Optimal training over the Gauss-Markov fading channel: a cutoff rate analysis

We consider the problem of optimal allocation of resources between training and data for transmission over a Gauss-Markov fading channel. Inaccurate channel state information (CSI) is available at the receiver through periodic training. There is no feedback, so that CSI is not available at the transmitter. We study MMSE estimators that predict the current channel state based on: all past pilot observations; only the most recent pilot observation; and the most recent and next in the future pilot observations. We analyze the optimal training energy and periodicity for each of these estimators. We show that optimizing the energy and periodicity of training results in significant energy savings over a sensible, but unoptimized, approach, particularly for rapidly varying channels.