Research On Parallel Filtering Algorithms And Systolic Structures

With the continuous development of modem signal processing technologies, the signal to be processed becomes more diversified and complex. The requirements for the systems speed have got more rigorous than ever. In the fields of radar and communications, the high processing speed of a filter is usually expected, and, in the other fields, such as speech signal processing, the high order filters are in use. All these applications make it necessary to introduce the filters with high processing speed ability, and as for adaptive processing, fast convergence and small stationary error are also required. How to raise the speed of convergence and find out an applicable adaptive algorithm and structure in the design of an adaptive filter is a primary problem faced by the engineers. The dissertation focuses on the researches into the parallel implementation of adaptive algorithms and the structure of high-speed FIR/hR filters and adaptive filters. We have made several contributions in this thesis. They are summarized as follows: > A novel parallel systolic structure is given. This structure has the advantages of both the parallel property and high pipeline property of the systolic structure. Compared with the ordinary systolic implementation of the FIR filter, it has the same systolic frequency as the input-output. While compared with the ordinary parallel structure of the FIR filter, it has the high pipeline property, which makes it maximize the processing speed of a system. Wavelet transform has found wide application in advanced signal processing. But most of research works are focuses on the theory study and few investigations have been made into the realization of high-speed wavelet filter by using VLSI technologies. In this paper, combining with Mallat algorithms, a parallel systolic structure is given based on wavelet analysis and synthesis. By using the method of input data rearrange, a pair of filters can be used as multistage filters, which simplifies the design of a system and reduces the cost of the products. > The parallel expression, which is based on polynomial decomposition and isapplicable to MA model, is generalized to ARMA model. Both the parallel structure and the parallel systolic structure of the hR filter are designed. The parallel systolic structure has the advantages of high-speed and low power consumption over the IIR filter structure available. An analysis of the principle of parallel adaptive LMS algorithm has been made, which shows that the parallel adaptive filter has the same the optimum weights format as the wiener solution to the traditional adaptive filters. Their difference lies only in the selection of the correlation matrix of the input signal and the correlation vector of the input and reference signal. So we make an analysis of the performance of the parallel LMS algorithm following the derivation of the algorithm ground on gradient. Going on the premise of convergence, parallel LMS algorithm has a wider selecting range of step factor than the traditional ones and they have the identical misadjustment in stationary state. There are three ways to implement the parallel filters and these parallel algorithms are better than those ordinary ones in the performance of convergence. Even in the case that the input signal are irrelevant, the third method still excels to the ordinary and the other two have the same speed of convergence as the ordinary method, Moreove...