A High-Resolution Detector in Multipath Environments

In this paper we introduce a high-resolution CFAR detector for the linear data model. The detector is derived using the generalized likelihood-ratio test (GLRT) framework. The resulting detector has a number of distinctive properties. One property of the detector is that it is most useful when the training and testing data contain the signal-of-interest. In fact, this detector does not need training data although it can utilize it if provided. This is in contrast to most methods, which in order to prevent signal suppression, require a set of signal-free training data to estimate the noise/interference background. Another important property of the detector is that the probability of detection is invariant to the correlation of the signal-of-interest with other signals. This can be particularly important in array processing, where the signal-of-interest is often correlated with one or more multipath components. If the objective of the detector is to discriminate between sources that are very close in terms of the spans of their signal subspaces, then the detector derived herein has very good resolution properties. In array processing, this translates to very good spatial resolution.