Feasibility Study of Device Synthesis of Non-Linear Filters.

Abstract : Efforts to date have shown that with careful tailoring to the problem at hand and by taking maximum advantage of recent advances in parallel/pipeline array processor architecture, the nonlinear filter is a practical estimation technique. The computational effort is significantly greater than for conventional linear or linearized (e.g. extended Kalman filtering) techniques, but the performance advantage may be significant where the observation is a significantly nonlinear function of the estimated states, and/or where the observation noise (or plant noise) is significantly non-Gaussian. The research has concentrated on the phase estimation problem which has proven to be an ideal test problem. However, with the techniques that have now been developed and proven, it is appropriate to consider a larger class of problems. Potential applications should meet the following criteria: (1) Performance advantages are of significant economic value, (2) Observations are essentially nonlinear functions of the states that have to be estimated where the linear filter leads to unacceptable performance, (3) Measurement (or state) noise is highly non-Gaussian. Excellent example meeting all the above criteria would be the ELF communications problem. Other examples occur in deep space communication and in detection and tracking problems in general. (Author)