A handoff algorithm for wireless systems using pattern recognition

A new handoff algorithm based on recognition of patterns in the received signal power is presented. The handoff algorithm uses the constancy of the large scale signal variation with respect to a base station to improve handoff performance in comparison with the conventional hysteresis rule. A probabilistic neural network (PNN) is used for a pattern classifier. For a given environment, a training run is performed. A set of training patterns consists of averaged signal power samples (in dB) from nearby base stations within adjacent spatial windows. A probability of failure based on signal strength is defined and determines the possible serving base station(s) for each training class. A handoff is performed if a nearby class in the sequence of classes requires a handoff. Performance comparison of this handoff algorithm and the hysteresis rule is made for environments with four and five base stations. Simulation results indicate that for a fixed probability of failure, the pattern recognition-based handoff algorithm results in considerably fewer handoffs in comparison to the hysteresis rule. This reduction of unnecessary handoffs decreases the signalling load