Work

Interplay between information theory and optimization: Applications to multi-user wireless communications

Year: 2023
Type: dissertation
Abstract: The ever increasing number of communicating devices as well as the enormous growth of high data rate applications are driving the transition to future generation of wireless networks.As such, many challenges arise in terms of users' connectivity, energy-and spectral-efficiency as well as managing the interference.To tackle such challenges in multi-user wireless networks, the aim of this HDR thesis is to study the interaction between information theory and optimization.Many investigations showed that in heterogeneous and multi-user networks, the interference exhibits an impulsive nature, which is not encompassed by the Gaussian model.In the first part of this HDR thesis, we focus on approximating the resulting non-linear log-likelihood ratios (LLR) using an information-theory based criterion.Since sending pilots decreases the useful data rate, we propose an unsupervised LLR estimation in both the asymptotic and the short blocklength regime, without requiring any prior on the input and noise distribution.Our proposed LLR estimation method, numerically evaluated using LDPC codes, was shown to be able to cope with various noise models.In the second part of this HDR thesis, we focus on resource optimization for cooperative multi-user communications, where we first derive the achievable rate regions of the considered networks.On the one hand, we investigate a backscatter enhanced multi-user downlink non orthogonal multiple access (NOMA) transmission, whose energy-efficiency is maximized under user minimum quality of service (QoS) constraints.For this, we consider both the perfect channel state information (CSI) case and the other extreme case, where no CSI is available at the transmitter side and for which this lack of knowledge is compensated by the use of reinforcement learning.On the other hand, we study a relay-aided cognitive radio network, where the opportunistic achievable rate is maximized under a primary QoS constraint.While under NOMA the optimal resource allocation policy was derived in closed-form in the perfect CSI case, deep learning techniques were used in the cooperative cognitive network because the non-linear relay operations rendered the optimization problem non-convex and difficult to solve.The HDR thesis is concluded by discussing some open and future research directions.
Source: HAL
Author: Savard, Anne
Language: English
Open Access status: green
OpenAlex ID: W4392498132
Cites: 176
Cited by:
Related to: 10