Open Access
Description:
In this thesis, we investigate the designs of pragmatic data detectors and channel decoders with the assistance of deep learning. We focus on three emerging and fundamental research problems, including the designs of message passing algorithms for data detection in faster-than-Nyquist (FTN) signalling, soft-decision decoding algorithms for high-density parity-check codes and user identification for massive machine-type communications (mMTC). These wireless communication research problems are addressed by the employment of deep learning and an outline of the main contributions are given below. In the first part, we study a deep learning-assisted sum-product detection algorithm for FTN signalling. The proposed data detection algorithm works on a modified factor graph which concatenates a neural network function node to the variable nodes of the conventional FTN factor graph to compensate any detrimental effects that degrade the detection performance. By investigating the maximum-likelihood bit-error rate performance of a finite length coded FTN system, we show that the error performance of the proposed algorithm approaches the maximum a posterior performance, which might not be approachable by employing the sum-product algorithm on conventional FTN factor graph. After investigating the deep learning-assisted message passing algorithm for data detection, we move to the design of an efficient channel decoder. Specifically, we propose a node-classified redundant decoding algorithm based on the received sequence’s channel reliability for Bose-Chaudhuri-Hocquenghem (BCH) codes. Two preprocessing steps are proposed prior to decoding, to mitigate the unreliable information propagation and to improve the decoding performance. On top of the preprocessing, we propose a list decoding algorithm to augment the decoder’s performance. Moreover, we show that the node-classified redundant decoding algorithm can be transformed into a neural network framework, where multiplicative tuneable weights are attached to the decoding ...
Publisher:
UNSW, Sydney
Year of Publication:
2021
Document Type:
doctoral thesis ; http://purl.org/coar/resource_type/c_db06 ; [Doctoral and postdoctoral thesis]
Language:
EN
Subjects:
Deep learning ; Data detection ; Channel decoding
Rights:
open access ; https://purl.org/coar/access_right/c_abf2 ; CC BY-NC-ND 3.0 ; https://creativecommons.org/licenses/by-nc-nd/3.0/au/ ; free_to_read
Terms of Re-use:
CC-BY-NC-ND
Content Provider:
UNSW Sydney (The University of New South Wales): UNSWorks  Flag of Australia
Loading ...
Loading ...
Loading ...