Estimation of mobile terminal position in cellular networks

Mobile terminal location has attracted much interest for emergency communications, location sensitive browsing, and resource allocation. The topic of this thesis is location estimation based on propagation distance measurements from fixed location base stations. The relationship between the measurements and terminal location is complicated by Non Line of Sight (NLOS) propagation when the shortest, distance straight line path from receiver to transmitter is obstructed, multipath propagation, receiver noise, and interference noise. This thesis introduces non-parametric estimation and dynamic filtering for accurate location estimation.; Non-parametric estimators perform better than the previously used parametric estimators for a simulated urban micro-cell environment. The new estimators are robust with respect to NLOS propagation and variations of the measurement noise. An accurate dynamic model for mobile terminal motion with a state space model describing the physical rules governing motion and a control model for the human control input is developed based on vehicular traffic engineers' observations. A dynamic filter using the dynamic model is presented that combines information from measurements made at different times to create improved location estimates. A novel generalized multiple model filter is created incorporating the dependency of the switching probabilities of the control input on the location state into the filtering algorithm. This filter is more accurate than previous location techniques and is robust to variations in the motion models and the propagation environment.; Lower bounds are calculated for the performance of location estimators. A lower bound that considers NLOS propagation measurements on location estimation is developed. A novel bound on the performance of multiple model filter algorithms is also presented.; The location methods presented reduce the root mean square location error from 100 meters, for the previous methods, to 10 meters for a range error standard deviation of 15 meters. They allow for location prediction in resource allocation algorithms to facilitate efficient cellular networks to carry more data using less bandwidth.