BNET: A Neural Network Approach for LLR-Based Detection in the Presence of Bursty Impulsive Noise

Abstract

Ensuring reliable wireless communications is often a major challenge in environments characterized by bursty impulsive noise. Unfortunately, effective receivers adapted to this problem employ recursive techniques that are computationally complex and challenging to implement where computing and battery resources are limited. As a solution, we propose a computationally efficient neural network-based framework to approximate the log-likelihood ratio function produced by the Bahl Cocke Jelinek Raviv (BCJR) algorithm for single-carrier communication systems. The results show that our method can match the BCJR algorithm’s BER performance in many scenarios and even outperform it when some noise model parameters change over time.