Channel Estimation and Prediction for MIMO OFDM Systems : Key Design and Performance Aspects of Kalman-based Algorithms

Abstract: Wireless broadband systems based on Orthogonal Frequency Division Multiplexing (OFDM) are being introduced to meet demands for high data transfer rates. In multiple users systems, the available bandwidth has to be shared efficiently by several users. The radio channel quality will fluctuate, or fade, as users move. Fading complicates the resource allocation, but channel prediction may alleviate this problem. A flexible and computationally inexpensive state space representation of fading channels is here used in conjunction with a Kalman filter, operating on special-purpose reference signals, to track and predict fading OFDM channels. The thesis investigates key design and performance aspects of such estimators. Taking a probabilistic approach, we interpret the output of the Kalman filter as a full representation of a state of knowledge about the fading channels, given whatever information is at hand. For systems analysis, this permits conclusions to be drawn about channel estimation and prediction performance based on only vague information about the fading characteristics of the channel rather than on actual channel measurements. This is an alternative to conducting classic simulation studies. Various reference signal designs are studied and good design choices are recommended. Superimposed reference signal schemes are also proposed for and evaluated in cases where multiple signals are received, e.g. in multi-user (MU), multi-input multi-output (MIMO), or coordinated multi-point (CoMP) settings. By using time-varying reference signals, channel estimation and prediction performance is shown to be improved considerably in crowded frequency bands. The variation of prediction performance with prediction range and Doppler spectrum characteristics is investigated. For link adaptation, we derive the appropriate metric on which adaptation decisions should be used. The probability density function for this metric is derived for general MIMO channels. Link adaptation is studied for a single link system when channel prediction and estimation errors are present, both for uncoded systems and systems using large block codes with soft decoders. Various aspects of channel model acquisition are addressed by conducting studies on measured channels. Owing to the use of special matrix structures and fast convergence to time-invariant or periodic solutions, we find the Kalman filter complexity to be reasonable for future implementation. Finally, expressions for the impact of modelling errors are derived and used to study the impact of modelling errors on channel prediction performance in some example cases.