An RNN-based decoding scheme for IM/DD visible light communication systems

In Visible Light Communication (VLC) systems, the integration of Forward Error Correction (FEC) and Run-Length Limited (RLL) coding is essential for ensuring transmission reliability and flicker-free operation. Despite their efficacy, traditional joint decoding schemes often suffer from prohibitive computational overhead. This paper introduces a Deep Learning-based decoding framework designed to replace the conventional concatenated RLL-FEC decoding chain. The proposed architecture employs a single-layer Recurrent Neural Network (RNN) followed by a dense fully-connected layer. Furthermore, a confidence-bit demodulation strategy is implemented to refine the input features for the RNN. Empirical results demonstrate that for short-packet scenarios, the proposed RNN-centric decoder achieves error-correction performance on par with state-of-the-art benchmarks while delivering a significant leap in processing throughput.