A Novel Learning-Based Estimation Scheme for Communication Over Impulsive Noise Channels

Abstract

This paper proposes a novel statistical hybrid neural network (S-HNN) based estimation of impulse noise infested wireless communication channels. Spatial fading characteristics are found using a convolutional neural network (CNN), while long short-term memory (LSTM) network extracts temporal information over subsequent time horizons. Finite lag samples are employed to extract the channel gain distribution based on multiple recycling of the CNN-LSTM network. The proposed S-HNN framework for orthogonal frequency division multiplexed (OFDM) communication channel with subcarrier spacing of 15 kHz, sampling rate of 15.36 MHz, IFFT size of 1024, and various pilot density deployments is shown to outperform the existing state-of-the-art channel estimation techniques in terms of 50% reduced training length and nearly 49% saving in training time.