Noise Reduction Techniques

Signal Processing based on higher-order cumulants is a powerful tool to detect and estimate non-Gaussian signals in the presence of colored Gaussian noise of unknown power spectral density. Since signals used in communication systems are symmetrically distributed, the third-order cumulants (or bispectra) are theoretically zero, and hence, we are forced to use fourth-order cumulants (or trispectra). The blind equalization problem, in the presence of colored Gaussian noise, will be studied in this proposal. Existing algorithms, as well as proposed theoretical and practical extensions will be investigated. Both transversal as well as lattice structures, and algorithms based on conventional cumulants, as well as the trispectrum will be investigated. A candidate architecture amenable to implementation on existing signal processing hardware will be developed. Promising algorithms will be selected based on their theoretical performance analysis, the computational complexity, and their suitability to hardware implementation. Algorithms will be validated via simulations, and possibly against data to be obtained from the DoD.