Detection in colored non-Gaussian noise using cumulants

The authors consider detection in colored linear non-Gaussian noise of unknown covariance. Higher-order cumulants of non-Guassian processes are estimated in the presence of unknown deterministic and/or Gaussian signals. This allows either parametric or non-parametric estimation of the covariance of the non-Gaussian process via its cumulants. The approach to detection is to augment methods based on second-order statistics using cumulants. A solution is proposed for detection of deterministic signals based on matched filters which incorporate cumulants, where the resulting solutions are valid under either detection hypothesis. This allows for single record detection and obviates the need for prewhitening using noise-only training records. Detection of random Gaussian signals in non-Gaussian noise is also considered.<>