Least Squares Prediction Applied to Adaptive RAKE Receivers

Application of a RAKE receiver to a multipath mobile communication environment can improve the bit error rate of a receiver through the combination of multipath components of a received spread spectrum signal In order to achieve this, some estimate of the channel impulse response must be made prior to receiving data. Estimates of the channel impulse response for previously received data may be made, albeit with some uncertainty in the case of estimated data. Assuming a linear relationship between the previous channel impulse response estimates and the current channel impulse response, least squares prediction has been applied to this problem in order to reduce the bit error rate from that achieved by a non-predictive algorithm. Estimation of the prediction weights is often achieved through the application of training periods where known data is transmitted. However, this training must be repeated from time to time to cope with a non-stationary channel, and therefore reduces the bandwidth available to the communicating entities. this paper describes various algorithms for performing the updating of the prediction weights using the data decisions previously made. The result of this is that the prediction algorithm will also apply to non-stationary channel situations with a limited degree of continuous retraining from "training blocks" being required.