Channel-Attention-Enhanced LSTM Neural Network Decoder and Equalizer for RSE-Based Optical Camera Communications

Abstract

In an RGB-LED-based optical camera communication system, it is an essential goal to have better performance in the data rate and BER. However, in a higher symbol rate, due to the conventional sampling algorithm, the deterioration of transmission performance brought by the inter-symbol interference and inter-channel interference is significant. Innovatively, in this paper, the sub-image obtained by a captured frame of received video is encoded by a channel-attention-Net-based encoder to generate a descriptor without existing sampling methods. Moreover, we propose an LSTM-based equalizer to decode the descriptor and mitigate transmission performance deterioration. Utilizing the long-short-term memory of an LSTM unit, an equalizer not only can reduce bit error rates but also increase the data rate. The experimental results show that at a symbol rate of 46 kbaud/s, a record-high data rate at 44.03 kbit/s is achieved under random data transmission while still meeting the pre-forward error correction requirement.