by Khorbotly, Sami

Abstract (Summary)
A low cost de-noising system that is appropriate for de-noising real-time control signals is investigated in this work. In the existing research, signal de-noising has been addressed exclusively from a performance point of view. The main focus of researchers has been on improving the noise reduction, regardless of the cost at which this noise reduction is achieved. The cost of de-noising is a two-fold issue, involving both the hardware size and the time delay needed to achieve the de-noising. This work addresses signal de-noising for feedback control system applications, where the information encoded in the noisy signal is time varying and must therefore be fed back to the controller in a timely manner. In these situations, a conventional de-noising system introduces an unacceptable amount of delay. Moreover, the statistical techniques used to set some of the parameters in conventional de-noising systems are too complex for implementation in feedback control loops. This work considers embedded, short delay, real-time signal de-noising systems. A variety of design aspects must be considered in order to develop an effective system. Careful consideration is given to the sampling rate of the signal to be de-noised, as it is shown that appropriately sampled signals are easier to de-noise with lower cost denoising systems. Also discussed in this work is a hybrid thresholding technique that is neither soft nor hard. In this thresholding technique, the threshold value is selected based on real-time calculations without the need for any sophisticated statistical procedure. Another approach that is suggested to improve de-noising results without increasing the costs is application-aware de-noising; this approach considers the way in which a signal is processed and used in deciding how best to de-noise it. The results show that application-aware de-noising results in more noise reduction than can be obtained using a more conventional approach of similar cost. The dissertation focuses on de-noising for specific real-time applications, and a low cost technique for de-noising signals of known shape and dynamic range. In this technique, signal thresholding takes place in the derivative domain of the signal. This derivative domain de-noising technique is combined with another technique to develop a custom made de-noising system that is specifically designed to de-noise motor position control signals. The system is implemented and hardware performance results are reported. The results obtained in this work show that real-time system designers, who usually have limited resources, do not necessarily have to deal with the noise in their systems. While it is true that they are unable to afford the near-optimum de-noising systems presented in the literature, it is always possible to design affordable, low cost denoising systems that achieve significant noise reduction.
Bibliographical Information:


School:The University of Akron

School Location:USA - Ohio

Source Type:Master's Thesis

Keywords:signal de noising electric motor drives wavelet processing digital


Date of Publication:01/01/2007

© 2009 All Rights Reserved.