A joint design of code and training sequence for frequency-selective block fading channels with partial CSI

In this paper, we propose an iterative algorithm to jointly design codes and training sequences for frequency-selective block fading channels with partial channel state information (CSI) at the receiver. After showing that the maximum-likelihood (ML) decoding metric over channels with partial CSI can be well approximated by the joint maximum-likelihood (JML) decoding metric for combined channel estimation and data detection, we propose to use the JML criterion to search for good codes and training sequences in an iterative fashion. Simulations show that the code and training sequence found by our method can outperform a typical system using a channel code with a separately designed training sequence, in particular when codes of low rates are considered.