Adaptive signal processing and its application to infrared detector systems

White, P.R (1992) Adaptive signal processing and its application to infrared detector systems. University of Southampton, Doctoral Thesis.

Abstract

This thesis deals with several general aspects of adaptive filtering as well as the application of adaptive techniques to specific problems associated with infrared detectors. Two infrared detection problems provide the motivation for the majority of the work considered herein. The first problem is that of reducing microphony, i.e. vibration induced signals, in a Pyroelectric detector. The approach proposed in this thesis uses a form of adaptive noise cancellation. The second is that of enhancing/detecting pulse-like periodic signals in broadband noise. Once again an adaptive solution is sought to this problem. With these objectives in mind a general study of adaptive filtering is presented. The classical Least Mean Squares (LMS) algorithm is discussed. Its performance in the microphony cancellation problem is limited by the large eigenvalue spread of the data, which results in slow convergence times. The alternative approaches of Gradient Adaptive Lattices (GALs) and Exact Least Squares (ELS) methods are considered. The ELS algorithms suffer from two problems; a large computational burden and numerical instability. The computational loading can be reduced by the use of `fast' ELS algorithms but this only serves to exacerbate the numerical instability. This thesis makes a minor study of this instability and proposes a new algorithm which solves this problem, only in part. The LMS algorithm in the line enhancement problem suffers from poor performance due to the noisy nature of the updates used. Several methods are proposed for reducing this noise, including the use of structures containing two adaptive filters. Two of these structuresare highlighted; the error filter routine and the balancedfilter. Attempts are made to analyse the performance of both of thesestructures but the majority of the theoretical results pertain to the balanced filter. It is shown that such a filter requires constraints and the consequence of applying these constraints is examined. The adaptive algorithm used to update the balanced filter is analysed and it is shown that convergence to the constrained optimum occurs. Unfortunately, this constrained optimum possesses some undesirable properties and as a result the balanced filter is of limited use. Finally, the application of these techniques to data from actual infrared systems is performed. The results from the off-linemicrophony simulations are encouraging. Although, ideally, one would like to apply an alternative algorithm, the pragmatic issues of available hardware and expense limit on-line applications to using the LMS algorithm. The off-line simulations for the line enhancement problem indicate that the balanced filter has good detection capabilities but the problems already mentioned make it unattractive. Once again the LMS algorithm is implemented in real time to perform this task.

This record has no associated files available for download.

Contributors

Download statistics