POWDER-RENEW

Mobile and Wireless Week 2023

Mini-project presentations

References

[1]. Qian et al, “MilliMirror: 3D Printed Reflecting Surface for Millimeter-Wave Converage Expansion” , ACM MobiCom 22.

​

POWDER-Twin

Maxwell McManus, Yuqing Cui, Josh Zhang, Tahenan H S

State University of New York at Buffalo

{memcmanu, yuqingcu, zhaoxizh, tahenanh}@buffalo.edu

​

POWDER Mobile Wireless Week (MWW) 2023

January 27th, 2023

Overview: Digital Twin

Motivation

• Relax resource constraints
• No limits on spectrum use
• Only limited by computational capabilities (no need to share radios!)
• No risk to hardware during policy iteration
• Accelerating AI for wireless
• Learning algorithms can take 1000’s of iterations (30 min vs 3 hrs)
• Train on multiple scenarios simultaneously

​

Goal: develop digital twin of POWDER platform and generate benchmarks to accelerate future research using POWDER.

POWDER-Twin

UBSim

Universal Broadband Simulator

• Open source - written in Python 3.8
• Fully configurable between microwave, millimeter-wave, and THz bands
• Static, mobile ground, and flying node types
• Default protocols: LTE, multi-user CDMA, CW
• Low fidelity - analytical model with limited physics
• NET-layer simulation with estimated PHY/MAC behaviors
• High speed - low fidelity allows for accelerated data generation
• Accelerates control policy generation (30 min vs 3 hrs on testbed)

​

POWDER-Twin

Digital Twin of UB-NeXT

Snapshot of UB-NeXT testbed.

UBSim GUI of virtual model.

POWDER-Twin

Digital Twin of SOAR

Snapshot of UB SOAR facility.

UBSim GUI of virtual model.

POWDER-Twin

Digital Twin Design Process

1. Campus layout
• Building locations, dimensions (UBSim can only recognize AABB)
• Taken from: https://powderwireless.net/map
2. Environment topography
• Elevation slices based on GIS data
3. POWDER deployment
• Node type and coordinates
• Copy TX/RX parameters in netconfig API
4. Generate benchmarks
• Compare OTA data with simulation results for various node configurations
• Limited to SINR based on time/resource availability

POWDER-Twin

Physical Environment Definition

Color-coded elevation map. [1]

Generate AABB set based on map geometry.

Rotate map so majority of buildings are axis-aligned.

Start with POWDER real-time deployment map.

POWDER-Twin

[1] Utah Geological Survey topographic maps: https://ngmdb.usgs.gov/topoview/viewer/#15/40.7649/-111.8436

POWDER-Twin Deployment

AABB map imported to UBSim with node locations.

AABB map of all relevant campus buildings.

Performance measurements

Measured SINR (dBW) between selected rooftop nodes (3425 MHz CW):

POWDER-Twin

Performance measurements

Measured SINR (dBW) between rooftop nodes and endpoints (3425 MHz CW):

POWDER-Twin

Raw Data

Conclusions and Future Work

POWDER-Twin can improve offline experience with POWDER platform

• UBSim can provide experimental data when platform is unavailable
• Converged policies, experimental targets can be tested on hardware when available

Future work

• Expand POWDER-Twin to include protocol benchmarks
• Reconfigure existing profile (“shout-iface-node” or “shout-mww2023”)
• Add LoRa model to UBSim
• Experimental breadth improves robustness of POWDER-Twin
• Add support for OBB and AABB in UBSim to improve accuracy

POWDER-Twin

Skier: Powder Automation

Jacob Bills: University of Utah

Project Goal

• Resources, Resources, Resources…
• Able to start profile automatically
• Setup for scheduled collection
• Collect a 10 minutes of iq data
• Process data looking for radar signals
• Log current weather conditions and freq
• Move captured data to long term storage in powder
• Release rooftop back for other users

Project Goal

• Used python API for auto start
• Instantiate profile daily if resources available
• Detects startup failures and recovers
• Terminates after workflow completion releasing resources
• Startup script for workflow
• Works when run by hand
• Does not successfully complete the RX capture for some reason
• Logs weather conditions from openWeather API (600 today)
• Uses FFT to determine operating frequency (2855MHz)

Larger Integration

• Automatic Radar RX
• Capture state of radar daily
• Log weather and frequency
• Create database over several months
• Spectrum Sharing
• Use observations to look for longterm spectrum sharing opportunities with NexRad Radar between 2.7-2.9GHz
• Develop monitoring mechanism to ensure POWDER usage doesn’t interfere with KMTX installation NW of SLC

Massive MIMO with Low-Resolution ADC’s

Abhijith Atreya, University of California, Santa Barbara

Canan Cebeci, University of California, Santa Barbara

Motivation

• For operating in mmWave and sub-THz bands, massive MIMO’s need hundreds of antennas
• More ADC output bits require more power
• Decreasing ADC bits reduces energy consumption and hardware complexity while introducing severe nonlinearity
• Struder et al. claim : For 1-bit ADC case, estimating channel with LS and MRC or ZF detection is sufficient for supporting reliable multi-user transmission.

​

​

BEACHES: Beamspace Channel Estimation for Multi-Antenna mmWave Systems and Beyond, Struder et al.

Massive MIMO with Low-Resolution ADC’s, Abhijith, Canan

​

​

Results

Massive MIMO with Low-Resolution ADC’s, Abhijith, Canan

​

Next Steps

• Increase the number of
• pilot symbols
• antennas

​

• Change SNR (decreasing SNR is shown to help in severely quantized systems), modulation type, detection technique (try a nonlinear detection technique).

​

Massive MIMO with Low-Resolution ADC’s, Abhijith, Canan

​

Mini-Project: Commercial LTE Transmitter Localization

Team NightOwls: Nishant, Sravan, Sergei, Keerthana, Bhaskar�MWW2023, January 2023

NightOwls Localization

Goal: Localize Surrounding LTE Base Stations

31

AT&T

NightOwls Localization

Our Approach

32

NightOwls Localization

AT&T cell tower locations

Chosen cellsdr rooftop nodes

LTE band 2

Downlink center freq = 1.98 GHz

Cell bandwidth = 20 MHz

​

• Obtained AT&T cell tower coordinates from CellMapper
• Custom POWDER profile to receive LTE band 2 signals
• Recording for 0.2 s
• TDOA localization algorithm
• Different combinations of 3 rooftop nodes
• Different time-series lengths

Results

33

NightOwls Localization

Received signal power spectral densities

Bandwidth = 20 MHz

Rooftop node combination 1

Results

34

NightOwls Localization

Received signal power spectral densities

Bandwidth = 20 MHz

Rooftop node combination 2

AT&T cell tower locations

Chosen cellsdr rooftop nodes

Moving Forward …

• Datasets over longer time durations & more diverse collection geometries
• Datasets with “ground truth” cell tower locations
• Algorithms to localize in presence of multipath & multiple, unknown sectored transmissions
• Localization accuracy comparison across multiple bands
• “Bandwidth stitching” to improve localization accuracy

Vision: Opportunistic positioning & navigation with ambient OTA signals without GPS or being subscribed to network

35

NightOwls Localization

Measurements

Ashton Palacios [Brigham Young University]

Nawel Alioua [UC Santa Barbara]

Alicia Esquivel [University of Missouri - Columbia]

POWDER-RENEW Mobile and Wireless Week 2023

Measurements

Goals

• Investigate performance of LoRa transmissions on the POWDER network
• Frame Reception
• Received Signal Strength

37

Measurements

Testbed deployment details

• First configuration:
• TX: browning
• RX: ustar, smt, garage

​

• Second configuration:
• TX: ustar
• RX: browning, smt, garage

​

• Third configuration:
• TX: smt
• RX: browning, ustar, garage

​

38

• Transmitting 50 frames

Measurements

Results

39

Measurements

Results

40

Measurements

Results

41

Measurements

Results

42

Measurements

Conclusion and future work

• Investigate performance of LoRa transmissions on the POWDER network
• Frame Reception → Success
• Received Signal Strength → Success

​

• Signal Strength Based Localization Using Gaussian Process Regression in GPS-denied Environments

43

Dynamic Slice Allocation to Ensure Promised Quality of Service

01/27/2023

Syed Ali Nawazish, University of Utah

Muhammad Basit Iqbal, University of Utah

Raja Hasnain Anwar, University of Arizona

Project/Team Name: Shaheens (7)

Project/Team Name: Shaheens (7)

Motivation

• Weather conditions induce great levels of noise.
• Throughput drops, and potentially starves critical applications
• We need dynamic slice allocation to prioritize applications based on their demand.
• Deliver promised Quality of Service (QoS)
• AR/VR and 5G have enabled fast communication and remote delivery of services.
• The future promises even more reliance on remote services that are extremely time-critical.
• Defense
• Healthcare
• Emergency Services
• Our Solution: rApp-based closed loop policy.
• Weather-sensitive slice allocation
• Prioritize time-critical applications (UEs) after interference

Project/Team Name: Shaheens (7)

Experiment Setup

Configuration

• Profile: O-RAN
• Hardware: d430
• RIC Release: dawn
• No. UEs: 1

Implementation

• Deployed NexRAN
• Metrics: DL-bytes
• Additive White Gaussian Noise
• Updated Closed-loop Policy
• policy.cc
• Python Script

Project/Team Name: Shaheens (7)

Results

Project/Team Name: Shaheens (7)

Slice allocation in Normal Conditions – old policy.

Slice allocation in Noisy Conditions – old policy.

Slice Adjustment

Results

Project/Team Name: Shaheens (7)

Slice allocation per old policy.

Slice allocation per our policy.

Slice Adjustment

Future Directions

• Multiple UEs with dynamic demand
• Over the Air execution
• Additional metrics for more intelligent resource allocation
• Scheduling for multiple critical applications

Project/Team Name: Shaheens (7)

Thank you!

Q & A

Gold Sequence Measurements

Davide Villa, Gabriele Gemmi, Maria Tsampazi

Overview and Objectives

• Goal: POWDER Coverage Map
• Static and Mobile Nodes

​

• Tool: Shout Framework
• Gold Sequences
• With some changes: TX/RX Gains, Power Amplifiers.

​

POWDER Ideal Coverage Map

Accomplishments and Issues

• Only a subset of nodes available (3 RTs, 3 Denses).

Accomplishments and Issues

• Only a subset of nodes available (3 RTs, 3 Denses).

Next Steps and Long-Term Goals

• Get all nodes measurements.
• Leverage Shout on our emulator testbed at NEU (Colosseum).
• Create a channel model of POWDER with ray-tracing.
• Validate the model using Shout.

​

​

Thank You!

Davide Villa, Gabriele Gemmi, Maria Tsampazi

ML Based Adaptive Modulation

for Massive MIMO

Rice Warriors

Mehdi, Qing An, Jingyi Miao

Motivation and Background

• What is MCS?
• Modulation and coding scheme
• Why not fixed MCS?
• Time-varying channels
• Why choose adaptive MCS?
• Adapting the MCS to the instantaneous quality of the radio link
• Objective: higher throughput

Motivation and Background

[1] E. Bobrov, D. Kropotov, H. Lu and D. Zaev, "Massive MIMO Adaptive Modulation and Coding Using Online Deep Learning Algorithm," in IEEE Communications Letters, vol. 26, no. 4, pp. 818-822, April 2022, doi: 10.1109/LCOMM.2021.3132947.

Deep Q-learning (DQN) and DNN Model

• DQN model
• Specifics:
• State Space: raw CSI (real and imaginary parts)
• Action Space: modulation selection (from QPSK, 16-QAM and 64-QAM)
• Reward: throughput
• DNN:
• Specifics:
• Data: raw CSI (real and imaginary parts)
• Label of 3: QPSK, 16-QAM and 64-QAM

​

​

Experiment Setup

• Experiment Platform:
• RENEW massive MIMO
• System Configuration:
• Network configuration: 64 BS antennas & 4 UEs
• Modulation: QPSK, 16-QAM and 64-QAM
• Datasets:
• More than 20 datasets (>32GB), collecting different channels
• More than 45,000 observations for training & test
• ML Hyper-parameters:

​

Performance Comparison

• Channel frequency response

​

Distribution of Throughput

QPSK

16QAM

64QAM

Model Training Performance

• DQL Convergence
• Epsilon-greedy algorithm
• Epochs of 200 and converges at 100 epochs
• DNN Loss Decay
• Epochs of 500

​

Performance Comparison

• Results:
• DNN and DQN outperform the performance of all fixed modulations
• Both ML models can effectively learn the adaptive modulation pattern
• Choosing the best modulation based on the channel response

Thank you!

Localize Cellular Tower by TDoA in POWDER

Localize Cellular Tower by TDoA in POWDER

Why choice?

​

• Interest of our research and

​

Goal:

​

• the accuracy of locking a signal source by TDoA when the SDR are highly synchronized in an outdoor scenario.

​

​

​

Overview of this mini-project

Localize Cellular Tower by TDoA in POWDER

Workflow

• Build on the tutorial given at ‘Localization session’

​

• Circle three rooftop radio (hospital, SMT, Ustar) covered by the same cellular tower.

​

• Utilize POWDER platform to get the reception from the selected CBRS radio.

​

• Process data to compute DToA and target the source.

Localize Cellular Tower by TDoA in POWDER

Our Step-in for data processing

Localize Cellular Tower by TDoA in POWDER

Localization result from over-the-air

Single localization

Multi localization

Localize Cellular Tower by TDoA in POWDER

Future work

​

• Motivated by the results of multi-test, set threshold of RSS and correlation to filter out unreasonable position.

​

• apply clustering method to locate more than one signal source.

​

• Localize a known signal source if the spectrum available…

​

Localize Cellular Tower by TDoA in POWDER

​

Thank you

​

POWDER team and all attendees

5G Physical Downlink Shared Channel (PDSCH) Over-the-Air (OTA) Measurement Aided by Wi-Fi Preamble

Zhihui Gao, Lyutianyang Zhang, Liu Cao

Jan. 27th, 2023

UW Huskies

Background and Motivation

​

• The PDSCH demodulation in 5G requires precise alignment between TX and RX in
• a) time domain (5us) and
• b) frequency domain (100Hz).
• A bad alignments on the captured 5G waveform incurs serious inter-symbol interference (ISI) and inter-carrier interference (ICI), respectively,
• and thus result in high error vector magnitude (EVM) and bit error rate (BER).
• However, traditional 5G protocols requires heavy overheads.
• A novel algorithm aided by Wi-Fi preambles on the TX and RX on the POWDER testbed is yet to be proposed.

​

UW Huskies

System Design

PDSCH

PDSCH

DMRS

DMRS

PDSCH

PDSCH

PDSCH

PDSCH

PDSCH

PDSCH

DMRS

DMRS

PDSCH

PDSCH

Wi-Fi Preamble

Wi-Fi Preamble

Wi-Fi Preamble

Wi-Fi Preamble

Time

UW Huskies

Time Alignment

UW Huskies

Frequency Alignment

• Higher Resolution -> Higher Error
• Lower Resolution -> Lower Error

Error

Resolution

UW Huskies

Experiment Setup

• POWDER setups
• Profile: gnuradio_otalab
• Radio: 2 USRP X310 (TX and RX)
• Node: d740
• Wireless setups
• Gain: 10dB
• Carrier Frequency: 3GHz
• Sampling Rate: 10MHz
• 5G setups
• Modulation: 64QAM
• Code Rate: 490/1024
• Bandwidth: 52RBs in MU1

UW Huskies

Experiment Pipeline

tx_samples_from_file

rx_samples_to_file

UW Huskies

Emulation Result

• SNR 14.9712dB
• EVM 44.6556%
• BER 15.81%

​

To have a better time alignment performance and a better demodulation result, we increase the signal power by 8 dB

​

​

UW Huskies

Simulation Result

• SNR 22.9628dB
• EVM 12.8332%
• BER 0.21%

The simulation result has also been validated by the data collected from Duke University X310 setup in both FR1 and FR2.

UW Huskies

Thoughts for POWDER

• Displaying the antenna properties attached to each port (TX/RX, RX2), including the supported frequency,
• especially for the TX/RX ports where full-duplexity can be implemented.
• Enabling for calibration functionality on X310s by regular maintenance, including
• DC offset calibration by uhd_cal_tx_dc_offset,
• IQ sample balancing by uhd_cal_tx_iq_balance and uhd_cal_rx_iq_balance.

UW Huskies

Future Work

• Various modulations (QPSK/16QAM/256QAM), code rates, SNRs and bandwidths;
• Performance Validation between D2D communication by USRP and 5G NR sidelink link-level simulation by link-level simulator [1];
• PHY layer abstraction for the USRP experiment for fast and reliable simulations in the network simulator [2];
• Real-world implementation and evaluation of decentralized dynamic spectrum access (DSA) systems under 5G protocols [3].

[1] P. Liu, C. Shen, C. Liu, F. Cintron, L. Zhang, L. Cao, R. Rouil, S. Roy, “5G New Radio Sidelink Link Level Simulator and Performance Analysis,” in ACM MSWiM, 2022.

[2] L. Cao, L. Zhang, S. Jin, and S. Roy, “Efficient MIMO PHY abstraction with imperfect CSI for fast simulations,” IEEE Wireless Communications Letters, 2023.

[3] Gao, Z., Li, A., Gao, Y., Wang, Y., & Chen, Y. (2021). Hermes: Decentralized dynamic spectrum access system for massive devices deployment in 5G. arXiv preprint arXiv:2101.02963.

UW Huskies

Content

• Motivation
• Experimental Setting
• Disk image && Profile
• Deployment && Experimental Result
• Video Demo
• Future

85

Team: BlueDevils

Sounder/Agora-PAWR:

A software facilitating open-access mmWave and massive MIMO research in the broader community

Team: BlueDevils

Members: Zhenzhou (Tom) Qi, Jie Wang, Yanyu Hu

Duke University - FuNCtions Lab

WashU - SPAN Lab

U of U - C^3 Lab

01/27/2022

Motivation

Development of open-source software and tutorials that can facilitate open-access mmWave and massive MIMO research in the broader community leveraging the advanced programmable radios in PAWR testbeds

​

​

87

Team: BlueDevils

Motivation

• Initial sync problems with SoapyUHD with Sounder and Agora

88

SoapyUHD

Sync Error

Team: BlueDevils

Overview Sounder-PAWR Design

89

Team: BlueDevils

Sounder-PAWR in POWDER - Setup

90

Team: BlueDevils

Sounder-PAWR in POWDER - Experimental Setting

91

ota-wb-a1/b1

ota-wb-a2/b2

Team: BlueDevils

Sounder-PAWR in POWDER - Configuration File

92

Team: BlueDevils

Sounder-PAWR in POWDER - Example

93

Team: BlueDevils

HDF5 Data Visualization

94

Team: BlueDevils

Constellation (QPSK, 16QAM, 64QAM)

95

QPSK - ref-frame = 250

16QAM - ref-frame = 250

64QAM - ref-frame = 250

Team: BlueDevils

More Results [gain setting vs performance of constellation]

High SNR

Medium SNR

Low SNR

​

96

Team: BlueDevils

Video Demo

97

Team: BlueDevils

Special Thanks

​

​

Andrew Sedlmayr

​

​

​

​

Kirk Webb

​

​

​

98

Team: BlueDevils

5G OAI Network Stress Testing with MGEN and Multiple UEs

Anil Gurses (North Carolina State University)

Abhradeep Roy (Arizona State University)

Bryson Schiel (Brigham Young University)

Frank Yao (University of Utah)

5G OAI Network Stress Test 1

Team Goal

Compare performance for multiple UE’s on the 5G wireless network, first independently and then conjointly

5G OAI Network Stress Test 2

Experiment Tools

• Hardware
• Wasatch Dense Base station
• Two COTS UEs
• Software
• 5G-OAI Profile
• MGEN (For Traffic Generation)

5G OAI Network Stress Test 3

About MGEN

Multi-Generator (MGEN) Network Test Tool

• Network traffic generator for performance testing
• Like iperf, but more intensive
• Developed by the U.S. Naval Research Laboratory

5G OAI Network Stress Test 4

Experiment Results - Single UE, different traffic patterns

103

5G OAI Network Stress Test 5

Experiment Results - Multiple UE’s, normal traffic pattern

5G OAI Network Stress Test 6

Comparison (RSRP v.s Throughput)

5G OAI Network Stress Test 7

Discussion and Future Work

• Gathering more data at more locations
• Seeing how different locations affect each other
• Testing different traffic patterns(Clone, Poisson, etc.)

106

Thank You

Any questions?

5G OAI Network Stress Test 9

Syed Ayaz Mahmud

University of Utah

​

Meles G. Weldegebriel

Washington University in St. Louis

​

Pseudonymetry for POWDER

​

Pseudonymetry for POWDER

Goal

Identify transmit node without having access to POWDER monitoring system.

Ref: M. G. Weldegebriel, J. Wang, N. Zhang and N. Patwari, "Pseudonymetry: Precise, Private Closed Loop Control for Spectrum Reuse with Passive Receivers," 2022 IEEE International Conference on RFID (RFID), Las Vegas, NV, USA, 2022, pp. 91-96, doi: 10.1109/RFID54732.2022.9795976.

Pseudonymetry for POWDER

• Make use of received signal fluctuation.
• RXs are robust to such changes
• Exp. on POWDER
• TX: USTAR
• RX: Browning

Pseudonymetry for POWDER

Results

Future work

Can we stop an offending transmitter?

Chunks of 6000 samples

RMS

Power signature on each 6000 chunk representing binary bits (USTAR)

Thank YOu!

Adaptive Modulation for Shout

Aarushi Sarbhai, Kaitlyn Graves, Serhat Tadik

Yellow Jackets

MWW 2023

Goal

Implement an adaptive M-PSK modulation scheme on Shout in order to minimize BER and maximize throughput

• Modify ‘shout-long-measurements’ profile to switch between pre-generated IQ files of different modulation types for transmission based on received SNR

​

Yellow Jackets

Figure Source: G. Durgin (2022). VID6: Dynamic Digital Modulation Lecture Notes [PDF document]

Completed Work

• Generation and OTA verification M-PSK IQ files
• Between Hospital and SMT rooftop nodes
• OTA verification included a lot of debugging
• Sweeping Tx gain whole transmitting to vary SNR
• Freq: 3520 MHz
• 10-60 dB in 10 dB steps
• Real-time switching between differently modulated preloaded IQ files

Yellow Jackets

Next Steps

• Demodulating and analyzing the received signal which is a combination of different modulation types
• Incorporating adaptive coding along with modulation
• Changing the rule-based modulation-selection to a machine-learning based throughput maximizing/BER minimizing selection

Yellow Jackets

Inter-Numerology Interference Characterization

Jianxiu Li, Talha Bozkus

University of Southern California

Mostafa Ibrahim

Oklahoma State University

Team: INI-5G

Inter-Numerology Interference

• Mixed-Numerology [1, 2]

​

​

​

​

​

​

​

​

​

[1] J. Mao et al, “Characterizing Inter-Numerology Interference in Mixed-Numerology OFDM Systems,” arXiv, 2020.

[2] A. A. Zaidi et al., “Waveform and Numerology to Support 5G Services and Requirements,” in IEEE Communications Magazine, vol. 54, no. 11, pp. 90-98, November 2016.

​

​

gnuradio/tx_ofdm.grc at master · gnuradio/gnuradio · GitHub

​

​

​

​

​

Team: IN-5G

Setting & Resources

• Three nodes: TX: hospital and smt, RX: honors

​

Team: INI-5G

[3] gnuradio/tx_ofdm.grc at master · gnuradio/gnuradio · GitHub

​

​

​

​

​

120

GNURADIO Design

Team: INI-5G

121

Sliding Window Analysis

μ₁ Symbol 1

CP

Symbol 2

CP

Symbol 3

CP

Symbol 4

CP

μ₂ Window

Team: INI-5G

Experimental Results

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

[1] B. Farhang-Boroujeny, “OFDM Versus Filter Bank Multicarrier,” in IEEE Signal Processing Magazine, vol. 28, no. 3, pp. 92-112, May 2011.

​

​

​

​

​

Ambiguity function of the square-root raised-cosine filter for a roll-off factor 0.5 [4]

​

​

​

​

​

Team: INI-5G

Problems

Future work

• Synchronization among transmitters.
• Low dynamic range (we needed a better communication link).
• Other interference sources.

​

​

​

​

​

• Demodulating the victim OFDM numerology symbols, and get BER characterization.
• Extend the study to asynchronous OFDM symbols, and characterize with the level of asynchronicity.

​

​

​

​

Team: INI-5G

THANK YOU

Team: INI-5G

125

​

​

​

5G latency measurement

Dehkontee Cuppah, Xiangbo Meng,

Chih-Hsuan Sun, Yilong Li

Overview

Latency measurement is getting more attention since it is critical to AR/VR, web conferencing applications and self-driving applications.

​

• Uplink/Downlink latency comparison

​

• Test Profile: oai-indoor-ota (1 gNB + 4 COTS UEs)

​

• Latency for different number of UEs

​

​

5G latency measurement

​

Setup

(1) Deployment: 1 CN + 1 gNB + 4 UEs

(2) gNB is connected, and one UE is registered.

(3) UE is pinging the core network.

(4) tcpdump into pcap files at both UE and CN.

5G latency measurement

​

​

Measurements and results

(Top) Capture at CN.

(Bottom) Capture at UE.

Uplink = 4x Downlink

5G latency measurement

​

​

Future work

​

• Two way latency measurement using TWAMP
• Figure out the reason of the difference between UL/DL
• Latency for different number of UEs

​

​

5G latency measurement

​

​

Avg. response time for different number of UEs

Number of UEs

ms

5G latency measurement

​

​

Questions?

OpenAirInterface 5G COTS UE fingerprinting

Exploring 5GC Network Slicing with OAI

• Aditya Sathish, Virginia Tech
• Francis A. Gatsi, University of Notre Dame

Goal

• To acquire the experience of implementing core network slicing using actual SDRs on POWDER
• To explore the usage of Linux Traffic Controller
• To explore network performance measurement based on specific metrics:
• Throughput
• TCP Retransmissions

​

139

Topological Diagram

140

Network Architecture Implemented

141

SMF1

SMF2

SMF3

UPF1

NRF

UPF2

UPF3

AMF

NSSF

UE

DNN

Project Setup & Methodology

• Configure the 5GC profile without slicing
• Configure the 5GC profile with two slices as follows
• Slice1: Fast data rate → Regular Link
• Slice2: Slow data rate →Noisy and delated
• gNB Setup
• COTS UE Setup
• OAI Setup
• Performance comparison between Slice1 and Slice2

142

Lessons Learned & Future Work

• The CN5G kept works best with low bandwidth
• ​

143

Lessons Learned & Future Work

• The CN5G kept works best with low bandwidth
• ​

144

Performance Comparison: Slice1 vs. Slice2

145

Conclusion

• Successful implementation of 3 CN slices with one 1 gNB and 1 UE with real SDRs
• Used traffic controller to differentiate IP links to different UPF slices
• Measured performance using throughput and TCP retransmission metrics

​

146

Future Work

• Network slice orchestration with QoS correlation with RAN slices
• Implementation of ML algorithms for slice selection in the UE
• Implementing e2e network slicing
• Session management continuity for UEs through slices

​

​

​

147

OAI Experiments with POWDER Mobile Nodes

Innocent Obi and Sudheesh Singanamalla

​

inoobi@cs.washington.edu, sudheesh@cs.washington.edu

Team: Huskies on Skis

Context: We run a CBRS-based LTE network in Seattle!

Team: Huskies on Skis

Context: We run a CBRS-based LTE network in Seattle!

Goal: Dynamic Spectrum Coordination among Untrusted Cores

Higher level research effort towards decentralizing the SAS for individual community network deployment cores.

Some day

Team: Huskies on Skis

Builds on top of the efforts at Decentralized Authentication in Distributed Community Cellular Network

First Baby Steps: Run measurements on POWDER Mobile

Goal:

• When, and where, and how long do mobile UEs on the buses attach?

​

Team: Huskies on Skis

Setup - Dense Nodes and Mobile Endpoints

Team: Huskies on Skis

COTS UEs are flaky!

Learnings:

​

Mobile nodes and Quectel-CM are extremely flaky!

That’s an understatement!

Challenges:

​

1. Automating AT+QSCAN results in very inconsistent behavior
2. Preliminary measurements look like the results on the right
3. Tool failed to scan periodically and report data to the server.

Team: Huskies on Skis

What could we do? Multiple 5GC coordination

Learnings:

​

Mobile nodes and Quectel-CM are extremely flaky!

Lays the ground-work for development and measurement of handover (S1/XN)

Challenges: Resource Allocation

Results

What’s Next

1. We’ve learnt a lot of things here!
2. Improve challenges with flakiness of the COTS UEs with Quectel-CM and continue gathering serving cell, and neighbor cell metrics
3. Extend the measurement setup for spectrum measurement
4. Enable individual cores to coordinate spectrum usage.

Channel-adaptive Slice Sharing

&

E2 Latency Monitoring

Walaa Alqwider, MSU

and

Vikas Krishnan, Virginia Tech

Team RIC-sters

Team RIC-sters

Objectives:

Channel-adaptive Slice Sharing

• Create a new sharing policy
• Get buffer status metric
• Update the slice resources based on the channel conditions

​

​

E2 Latency Monitoring

• Measure the control loop latency over the E2 interface
• Trigger control loop threshold violation ( 10ms - 1s for the near-real-time control loop)
• Monitor (visualize) latency as a historical time-series graph

Experiment Setup (POWDER Profiles):

Profile 1: ORAN-RIC

(simulated eNodeB)

Profile 2: srsLTE-indoor-ota

(real SDR nodes for the eNodeB and two UEs)

Team RIC-sters

Results - I:

Team RIC-sters

Results - II:

Team RIC-sters

E2 Latency Monitoring

• Use the control loop latency to predict/detect DoS attacks on the RIC platform
• Perform Near-RT RIC load balancing (E2 Node handover) to ensure strict latency requirements for URLLC applications

Future Work:

Channel-adaptive Slice Sharing

• Get the buffer metrics from the UEs.
• Adjust the sharing based on the buffer metrics and channel quality.
• Build a machine learning based sharing algorithm
• Make the sharing based on the physical resource block within the subframe.

Team RIC-sters