Dissertation
Efficient Deep Learning-based Equalization and Detection Systems for Digital Communications
Washington State University
Doctor of Philosophy (PhD), Washington State University
01/2021
DOI:
https://doi.org/10.7273/000002429
Handle:
https://hdl.handle.net/2376/121867
Abstract
This first part of this dissertation considers the problem of equalization and detection for multilayer magnetic recording (MLMR) comprising two layers. To this end, we use MLMR waveforms generated using a grain switching probability (GSP) model that is trained on realistic micromagnetic simulations. We propose convolutional neural network (CNN) based systems for equalization and detection. We show that the CNN-based systems outperform the conventional linear equalizer followed by a Viterbi detector system. Furthermore, by interfacing the CNN detector with a channel decoder, we show that an areal density gain of 16.2% can be achieved by a two-layer MLMR system over a one-layer system.
This second part of this dissertation investigates candidate reduced complexity neural network architectures for equalization over two-dimensional magnetic recording (TDMR). We test the performance on readback signals measured over an actual hard disk drive with TDMR technology. The multilayer perceptron (MLP) non-linear equalizer achieves significant gains over the linear equalizer. But the MLP’s complexity is 6.6 times the complexity of the linear equalizer. To reduce the complexity, a reduced complexity MLP (RC-MLP) equalizer is proposed. RC-MLP consists of finite-impulse response filters, a non-linear activation, and a hidden delay line. A proposed RC-MLP variant entails a complexity of only 1.59 times the linear equalizer's complexity, while achieving most of the performance gains of the MLP.
This third part of this paper considers the encoding and decoding of transmitted sequences in a time-asynchronous non-orthogonal multiple access (NOMA) downlink wireless channel with faster than Nyquist signaling. As a baseline, we use a singular value decomposition (SVD)-based scheme. Although this SVD-based scheme achieves reliable communications, its time complexity is quadratic in the length of the transmitted sequence. We propose a CNN auto-encoder (AE) for encoding and decoding with linear time complexity. In a two-user, time-asynchronous NOMA system, the CNN AE outperforms the SVD method by about 2 dB using a lower implementation complexity. We also show that the CNN AE system is more robust to timing error and imperfect channel estimation than the SVD method.
Metrics
55 File views/ downloads
140 Record Views
Details
- Title
- Efficient Deep Learning-based Equalization and Detection Systems for Digital Communications
- Creators
- Ahmed Aboutaleb
- Contributors
- Benjamin Belzer (Advisor)Krishnamoorthy Sivakumar (Advisor)Roger Wood (Committee Member)Thomas Fischer (Committee Member)
- Awarding Institution
- Washington State University
- Academic Unit
- Electrical Engineering and Computer Science, School of
- Theses and Dissertations
- Doctor of Philosophy (PhD), Washington State University
- Publisher
- Washington State University
- Number of pages
- 152
- Identifiers
- 99900606756801842
- Language
- English
- Resource Type
- Dissertation