Imperceptible Visible Light Communications Based on Modified Just Imperceptible Difference and Aided by Deep Neural Network

Abstract

Indoor visible light communications (VLCs) are not supported in lighting restricted scenarios such as theater, cinema, dim sickroom or bedroom. Thus, different from radio frequency (RF) based communication technologies, such as WiFi, VLC is not “always on”. The “always on” VLC named imperceptible VLC (iVLC) has been proposed, where human cannot perceive glaring nor flicker during the communications. The flicker problem can be solved by increasing the light pulse frequency. In this paper, we propose a two-dimensional characteristic channel analysis structure by considering the different features of communication and light perception channels in iVLC system. The modified just imperceptible difference (JID) has been derived. Based on the modified JID, the upper bounds of average optical power are derived in both direct and reflected light perception scenarios. To reduce the impacts of indoor multiple reflection channel interference and light-emitting diodes (LEDs) transient behaviour in iVLC system where communication signals are modulated in ultra-short pulses, we propose the multi-quadric kernel and deep neural network (DNN) based hard-max pulse position classifier (MQK-DNN-HPPC). Numerical results show that the bit error rate (BER) and synchronization performances of iVLC system are improved by applying MQK-DNN-HPPC compared with the soft-max based DNN algorithm and traditional detection algorithm.