Machine-Learning Based Equalizers for Mitigating the Interference in Asynchronous MIMO OWC Systems

Abstract

The error performance of the optical wireless communication (OWC) link suffers from the effects of atmospheric turbulence and pointing errors. The multi-input-multi-output (MIMO) system can combat the damage by transmitting diverse replicas of symbols to the receivers, i.e. the spatial diversity. However, different delays between the transceivers can introduce inter-symbol interference (ISI) which degrades the system performance. The delays during transmission are mainly caused by placing locations and optical path differences. This article proposes algorithms to mitigate the ISI based on machine learning techniques, including both neural networks and the genetic algorithm. In the case of multi-input-single-output (MISO) system, we propose an algorithm based on a bidirectional long short-term memory (LSTM) recurrent neural network (RNN), which works as an equalizer. In the case of single-input-multiple-output (SIMO) system, additional delayers are utilized to align the signals in different apertures. The problem of deducing the values of the delayers is considered as seeking the minimum value in a high-dimensional space. With the help of the genetic algorithm, optimal values of the delayers are maintained, which is named as GAD (genetic algorithm-based delayers). In a more general MIMO case, the GAD and LSTM equalizers are further combined to deal with the ISI issue in the asynchronous MIMO OWC systems, (i.e. GAD-LSTM). Both experimental and simulation results show the remarkable performance improvement of the proposed method over conventional methods.