Application of canonical correlation analysis in detection in presence of spatially correlated noise

摘要

Abstract: A new approach to the problem of detecting the numberof signals in unknown colored noise environments ispresented. Based on an assumption that the noise iscorrelated only over a limited spatial range, theprinciple of canonical correlation analysis is appliedto the outputs of two spatially-separated arrays. Thenumber of signals is determined by testing thesignificance of the sample canonical correlationcoefficients. The new method is shown to work well inboth white and unknown colored noise situations anddoes not require any subjective threshold setting.Instead, a set of threshold values are generatedaccording to a specified or desired false alarm rate.Simulation results are included to illustrate thecomparative performance of the proposed canonicalcorrelation technique (CCT), versus the well- known AICand MDL criteria, in colored noise. It is found thatthe performance of the AIC and MDL criteria degradevery rapidly as the degree of color in the noiseincreases. On the other hand, the performance of theCCT method is relatively insensitive with respect tovariations in degree of color.!27