Dissertation
Two-Dimensional Turbo-Equalization with Machine Learning Estimation for Doubly-Dispersive Channels
Washington State University
Doctor of Philosophy (PhD), Washington State University
12/2024
DOI:
https://doi.org/10.7273/000007171
Abstract
On doubly-dispersive channels signals unavoidably spread in frequency due to the Doppler effect, and in time due to severe multi-path. These lead to inter-carrier interference (ICI) and inter-symbol interference (ISI) on multi-carrier communication systems. While ISI can be avoided by appending time guards to every multi-carrier symbol, in long delay channels, this consumes a significant fraction of the symbol
duration, reducing the effective data rate. In this dissertation, a novel time guard free ICI-ISI estimation-equalization system is presented, which demonstrates substantial improvements over previous literature. A joint equalization is performed by employing two maximum a-posteriori (MAP) equalizers that process data both frequency- and time-wise using the BCJR algorithm. The equalizers are paired together
into a turbo architecture with a channel decoder, in which soft-bit information is exchanged to boost performance. It is shown that the proposed system reduces bit error rate (BER) by exploiting Doppler and time diversity. Moreover, theoretical two-dimensional maximum-likelihood bounds for ICI-ISI equalization are derived, and these show diversities as well. By making use of these diversities, the proposed
system is able to perform nearly close to the full time guard case for certain channels. The system is initially presented for doubly-dispersive Rayleigh fading channels, with perfect channel state information. Subsequently, the system is extended to incorporate channel estimation, with a focus on high-spread sparse channels where compressed sensing techniques are applicable. An innovative approach that enables
both ICI-ISI estimation and equalization is introduced by using the OMP algorithm. When compared to one-dimensional ICI-only equalization with time guards, the time guard free ICI-ISI equalizers achieve significant BER reductions for very high-spread channels. This dissertation also addresses ISI equalization for two-dimensional magnetic recording (TDMR). In TDMR, media-dependent noise presents a significant obstacle. A media noise estimator based on convolutional neural networks (CNNs) is presented. By incorporating skip connections into the network architecture, the estimation accuracy is improved while achieving a substantial complexity reduction.
Metrics
1 File views/ downloads
3 Record Views
Details
- Title
- Two-Dimensional Turbo-Equalization with Machine Learning Estimation for Doubly-Dispersive Channels
- Creators
- Jorge Andres Pires
- Contributors
- Benjamin J Belzer (Co-Chair)Krishnamoorthy Sivakumar (Co-Chair)Thomas R Fischer (Committee Member)Mohammad Torabi (Committee Member)Yan Yan (Committee Member)
- Awarding Institution
- Washington State University
- Academic Unit
- School of Electrical Engineering and Computer Science
- Theses and Dissertations
- Doctor of Philosophy (PhD), Washington State University
- Publisher
- Washington State University
- Number of pages
- 179
- Identifiers
- 99901195200401842
- Language
- English
- Resource Type
- Dissertation