2019/05/21 by Avi Caciularu, Caciularu, Avi, David Burshtein +1
Computer Science · Biochemistry, Genetics and Molecular Biology · #Blind Source Separation Techniques #Fractal and DNA sequence analysis #Algorithms and Data Compression
paper · pdf · doi:10.48550/arxiv.1905.08795
A new approach for blind channel equalization and decoding, variational\ninference, and variational autoencoders (VAEs) in particular, is introduced. We\nfirst consider the reconstruction of uncoded data symbols transmitted over a\nnoisy linear intersymbol interference (ISI) channel, with an unknown impulse\nresponse, without using pilot symbols. We derive an approximate maximum\nlikelihood estimate to the channel parameters and reconstruct the transmitted\ndata. We demonstrate significant and consistent improvements in the error rate\nof the reconstructed symbols, compared to existing blind equalization methods\nsuch as constant modulus, thus enabling faster channel acquisition. The VAE\nequalizer uses a convolutional neural network with a small number of free\nparameters. These results are extended to blind equalization over a noisy\nnonlinear ISI channel with unknown parameters. We then consider coded\ncommunication using low-density parity-check (LDPC) codes transmitted over a\nnoisy linear or nonlinear ISI channel. The goal is to reconstruct the\ntransmitted message from the channel observations corresponding to a\ntransmitted codeword, without using pilot symbols. We demonstrate improvements\ncompared to the expectation maximization (EM) algorithm using turbo\nequalization. Furthermore, unlike EM, the computational complexity of our\nmethod does not have exponential dependence on the size of the channel impulse\nresponse.\n