Neural Network-Optimized Channel Estimator and Training Signal Design for MIMO Systems with Few-Bit ADCs

Abstract

This paper is concerned with channel estimation in MIMO systems with few-bit ADCs. In these systems, a linear minimum mean-squared error (MMSE) channel estimator obtained in closed-form is not an optimal solution. We first consider a deep neural network (DNN) and train it as a non-linear MMSE channel estimator for few-bit MIMO systems. We then present a first attempt to use DNN in optimizing the training signal and the MMSE channel estimator concurrently. Specifically, we propose an autoencoder with a specialized first layer, whose weights embed the training signal matrix. Consequently, the trained autoencoder prompts a new training signal design that is customized for the MIMO channel model under consideration.

arXiv comment5 pages, 3 figures, to appear in IEEE Signal Processing Letters

References in corpus (7)

• Scaling up MIMO: Opportunities and Challenges with Very Large Arrays
• Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation
• DoReFa-Net: Training Low Bitwidth Convolutional Neural Networks with Low Bitwidth Gradients
• Binarized Neural Networks
• Trained Ternary Quantization
• Channel Estimation and Performance Analysis of One-Bit Massive MIMO Systems
• Uplink Performance of Wideband Massive MIMO with One-Bit ADCs

Cited by in corpus (2)

• SVM-based Channel Estimation and Data Detection for One-Bit Massive MIMO Systems
• Intelligent Radio Signal Processing: A Survey