Design Of Efficient Digital Fir Filter And Filterbank

Digital filter is widely used for signal denoising and extraction in digital signal processing system. The filter complexity is inversely proportional to transition bandwidth. When the transition bandwidth is becoming narrower, the system complexity will increase significantly, resulting in larger power consumption. Frequency response masking (FRM) technique employs the subfilters with sparse filter coefficients to reduce its complexity.Filterbank is one of the most important parts in digital signal processing system. The main objective is to extract sub-bands from the whole signal frequency band. Discrete cosine modulated filterbank, discrete fourier transform modulated filterbank and the filterbank based on FRM are frequently used in current systems, such as software defined radio, communication systems and audio signal processing etc. In this thesis, we mainly focus on the application of filterbank in hearing aids system.For my contribution, much work was done in the low-complexity filter design and reconfigurable filterbank design with low group delay property, specifically:1. Propose a novel two-stage FRM filter structure and design method. FRM technique is a very efficient technique to realize FIR filters with narrow transition band. When the transition bandwidth is extremely small, multi-stage FRM is employed to further reduce the complexity. For the conventional two-stage FRM, there exist two constrains:1) the interpolation factor of the prototype filter must be the integer multiples of the interpolation factors of the masking filters; 2) the interpolation factor of masking filters in the second stage must be the same. These constraints were removed in the proposed structure. Therefore, new principles and equations were deduced to determine the design parameters. The subfilters were then jointly optimized using non-linear optimization. Suppose, for the solution of the proposed FRM, the interpolation factors of the prototype filter and the two masking filters are M, P and Q, respectively. When P=Q and M=kP, the solution is the same as the conventional two-stage FRM algorithm. When one of the above equations does not hold, the proposed algorithm provides a better solution with less number of filter coefficients than the conventional two-stage FRM, which can lead to a reduced hardware cost in applications.2. Propose a reconfigurable filterbank with low group delay used in hearing aids system. The previous work in the design of filterbank for hearing aid is mainly focused on the fixed structure which can only supply single compensation scheme for hearing loss patient. It’s much inconvenient for different hearing impairment cases. A reconfigurable non-uniform filterbank with low group delay is presented in this paper. Firstly, uniform filterbank is formed by cosine modulation. Then, with the transformation from low-pass filter to band-stop filter, non-uniform filterbank is realized based on the uniform filterbank. Finally, by adjusting the control parameter, the reconfigurable filterbank is implemented without changing the structure of filterbank. With the design example, the proposed method has a better performance in terms of matching result and group delay compared with the existing reconfigurable filterbanks.