps.
3.1.2 Arbitrary Waveform Generator Jitter
The AWG appeared to work correctly in the beginning. However, there was a hidden jitter pattern hiding
in the data that became apparent only after measuring roughly 1 TB of channel data. This jitter was only
affecting 0.1-1% of the data. More importantly, it had the chance to appear any time the AWG was triggered.
The AWG would show jitter bursts in between preiods of no jitter. The way of finding the root cause of
this issue and the solution used to overcome it are explained next.
3.1.2.1 Characterization
The jitter pattern was first identified after careful observation of the temporal PDPs. Fig. 3.2 shows an
example of a section of a temporal PDP where despite the TX and RX ends of the sounder being completely
static, a movement of approximately 1 delay bin can be seen for some of the temporal samples.
47

Figure 3.2: Temporal PDP showing delay bin movements due to AWG jitter.
The troubleshooting began by isolating the timing subsystem from the rest of the channel sounder
subsystems. The timing subsystem tests started with a direct connection from AWG to digitizer to see if
the problem was present there, or if it was any part of the custom timing circuitry.
The direct connection showed the problem upon closer inspection. Fig. 3.3 shows the overlapping of
multiple sounding sequence repetitions in the sample domain (equivalent to time domain), for multiple
zoom levels. Notice how in Fig. 3.3 (a) there is apparent overlap of all 3288 sequence repetitions, (b) shows
the start of the waveform without apparent distortions due to the high sampling rate, but (c) shows 2 states
of overlap. The majority of the repetitions are overlapping in the same sample index, but there is another
set of them that are overlapping exactly 8 samples later.
A trilateral test, i.e. a test involving all 2-device combinations between a set of 3 devices, was performed
and it was concluded that this 8 sample deviation came from the AWG. The AWG specification was not
sufficiently precise for the triggering system behavior, but it stated a strong temperature dependence and
that the trigger was detected every 8 clock samples. This means that for some of the sequence repetitions
the trigger was being detected 1 trigger detection event later than the target.
48

(a) (b) (c)
Figure 3.3: AWG Jitter as shown by static measurement. (a) shows received time domain waveform for a
single MIMO snapshot slot. (b) zoom in time to show apparent wave overlap. (c) shows overlap discrepancy
of 8 samples.
The natural next step was to take a close look at the quality of the trigger signal generated by the
custom timing circuitry. The default trigger level for the AWG is 1.2 V so special attention was put into that
region of the trigger pulse. The trigger pulse was observed with an oscilloscope and it was observed that
the rising edge was not monotonically increasing. In fact, it had a local maximum and minimum around
the 1.2 V level. This was considered a highly likely reason for the AWG jitter so the trigger threshold was
moved away from the problematic area. The direct test was then performed again in the lab and the jitter
appeared corrected.
In Fig. 3.4 the starting location of the first 411 sequence repetitions are shown. Fig. 3.4 shows the
behavior when the trigger threshold level is 1.2 V. Notice that 3 of the repetitions have their starting
points 8 samples later than the rest. This is not visible in (b), where the trigger threshold level has been
increased to 1.4 V (cleaner rising edge area).
A software routine was created to check successful resolution of this issue while performing measurements on-site. After numerous measurements it was concluded that the problem was not solved.
Moreover, the problem was uncorrelated to the trigger threshold level. Furthermore, the trigger jitter was
uncorrelated to ambient conditions like temperature, wind gusts, etc.
49

(a) (b)
Figure 3.4: Sounding sequence repetition start location as a relative count of number of samples.(a) shows
results for AWG trigger threshold set to 1.2 V. (b) shows the same for trigger threshold level 1.4 V.
3.1.2.2 Solution
Since the problem was presented at random and it wasn’t correlated to any parameter under control, the
system had to be made robust against this specific problem. This required a two-sided approach: part of
the problem needed to be solved inside a MIMO snapshot, i.e. within one block of sounding sequence
repetitions block, and part of the problem in between MIMO snapshots. This due to the trigger being
activated intra- and inter-snapshot. The intra solution required hardware changes, while the inter solution
was software-based.
Hardware Solution The main objective of this part of the solution is to make the AWG trigger only
once per MIMO snapshot. The initial configuration makes the AWG trigger 7 times per MIMO snapshot,
which means that even if the first trigger is correct, there might be issues with some angular combinations
being delayed an extra 8 samples. This is due to the fact that the total amount of samples of the MIMO
snapshot does not fit in the AWG memory. However, the device supports waveform segment repetitions
of the same memory regions. This allows the generation of a sequence of arbitrary length, as long as it
meets some requirements. We thus modify the waveform to use a repetition scheme and overcoming thus
50

the memory limits. However, careful attention needs to be paid to the custom timing circuitry constraints
to ensure compatibility with longer MIMO snapshot cycles. For long MIMO snapshots, the requirements
are as follows:
• Sampling rate should be higher than 2.4 GSa/s: This is due to the baseband waveform having a
maximum frequency of 1.2 GHz. Sampling theorem requirements need to be met.
• Sounding sequence repeat block must have granularity 64. This means that the total number of
samples in the block need to be a multiple of 64. The AWG cannot repeat waveforms endlessly
unless this condition is met.
• Number of samples in the sequence repeat block must be compatible to 10 MHz. This means that
the system needs to fit an integer amount of samples in 100 ns. Otherwise, the 10 MHz clock used
for the custom timing circuitry will drift with respect to the AWG internal sample clock.
The originally established sample clock of 5 GSa/s cannot meet all this requirements for long measurements (more than 784 angular slots, or sequence repetitions). A new AWG sample clock of 3.2 GSa/s was
used when long MIMO snapshots were used. Note that this doesn’t require modification of the digitizer
sample clock.
Software Solution The hardware solution fixes possible jitter within one MIMO snapshot. However,
multiple MIMO snapshots need to be triggered multiple times due to switch time guards at the end of
them. This creates the chance of trigger jitter at the start of each MIMO snapshot. Here, three facts can be
exploited:
• The jitter is equivalent to a real scatterer speed much higher than what’s physically possible in our
measurements. The duration of 8 samples is equivalent to a delay of 1.6 ns at 5 GSa/s. This can be
51

interpreted as a movement of 0.48 m in 1/411 s. The equivalent speed is thus 197.28 m/s. The 3.2
GSa/s case has an even higher equivalent speed.
• The waveform can only get delayed by 8 samples. This means that one MIMO snapshot can be -8,
0, 8 samples away from the previous one. No other values are possible due to AWG jitter.
• Within one MIMO snapshot, there will be at least one angular slot (sequence repetition) that has
significant SNR. If there are none, then that measurement is not valid data.
This means that correlation analysis can be used to overcome the jitter delays. After acquiring the
data, it is divided into MIMO snapshots and then they are iterated through. We take one snapshot and
calculate its correlation with three different versions of the previous snapshot: one shifted -8 samples,
one without shifting, and one shifted 8 samples. The highest correlation will indicate the relative delay
between snapshots even if the scenario is changing.
The software method described here proved to be sufficient to solve 99% of the delayed repetitions. A
manual pass for each scenario was performed to check cases where the correlation method failed and thus
all data can be corrected.
3.2 Elevation Scanning
The verification measurements from Chapter 2 served as a proof of all critical systems for sounding. Among
others they showed that the array panel beam scanning was operational at a very high speed required for
mmWave channel sounding. However, those measurements assume that if the azimuth scanning work, so
does the elevation scanning, since it relies on the same chipset architecture. This proved not to be true:
The elevation beam forming did not reliably produce acceptable beam shapes. In this section, the issue is
characterized and a solution is presented.
52

Figure 3.5: Measured beam pattern for scanning angle −10◦
in elevation and 0
◦
in azimuth.
3.2.1 Characterization
Fig. 2.16 shows the shapes for the azimuth beams. The characteristic radiation pattern shape for directive
antenna can be seen, and is evidence of correct beamforming. However, when an elevation beam was
configured in the antenna phased shifters a very different results was observed.
Fig. 3.5 shows the resulting elevation beam with scanning angle -10◦
in the RX arrays. The phase
payload that is uploaded to each antenna’s phase configuration register is calculated with the same method
as the correct azimuth beams. It is also transmitted using the same electronics as the azimuth beams, but
the observed antenna array radiation pattern has no mainlobe at mechanical angle -10◦
. In fact, it has a
null near that angle. Furthermore, there are three major lobes with a maximum relative power difference
of 7 dB.
The array scanning principle says that in order to steer the beam of an array, the antenna phases must
be arranged in such a way that the phases of all antennas are identical when the antenna points to the
scanning angle. This was the motivation to test the phase response of the antenna elements. The RX arrays
were moved to an anechoic chamber and, to speed up the process, its 4 rows were measured with a VNA to
53

(a) (b)
Figure 3.6: Measured phase for each row of the RX array when the scanning angle is -10◦
. (a) Phases
leading to erroneous beam shape. (b) Corrected phases.
determine their relative phases as a function of mechanical rotation angle. Any phase differences for each
of the rows’ elements get averages by a factor of 8, but only a factor of 4 in the columns. This motivated
the choice of measuring phase by row.
3.2.2 Solution
The measured phases are shown in Fig. 3.6. Fig. 3.6 (a) shows the phases measured for the problematic
beamshape. Note how rows 1 and 2 have intersecting phases at the angle of interest but 3 and 4 intersect at
a different place. The process to fix phases was cumbersome: the phases were modified according to their
apparent distance before modification. Then they were measured in the anechoic chamber again and the
process repeated until good phase agreement between rows was found at the scanning angle of interest.
Fig. 3.6 (b) shows the result of the iterative process for the measured phases. Notice how the phases of all
antenna element rows intersect at -10◦
.
The resulting beam shape after the iterative fixing process is shown in Fig. 3.7. The mainlobe of
the radiation pattern now is located at the angle of interest. Sidelobes are at least 14 dB lower than the
mainlobe.
54

Figure 3.7: Beam shape after fixing phase intersection
Fortunately, this iterative method of fixing was enough to fix all 4 array panel of the RX array box.
This means that the phase variations needed to get the expected beam shapes were the same for all panels.
Furthermore, this modifications to the antenna phase shifter payloads are totally independent from the
azimuth case and thus do not modify the array performance on that axis.
3.3 Self Propelled Sounder
A commonly overlooked aspect of sounding is how to move the sounder itself. This of course does not correspond to the discipline of electrical engineering, but it’s a key design aspect in high-complexity sounders,
like the one presented in this work, due to their high weight and physical volume. Moreover, this sounder
is designed to operate in potentially continuous movement so it is not possible to rely on human power for
this purpose. The sounder needs to move in trajectories that have a stepwidth of a couple of centimeters
for precision measurements to distances on the order of hundreds of meters. This also makes powering
the sounder with a cable difficult, so batteries are needed for the moving cart that hosts the sounder and
also for the sounder power consumption.
55

Figure 3.8: 3D rendering of cart design. Two units were built: one for the TX and one for RX.
A joint effort with the Diamond Engineering Company resulted in the design and construction of a
self-powered cart that is capable of hosting the 200 kg of equipment, moving at a 1.5 m/s top speed, and
having a battery autonomy of about 8 hours of movement on flat. The cart also has automatic braking for
safety, can go most up-slopes around measurement sites, and an elastomer suspension system designed to
prevent shocks to the precision measurement equipment it carries. The cart design is shown in Fig. 3.8.
3.4 Movement Tracking for Ground Truth
3.4.1 Location Tracking Systems for Channel Sounding
To characterize the performance of mmWave systems, specially for localization purposes, it is important
to have a solid ground truth for the location of the sounder in the environment. mmWave systems are
expected to have localization errors on the order of centimeters. This in turn results in strict requirements
for the location tracking systems. Usual approaches to solve this issue are the use of click wheels, encoders,
and Inertial Measurement Unit (IMU)s. Click wheels are hand-held devices mostly designed for land surveying and construction. They have significant measurement errors in the order of 10 cm for the distances
56

involved in channel sounding. This is due to their measurement being based on the angular variations
between the handle and a wheel that is tangentially tracking the ground. This means that the error not
only violates the requirements for the application of interest here, but also significantly depends on user
handling, since the system allows for human errors to be introduced in the measurement. An additional
challenge of this method is that it is usually no possible to place the click wheel in the location where
it needs to be to allow for lossless measurement of the antenna location. The click wheel needs to make
ground contact exactly on the projection of the antenna location over the 2D surfaces where it moves. Otherwise only linear trajectories are allowed to be measured without errors. The second option, encoders,
are a family of sensors based on different physical principles that can measure angular variations between
two point of a rotating mechanical contraption. They are equivalent to "electronic click wheels" and thus
exhibits a lot of the same issues. The carts detailed in the previous section have magnetic encoders, but
since they are moving on non-linear trajectories and are mounted on a wheel and not the projection of the
antenna location over the corresponding surface, they cannot be used for location tracking of the sounder.
The third method, widely used in the field of robotics is the use of IMUs. These are compound measurement units usually consisting of multiple accelerometer, gyroscope, and magnetic compass sensors.
IMUs are widely available in 6 and 9-axis variants. This means that the most common of these devices
usually have a 3 axis accelerometer plus a 3 axis gyroscope, with some of them having a 3 axis magnetic
compass. The advantage here is that they can track movement in 3D space, and they usually have very
high sampling rates. On the other hand they usually need developing a sensor fusion algorithm or, at the
very least, tuning parameters of an existing algorithm. Furthermore, they exhibit a phenomenon called
"drifting". Drifting occurs because IMUs obtain location by double-integrating acceleration data and then
using the other axes to improve the estimation. When double-integrating the acceleration data, the accelerometer error is also integrated and thus the data for location drifts. This drift is usually corrected in
57

Figure 3.9: Intel RealSense T265 Camera.
post-processing since it is small for short time periods. For tracking in long time frames, like the objective
of this work, more sophisticated method of drift correction are needed.
3.4.2 Location Tracking Solution for mmWave
Intel has released one of these solutions. Their Real Sense T265 camera (Fig. 3.9) is a location tracking
device with a 9-axis IMU plus 2 video cameras for stereo vision. The cameras are used to estimate relative
movement from frame to frame, and thus any drift informed by the onboard IMU can be corrected. The
hardware and software are discontinued by Intel, but the cameras can be still found online and old versions
of the software SDK are available from mirrors online. The system does not require any hardware work
to use, but the software needs to be adapted to the specific application. Furthermore, this work requires a
fully autonomous solution for location tracking so the camera host devices is designed around a Raspberry
Pi computer. A complete location measurement software suite was developed for location tracking. The
camera achieves an accuracy of around 5%.
58

3.4.3 Additional Considerations
The camera exhibits significant light-condition sensitivity. Despite indoor tests at development time, errors appeared when measuring in real scenarios. Multiple measurement runs were discarded due to the
erroneous position estimation. The main problem being that measurements are usually taken at night
time, when traffic allows channel measurements, but night light conditions are unfavorable for stereo vision edge detection. Unlike an indoor lighting context, where lighting is uniform and exhibits low dynamic
range, night time lighting usually consists on big areas of darkness with a few small but very intense light
sources (e.g. lamp posts). The stereo camera from the Intel T265 does not have enough dynamic range to
cover the measurement environment of this work.
As a first solution, a strong light source is mounted to the sounding cart. This proves to be ineffective
since the light does not uniformly light the environment. A partial solution for this is to use a light diffuser
(like those from photography) to scatter the light and improve it’s uniformity. This helps in most cases but
is not enough to ensure 0% estimation error.
Location of the camera proved to be another method of improving location tracking accuracy. By
pointing straight down, the camera is only exposed to the dynamic range of the ground, which does not
involve high dynamic range (there are no lamp posts on the ground). Additionally, the ground can be more
easily lit by a lamp.
Lastly, a set of hard edge objects are placed besides the cart trajectory before every run to help with
edge detection. In addition to the camera, a tape measure was used to record XY coordinates of the cart’s
start and stop positions. With this information we can counteract the residual 5% errors from the camera.
The implementation of all these additional consideration led to a robust ground truth for localization
performance.
59

Figure 3.10: Custom Positioning Arm for Calibration.
3.5 Calibration Positioner
The array boxes that the sounder is based on are too big and heavy to be calibrated inside the anechoic
chamber using the chamber’s positioning arm. When calibrating, it is important to make the azimuth
and elevation axes align with the phase center of the array panels being calibrated. For a phased array
this means the box rotation axes need to cross the array panel surface [32]. Given the shape of the array
boxes (Fig. 2.2), this means that a considerable amount of torque needs to be applied for rotating the array.
Besides this, the existing positioners have mounting brackets that do not fit and are not big enough for the
array box. A custom arm was therefore designed and built, shown in Fig. 3.10. This positioner is based on
a one stepper motor per axis, plus a turntable with an inner reduction gearbox to act as a torque multiplier.
This reduced the size and current requirement of the stepper motors.
The positioner is designed to be fully compatible with the anechoic chamber positioning control system. Its mechanical interface matches the one available in the chamber and is electrically compatible with
the control software and devices. Fine tuning of the software’s control parameters is needed to ensure that
the stepper motors do not miss steps. Step missing leads to errors in the antenna radiation pattern since
60

the positioner loses track of where it is located. The system was tested and tuned until no missing steps
occur.
3.6 Control Signal Integrity and Overhead Reduction
3.6.1 Signal Integrity
The array boxes and single panels offer high levels of flexibility. VGAs and phase shifters inside the box
can be controlled in real time using SPI and QSPI buses. However, given the large number of antenna
elements on each array a fast SPI clock speed is needed. The arrays work at 50 MHz SPI clock speed. The
clock speed and the high number of devices inside the box create significant challenges for signal conditioning. The clock harmonics appear at odd multiples of 50 MHz and thus the circuit becomes sensitive
to multiple parameters like track length, via capacity. This mandates careful PCB design to host all SPI
related components. Custom PCB boards were design and manufactured. All fast lines on the PCBs were
routed on a single side with no vias. Additionally, buffer circuitry was added to improve signal strength.
Fig. 3.11 (a) shows the poor performance of the SPI signals as captured at the array input, where they
should be clean. Fig. 3.11 (b) shows the SPI signal bundle at the array control input. The array works
reliably with the added circuitry.
3.6.2 Control Signal Overhead
Communication using SPI is a popular way of interfacing a wide range of electronic devices. This means
that there is a large collection of control devices including SPI functionality in the market. These go
from simple serial/parallel communication circuits to microcrontollers and FPGAs IP. As stated before, the
complexity of the sounder requires the use of MCUs. These devices usually come with SPI functionality in
hardware as well as the firmware libraries required to run them. There are a few of these controllers that
can run at 50 MHz. However, most of them are not designed for fast speed continuous operation and leave
61

(a)
(b)
Figure 3.11: (a) Unbuffered signals for data and clock lines of SPI array control. Low slew rate leads to high
distortion and array does not work. (b) Buffered signals for data and clock lines of SPI array control. Slew
rate is high enough to enable 50 MHz SPI communications. Enabled by custom PCB with careful design
contraints.
62

large gaps in between bytes. These gaps can be up to 10 times as long as the time needed to transmit the
communication payload. After iterating over more than 10 MCU technologies, it was concluded that the
STM32 family of MCUs [1] was capable of driving the arrays fast enough. However, the provided firmware
libraries add big overheads to the communication, so custom firmware was developed and coded at the
register level for fast operation.
63

Chapter 4
mmWave Localization Measurements
The sounder described in the previous sections was used to acquire data to understand localization in
mmWave channels. This data set achieves three important objectives: (a) it serves as exploration data
to better understand the channel aspects relevant for mmWave localization, (b) it enables validation and
development enabler for mmWave localization algorithms, and (c) it can help with Artificial Intelligence
(AI) model training for localization applications. The specific objectives, measurement details, and results
are explained below.
4.1 Measurement Design
The measurements were carried out in the University of Southern California University Park Campus.
The sounder was deployed in distributed mode, i.e. with three TXs located in different points in space,
about 10 m from each other. The three anchor system allows to capture data that can help in trilateration
contexts, but also in single anchor localization instances. Three scenarios were chosen, ensuring they cover
a representative mix of LOS and NLOS conditions as well as transitions between these. The scenarios are
chosen to maximize the probability of at least 3 MPCs for each anchor at any point in the trajecotry. This
is to guarantee that a single anchor case can be evaluated. Additionally, one of the anchors was in LOS
64

for the whole trajectory of the RX in order to test and correct possible AWG trigger deviations between
adjacent MIMO snapshots, as explained in Chapter 3.
The TX and RX heights are 1.8 m, and they are scanning in azimuth. Each TX (or "anchor") covers a
90◦
sector with 7 beams separated by 15◦
. The RX covers 360◦ with the same angular separation. The total
MIMO snapshot time is 2.4 ms.
4.1.1 Scenario 1
Scenario 1, pictured in Fig. 4.1, corresponds to an open plaza surrounded by buildings and three access
streets. One of the streets enters the plaza through a tunnel. The RX trajectory starts inside this tunnels
and then follows a straight path. When the RX is around the center of the plaza, it turns left to connect
with its exit street. Finally, the RX trajectory follows the exit street orientation. A few lamp posts, small
carts and trash containers are present in the scenario. There are some hanging branches from trees above
the antenna height (not pictured).
Anchor A2 is visible throughout the experiment with LOS. Anchor A4 is outside the plaza area in one
of the streets leading to it, so it starts in NLOS and will become visible for some time interval when the
cart is roughly halfway through its trajectory. A3 is inside the open area, pointing to the exit street. This
means the RX starts in NLOS with it (RX cart behind A3’s area of coverage) but then it established LOS as
it exits the open area.
4.1.2 Scenario 2
Scenario 2, pictured in Fig. 4.2, corresponds to a street canyon intersection. The intersection has four
major buildings with fairly flat walls on each corner. There is some low height vegetation surrounding the
area and a couple of trees with thick trunks and high foliage. No scattering vegetation exists at antenna
height. A few tree trunks surround the experiment site.
65

Figure 4.1: Scenario 1 trajectory (in blue) and most significant MPCs.
The RX trajectory begins in one of the intersecting streets and then describes a 90◦ onto the adjacent
street. This turn is taken in two steps, in order not to get the RX too close to the A3 anchor (maximum
power limitations).
Anchor A3 is visible by the RX throughout the run. A2 and A4 start in NLOS condition with the RX.
A4 is blocked by a building at first, while A2 is facing the opposite way, respect to the RX start location.
As the Rx moves it transitions into LOS with A4 first, and then A2. After all anchors are in LOS, the cart
describes a straight trajectory near one of the flat walls of the exit canyon.
4.1.3 Scenario 3
Scenario 3, pictured in Fig. 4.3, corresponds to a parking lot that is adjacent to a street canyon. There are
four major buildings with fairly flat walls surrounding the parking lot and also on the sides of the canyon.
There is some low height vegetation surrounding the area and a couple of trees with thick trunks and high
foliage. No scattering vegetation exists at antenna height, the flat walls are mostly free of scatterers at
66

Figure 4.2: Scenario 2 trajectory (in blue) and most significant MPCs.
antenna height. A few material discontinuities are present in the wall: some windows and a big ornamental
mirror of around 60m2
.
The RX trajectory begins inside the parking lot, next to one of the surrounding buildings. The RX
trajectory enters the street canyon taking a right hand turn. This turn is taken in two steps, in order not
to get the RX too close to the A3 anchor.
Anchor A2 is visible by the RX throughout the run. A3 and A4 start in NLOS condition with the RX. A3
is blocked by a building at first, while A4 is facing the opposite way, with respect to the RX start location.
As the Rx moves it transitions into LOS with A3 first, and then A4. After all anchors are in LOS, the cart
moves on a straight trajectory near one of the flat walls of the exit canyon. A few tree trunks block A2
and A4 in some short sections of the trajectory.
4.2 Data and Results
The measurement campaign resulted in 176 s of continuous capture of channel data. The data corresponds
to 411 MIMO snapshots per second, for a total of 72336 MIMO snapshots. Each snapshot contains 196
67

Figure 4.3: Scenario 3 trajectory (in blue) and most significant MPCs.
AoD/AoA combinations for each of the three anchors. The bandwidth is 1 GHz, giving a delay resolution
of 0.3 m. This means we can generate a PDP with absolute delay at 411 Hz sampling rate to estimate the
location of the RX with respect to any of the three anchors. Additionally, we can look into the angular
domain results to support the location estimation. In all scenarios, there are multiple MPCs than can be
exploited, for example, to locate the RX unit with the information of a single anchor.
For each scenario, the 72336 PDPs are split into three plots corresponding to a single anchor and shown
below. A representative APS is shown for Scenario 1 due to space constraints.
4.2.1 Scenario 1
Fig. 4.4 shows the time-variant PDPs for Scenario 1. A2 shows significant intensity for the LOS component
throughout the trajectory of the RX. This works as expected, since A2 was designed to be in LOS at all
time. More interestingly, there is a significant MPC component that has a delay slope that is equal to the
inverse of the delay slope of the LOS component. This means that as the RX get closer to A2, its path
length decreases, but the path length for the MPC gets longer by the same amount. This is characteristic
of reflection behind a mobile unit that is approaching an anchor. By looking at our 360◦ video for the
68

measurement run, we confirm this reflection happened at a flat wall behind the starting position of the
RX. As expected, A3 and A4 start the run with significant path loss. This is due to the fact that they are both
in NLOS conditions. The RX enters A3’s LOS at around the 25 s mark and continues to show significant
power until the end of the run. On the other hand, A4 appears to transition to LOS at around the same time
but shows intermittent behavior towards the end. This is explained by a thick tree trunk that intersects
the LOS as the RX moves. All anchors show multiple MPCs that can be exploited in the single anchor case.
APSs are available for the 72336 snapshots. To preserve space, only one of them is shown here to serve
as an example. Fig. 5.5 shows nine snapshots, each 8 seconds apart for the measurement run in scenario
1. Each subplot shows a grid of 12 array panel combinations and each combination has 49 AoA and AoD
combinations. Rows of array panel combinations correspond to a single anchor case; columns are RX array
panels. Fig. 5.5 (a) reflects what is also observed in the time-variant PDP case: at the start of the run only
A2 has significant MPCs (LOS), A3 and A4 are obstructed or out of range. Fig. 5.5 (b) and (c) show how
the received power increases as the RX moves within LOS of A3 and A4. This trend continues and in Fig.
5.5 (e) we see that all anchors show significant path gain. This is the dominant trend until the end of the
run, minus a few instances of obstructions from poles that obstruct the LOS between the RX and A2 (also
seen in the time-variant PDP case).
4.2.2 Scenario 2
Fig. 4.6 shows the time-variant PDPs for Scenario 2. A3 remains in LOS for the duration of the experiment,
but before the turn it shows weaker behavior. This is explained by a difference in slope between the
streets. Impulse responses in scenario 2 are, in general, more sparse than Scenario 1 due to high presence
of vegetation a few meters above and a few meters below the antenna heights. Waves reflecting upwards
or downwards have low likelihood of reaching the RX. There is no significant secondary MPC for the
69

(a)
(b)
(c)
Figure 4.4: Time-variant PDPs for Scenario 1. The horizontal axis is time in seconds, the vertical axis is
delay in meters. Path gain intensity is shown as a color scale. (a) shows the time-variant PDP for anchor
A2. (b) shows the time-variant PDP for anchor A3. (c) shows the time-variant PDP for anchor A4.
70

(a) (b) (c)
(d) (e) (f)
(g) (h) (i)
Figure 4.5: APS snapshots for Scenario 1. On each subplot, the horizontal axis corresponds to AoA while
the vertical axis is AoD. For a particular AoA/AoD combination, the path gain is shown as a color-mapped
intensity. The multiple plots correspond to multiple time instants: (a) t = 0 s. (b) t = 8 s. (c) t = 16 s. (d)
t = 24 s. (e) t = 32 s. (f) t = 40 s. (g) t = 48 s. (h) t = 56 s. (i) t = 64 s.
71

A2 case, until time mark 42 s. This corresponds to a steady reflection at the nearest wall that follows the
trajectory of the RX. A3 and A4 show a significant increase in path gain after they turn into the street.
4.2.3 Scenario 3
Fig. 4.6 shows the time-variant PDPs for Scenario 3. In this case A2 is the antenna with permanent LOS.
This scenario exhibits rich MPC content. A3 starts in NLOS, blocked by a building corner and then goes
into LOS. The RX starts in the back of the sector covered by A4 but enters its LOS around the 12 s mark.
Besides the LOS component, the most significant MPC in the A3 and A4 case is the reflection of the nearest
wall the RX sees after the turn. Moreover, the big mirror in this wall shows significant path gain when the
angles between it, the RX and A4 match those required for reflection. This is visible starting at the 50 s
mark. There are multiple MPCs that can be exploited for single anchor localization.
72

(a)
(b)
(c)
Figure 4.6: Time-variant PDPs for Scenario 2. The horizontal axis is time in seconds, the vertical axis is
delay in meters. Path gain intensity is shown as a colorscale. (a) shows the time-variant PDP for anchor
A2. (b) shows the time-variant PDP for anchor A3. (c) shows the time-variant PDP for anchor A4.
73

(a)
(b)
(c)
Figure 4.7: Time-variant PDPs for Scenario 3. The horizontal axis is time in seconds, the vertical axis is
delay in meters. Path gain intensity is shown as a color scale. (a) shows the time-variant PDP for anchor
A2. (b) shows the time-variant PDP for anchor A3. (c) shows the time-variant PDP for anchor A4.
74

Chapter 5
mmWave Microcell Measurements
Measurements were carried out to study the mmWave channel for communications in urban microcell
environments. In the measurements the the TX is elevated, emulating a BS located on top of a building/post
that covers a block and its surrounding streets. The scenarios are chosen so a mixture of LOS and NLOS
situations are present. Additionally, the difference in height mandates the enabling of elevation scanning
on the TX and RX ends. This is in addition to the azimuth scanning discussed in Sec. II. The modification
to the sounder enabling these measurements, the scenarios, and some results are explained/shown next.
5.1 Measurement Design
The measurements need to allow the TX to be placed at around 15 m height and the RX needs to move in
a trajectory at street level until a maximum TX/RX separation of about 150 m. LOS conditions are enabled
by measurements along streets adjacent to the TX location while NLOS situations are found by turning
around the corner and also from thick vegetation found along the streets that block the signals.
Since the TX is mounted in an elevated position, it does not need to scan positive elevations (looking
to the sky). The 7 azimuth beams are preserved (as used in Chapter 4) but this time they are scanned for
4 different elevations starting from directly pointing to antenna broadside and then descending in 12.5◦
steps. This leads to a total of 28 beams in the TX side. On the other hand, the RX needs to steer up and
75

down in the elevation domain. Again, the 7 existing azimuth scans are preserved but this time they are
repeated for 7 elevations from +30◦
to -30◦
every 10◦
. This brings the total beams for the receiver (across
4 panels) to 196. A full MIMO snapshot is equal to 5488 angular combinations scanning in both azimuth
and elevation.
5.1.1 Sounder Modifications
The scheduling diagram shown in Fig. 2.11 allows a maximum of 28 beam on the TX and 28 beams on
the RX end of the sounder. Two principles are used to allow the required MIMO snapshot structure for
the microcell case. In one hand, the MRK cycle is kept but this time only one panel is active on the TX
side. The rest of the time the panel is either loading its beam (roughly 50% of the time) or idle (around
25% of the time). This is shown in the upper part of Fig. 5.1, with the modified MRK cycle. On the other
hand, the overall event grid from the timing structure of the sounder is kept, but trigger pulses are added
or suppressed in order to make room for a longer MIMO snapshot. This corresponds to the second level
of Fig. 5.1 where the pink time slot at the end of 7 TX beams is silent for 3 repetitions and active on the
4th one, when all 28 TX beams have been measured. This gives the resulting block in the 3rd level from
top to bottom in the diagram and this block is 1 of 49 blocks were the RX changes beam once. The second
to last level of Fig. 5.1 shows, finally, a full MIMO snapshot. This snapshot has a duration of 67.62 ms so it
can be repeated 14 times per second (bottom level of Fig. 5.1).
In addition to the timing modifications, a new sounder calibration is performed. The overall procedure
to get the system calibration is identical to the one described in Chapter 2, but this time 28 TX beams, and
49 RX beams per panel need to be calibrated inside the anechoic chamber. This is equivalent to getting the
pattern of 224 antennas. Once the radiation patterns for each beam on each array panels are obtained, the
full response of the system can be calculated and then removed from measured data.
76

5.1.2 Scenarios
Measurements took place in the University of Southern California University Park Campus, Los Angeles,
CA. Four different trajectories representative of a UE moving inside a mmWave microcell are measured.
Fig. 5.2 shows these routes: a straight trajectory (route 1) along a street with LOS and vegetation-obstructed
locations (Fig. 5.2 (a) left); the same street measured on the opposite side of the street (route 2) where there
is less vegetation blockage, but this time turning around the corner into vegetation blockage (Fig. 5.2 (a)
right); a trajectory along a street and a turn around the corner (route 3) for building blockage ((Fig. 5.2
(b))); and a mostly LOS trajectory (route 4) along a street (Fig. 5.2 (c)).
The TX was placed on a parking structure buildings with wide wall openings to avoid obstruction,
although some tree branches might obstruct specific measured points. The TX height was between 11
and 14 m above street level, which was below rooftop height of the surrounding buildings, in line with
the standard definition of microcells. The RX is placed at points along the trajectory that are 0.5 to 1 m
apart, from a minimum TX-RX distance of 30 m to a maximum of 150 m. Note that due to the longer
measurement time for each MIMO snapshot, a continuous movement of the RX is not possible in this
setup. Fig. 5.2 shows the trajectories, TX placements and general building configuration of the measured
areas.
5.2 Data and Results
The captured data are processed in order to remove the system response and thus obtain the channel
response for the 566 points measured. At each point, 1 GHz bandwidth was measured resulting in 0.3 m
delay resolution, and 5488 angular combinations are recorded. PDPs, Path Gain and APSs are discussed
next. At each point, 10 repetitions of the sounding sequence were measured. This can help get 10 dB
77

Time (us) 0 2.7 5.4 8.1 11 14 16 19 22 24 27 30 32 35 38 41 43
TX1 L W W W W W W W W W L L L L L L L W 2 TX BEAM_TRIG
TX2 L W W W W W W W W W L L L L L L L W 1 MRK1 pulse
TX3 L W L L L L L L L W 1 1 1 1 W W W W
TX4 L W L L L L L L L W W W W W W W W W
RX1-VL L 1 W W W 1 W W W 1 W W W 1 W W W
RX2-VL L W 1 W W W 1 W W W 1 W W W 1 W W
RX3-VL L W W 1 W W W 1 W W W 1 W W W 1 W
RX4-VL L W W W 1 W W W 1 W W W 1 W W W 1
TX(1) RX(1) TX(2) RX(1) TX(3) RX(1) TX(4) RX(1) TX(5) RX(1) TX(6) RX(1) TX(7) RX(1) 14 TX BEAM_TRIG
0.3 7 MRK1 pulse
1 AWG_TRIG
56 TX BEAM_TRIG MRK1 x7 MRK1 x7 MRK1 x7 MRK1 x7 MRK1 x7 MRK1 x7 MRK1 x7
28 MRK1 pulse TX BT x14 TX BT x14 TX BT x14 TX BT x14 TX BT x14 TX BT x14 TX BT x14
4 AWG_TRIG
1 RX BEAM_TRIG
392 TX BEAM_TRIG
196 MRK1 pulse
28 AWG_TRIG
7 RX BEAM_TRIG
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 2744 TX BEAM_TRIG
1372 MRK1 pulse
196 AWG_TRIG
49 RX BEAM_TRIG
1 MIMO SNAP
1 2 3 4 5 6 7 8 9 10 11 12 13 14 38416 TX BEAM_TRIG
19208 MRK1 pulse
2744 AWG_TRIG
686 RX BEAM_TRIG
14 MIMO SNAP
1 PPS
RX(1)
1380
2415
946680
1000000
9660
67620
RX(2) RX(3) RX(4) RX(5) RX(6) RX(7)
RX(1)
1380 345
2.415 ms waveform 2.415 ms waveform 2.415 ms waveform 2.415 ms waveform
43.2
304.5
345
2.415 waveform
1 2 3 4
Figure 5.1: Marker cycle for microcell measurements.
(a) (b) (c)
Figure 5.2: Trajectories for the microcell measurement campaign.
78

(a) (b)
Figure 5.3: Example PDP and synthetic PDP omni. (a) is a PDP. (b) is a synthetic omnidirectional PDP.
additional SNR by averaging the results. The evaluations presented here are preliminary, and more detailed
investigations will be done in future work (see also Chapter 6).
5.2.1 PDP
For each of the 566 measured points the PDP for each angular (azimuth/elevation) combination can be calculated. The mmWave microcell channel is in general very sparse, according to these measurements. As
an example, a PDP for one of the measured points is shown. The omnidirectional PDP is also synthesized
[18]. Note that in the omnidirectional case, the noise floor is higher than in the singular angular combination case. This is due to the use of the max function in the procedure to generate the omnidirectional PDP
[15]. Outside the range shown in 5.3, there are no other significant multipath components in this case.
5.2.2 Path Gain
From the collection of PDPs for each measured point, the measured path gain is obtained. For each measured point, the distance is recorded and then the sum of all MPC over powers over the noise threshold
[15] is calculated. The results are condensed in Fig. 5.4. The routes are shown separately and also the free
space gain as calculated from Friis’ formula.
79

Figure 5.4: Measured path gain for all routes.
5.2.3 APS
The APSs are also calculated. Each measured point consists of 5488 angular combinations. As an example,
and to preserve space, only one of them is shown. Fig. 5.5 shows all azimuth/elevation combinations for a
single measurement point. As already seen in the omni PDP case, the channel appears to be very sparse.
80

(a) (b) (c)
(d) (e) (f)
(g)
Figure 5.5: APS snapshots for microcell measurement route 1. On each subplot, the horizontal axis corresponds to azimuth AoA while the vertical axis is azimuth AoD. Rows of subplots correspond to a fixed
elevation on the transmit side. For a particular AoA/AoD combination, the path gain is shown as a colormapped intensity. The multiple plots correspond to fixed RX elevations: (a) -30◦
. (b) -20◦
. (c) -10◦
s. (d) 0◦
.
(e) 10◦
. (f) 20◦
. (g) 30◦
.
81

Chapter 6
Conclusions and Future Work
This thesis presented an ultrawideband, double-directional channel sounder for the 28 GHz band that
can operate in both compact and distributed spatial modes. The sounder is based on the switched-beam
principle, using phased arrays that are controlled by low-cost non-proprietary timing circuitry.
The sounder can acquire the channel response for up to 768 beam combinations in 2.4 ms. In the current
configuration the sounder scans 360 in azimuth on TX and RX side when in the compact configuration, or
3x90 TX and 360 in RX for the distributed case. These capabilities make it well suited for measurement of
channels along trajectories, enabling to both acquires large (statistically relevant) amounts of data, and to
investigate the validity of wide-sense stationarity and uncorrelated scattering assumption. The distributed
setup enables verification of both trilateration for localization and the effectiveness of various Coordinated
Multi-Point (CoMP) schemes. We will perform such measurement campaigns in future work.
The overall structure is flexible and numerous tradeoffs could be implemented. Firstly, the number
of beams can be increased at the expense of an increase in the scanning time. This will be relevant for
extending the sounder for dual polarization measurements or to implement scanning in elevation angle. If
the resulting scan time exceeds the coherence time of the channel, one can either (i) use multiple digitizers
to receive the signals from the different RX boards in parallel, or (ii) use non-uniform scanning to eliminate ambiguities in the Doppler domain [47]. The phase stability of the sounder allows high-resolution
82

parameter estimation to evaluate the measurement results. As an example, the MUSIC algorithm was
implemented to show how the angular resolution can be improved beyond Fourier limits.
The sounder was then taken to real scenarios for localization to gather statistically sufficient amounts of
data on 3 different environments representative of these use cases. The scenarios contain a varied mixture
of LOS and NLOS situations. A first-in-kind measurement campaign of almost 3 minutes of gapless data
for a 3 anchor system steering in azimuth helps understand the localization channel in trilateration and
single anchor cases. This data will help the development of localization algorithms that exploit the specific
propagation characteristics of the mmWave band.
The sounder was then modified to operate with elevation steering. This was a necessary step in order
to enable microcell communication measurements in the mmWave band. The sounder was deployed on
4 routes representative of microcell scenarios where there is a base station at height that is surrounded
by streets. Different conditions for LOS and NLOS links were measured, with and without vegetation and
building blockage. A total of 566 points were measured, with each point having 5488 angular combinations
to study the directional characteristics of this channel. The delay resolution is 0.3 m and the delay range
is around 800 m.
Future work will fall into two categories: (i) the acquired data can be evaluated in greater detail, including with high resolution parameter estimation, to obtain additional insights of the channels and exploit
the large amount of raw data collected. (ii) the sounder will be used in future explorations of the mmWave
band in order to establish the optimal ways of operation in it for communications and localization. Possible use scenarios include D2D measurements, Indoor-to-indoor channels, Outdoor-to-indoor channels, and
vehicular communications. Additionally, the data acquired in these future campaigns will enable multiple
studies and algorithm development including modelling the communication channel, model validation,
capacity modelling, cooperative multi-point communications, channel stationarity, among others.
83

Bibliography
[1] url:
https://www.st.com/en/microcontrollers-microprocessors/stm32-32-bit-arm-cortex-mcus.html.
[2] Naveed A. Abbasi, Jorge Gomez-Ponce, Revanth Kondaveti, Ashish Kumar, Eshan Bhagat,
Rakesh N. S. Rao, Shadi Abu-Surra, Gary Xu, Charlie Zhang, and Andreas F. Molisch.
“Double-Directional Channel Measurements for Urban THz Microcellular Communications in a
Street Canyon”. In: ICC 2022 - IEEE International Conference on Communications. 2022,
pp. 2876–2881. doi: 10.1109/ICC45855.2022.9838551.
[3] Constantine A Balanis. Antenna theory: analysis and design. John wiley & sons, 2016.
[4] Celalettin Umit Bas, Rui Wang, Dimitris Psychoudakis, Thomas Henige, Robert Monroe,
Jeongho Park, Jianzhong Zhang, and Andreas F Molisch. “A real-time millimeter-wave phased
array MIMO channel sounder”. In: 2017 IEEE 86th Vehicular Technology Conference (VTC-Fall).
IEEE. 2017, pp. 1–6.
[5] Celalettin Umit Bas, Rui Wang, Seun Sangodoyin, Dimitris Psychoudakis, Thomas Henige,
Robert Monroe, Jeongho Park, Charlie Jianzhong Zhang, and Andreas F. Molisch. “Real-Time
Millimeter-Wave MIMO Channel Sounder for Dynamic Directional Measurements”. In: IEEE
Transactions on Vehicular Technology 68.9 (2019), pp. 8775–8789. doi: 10.1109/TVT.2019.2928341.
[6] Xuesong Cai, Erik L. Bengtsson, Ove Edfors, and Fredrik Tufvesson. “A Switched Array Sounder
for Dynamic Millimeter-Wave Channel Characterization: Design, Implementation, and
Measurements”. In: IEEE Transactions on Antennas and Propagation 72.7 (2024), pp. 5985–5999. doi:
10.1109/TAP.2024.3412187.
[7] Derek Caudill, Jack Chuang, Sung Yun Jun, Camillo Gentile, and Nada Golmie. “Real-Time
mmWave Channel Sounding Through Switched Beamforming With 3-D Dual-Polarized
Phased-Array Antennas”. In: IEEE Transactions on Microwave Theory and Techniques 69.11 (2021),
pp. 5021–5032. doi: 10.1109/TMTT.2021.3104278.
[8] Derek Caudill, Peter B. Papazian, Camillo Gentile, Jack Chuang, and Nada Golmie.
“Omnidirectional Channel Sounder With Phased-ArrayAntennas for 5G Mobile Communications”.
In: IEEE Transactions on Microwave Theory and Techniques 67.7 (2019), pp. 2936–2945. doi:
10.1109/TMTT.2019.2910109.
84

[9] Aditya Chopra, Andrew Thornburg, Ojas Kanhere, Saeed S. Ghassemzadeh, Milap Majmundar, and
Theodore S. Rappaport. “A Real-Time Millimeter Wave V2V Channel Sounder”. In: 2022 IEEE
Wireless Communications and Networking Conference (WCNC). 2022, pp. 2607–2612. doi:
10.1109/WCNC51071.2022.9772001.
[10] Aditya Chopra, Andrew Thornburg, Ojas Kanhere, Abbas Termos, Saeed S. Ghassemzadeh, and
Theodore S. Rappaport. “Real-time Millimeter Wave Omnidirectional Channel Sounder Using
Phased Array Antennas”. In: GLOBECOM 2020 - 2020 IEEE Global Communications Conference.
2020, pp. 1–7. doi: 10.1109/GLOBECOM42002.2020.9322491.
[11] Jean-Marc Conrat, Patrice Pajusco, and Jean-Yves Thiriet. “A Multibands Wideband Propagation
Channel Sounder from 2 to 60 GHz”. In: 2006 IEEE Instrumentation and Measurement Technology
Conference Proceedings. 2006, pp. 590–595. doi: 10.1109/IMTC.2006.328629.
[12] Jinfeng Du, Dmitry Chizhik, Reinaldo A. Valenzuela, Rodolfo Feick, Guillermo Castro,
Mauricio Rodriguez, Tingjun Chen, Manav Kohli, and Gil Zussman. “Directional Measurements in
Urban Street Canyons From Macro Rooftop Sites at 28 GHz for 90% Outdoor Coverage”. In: IEEE
Transactions on Antennas and Propagation 69.6 (2021), pp. 3459–3469. doi:
10.1109/TAP.2020.3044398.
[13] Fatih Erden, Ozgur Ozdemir, and Ismail Guvenc. “28 GHz mmWave Channel Measurements and
Modeling in a Library Environment”. In: 2020 IEEE Radio and Wireless Symposium (RWS). 2020,
pp. 52–55. doi: 10.1109/RWS45077.2020.9050106.
[14] M. Friese. “Multitone signals with low crest factor”. In: IEEE Transactions on Communications 45.10
(1997), pp. 1338–1344. doi: 10.1109/26.634697.
[15] Jorge Gomez-Ponce, Naveed A. Abbasi, Zihang Cheng, Shadi Abu-Surra, Gary Xu,
Jianzhong Zhang, and Andreas F. Molisch. “Impact of Noisy Measurements With Fourier-Based
Evaluation on Condensed Channel Parameters”. In: IEEE Transactions on Wireless Communications
23.8 (2024), pp. 8136–8152. doi: 10.1109/TWC.2023.3342733.
[16] Jorge Gomez-Ponce, Daoud Burghal, Naveed A. Abbasi, Arjun Hariharan, Gopal Jakhetia,
Praveen Chaganlal, and Andreas F. Molisch. “Directional Delay Spread and Interference Quotient
Analysis in sub-7GHz Wi-Fi Bands”. In: GLOBECOM 2020 - 2020 IEEE Global Communications
Conference. 2020, pp. 01–06. doi: 10.1109/GLOBECOM42002.2020.9322252.
[17] Hussein Hammoud, Yuning Zhang, Zihang Cheng, Seun Sangodoyin, Hofer Markus, Faruk Pasic,
Thomas M. Pohl, Radek Závorka, Alex Prokes, Thomas Zemen, Christoph Mecklenbräuker, and
Andreas F. Molisch. “A Novel Low-Cost Channel Sounder for Double-Directionally Resolved
Measurements in the MmWave band”. In: 2024 IEEE International Conference on Communications
(ICC-Spring). IEEE. Submitted.
[18] Katsuyuki Haneda, Sinh Le Hong Nguyen, Jan Järveläinen, and Jyri Putkonen. “Estimating the
omni-directional pathloss from directional channel sounding”. In: 2016 10th European Conference
on Antennas and Propagation (EuCAP). 2016, pp. 1–5. doi: 10.1109/EuCAP.2016.7481455.
[19] F.J. Harris. “On the use of windows for harmonic analysis with the discrete Fourier transform”. In:
Proceedings of the IEEE 66.1 (1978), pp. 51–83. doi: 10.1109/PROC.1978.10837.
85

[20] Shihao Ju, Yunchou Xing, Ojas Kanhere, and Theodore S. Rappaport. “Millimeter Wave and
Sub-Terahertz Spatial Statistical Channel Model for an Indoor Office Building”. In: IEEE Journal on
Selected Areas in Communications 39.6 (2021), pp. 1561–1575. doi: 10.1109/JSAC.2021.3071844.
[21] Ojas Kanhere, Aditya Chopra, Andrew Thornburg, Theodore S. Rappaport, and
Saeed S. Ghassemzadeh. “Performance Impact Analysis of Beam Switching in Millimeter Wave
Vehicular Communications”. In: 2021 IEEE 93rd Vehicular Technology Conference (VTC2021-Spring).
2021, pp. 1–7. doi: 10.1109/VTC2021-Spring51267.2021.9448914.
[22] Mohsen Khalily, Sohail Taheri, Sohail Payami, Mir Ghoraishi, and Rahim Tafazolli. “Indoor
wideband directional millimeter wave channel measurements and analysis at 26 GHz, 32 GHz, and
39 GHz”. In: Transactions on Emerging Telecommunications Technologies 29.10 (2018). e3311
ett.3311, e3311. doi: https://doi.org/10.1002/ett.3311. eprint:
https://onlinelibrary.wiley.com/doi/pdf/10.1002/ett.3311.
[23] Junghoon Ko, Yeon-Jea Cho, Sooyoung Hur, Taehwan Kim, Jeongho Park, Andreas F. Molisch,
Katsuyuki Haneda, Michael Peter, Dong-Jo Park, and Dong-Ho Cho. “Millimeter-Wave Channel
Measurements and Analysis for Statistical Spatial Channel Model in In-Building and
UrbanEnvironments at 28 GHz”. In: IEEE Transactions on Wireless Communications 16.9 (2017),
pp. 5853–5868. doi: 10.1109/TWC.2017.2716924.
[24] Mengting Li, Fengchun Zhang, and Wei Fan. “Measurement-based Channel Characterization Using
Virtual Array with Directional Antenna”. In: 2023 17th European Conference on Antennas and
Propagation (EuCAP). 2023, pp. 1–5. doi: 10.23919/EuCAP57121.2023.10132976.
[25] Qian Ma, Hyunchul Chung, Yusheng Yin, Tumay Kanar, Samet Zihir, and Gabriel M. Rebeiz. “A 5G
24-30 GHz 2x32 Element Dual-Polarized Dual-Beam Phased Array Base-Station for 2x2 MIMO
System”. In: 2019 IEEE Global Communications Conference (GLOBECOM). 2019, pp. 1–5. doi:
10.1109/GLOBECOM38437.2019.9013793.
[26] George R. MacCartney and Theodore S. Rappaport. “A Flexible Millimeter-Wave Channel Sounder
With Absolute Timing”. In: IEEE Journal on Selected Areas in Communications 35.6 (2017),
pp. 1402–1418. doi: 10.1109/JSAC.2017.2687838.
[27] George R. MacCartney and Theodore S. Rappaport. “Millimeter-Wave Base Station Diversity for
5G Coordinated Multipoint (CoMP) Applications”. In: IEEE Transactions on Wireless
Communications 18.7 (2019), pp. 3395–3410. doi: 10.1109/TWC.2019.2913414.
[28] T. Manabe, K. Taira, K. Sato, T. Ihara, Y. Kasashima, and K. Yamaki. “Multipath measurement at 60
GHz for indoor wireless communication systems”. In: Proceedings of IEEE Vehicular Technology
Conference (VTC). 1994, 905–909 vol.2. doi: 10.1109/VETEC.1994.345222.
[29] Allan Wainaina Mbugua, Wei Fan, Yilin Ji, and Gert Frølund Pedersen. “Millimeter Wave
Multi-User Performance Evaluation Based on Measured Channels With Virtual Antenna Array
Channel Sounder”. In: IEEE Access 6 (2018), pp. 12318–12326. doi: 10.1109/ACCESS.2018.2812304.
[30] Andreas F. Molisch. Wireless Communications: From Fundamentals to Beyond 5G, 3rd Edition.
Wiley-IEEE Press, 2023.
86

[31] Alfred Mudonhi, Raffaele D’Errico, and Claude Oestges. “Indoor mmWave Channel
Characterization with Large Virtual Antenna Arrays”. In: 2020 14th European Conference on
Antennas and Propagation (EuCAP). 2020, pp. 1–5. doi: 10.23919/EuCAP48036.2020.9135628.
[32] Ahmed Nafe and Gabriel M. Rebeiz. “On the phase center analysis of linear phased-array
antennas”. In: 2017 IEEE International Symposium on Antennas and Propagation and USNC/URSI
National Radio Science Meeting. 2017, pp. 2023–2024. doi: 10.1109/APUSNCURSINRSM.2017.8073054.
[33] Nikolaos Ntetsikas, Nithin Babu, Muhammad Haroon Tariq, Constantinos B Papadias, Jinfeng Du,
Dmitry Chizhik, Reinaldo Valenzuela, Mauricio Rodriguez, and Rodolfo Feick. “60 GHz Outdoor to
Indoor (O2I) Propagation Measurements in a University Campus”. In: 2022 IEEE 23rd International
Workshop on Signal Processing Advances in Wireless Communication (SPAWC). 2022, pp. 1–5. doi:
10.1109/SPAWC51304.2022.9833968.
[34] Peter Papazian, Derek Caudill, Camillo Gentile, Jack Chuang, and Nada Golmie. “62.5-GHz
Phased-Array Channel Sounder for Double-Directional Angle Estimation”. In: 2019 13th European
Conference on Antennas and Propagation (EuCAP). 2019, pp. 1–5.
[35] 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. 2007, pp. 1–8. doi: 10.1049/ic.2007.1555.
[36] Sylvain Ranvier, Jarmo Kivinen, and Pertti Vainikainen. “Millimeter-Wave MIMO Radio Channel
Sounder”. In: IEEE Transactions on Instrumentation and Measurement 56.3 (2007), pp. 1018–1024.
doi: 10.1109/TIM.2007.894197.
[37] Theodore S. Rappaport, Felix Gutierrez, Eshar Ben-Dor, James N. Murdock, Yijun Qiao, and
Jonathan I. Tamir. “Broadband Millimeter-Wave Propagation Measurements and Models Using
Adaptive-Beam Antennas for Outdoor Urban Cellular Communications”. In: IEEE Transactions on
Antennas and Propagation 61.4 (2013), pp. 1850–1859. doi: 10.1109/TAP.2012.2235056.
[38] Ahmed Al-Saman, Michael Cheffena, Olakunle Elijah, Yousef A. Al-Gumaei,
Sharul Kamal Abdul Rahim, and Tawfik Al-Hadhrami. “Survey of Millimeter-Wave Propagation
Measurements and Models in Indoor Environments”. In: Electronics 10.14 (2021). issn: 2079-9292.
doi: 10.3390/electronics10141653.
[39] R. Schmidt. “Multiple emitter location and signal parameter estimation”. In: IEEE Transactions on
Antennas and Propagation 34.3 (1986), pp. 276–280. doi: 10.1109/TAP.1986.1143830.
[40] Anish Shastri, Neharika Valecha, Enver Bashirov, Harsh Tataria, Michael Lentmaier,
Fredrik Tufvesson, Michele Rossi, and Paolo Casari. “A Review of Millimeter Wave Device-Based
Localization and Device-Free Sensing Technologies and Applications”. In: IEEE Communications
Surveys & Tutorials 24.3 (2022), pp. 1708–1749. doi: 10.1109/COMST.2022.3177305.
[41] Gustavo Siles, Jose Riera, and Pedro García-del-Pino. “Atmospheric Attenuation in Wireless
Communication Systems at Millimeter and THz Frequencies [Wireless Corner]”. In: IEEE Antennas
and Propagation Magazine 57 (Feb. 2015), pp. 48–61. doi: 10.1109/MAP.2015.2401796.
87

[42] Christopher Slezak, Aditya Dhananjay, and Sundeep Rangan. “60 GHz blockage study using
phased arrays”. In: 2017 51st Asilomar Conference on Signals, Systems, and Computers. 2017,
pp. 1655–1659. doi: 10.1109/ACSSC.2017.8335640.
[43] Christopher Slezak, Vasilii Semkin, Sergey Andreev, Yevgeni Koucheryavy, and Sundeep Rangan.
“Empirical Effects of Dynamic Human-Body Blockage in 60 GHz Communications”. In: IEEE
Communications Magazine 56.12 (2018), pp. 60–66. doi: 10.1109/MCOM.2018.1800232.
[44] Ruoyu Sun, Peter B. Papazian, Jelena Senic, Yeh Lo, Jae-Kark Choi, Kate A. Remley, and
Camillo Gentile. “Design and calibration of a double-directional 60 GHz channel sounder for
multipath component tracking”. In: 2017 11th European Conference on Antennas and Propagation
(EUCAP). 2017, pp. 3336–3340. doi: 10.23919/EuCAP.2017.7928270.
[45] Peter Vouras, Benjamin Jamroz, Alec Weiss, Dylan F. Williams, Rodney Leonhardt, Damla Guven,
Robert Jones, Josh Kast, Kate A. Remley, Robert D. Horansky, and Camillo Gentile. “Wideband
Synthetic-Aperture Millimeter-Wave Spatial-Channel Reference System With Traceable
Uncertainty Framework”. In: IEEE Open Journal of Vehicular Technology 4 (2023), pp. 325–341. doi:
10.1109/OJVT.2023.3239617.
[46] Rui Wang, Celalettin Umit Bas, Zihang Cheng, Thomas Choi, Hao Feng, Zheda Li, Xiaokang Ye,
Pan Tang, Seun Sangodoyin, Jorge Gómez-Ponce, et al. “Enabling super-resolution parameter
estimation for mm-wave channel sounding”. In: IEEE Transactions on Wireless Communications
19.5 (2020), pp. 3077–3090.
[47] Rui Wang, Olivier Renaudin, C. Umit Bas, Seun Sangodoyin, and Andreas F. Molisch. “On Channel
Sounding With Switched Arrays in Fast Time-Varying Channels”. In: IEEE Transactions on Wireless
Communications 18.8 (2019), pp. 3843–3855. doi: 10.1109/TWC.2019.2919023.
[48] Alec J. Weiss, Jeanne Quimby, Rod Leonhardt, Ben Jamroz, Dylan Williams, Kate Remley,
Peter Vouras, and Atef Elsherbeni. “Configuration and Control of a Millimeter-Wave Synthetic
Aperture Measurement System with Uncertainties”. In: 2020 95th ARFTG Microwave Measurement
Conference (ARFTG). 2020, pp. 1–4. doi: 10.1109/ARFTG47271.2020.9241381.
[49] Fuxi Wen, Henk Wymeersch, Bile Peng, Wee Peng Tay, Hing Cheung So, and Diange Yang. “A
survey on 5G massive MIMO localization”. In: Digital Signal Processing 94 (2019). Special Issue on
Source Localization in Massive MIMO, pp. 21–28. issn: 1051-2004. doi:
https://doi.org/10.1016/j.dsp.2019.05.005.
[50] Yusheng Yin, Zhe Zhang, Tumay Kanar, Samet Zihir, and Gabriel M. Rebeiz. “A 24-29.5 GHz
256-Element 5G Phased-Array with 65.5 dBm Peak EIRP and 256-QAM Modulation”. In: 2020
IEEE/MTT-S International Microwave Symposium (IMS). 2020, pp. 687–690. doi:
10.1109/IMS30576.2020.9224031.
88

Abstract (if available)

Abstract The mmWave channel offers significant advantages for both communication and localization applications of wireless systems. For this reason, the mmWave band is a key component of fifth generation(5G) and future sixth generation (6G) communication systems. The large bandwidth available at mmWave frequencies provides the potential for ultra-high data rate communications; it also results in higher delay resolution that greatly improves the accuracy of localization when compared against sub-6GHz systems. As for all wireless communication and localization systems, also those operating in the mmWave band require deep knowledge of the physical characteristics of the channel to ensure good performance. The channel characteristics create significant channel-sounding challenges in the case of mmWave, since the channel exhibits fast fluctuations and losses can be higher than in sub-6GHz bands if antenna gain is not exploited. More generally, there is a fundamental technological trade-off for existing sounders in the gain-speed-bandwidth triangle. In this work, we developed a state-of-the-art double-directional phased array channel sounder that has high EIRP, is real-time and can capture mmWave channel dynamics, and has a bandwidth of 1 GHz. With this, we have captured statistically significant amounts of channel data that can address the problems of mmWave communications modelling for single link and cooperative multipoint systems, as well as localization with and without multiple anchor nodes.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Signal processing for channel sounding: parameter estimation and calibration

PDF

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

PDF

Double-directional channel sounding for next generation wireless communications

PDF

Channel sounding for next-generation wireless communication systems

PDF

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

PDF

Propagation channel characterization and interference mitigation strategies for ultrawideband systems

PDF

Hybrid beamforming for massive MIMO

PDF

Enabling virtual and augmented reality over dense wireless networks

PDF

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

PDF

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

PDF

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

PDF

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

PDF

High capacity optical and RF/mm-wave communications utilizing orbital angular momentum multiplexing

PDF

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

PDF

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

PDF

Design and analysis of reduced complexity transceivers for massive MIMO and UWB systems

PDF

Exploiting side information for link setup and maintenance in next generation wireless networks

PDF

Achieving high data rates in distributed MIMO systems

PDF

Multichannel data collection for throughput maximization in wireless sensor networks

PDF

Enhancing collaboration on the edge: communication, scheduling and learning

Asset Metadata

Creator Castro Carranza, Guillermo Andres (author)

Core Title mmWave dynamic channel measurements for localization and communications

School Viterbi School of Engineering

Degree Doctor of Philosophy

Degree Program Electrical Engineering

Degree Conferral Date 2024-12

Publication Date 09/26/2024

Defense Date 08/30/2024

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

Tag 6G mobile communication,channel sounding,distributed MIMO,millimeter wave communication,Millimeter wave measurements,OAI-PMH Harvest,wireless communications

Format theses (aat)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Molisch, Andreas F. (committee chair), Govindan, Ramesh (committee member), Sideris, Constantine (committee member)

Creator Email gcastro6@usc.edu,visques@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC11399B9YO

Unique identifier UC11399B9YO

Identifier etd-CastroCarr-13552.pdf (filename)

Legacy Identifier etd-CastroCarr-13552

Document Type Dissertation

Format theses (aat)

Rights Castro Carranza, Guillermo Andres

Internet Media Type application/pdf

Type texts

Source 20240926-usctheses-batch-1214 (batch), 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 author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.

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

Repository Email cisadmin@lib.usc.edu