Teleconferencing, system identification and array processing

Detta är en avhandling från Uppsala universitet

Sammanfattning: The area of teleconferencing has long yielded great interest in the signal processing community. The main reasons for this are probably the huge interest from the industry and the challenging problems of the topic. The problems of teleconferencing are relevant for several different disciplines in signal processing. Three of these are Acoustic Echo Cancellation, System Identification and Sensor Array Signal Processing. In this thesis some problems related to these disciplines are studied. The thesis is divided into 6 parts, one for each paper included.In the first part a new adaptive algorithm is applied to the acoustic echo cancellation problem. It is shown to perform much better than the Normalized Least Mean Squares (NLMS) algorithm and while it performs worse than the standard Recursive Least Squares (RLS) algorithm it is shown to be computationally simpler than this.In the second part the hierarchical RLS algorithm is analyzed. The extraordinary results presented for this algorithm in previous papers are discussed and explained.In the third part a new initialization method for RLS is presented that yields the exact Least Squares estimates while not being computationally more demanding than RLS. This is an important contribution since the standard initialization of the RLS algorithm is somewhat arbitrary.In the fourth part a method is presented that deals with the problem of estimating the common factors out of an arbitrary number of polynomials. Two problems of array processing and system identification are stated as problems for common factor estimation and the presented method is applied to these. For these two problems the method is shown to perform better than existing methods.In the fifth part a method for beamforming using few sensors is presented. Data-dependent beamformers usually perform badly when there are few sensors in the array, particularly when the beamformer constraints are numerous. The method presented deals with this problem by approximately fulfilling the beamformer constraints and hence getting extra degrees of freedom for suppressing interferences.In the sixth part the previously unsolved problem of array processing of non-zero mean signals is solved for the colored noise case. Methods are presented both for the estimation problem and the detection problem and are shown to perform well in numerical examples.