Impulsive Noise Parameter Estimation: A Deep CNN-LSTM Network Approach

Abstract

Impulsive noise is a widespread phenomenon that can hinder the performance of wireless communication systems, especially given the wireless medium’s dynamic channel characteristics. To alleviate the effects of the noise, several mitigation techniques can be resorted to. In this context, information on the impulsive noise’s statistical parameters is generally required in order to optimize the mitigation technique performance. To this end, this study proposes a deep learning approach for the estimation of the statistical parameters of impulsive noise with memory where the received signal is impaired by Rayleigh fading and two-state Markov-Gaussian impulsive noise. A deep Convolutional Neural Network - Long-Short Term Memory (CNN-LSTM) model is designed to extract this information. Provided results demonstrate that the model outperforms baseline approaches and is able to efficiently learn and infer the impulsive noise parameters from a relatively small number of symbols, making it suitable for impulsive noise detection and mitigation in dynamic environments.