Fast variational Bayesian learning for channel estimation with prior statistical information

This work addresses the issue of incorporating prior statistical information about the channel into the pilot-assisted OFDM equalisation process for the purpose of increasing performance and speed. This is performed by considering certain informative prior distributions for the channel coefficients. Assuming a sparse multipath channel, the equalisation problem is formulated in a Bayesian setting and inference is performed in the well-known framework better known as Sparse Bayesian Learning (SBL). The previously proposed Fast Variational SBL (FVSBL) algorithm is capable of efficient inference in a true Bayesian setting but only in the case of uninformative prior distributions. We use a set of extensions to the FVSBL approach to mitigate these problems. These modifications stem from a refined fixed-point analysis. Empirical evidence supports the proper function of the proposed method. Results from a real-world channel estimation problem suggest that the proposed method achieves excellent performance.