Deep Learning Based Design of Binary Signalling for Optical Wireless Communication Systems With 2D Receiver

Abstract

Recently, deep learning (DL) techniques have been increasingly applied to communication system design, owing to their powerful capabilities in handling complex channel characteristics. We apply a DL technique to binary signalling design for optical wireless communication system, which incorporates a $2\times 2$ light-emitting diode (LED) transmitter array. The $2\times 2$ LED transmitter array is adopted for two-dimensional (2D) arrayed photodiode receiver. The system encodes binary bit streams into $2\times 2$ LED lighting patterns and the 2D received lighting patterns are interpreted as 2D images, which are decoded for retrieving the binary bit stream. For conventional on-off keying (OOK) signal for optical wireless communication system, it is necessary to choose appropriate 2D lighting patterns to represent binary logic symbols effectively. In this paper, we propose a design algorithm for binary signalling to generate appropriate 2D binary symbols. The DL-based signalling design algorithm is implemented as an autoencoder (AE) structure. It is trained with the set of transmitted and received signal patterns over the physical channel model with additive noise. To validate the proposed signalling design scheme, we adopt a two-step approach. Firstly, the signalling design algorithm generates binary LED signal sets after appropriate training processes. Secondly, from the generated signal sets, the LED patterns of interest are investigated considering their symbol error rate (SER) performance. It is confirmed that the proposed design algorithm provides binary signalling sets that meet the required SER performance. Through this study, it is demonstrated that DL-based signalling design is feasible, and the results are expected to contribute to further research aimed at extending the approach to more practical and complex system scenarios.