Sparse Bayesian RVM regression based channel estimation for IM/DD OFDM-VLC systems with reduced training overhead

We propose a novel channel estimation technique for intensity modulation/direct detection (IM/DD) based orthogonal frequency division multiplexing visible light communication (OFDM-VLC) systems, utilizing sparse Bayesian dual-variate relevance vector machine (RVM) regression. By exploiting sparse Bayesian framework, dual-variate RVM regression can provide accurate estimation of the real and imaginary parts of the complex channel response, and therefore the channel response can be estimated to perform channel compensation. Simulation results show that a 200 Mb/s OFDM-VLC system using sparse Bayesian RVM regression based channel estimation with only one complex training symbol (TS) achieves nearly the same bit error rate (BER) performance as the system using conventional time domain averaging (TDA) based channel estimation with a total of 20 complex TSs, indicating a significant reduction of training overhead. Moreover, by employing a fast marginal likelihood maximization method, the sparse Bayesian RVM regression based channel estimation can be computational efficient for practical application in high-speed OFDM-VLC systems.