Detecting and Tracking Separating Objects Using Direct Monopulse Measurements

A significant complication to the ballistic missile tracking problem is the tendency for target "spawn": one ballistic object becomes two, several or several hundred, of which only one or a few are of concern (warheads), while uninteresting objects such as chaff and boost-rockets comprise the rest. To discern which is which takes time; and during that time it remains vital to maintain surveillance on all objects. Unfortunately these objects are closely-spaced (CSOs), and detection/tracking for CSOs is notoriously difficult. One approach is to use radar waveforms of such high resolution that all objects are "resolved" by a sharp ambiguity function peak. However, not all radar systems have the required bandwidth or operational modes, and inclusion of those modes in new radars would impact the cost. It appears that with unresolved targets there is little that can be done to improve tracking until the targets become resolved. Nonetheless, recent work with advanced statistical modeling of monopulse radar returns has questioned the need for such high resolution: signal processing techniques can "see" inside a previously-closed resolution cell in a similar way that a mis-formed Hubble mirror can be corrected to see clearly.In this presentation we proffer two approaches. In the first, we discuss the maximum-likelihood approach: it turns out that it can be shown that up to 5 targets can be resolved from between two matched filter samples, rather than the one-per- sample case that might otherwise be thought. We also will present an even more advanced statistical approach known as particle filtering. The standard tracking approach is to work with detection "hits"; that is, to separate the signal processing and tracking. The particle filter is a more natural integrated approach, and we have found tracking errors reduced by 50% or more by its use along with significant increase in speed of detection of multiple target spawn.