Robust structured interference rejection combining

The problem of suppressing interference and jamming in a wireless communications receiver is a fundamental one. Many existing algorithms for this task are based on prewhitening of the received data so that the interference appears as white noise, and a conventional detector can be applied. However, such methods, in general, need to know the structure of the interference signal (for example, the rank of its autocorrelation matrix, m). The classical approach to the latter problem is either to assume that m is fixed (known), or to use rank detection techniques to estimate m first and then apply prewhitening in an adaptive fashion. This is suboptimal for two reasons: First, information is discarded when one takes a decision on m; second, no optimal rank detection methods exist. We present a new approach to the problem that circumvents the estimation of m. The key idea is to use a trained mixture model ("mixture of experts") to describe the received data, where each mixture component is associated with a particular rank, m. The optimal receiver for this model consists of a bank of detectors (one for each possible m) whose outputs are combined. One can think of this scheme as a way of taking "soft decisions" on the rank, m. Via numerical examples, we show that the new method significantly outperforms methods based on rank detection, at a modest increase in complexity.