Airborne/spacebased radar STAP using a structured covariance matrix

It is shown that partial information about the airborne/spacebased (A/S) clutter covariance matrix (CCM) can be used effectively to significantly enhance the convergence performance of a block-processed space/time adaptive processor (STAP) in a clutter and jamming environment. The partial knowledge of the CCM is based upon the simplified general clutter model (GCM) which has been developed by the airborne radar community. A priori knowledge of parameters which should be readily measurable (but not necessarily accurate) by the radar platform associated with this model is assumed. The GCM generates an assumed CCM. The assumed CCM along with exact knowledge of the thermal noise covariance matrix is used to form a maximum likelihood estimate (MLE) of the unknown interference covariance matrix which is used by the STAP. The new algorithm that employs the a priori clutter and thermal noise covariance information is evaluated using two clutter models: 1) a mismatched GCM, and 2) the high-fidelity Research Laboratory STAP clutter model. For both clutter models, the new algorithm performed significantly better (i.e., converged faster) than the sample matrix inversion (SMI) and fast maximum likelihood (FML) STAP algorithms, the latter of which uses only information about the thermal noise covariance matrix.     

Convergence bounds of an SMI/Gram-Schmidt canceler in colored noise

The performance of the sampled matrix inversion (SMI) adaptive algorithm in colored noise is investigated using the Gram-Schmidt (GS) canceler as an analysis tool. Lower and upper bounds of average convergence are derived, indicating that average convergence slows as the input time samples become correlated. When the input samples are uncorrelated, the fastest SMI algorithm convergence occurs. When the input samples are correlated then the convergence bounds depend on the number of channels N, the number of samples per channels K , and the eigenvalues associated with K×K correlation matrix of the samples in a given channel. This matrix is assumed identical for all channels     

Low sidelobe radar waveforms derived from orthogonal matrices

Novel waveforms are described that have low sidelobes when individual or multiple waveforms are approximately processed. They are related to orthogonal matrices that may be associated with complementary sequences and also with periodic waveforms having autocorrelation functions with constant zero-amplitude sidelobes. Also described are sets of sequences whose cross-correlation functions sum to zero everywhere. A potential application is the elimination of ambiguous range stationary clutter     

Convergence properties of Gram-Schmidt and SMI adaptive algorithms

The open-loop Gram-Schmidt (GS) canceler is shown to be numerically identical to the sampled matrix inversion (SMI) algorithm in the transient state if infinite numerical accuracy is assumed. Two forms of the GS canceler are discussed and analyzed: concurrent and nonconcurrent processing. Results for concurrent and nonconcurrent SMI cancelers have been obtained in the past by I.S. Reed, J.D. Mallet, and E. Brennan (see ibid., AES-10, p.853-63, 1974) under the assumption that the inputs are Gaussian. Many of those results are reproduced here using the GS structures as an analysis tool. In addition, new results are obtained when the input noises are not Gaussian. The deleterious effect of overmatching the degrees of freedom is discussed     

A comparison of two adaptive detection schemes

Two schemes for adaptive detection are compared: Kelly's generalized likelihood ratio test (GLRT) and the mean level adaptive detector (MLAD). Detection performance, P_D, is predicted for the two schemes under the assumptions that the input noises are zero-mean complex Gaussian random variables that are temporally independent but spatially correlated; and the amplitude of the desired signal is Rayleigh distributed. P_D is computed as a function of the false alarm probability, the number of input channels, the number of independent samples per channel, and the matched filtered output signal-to-noise (S/N) power ratio. In this analysis the GLRT is shown to have better detection performance than the MLAD. The difference in detection performance increases as one uses fewer input samples. However, the required number of samples necessary to have only a 3 dB detection loss for both detection schemes is approximately the same. This is significant since for the present, the MLAD is considerably less complex to implement than the GLRT