Neural network techniques for multistatic hyperbolic vehicle positioning

Multistatic hyperbolic position location systems can accurately predict vehicle location in Intelligent Transportation Systems (ITS) applications. The hyperbolic position location (PL) techniques described can provide centimeter accuracy for short range vehicle tracking and control applications. These PL techniques can also be used in wireless E-911 and MayDay applications providing geolocation of mobile users within cellular and PCS systems. Hyperbolic position location systems locate a source by measuring the time difference of arrival (TDOA) of a sources signal between multiple receivers. These TDOA measurements define a set of range difference measurements which describe hyperbolas with the receivers at the foci. The intersection of the hyperbolas provides the position location estimate of the source. If there are no timing errors, a unique position location solution exists. When there are timing errors, more sophisticated algorithms must be used to provide a "best fit" to the measurement data. This paper compares two methods of solving this problem. The first method calculates the cross points of all possible pairs of hyperbolas having a common focus. These points are clustered and centroided to obtain the final solution. Additional tracking can be done to eliminate false alarms and improve accuracy. The second method uses a neural network approach to solve the problem. This approach has some advantages over the first approach. First, it can process all the information at once so that there is no need to do the centroid process. Second, the location and tracking processes can be done by a single network. Third, the trained neural network will be able to process information faster than the first technique.