EE Seminar: Zero-Shot Linear and Nonlinear Channel Equalization and Decoding using Variational Autoencoders
Speaker: Avi Caciularu
M.Sc. student under the supervision of Prof. David Burshtein
Wednesday, May 15th, 2019 at 15:30
Room 011, Kitot Bldg., Faculty of Engineering
Zero-Shot Linear and Nonlinear Channel Equalization and Decoding using Variational Autoencoders
Abstract
A new maximum likelihood estimation approach for zero-shot unsupervised blind channel equalization and decoding, using variational autoencoders (VAEs), is introduced. We first consider the reconstruction of uncoded data symbols transmitted over a noisy intersymbol interference (ISI) channel. Significant and consistent improvements in the error rate of the reconstructed symbols, compared to existing blind equalization methods such as constant modulus equalization, are demonstrated. In fact, for the channels that were examined, the performance of the new VAE equalizer that does not require a pilot signal, was close to the performance of a non-blind adaptive linear minimum mean square error equalizer that requires a pilot signal. The new equalization method enables a significantly lower latency channel acquisition compared to other algorithms. The VAE equalizer uses a convolutional neural network with two layers and a very small number of free parameters. Although the computational complexity of the VAE equalizer is higher compared to CMA, it is still reasonable, and the number of free parameters to estimate is small. The results have also been extended to unsupervised equalization over non-linear channels, and to coded communication using low-density parity-check (LDPC) codes, with substantial improvements compared to baseline methods, e.g. expectation maximization (EM) using Turbo equalization.
