Detection and estimation methods for non-stationary signals
Peter O’Shea · The University of Queensland · 1991
The areas of detection and estimation have long been seen as important in the field of Signal Processing. Both areas have seen significant scientific research effort, with the result that there is a large core of well documented knowledge on them, There are many aspects of both areas, however, which are not adequately covered by this classical theory. The treatment of non-stationary signals is a particularly poignant example. Typically in the past, signal theorists have been quick to adopt assumptions of stationarity, or at least of quasi stationarity to characterise their signals. This, however, is overly restrictive in many instances. This thesis seeks to address this shortcoming by considering a number of problems in detection and estimation without making restrictive assumptions about stationarity. A number of different problems are considered - instantaneous frequency estimation, time-varying envelope estimation, detection of narrowband non-stationary signals, detection of non-stationary signals with significant harmonic content, pattern recognition techniques for signals with arbitrary time-varying spectral features, etc. In presenting this material there were two conflicting aims. One of the author's aspirations was to illustrate the many areas in which recent advances in non-stationary signal processing can be applied to enhance existing detection and estimation methods. His other aim was to cover all these enhancements in depth. It became clear that not all topics presented could be dealt with in great depth, and so some measure of priority was given to the treatment of the various areas. Some problems, such as that of estimating the instantaneous frequency of a monocomponent signal, are covered in a lot of detail. Other topics range in depth of treatment, with some receiving moderate coverage, and others less so.