A Measurement Correlation Algorithm for Line-of-Bearing Geo-Location
Passive geo-location of ground targets is commonly performed by surveillance aircraft using Direction Finding (DF) angles. These angles define the line-of-sight from the aircraft to the target and are computed using the response of an antenna array on the aircraft to the target's RF emissions. Aircraft that depend entirely upon DF angles for geo-location will often convert each DF angle measurement to a Direction of Arrival (DOA) angle measurement and use these values for geo-location. DOA is the angle equivalent to antenna azimuth when defined relative to a local-level coordinate frame at the current aircraft position. DOA is computed using antenna azimuth, an estimate of the elevation angle to the target, the antenna array mounting angles on the aircraft, and aircraft navigation system output. Associated with each angle measurement is a Line-of-Bearing (LOB) that originates at the aircraft and, if perfect, would pass through the target's position.
Most geo-location scenarios faced by surveillance aircraft involve the existence of multiple target signal sources within the area of operation. As a result, each new LOB must be correlated with a specific target emitter before that target's position estimate can be updated by the geo-location algorithm. Geo-location performance is typically degraded in a dense emitter environment due to the difficulty of correlating each LOB with the correct target, and of preventing the generation of numerous false or "ghost" targets. Many LOB correlation algorithms form measurement clusters by evaluating all possible combinations of measurements using some performance metric. Also, many correlation algorithms will simply compute the target position estimate using the cluster having the most measurements.
This algorithm is based on statistical clustering of measurements but does not require that the number of clusters be specified in advance, as with methods such as k-means clustering. This algorithm determines the cluster of LOBs that maximizes the target position log-likelihood function when compared to all candidate clusters. The candidate clusters are those that pass a Mahalanobis distance association criterion - an exhaustive search over all possible measurement combinations is not performed. Once the optimal cluster of LOBs has been determined, a target position estimate is computed using the cluster, the cluster is removed from the set of all measurements, and the process is repeated. This continues until no additional clusters can be formed.
A detailed paper on the technique is available upon request.
Mr. M. T. Hickman