Some results on the linear prediction of bandlimited fading processes

Channel state information at the transmitter (CSIT) is useful in many wireless communication systems, and is usually obtained by feedback in frequency division duplex (FDD) systems. To compensate for the feedback delay, prediction techniques are often used at the transmitter or the receiver side. We study the design and performance analysis of linear prediction of a bandlimited complex Gaussian fading process. We consider prediction based on both noiseless and noisy past samples of the fading process and study both finite and infinite linear minimum mean square error (MMSE) prediction filters. In particular, we derive the MMSE achievable by an infinite length linear filter in the case of prediction based on noisy past samples. We also study the design of robust prediction filters whose coefficients are independent of the second order statistics of the fading process.