In recent years, Visible light communication (VLC) has become a mainstream of optical wireless communication (OWC) and attracted considerable interest in both academia and industry. To ensure reliable data transmission and mitigate flicker, forward error correction (FEC) and run-length limited (RLL) codes are serially implemented in the VLC systems.
In this thesis, I first propose the use of RLL code in serial concatenation with the Polar code as an FEC for VLC. To improve error-correction capability, I employ the soft-input soft-output RLL decoding scheme which produces soft-information for Polar decoder afterward. This soft RLL-Polar decoding scheme obtains best bit error rate (BER) compared with previous works.
Improving the reliability with soft RLL-Polar decoder requires accurate knowledge of the VLC channel, which is challenging to acquire in practical VLC system. I then propose a low-bit demodulator to calculate soft-input for RLL-FEC decoding scheme. This demodulator makes use low-bit analog to digital converter (ADC) to determine the confidence bits of the received signal and statistically characterize the VLC channel to generate soft information to be provided to the soft RLL-FEC decoder.
Finally, I propose deep learning based decoders which replace the concatenation of RLL and FEC decoder to reduce the complexity. The proposed approach consists of a one-layer recurrent neural network (RNN) followed by a fully-connected layer. I also propose the use of low-bit demodulator to generate the input data for the RNN structure. The numerical results show that the proposed RNN-based decoders for short block lengths are capable of achieving close error-correction performance to the state-of-the-art decoders while the processing speed is improved significantly.