POWDER-RENEW MWW 2019

Mini-projects Presentations

Study of Wi-Fi Channel Occupancy over Time

Lianjun Li, Muhammad Iqbal Rochman, Vanlin Sathya Robinson

Motivation

1. Identify spectrum usage over a long period of time.
2. Gathering real network data for various future works:
1. ML-based classification
2. Identifying illegal usage of any/specific spectrum

​

Methodology Overview

1. Set USRP to listen to Wi-Fi 2.4 GHz channel 11 (center freq 2.462 GHz, BW 20 MHz)
2. Convert raw IQ data to magnitude
3. Write to file
4. Plot the time series data
5. Record Wireshark data as�comparison

Spectrum Monitor

Wireshark@ Source for comparison

Wi-Fi @ Channel 11

NUC & B210 for sensing

Flux group lab, 2nd floor of MEB

GNURadio Companion

• Simple USRP source -> convert to magnitude -> write to file
• Write our own block that outputs the data as string
• Smaller file but less accuracy
• Float32 vs string with 9 digits after decimal point

Custom block

Data processing

1. Raw data: raw IQ measurement in magnitude (20M samples per second)
2. Average raw data every millisecond (1k samples per second)
3. Activity Ratio:

magnitude > threshold -> active

magnitude < threshold -> inactive

Activity ratio = # active samples/ # total samples

1

2

Post-�processing

3

Plot results – Overall hourly data

22:00 to 6:00, only beacons broadcast by APs

One beacon every ~100ms, 3 APs, so:

Activity rate = 1/100 * 3 = 3%

Comparison with Wireshark Data

• Comparing our measurement vs Wireshark on the same time frame
• 11:36 - 11:54

Wireshark – # of received packets on each second

Our measurement

How this project help future research?

1. Identify the illegal usage of any/specific spectrum when its not allocated by the POWDER team.
2. Optimized way to identify the additional interference on the allocated spectrum (for eg: 20 MHz) using Machine Learning, random process, etc

​

Thank You

Q&A

Exploring PAPR trade-offs in OFDM using srsLTE

Ozgur Ozdemir, Mrugen Deshmukh

​

North Carolina State University

Project Overview

• Peak-to-Average-Power Ratio (PAPR) is one of the most difficult problems in OFDM.
• High PAPR results in distortions in the Tx signal and may affect the performance.
• One way to prevent this is to backoff the transmit power to avoid clipping.
• In this project, we allow clipping to occur to improve reception on the POWDER platform.

Exploring PAPR trade-offs in OFDM using srsLTE

​

Block Diagram

• Hardware: Use a pair of NUC+B210 via the attenuator matrix.
• Software: Use the srsLTE-SDR profile on POWDER platform.

​

Exploring PAPR trade-offs in OFDM using srsLTE

​

srsUE sample terminal output

Exploring PAPR trade-offs in OFDM using srsLTE

​

Results

Exploring PAPR trade-offs in OFDM using srsLTE

​

Results

Exploring PAPR trade-offs in OFDM using srsLTE

​

Results

Exploring PAPR trade-offs in OFDM using srsLTE

​

Thank You!

Exploring PAPR trade-offs in OFDM using srsLTE

​

Spectrum Occupancy Measurements

Syed Ayaz Mahmud (UofU)

Udit Paul (University of California SB)

Alireza Shams (NAU)

Project overview

• Sense the spectrum for unauthorized transmitting frequencies.
• Captures raw data and detects occupied/unoccupied frequencies.
• Calculates the received power level.
• Provides the user with information maximum power level on unauthorized frequencies.

GNU Radio Flow Graph

Power Spectral Density

Application-aware Scheduling Optimization

Can (John) Carlak, Xumiao Zhang

University of Michigan

Overview

• Application-aware resource allocation
• Fine-grained bandwidth allocation
• Higher bandwidth given to more demanding apps
• Do not compromise QoS
• Real time results vs. pure scheduling study
• Observe overhead introduced by the system
• POWDER-RENEW platform ROCKS!
• Related Studies
• Application-aware resource block and power allocation for LTE
• Context-aware resource allocation for cellular wireless networks
• QoS guaranteed resource block allocation algorithm in LTE downlink
• An Optimal Application-Aware Resource Block Scheduling in LTE

System Design

Implementation

• Two nodes
• sim-enb node: OAI UE + OAI eNB, with XRAN integrated
• epc node: NextEPC + ONOS
• Profile: cc-xran-e2e-lte
• Code
• Shell scripts
• Trigger iperf measurements (simulated apps)
• Send application type to server
• C program (~150 LoC)
• Accept app_types on an IP socket
• Inform XRAN controller about allocation decisions

Resource allocation strategy

• 25 resource blocks in total

[1] FCC Broadband Speed Guide. https://www.fcc.gov/reports-research/guides/broadband-speed-guide

Preliminary results

• Compare with the default allocation strategy

Demo Video

Future work & Conclusion

• Real radio devices
• Multiple UEs
• XRAN -> FlexRAN
• Resource Control -> Scheduling
• State-of-the-art scheduling
• Fine grained app/QoS info

Infrastructure-aware Self-adapting Cellular Network Utilizing 4G and 5G

Batuhan Mekiker

​

Project overview

• Why do we care?
• It is interesting because switching from one technology to another within the same device provides flexibility.
• Topology looks as follows:

​

Project overview

• It is useful because that flexibility extendeds usability
• It is challenging because setting up two technology within the same device requires careful configuration.
• The switch condition has to be implemented in a way that it doesn’t interrupt users interactions.
• With access to POWDER/RENEW, SDR’s, computational nodes and different locations, access to both hardware and software provides complete control over the system.

Project details

• From Day 1, combination of H2 and H3
• I was able to run two UEs in one node and two eNBs in another.
• Also running NextEPC on the third node.
• Network configurations and device configurations were the challenges.
• Q: Is there a POWDER profile related to your project?
• A: Yes! It is “double-link-cell-net”.

Steps

1. Setup UE, eNB and EPC on different nodes

​

• Add another link between UE and eNB

​

• Configure another stack in UE to use the second link

Demo

Future Work

• Turn wired links into wireless links

​

• Implement a controller/switch on UE to switch between stacks

​

• Use secondary transceiver to operate on another band

​

• Integrate our own protocol as a secondary stack to replace the existing one.

​

Thank You!

​

​

​

​

​

​

Questions?

802.11 with GNURadio

Hossein Pirayesh, Pedram Kheirkhah

​

University of Louisville

Project overview

• User-defined out-of-tree modules in GNU-Radio
• Several functions or whole Tx/Rx BB processing in one block

​

• Challenges
• Signal processing scripts ✔
• Loading the custom-built module ✔
• UHD compatibility 🗶

​

​

​

Powder-RENEW versus our Testbed

• Communication range
• Connectivity
• Frequency range

​

​

• Duration of experiments in

outdoor environments

• RF bandwidth

​

​

Tx Implementation

• Realization with source tree

​

​

​

​

• Tx implementation with out-of-tree module

​

Rx Implementation

• Realization with source tree

​

​

​

​

• Realization with out-of-tree module

​

Results: Constellation of Decoded Signals

• Realtime processing

Results: Constellation of Decoded Signals

• Offline processing (50 frames)

Thank You!

802.11 with GNURadio

​

Analysis of FCC licenses usage and spectrum activity

Lorenzo Bertizzolo, Leonardo Bonati, Hai Cheng, Guillem Reus Muns

Analysis of FCC licensed spectrum

• Measure-based spectrum characterization
• 3 Different locations:

​

​

​

​

​

​

• 2 Different band ranges:

Analysis of FCC licensed spectrum

​

• GNU Radio spectrum power receiver:

​

​

​

​

​

​

• For each band chunk (5MHz):
• Bursty max power, average received power

​

Analysis of FCC licensed spectrum

• Measure and match FCC licenses on LTE bands:

​

​

​

​

​

​

​

​

​

LTE band 65

(T060430152, WQKT248, WQTX351)

LTE band 7

(experimental?)

Analysis of FCC licensed spectrum

• LTE bands Rooftop vs Ground activity:

​

​

​

​

​

​

​

​

​

ISM activity (Wi-Fi, Bluetooth...)

Analysis of FCC licensed spectrum

• CBRS band activity: rooftop vs ground

������

• 3.5 GHz: 40+ FFC licenses

Thank You!

Modulation Classification

Project members:

Joshua Bassey(jbassey@student.pvamu.edu)

Ajeya Anand (ajeyaana@buffalo.edu)

Sabarish Krishna Moorthy (sk382@buffalo.edu)

​

Contents

• Why?
• What? Where?
• How?
• Next?

56

• Why?: History & Importance

• Research literature is abundant for modulation classification schemes [1][2]
• Why ? Modulation Classification enables
• Use of a universal receiver

​

​

[1] Arulampalam, Ganesh, et al. "Classification of digital modulation schemes using neural networks." ISSPA'99. Proceedings of the Fifth International Symposium on Signal Processing and its Applications (IEEE Cat. No. 99EX359). Vol. 2. IEEE, 1999.

[2]T. J. O’Shea, T. Roy and T. C. Clancy, "Over-the-Air Deep Learning Based Radio Signal Classification," in IEEE Journal of Selected Topics in Signal Processing, vol. 12, no. 1, pp. 168-179, Feb. 2018.

57

• What? Where?

58

Powder Profile : GNURADIO-SDR-X310-Pair

Receiver

Transmitter

• How?: GNURADIO & Data collection

59

Based on H8: Hands-on: QPSK over OFDM (Ayaz)

60

• How?: GNURADIO & Data collection (continued)

H8: Hands-on: QPSK over OFDM (Ayaz)

• How ? Tapping Data Out

61

H8: Hands-on: QPSK over OFDM (Ayaz)

• How? TX

62

H8: Hands-on: QPSK over OFDM (Ayaz)

Output Files

• How?: Machine Learning Classification & Results

• Model is ~100% in detecting modulation scheme by having access to only the received signal.
• Complex valued signal is not an issue for the network.
• Data required for training not big.
• Model performs automatic feature engineering.

63

• Results: Model Accuracy & Loss

64

• Next?: Future Research Possibilities

• Add more modulation schemes - we tried BPSK, QPSK, more to explore
• Gather more data - longer runs
• Refine ML
• Present and share repeatable experimental results

65

Thank You

POWDER Team and Attendees

​

66

Special Credits to

Ayaz Mahmud and Dr. Neal for their guidance

67

Mobile core network control plane 4G/5G comparison

Team Members:

Snigdhaswin Kar, Clemson University

Prabodh Mishra, Clemson University

Agenda

• Overview
• 4G and 5G Network Architecture
• Implementation using free5GC core
• Demo of 4G and 5G networks
• Results
• Future work

• 5G will enable ultra reliable low latency communication and massive machine type communication.
• Design end-to-end 4G and 5G networks using different open source stacks.
• Study the service based architecture of 5G core network.
• POWDER provides the ideal platform for implementing these cellular networks.

Overview

Implementation

• The UE and eNodeB is realized with simulated components provided by OAI on d430 nodes.
• EPC services are provided by NextEPC for 4G networks and by free5GC for 5G networks.
• Configured both the networks and established connectivity.
• Created POWDER profiles MiniProj4G and MiniProj5G.

​

​

4G Network Architecture

• UE: User Equipment (phone, etc.)
• eNodeB: Evolved Node B
• HSS: Home Subscriber Server
• MME: Mobility Management Entity
• PDN: Packet Data Network
• PCRF: Policy and Charging Rules Function

Free5GC Network Architecture

• AMF: Access and Mobility Management Function
• SMF: Session Management Function
• UPF: User Plane Function
• HSS & PCRF: From NextEPC Implementation
• DN: Data Network

Configuring 4G core using NextEPC

Configuring 5G core using Free5GC

Free5GC Testing

EPC vs 5GC Initialization

Future Work

• We can conduct experiments with upcoming implementations of 5G core network stacks such as NextEPC 5G, etc.
• Experiments can be conducted using real physical hardware for cellular networks.
• Design experiments for testing applications such as IoT, autonomous vehicular networks, etc.
• Further research in the areas of 5G architecture and protocol stack.

​

Thank you!

Antenna Design, Deployment and Testing

Chen Ye Lim, Iowa State University

Tianyi Zhang, Iowa State University

​

Project overview

• Design, fabricate and test a spiral antenna
• A broadband antenna for USRP SDR

​

Spiral Antenna

Covered bands

​

• EBS: 2496MHz to 2690MHz
• CBRS: 3550 MHz to 3700 MHz

​

​

R_0

R_spiral

Simulation result

Unoptimized Simulation Result

Simulation result

Optimized Simulation Result

How it looks like

Result: Our spiral antenna

Result: Taoglas wideband antenna

Potential Use Cases for Powder

​

• Wideband Experiments
• One size fits all wireless control plane

​

Challenges

​

• Sequential processes (e.g. simulation, fabrication, outdoor testing) that are time consuming
• Limitation of supported bands by the rooftop antenna

​

Future Works

• Design works to improve reflection coefficient
• Optimize antenna size to increase bandwidth

MMSE Beamforming

Brent Kenney, Liangping Ma, Hamed Hosseiny, Yongce Chen, Nima Taherkhani, John Kaewell

Project Overview

• Implement MMSE beamforming
• Modify code to increase number of antennas from 8 to 32
• Compare performance of ZF, Conjugate BF and MMSE

Test Scenario

• 64-element BS (roof of MEB)
• East-pointing
• 2 fixed UEs in POWDER lab/office (NE corner of MEB)
• Non-LOS
• Parameters
• Constellation size
• QPSK, 16-QAM, 64-QAM
• Detection methods
• ZF, Conj. BF, MMSE

POWDER Profile and Resources used

• Profile Used: gary-faros-renew
• Used during Communications Hands-on activities
• aka RENEW-TUT
• MATLAB and Python Driver
• Massive MIMO Skylark Equipment

Modifications to support 64 Antennas

Lorem Ipsum

​

• Hands-on demonstrations used 8 Iris SDRs from a single chain
• Communicated through first SDR in the daisy chain configuration
• To employ all 32 Iris SDRs, delay calculation and triggering must be managed through the hub
• Requires new driver files

ofdm_mimo.m

getRxVec.m

hub_py.m

hub_py.py

iris_py.m

iris_py.py

Trigger and Delay is performed by hub instead of daisy chain leader

Progress and Results

• Implement MMSE beamforming
• Constructed regularized detection matrix
• Used fixed values for SNR estimate
• Expanding to 64 antennas
• Hardware issues prevented new MIMO captures on Friday
• Second UE (sn 145 went down)
• Switched gears to a SIMO experiment with remaining UE (sn 060), but only had time to run the 1x2 example code
• Comparing ZF, Conj. BF, and MMSE
• All 3 were implemented, but we did not have a sufficient number of antennas to make Conj. BF useful

Enabling both antennas on an Iris board

• Changes in iris_py.py
• Flag: both_channels = True
• Add a buffer for the new outgoing streams on each UE
• buf_a = cfloat2uint32(data, order='QI')
• buf_b = cfloat2uint32(data, order='QI')
• self.sdr.writeRegisters("TX_RAM_A", replay_addr, buf_a.tolist() )
• self.sdr.writeRegisters("TX_RAM_B", replay_addr, buf_b.tolist() )
• Add a buffer for the new incoming stream for each Iris board on the BS
• rx_frames_a = np.zeros((in_len*max_frames), dtype=np.complex64)
• rx_frames_b = np.zeros((in_len*max_frames), dtype=np.complex64)
• rx_frames_a[m*in_len : (m*in_len + in_len)] = wave_rx_a
• rx_frames_b[m*in_len : (m*in_len + in_len)] = wave_rx_b
• Modify the return: return(rx_frames_a, rx_frames_b)
• Changes in ofdm_mimo.m
• Adjust the data length for each outgoing stream because two streams share a block RAM
• Add the ID’s of the new Iris boards for increasing the total number of operating antennas

MIMO-OFDM UL Beamforming: ZF vs. MMSE

• ZF

32-antenna Rx - 2-antenna Tx : SNR_X1=12.68 (dB), SNR_X2=12.19 (dB)

MIMO-OFDM UL Beamforming: ZF vs. MMSE

• MMSE

32-antenna Rx - 2-antenna Tx : SNR_X1= 12.66 (dB), SNR_X2= 13.83 (dB)

Channels Correlation

Future Works

• RF Chain Calibration
• Needed for Downlink Communication
• DL leverages reciprocal channel assumption
• Downlink Precoding
• Uses the estimated channel matrix
• Compare ZF, Conj. BF, and MMSE performance
• Increase number of UEs
• Each Iris SDR used for an UE has two receive chains
• Enable both chains to double number of UEs to 4

Thank You