Reducing the complexity of the space-time channel estimate at minimum risk

In a mobile communication system there is the need to define a channel length that could be useful to describe many different and varying propagation environments, over-parameterization is a simple solution. The channel matrix estimated when using an array of antennas may contain more parameters than those really needed to parsimoniously describe the space-time channel. The problem is even worse when a long channel has to be estimated from short training sequences. Classically the reduction of complexity is carried out by masking some samples of the channel estimate according to a threshold heuristically defined. We propose to adaptively classify (and mask) the estimated channel samples by minimizing the Bayes risk for space-time systems. The parameters of the probability densities are iteratively estimated from the estimated channel samples and contribute to define the optimum threshold. Simulation shows that in Rayleigh fading channels the adaptive threshold is close to the one that minimize the mean square error. The performance can improve by approx. 3-5 dB in signal to noise ratio when the method is applied to reduce the complexity of space-time channels in a CDMA system with realistic propagation environments.