Unfolding Neural Networks for Compressive Multichannel Blind Deconvolution

Abstract

We propose a learned-structured unfolding neural network for the problem of compressive sparse multichannel blind-deconvolution. In this problem, each channel's measurements are given as convolution of a common source signal and sparse filter. Unlike prior works where the compression is achieved either through random projections or by applying a fixed structured compression matrix, this paper proposes to learn the compression matrix from data. Given the full measurements, the proposed network is trained in an unsupervised fashion to learn the source and estimate sparse filters. Then, given the estimated source, we learn a structured compression operator while optimizing for signal reconstruction and sparse filter recovery. The efficient structure of the compression allows its practical hardware implementation. The proposed neural network is an autoencoder constructed based on an unfolding approach: upon training, the encoder maps the compressed measurements into an estimate of sparse filters using the compression operator and the source, and the linear convolutional decoder reconstructs the full measurements. We demonstrate that our method is superior to classical structured compressive sparse multichannel blind-deconvolution methods in terms of accuracy and speed of sparse filter recovery.

Iterative deep neural networks based on proximal gradient descent for image restoration

“…Zhang and Ghanem [25] achieved proximal mapping related to sparsity-inducing regularizer without handcraft parameter adjustment. Tolooshams et al [26] utilized an unfolding autoencoder neural network with an accelerated proximal gradient to learn compression matrix. Based on prior knowledge, model-based iterative networks with stationary layers are interpreted as the convolution and activation operations.…”

Interpretable deep learning for deconvolutional analysis of neural signals

“…Importantly, this generative model does not require detailed assumptions about the data: it provides domain knowledge infor-mation, without restricting the model's output in such a way that important features of the data would be missed. Following seminal work in algorithm unrolling [39], numerous applications have been developed across several fields, including computational imaging (e.g., super-resolution [42] and image deblurring [43]), medical imaging [44,45], identification of dynamical systems [46], remote sensing applications (e.g., radar imaging [47]) or source separation in speech processing [48].…”