Machine learning-enabled MIMO-FBMC communication channel parameter estimation in IIoT: A distributed CS approach

Abstract

Compressed Sensing (CS) is a Machine Learning (ML) method, which can be regarded as a single-layer unsupervised learning method. It mainly emphasizes the sparsity of the model. In this paper, we study an ML-based CS Channel Estimation (CE) method for wireless communications, which plays an important role in Industrial Internet of Things (IIoT) applications. For the sparse correlation between channels in Multiple Input Multiple Output Filter Bank MultiCarrier with Offset Quadrature Amplitude Modulation (MIMO-FBMC/OQAM) systems, a Distributed Compressed Sensing (DCS)-based CE approach is studied. A distributed sparse adaptive weak selection threshold method is proposed for CE. Firstly, the correlation between MIMO channels is utilized to represent a joint sparse model, and CE is transformed into a joint sparse signal reconstruction problem. Then, the number of correlation atoms for inner product operation is optimized by weak selection threshold, and sparse signal reconstruction is realized by sparse adaptation. The experiment results show that the proposed DCS-based method not only estimates the multipath channel components accurately but also achieves higher CE performance than classical Orthogonal Matching Pursuit (OMP) method and other traditional DCS methods in the time-frequency dual selective channels.