Enhancing Relay-Assisted Wireless Networks Through Learning and Optimization over Riemannian Manifolds

Authors

Imtiaz NasimFollow

Document Type

Dissertation

Degree

Doctor of Philosophy (PhD)

Major/Program

Electrical and Computer Engineering

First Advisor's Name

Ahmed S. Ibrahim

First Advisor's Committee Title

Committee Chair

Second Advisor's Name

Arjuna Madanayake

Second Advisor's Committee Title

Committee Member

Third Advisor's Name

Elias Alwan

Third Advisor's Committee Title

Committee Member

Fourth Advisor's Name

Jason Liu

Fourth Advisor's Committee Title

Committee Member

Keywords

Wireless Communications, Vehicular Systems, Riemannian Manifold, Machine Learning, Optimization, Throughout Enhancement, Clustering, Reinforcement Learning, Network Flowrate, Beam Tracking, Codebook Design, COSMOS, Relay Optimization

Date of Defense

3-16-2023

Abstract

Achieving high data rate for next generation mobile networks, including highly dynamic networks (e.g., vehicular networks), is essential for emerging applications such as extended reality. Maximizing the overall network flow data rate in such networks can be achieved through two complementary maximizations across the network and link levels. First at the network level, we find the optimal relay locations that maximize network flow rate through learning and optimization over Riemannian manifolds (i.e., curved surfaces). More specifically, we utilize the symmetric positive definite (SPD) structures of network topology and propose two different relay placement approaches- (i) Riemannian Multi-Armed Bandit (RMAB) that applies a reinforcement learning (RL) model, and (ii) Riemannian Particle Swarm Optimization. Compared to state-of-the-art benchmarks, our solutions achieve higher network flow rate and ensure a balanced network robustness in terms of connectivity. Once the best relay locations are found, we focus on maximizing the rate over individual link levels. To this end, we design an adaptive codebook for the previously deployed relays, making use of the observation that covariance matrices of wireless channels are SPD ones that can be represented over Riemannian manifolds. In particular, we propose three clustering solutions to design our beamforming codebook- (i) Riemannian K-Means that performs offline clustering of the channel covariance matrices, (ii) Riemannian Competitive Learning which is an online clustering solution, and (iii) Riemannian Dictionary Learning that leverages the sparse channel properties while designing the codebook. The performance of the proposed three solutions is compared with existing benchmarks. As the adaptive codebooks are established, we now focus on tracking the optimal beamforming vectors (beams) towards highly-mobile users and maximize their individual link rate. To do so, we first propose an RL based combinatorial MAB algorithm that tracks the optimal beams jointly for multiple users. Then, we propose a sequential MAB solution where the beams are selected in a one-by-one manner. Finally and through experimentations on large-scale testbed, namely, COSMOS, we validate the performance of our beam tracking solution for a single user case.

Identifier

FIDC011055

Previously Published In

i) I. Nasim and A. Ibrahim, “Millimeter wave beamforming codebook design via learning channel covariance matrices over Riemannian manifolds,” IEEE Access, vol. 10, Dec. 2022.

ii) I. Nasim, A. Ibrahim, and S. Kim, “Learning-based beamforming for multi-user vehicular communications: A combinatorial multi-armed bandit approach,” IEEE Access, vol. 8, Dec. 2020.

iii) I. Nasim, P. Skrimponis, A. S. Ibrahim, S. Rangan, and I. Seskar, “Reinforcement Learning of MillimeterWave Beamforming Tracking over COSMOS Platform,” to appear in ACM WiNTECH Conference, 2022.

iv) I. Nasim and A. Ibrahim, “Relay Placement for Maximum Flow Rate via Learning and Optimization over Riemannian Manifolds,” submitted to IEEE Trans. on Machine Learning in Com. and Networking, 2023

Recommended Citation

Nasim, Imtiaz, "Enhancing Relay-Assisted Wireless Networks Through Learning and Optimization over Riemannian Manifolds" (2023). FIU Electronic Theses and Dissertations. 5279.
https://digitalcommons.fiu.edu/etd/5279