Mohammad, Abdullahi;
(2022)
Deep Learning Designs for Physical Layer Communications.
Doctoral thesis (Ph.D), UCL (University College London).
Preview |
Text
Mohammad_10148915_Thesis.pdf Download (2MB) | Preview |
Abstract
Wireless communication systems and their underlying technologies have undergone unprecedented advances over the last two decades to assuage the ever-increasing demands for various applications and emerging technologies. However, the traditional signal processing schemes and algorithms for wireless communications cannot handle the upsurging complexity associated with fifth-generation (5G) and beyond communication systems due to network expansion, new emerging technologies, high data rate, and the ever-increasing demands for low latency. This thesis extends the traditional downlink transmission schemes to deep learning-based precoding and detection techniques that are hardware-efficient and of lower complexity than the current state-of-the-art. The thesis focuses on: precoding/beamforming in massive multiple-inputs-multiple-outputs (MIMO), signal detection and lightweight neural network (NN) architectures for precoder and decoder designs. We introduce a learning-based precoder design via constructive interference (CI) that performs the precoding on a symbol-by-symbol basis. Instead of conventionally training a NN without considering the specifics of the optimisation objective, we unfold a power minimisation symbol level precoding (SLP) formulation based on the interior-point-method (IPM) proximal ‘log’ barrier function. Furthermore, we propose a concept of NN compression, where the weights are quantised to lower numerical precision formats based on binary and ternary quantisations. We further introduce a stochastic quantisation technique, where parts of the NN weight matrix are quantised while the remaining is not. Finally, we propose a systematic complexity scaling of deep neural network (DNN) based MIMO detectors. The model uses a fraction of the DNN inputs by scaling their values through weights that follow monotonically non-increasing functions. Furthermore, we investigate performance complexity tradeoffs via regularisation constraints on the layer weights such that, at inference, parts of network layers can be removed with minimal impact on the detection accuracy. Simulation results show that our proposed learning-based techniques offer better complexity-vs-BER (bit-error-rate) and complexity-vs-transmit power performances compared to the state-of-the-art MIMO detection and precoding techniques.
| Type: | Thesis (Doctoral) |
|---|---|
| Qualification: | Ph.D |
| Title: | Deep Learning Designs for Physical Layer Communications |
| Open access status: | An open access version is available from UCL Discovery |
| Language: | English |
| Additional information: | Copyright © The Author 2022. Original content in this thesis is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) Licence (https://creativecommons.org/licenses/by-nc/4.0/). Any third-party copyright material present remains the property of its respective owner(s) and is licensed under its existing terms. Access may initially be restricted at the author’s request. |
| UCL classification: | UCL > Provost and Vice Provost Offices > UCL BEAMS > Faculty of Engineering Science UCL > Provost and Vice Provost Offices > UCL BEAMS > Faculty of Engineering Science > Dept of Electronic and Electrical Eng UCL > Provost and Vice Provost Offices > UCL BEAMS UCL |
| URI: | https://discovery.ucl.ac.uk/id/eprint/10148915 |
Archive Staff Only
 |
View Item |
