Increasing Throughput in 5 G Physical Layer Communication

Increasing Throughput in 5 G+ Physical Layer Communication Systems Prof. Brian L. Evans Feb. 11, 2021 Includes research results from Recent Ph. D graduates Dr. Jinseok Choi and Dr. Faris Mismar Ph. D students Mr. Yunseong Cho and Ms. Pooja Nuti 1

WNCG Faculty in ML + 5 G/6 G Faculty 5 G/6 G Machine Learning Cloud Comp. Vehicular Sys. Jeffrey Andrews Phy+Net Cellular Comm. , Alex Dimakis Physical Brian Evans Physical Cellular Comm. . Todd Humphreys Physical GPS; Drone Swarms Hyeji Kim Physical Cellular Comm. . Sriram Vishwanath Phy+Net Adversarial Nets Storage+Coding Comm+Radar Gustavo de Veciana Networking Cellular Networks Wireless Net. Sensing+Netwo rk Cellular Comm. Generative Models Storage+Coding INTRODUCTION Wireless Comm Nav+ADAS 2

Example Research Projects System 31 Ph. D and 13 MS alumni Contribution Cellular (LTE) large antenna array SW release Prototype Funding Lab. VIEW NI FPGA NI analog/digital beamforming Matlab Futurewei machine learning Python UT WNCG Smart grid commun. interference reduction real-time testbeds Matlab Freescale & TI modems IBM, NXP, TI Wi-Fi interference reduction Matlab NI FPGA Intel, NI Underwater large receive array Matlab Lake testbed ARL: UT Camera image acquisition Matlab DSP/C Intel, Ricoh video acquisition Matlab Android TI image halftoning Matlab C HP, Xerox video halftoning Matlab C Qualcomm Linux/C++ Navy sonar Navy, NI Display Design tools distributed computing INTRODUCTION 3

Deep Learning Predictive Band Switching q Band selection: How does a user choose a frequency band to improve their rate? q Problem [3 gpp 18] • Measurement gap reduces users’ effective Legacy achievable rates • Blindly switching a user eliminates need Blind for gap but risks rates Typical band switching with measurement gap q Solution • Main idea: rank bands based on their quality • Grant switch to band with highest rank if requested Data-driven approach to eliminate the “measurement gap” F. B. Mismar, A. Al. Ammouri, A. Alkhateeb, J. G. Andrews, and B. L. Evans, ``Deep Learning Predictive Band Switching in Wireless Networks'', IEEE Transactions on Wireless Communications, vol. 20, no. 1, Jan. 2021, pp. 96109. TOPIC 1: DEEP LEARNING PREDICTIVE BAND SWITCHING IN WIRELESS NETWORKS 4

more Proposed Solution Scenario A: 100% users start sub-6 GHz q Use spatial correlation between bands at user location q Train DNN using the Deep. MIMO ray-tracing dataset Scenario B: 100% users start in mm. Wave band Feature Description Bias term (equal to unity) Effective rate at sub-6 Effective rate at mm. Wave Source technology (0 mm. Wave, 1 sub-6 GHz) Latitude, longitude, height Band switch requested Band switch request band switching threshold Band switch granted estimated instantaneous rate based on other users TOPIC 1: DEEP LEARNING PREDICTIVE BAND SWITCHING IN WIRELESS NETWORKS 5

Rate Optimization for Reconfigurable Intelligent Surfaces (RIS) q Improve spectral efficiency by reconfiguring the wireless propagation path RIS q Design parameters Nonconvex unit-modular constraint (global optimum is not available) Multi-input single-output (MISO) case Work with Ms. Pooja Nuti (Ph. D student) TOPIC 2: RATE OPTIMIZATION FOR RECONFIGURABLE INTELLIGENT SURFACES (RIS) 6

Simulation Results: Spectral Efficiency and Run-Time Complexity [14] [15] more [14] [15] Downlink narrowband transmission in sub-6 GHz frequency bands. Assume exact knowledge of channels. Maximize the spectral efficiency (two methods) or equivalent channel power as a proxy (other five [14]methods) X. Yu, D. Xu, and R. Schober, “MISO wireless communication systems via intelligent reflecting surfaces, ” IEEE/CIC Intl. Conf. Comm. China, 2019, pp. All 735– 740. methods are iterative except semi-definite relaxation [15] which relaxes the unit modular [15] Q. Wu and R. Zhang, “Intelligent reflecting surface enhanced wireless network: Joint active and SURFACES passive beamforming design, ” IEEE Global Comm. 7 TOPIC 2: RATE OPTIMIZATION FOR RECONFIGURABLE INTELLIGENT (RIS) constraint.

Massive MIMO Systems with Multicell Coordinated Beamforming and Power Control q. Minimize total transmit power w/ communication System model & Configuration target • Quantization noise from low-resolution (low-power) ADCs and DACs • Thermal noise, intra-cell, and inter-cell interference • Narrowband wideband transmission q Optimize power control (uplink) and beamforming (downlink) • Minimize total transmit power for target SINR • Use strong duality between our uplink/downlink formulations J. Choi, Y. Cho, and B. L. Evans, ``Quantized Massive MIMO Systems with Multicell • Estimate inter-cell interference locally w/ooncoordination Coordinated Beamforming and Power Control'', IEEE Transactions Communications, 16 pages, accepted for publication with other cells. TOPIC 3: Quantized Massive MIMO Systems with Multicell Coordinated Beamforming and Power Control Derivation 8

Simulation Results and Future Work More DDPG RL + Hybrid Beamforming performance gain over per-cell method Work with Mr. Yunseong Cho (Ph. D student) and Dr. Jinseok Choi TOPIC 3: Quantized Massive MIMO Systems with Multicell Coordinated Beamforming and Power Control Analog beamformer/ combiner use ULA parametrized by angles Not a convex problem 9

Supplemental Slides 10

TOPIC 1 JOINT BEAMFORMING, POWER CONTROL, AND INTERFERENCE COORDINATION Included in Ph. D dissertation by Dr. Faris Mismar TOPIC 1: JOINT BEAMFORMING, POWER CONTROL, AND INTERFERENCE COORDINATION 11

BACKGROUND q Problem § User served by serving base station receives interference from neighboring base stations § Base station serving the user causes interference to other users q Goal § Maximize the SINR from serving base station to user q Design Parameters § Beamforming (BF) to create a virtual sense of a user-specific channel for data § Power Control (PC) to control the transmit power of the serving BS towards a user § Interference Coordination (IC) to control the transmit power of the neighboring BSs Use deep Q-network (DQN) with greedy policy TOPIC 1: JOINT BEAMFORMING, POWER CONTROL, AND INTERFERENCE COORDINATION 12

more SOLUTION & RESULT q Simulation result q Encoded Bearer selector actions 0: step down beamforming codebook 1: step up beamforming codebook up to 16 discrete actions q States q Reward function voice data if any constraint becomes inactive. if the target SINR is achieved. TOPIC 1: JOINT BEAMFORMING, POWER CONTROL, AND INTERFERENCE COORDINATION 13

SIMULATION Go back Communication System Parameters Deep Reinforcement Learning Hyperparameters (exhaustive search) JB-PCIC algorithm achieves upper bound on performance but without exhaustive search in action space TOPIC 1: JOINT BEAMFORMING, POWER CONTROL, AND INTERFERENCE COORDINATION 14

TOPIC 2 DEEP LEARNING PREDICTIVE BAND SWITCHING IN WIRELESS NETWORKS ML-driven approach Included in Ph. D dissertation by Dr. Faris Mismar TOPIC 2: DEEP LEARNING PREDICTIVE BAND SWITCHING IN WIRELESS NETWORKS 15

SIMULATION Go back q Impact of the band switching threshold on the performance Higher band switching thresholds cause the legacy approach performance to do worse. Higher band switching thresholds enable my proposed algorithm to do even better absence of measurement gap in optimal … and near perfect DNN classification decisions DNN Confusion Matrix (Scenarios A, and B) XGBoost Confusion Matrix (Scenarios A, and B) Scenario A: 100% users sub-6 GHz Scenario B: 100% users mm. Wave Trade-off of accuracy and run-tim Legacy approach performs better than blind in low throughput regime. TOPIC 2: DEEP LEARNING PREDICTIVE BAND SWITCHING IN WIRELESS NETWORKS 16

TOPIC 3 Rate Optimization for Reconfigurable Intelligent Surfaces (RIS) Ph. D research work of Ms. Pooja Nuti TOPIC 3: RATE OPTIMIZATION FOR RECONFIGURABLE INTELLIGENT SURFACES (RIS) 17

FOUR PROPOSED RIS CODESIGN ALGORITHMS Go back § Algorithm 1 (Power method) § Algorithm 2 (Gradient Ascent over RIS angles) § Algorithm 3 (Gradient Ascent) Extension may include MIMO systems and imperfect channel TOPIC 3: RATE OPTIMIZATION FOR RECONFIGURABLE INTELLIGENT SURFACES (RIS) 18

TOPIC 4 Quantized Massive MIMO Systems with Multicell Coordinated Beamforming and Power Control Low-resolution MIMO Ph. D research work of Mr. Yunseong Cho Collaborative work with Dr. Jinseok Choi TOPIC 4: Quantized Massive MIMO Systems with Multicell Coordinated Beamforming and Power Control 19

Go back Derived Results and Proposed Algorithm (NB) Uplink Total Transmit Power Minimization Problem Theorem 1: Duality separate problems Downlink Total Transmit Power Minimization Problem Corollary 1: Strong Duality considering non-negligible quantization error Optimal Uplink Solution Proposed Algorithm (WB) : proved the same results in wideband ca covariance matrix of received signal Optimal Downlink Solution computed from uplink solution convergence is also proved! TOPIC 4: Quantized Massive MIMO Systems with Multicell Coordinated Beamforming and Power Control 20

Go back Simulation Results performance gain over per-cell method validates strong duality verifies wideband case TOPIC 4: Quantized Massive MIMO Systems with Multicell Coordinated Beamforming and Power Control 21

BACKGROUND Go back Critic Actor Critic_target Actor_target Analog beamformer/combiner use ULA parametrized by angles Not a convex problem!! TOPIC 4: Quantized Massive MIMO Systems with Multicell Coordinated Beamforming and Power Control 22