A soft computing based approach for signal processing

Abstract

In the present mechanized world, digitalization has encroached almost every sphere of technology like biomedical information processing, defense applications, astrophysical data analysis, just to name a few. Biomedical information comprises of one dimensional or two dimensional signals which are received as output of different biomedical instruments. Defence technologies involve use of radar signals received as outputs of sensor devices. Watermarking is another application of digitalization in defense technologies that provides security to the secret information. Visible watermarking is also very much useful in the department of law to prove intellectual property rights. Astrophysical instruments acquire information generally in time series that is basically one dimensional series of data points indexed in order of time. The current globalized era is marked by a rapid increase in the use of wireless media to exchange information over globally distributed locations. This advancement and growth of technologically mediated information help provide medical care remotely by exchanging biomedical signals amongst various hospitals and diagnostic centers across the globe. It also helps in the area of defense by sharing the radar signals instantly via phone or internet. However, while transmitting, these signals may get affected by some unwanted components termed as noise which are adverse but inevitable. Removal of such unwanted components from the signals has remained challenging for the researchers since the earliest days of signal processing. For removing noise from signals, the use of digital filters has been proved to be more effective than the analog filters due to the flexibility of hardware efficiency provided by the digital filters. Among the two types of digital filters, Finite Impulse Response (FIR) filters are used more extensively compared to the Infinite Impulse Response (IIR) filters because of FIR filters' outstanding characteristic of stability and the capability of obtaining linear phase response. FIR filters take a discrete time signal as input and performs addition and multiplication operations to obtain the desired filtered discrete time output signal. Among various techniques proposed by several researchers to design FIR filters, the use of window functions is the most popular approach. For implementation of FIR filters, several soft computing approaches have also been proved to be effective. Limitations of conventional approaches such as window methods, frequency sampling methods used for filter design motivated us to use soft computing techniques to design FIR filters. In this thesis, an innovative combined approach has been proposed to de-noise biomedical signals using the most effective window function Kaiser Window and Genetic Algorithm, well known evolutionary approach. Kaiser Window with varying passband and stopband ripples is initially used for filtration of noisy heart sound signals. Genetic Algorithm is then used to obtain the least noisy signal. Optimization of parameters used to design a digital filter using Kaiser Window function is also performed in this thesis using an adaptive Ant Weight Lifting Algorithm. Another approach for FIR filter design is to use optimized set of coefficients. In this thesis, we also make a comparative study of a few traditional algorithms in optimizing filter coefficients. A novel algorithm, namely Global Best Steered Cuckoo Search Algorithm, is also proposed for the same purpose. This new algorithm is proved to be much more efficient in filter design compared to the traditional algorithms in terms of passband ripple and stopband attenuation of the filters. Nowadays, many battery operated devices such as mobile phones, hearing aids, FM radios, etc. also use FIR filters. Due to the power starving nature of these devices, implementation of FIR filters with as low power as possible is of utmost necessity. Therefore, aiming to address the need for perpetually demanding high speed and low power devices, innovative techniques for implementing hardware efficient FIR filter are also proposed in this thesis. These techniques involve the use of fixed length coefficients and two innovative algorithms are proposed to obtain optimized coefficients. A new quantum algorithm, namely Global Best Steered Quantum Inspired Cuckoo Search Algorithm, is proposed in this thesis for obtaining optimized filter coefficients capable of acquiring desired filter responses. This algorithm is also proved to be effective in reducing the number of SPT terms. With the help of common sub-expression elimination technique, required number of adders is further minimized for hardware efficient filter implementation. We also proposed another new algorithm, namely Fast Converging Flower Pollination Algorithm, for the same purpose. These new algorithms have been proved to be much more effective compared to their traditional versions which have also been proved to be efficient in the domain of filter coefficient optimization compared to the conventional approaches like window methods and frequency sampling methods.