Machine Learning-Based Interference Mitigation in Wireless Networks

Abstract

The congestion of the electromagnetic spectrum and the diminishing size of cells in wireless networks have made crosstalk between base stations and consumers a significant issue. While hand-crafted functional blocks and coding schemes are established methods for ensuring reliable data transport, deep learning-based techniques have recently garnered significant interest in communication system modeling [1, 2]. This research presents a Neural Network (NN) based signal processing approach that integrates with conventional DSP techniques to address the interference issue in real-time. This approach does not need any feedback mechanism between the receiver and transmitter, making it very appropriate for low-latency and high data-rate applications such as autonomy and augmented reality. Recent research has focused on using Reinforcement Learning (RL) at the control layer to limit interference; however, our technique is innovative since it incorporates a neural network for signal processing at the baseband data rate and inside the physical layer. We illustrate the "Deep Interference Cancellation" method with a convolutional LSTM autoencoder. The use of QAM-OFDM modulation to the data results in a substantial enhancement of the symbol error rate (SER). We moreover examine the hardware implementation, encompassing latency, power consumption, memory needs, and chip space