Wireless communication systems operating in industrial environments are often subject to bursty impulsive noise and frequency-selective fading that significantly degrade their performance. Under these conditions, conventional orthogonal frequency division multiplexing (OFDM) receivers struggle to maintain reliable communication. In this paper, we address these challenges by introducing a reliable receiver design consisting of two key components. First, we develop an adaptive memory-aware noise covariance estimation method using neural networks to dynamically predict the noise level for each received symbol. Second, we propose a channel state information (CSI) enhancement method using two-dimensional average pooling and interpolation (2DAPI) to smooth out inaccuracies in CSI caused by bursty impulsive noise. We further derive a theoretical bit error rate (BER) performance approximation and incorporate a lower bound benchmark for rigorous evaluation. The results of our extensive simulations demonstrate the extreme power efficiency of the proposed receiver, which requires less than 10% of the transmit power as compared to conventional OFDM receivers that use linear minimum mean square error (LMMSE) equalizers and least squares (LS) CSI estimation to achieve the same BER. In addition, performance analyses across various modulation schemes and fading models confirm the reliability and adaptability of our design in diverse operational environments.

In modern wireless communication systems, using high-order modulation is often key to achieving high data rates while maintaining a good spectral efficiency. By packing more bits into each transmitted symbol, more data can be sent over a given bandwidth, but that also makes the transmission more vulnerable to fading and noise. Moreover, traditional methods often isolate the design and optimization of transmitters and receivers, leading to poor error rate performances. To tackle these challenges, we propose a novel end-to-end (E2E) optimization strategy that includes a trainable constellation at the transmitter and a convolutional neural network (CNN)based demapper at the receiver. The transmitter and receiver are jointly trained to handle channel fading and detect symbols with minimal errors. Our findings demonstrate that the proposed transceiver architecture exhibits a significantly lower bit-error rate (BER) than traditional modulation methods and earlier dense layer-based models, especially for high-order modulation.

The problem of relay selection is pivotal in the realm of cooperative communication. However, this issue has not been thoroughly examined, particularly when the background noise is assumed to possess an impulsive characteristic with consistent memory as observed in smart grid communications and some other wireless communication scenarios. In this paper, we investigate the impact of this specific type of noise on the performance of cooperative Wireless Sensor Networks (WSNs) with the Decode and Forward (DF) relaying scheme, considering Symbol-Error-Rate (SER) and battery power consumption fair-ness across all nodes as the performance metrics. We introduce two innovative relay selection methods that depend on noise state detection and the residual battery power of each relay. The first method encompasses the adaptation of the Max-Min criterion to this specific context, whereas the second employs Reinforcement Learning (RL) to surmount this challenge. Our empirical outcomes demonstrate that the impacts of bursty impulsive noise on the SER performance can be effectively mitigated and that a balance in battery power consumption among all nodes can be established using the proposed methods.