B. E. McDonald
Naval Res. Lab., Washington, DC 20375-5000
For low signal-to-noise (S/N) data, an algorithm has been reported that is very effective when the shape of the noise spectrum is known and smooth as a function of frequency. The method uses the phases of the spectral components as free parameters in a variational problem. Improvements on the basic method to help relax the idealistic assumptions made have been investigated. Methods employing empirical orthogonal functions (EOFs) to exploit further degrees of freedom in the solution are presented. For cases of moderate S/N, an algorithm is presented that attempts to utilize signal-free intervals in improving S/N.