A New Structure for High Speed Sharp FIR Filters 
Using the Frequency Response Masking Approach

Digital signal processing techniques are being applied to digital TVs, cellular telephony, and multimedia systems. This has resulted in a demand for high-sample-rate narrow transition width linear phase FIR digital filters. The advantages of using FIR filters are their guaranteed stability and freedom from phase distortion. Their major drawback is that filters with sharp transitions have very high implementation complexity. One of the most efficient ways to design narrow transition-width arbitrary bandwidth FIR filters is to use frequency-response masking techniques. The basic idea behind the frequency-response masking technique is to compose the overall filter using several sub-filters. The first sub-filter is called the edge-shaping filter, which is up-sampled to form the sharp edge needed in a narrow transition-width filter. The second sub-filter is the complement of the edge-shaping filter, and helps to form the arbitrary bandwidth of the overall filter response. The other two sub-filters are masking filters that remove unwanted frequency components to form the stopband of the overall filter response. 

To implement a high speed digital FIR filter in an ASIC, the filter length needs to be kept as short as possible to reduce the quantization noise and hardware cost. It is well known that the filter length of each sub-filter becomes long for very sharp filters using a single-stage frequency-response masking approach. This is due to the fact that the transition-width of the edge-shaping filter is determined by the product of the transition-width of the overall filter and the up-sampling rate. If we want to reduce the complexity of the overall filter, one of the ways is to increase the sampling rate of the edge-shaping filter while the band edges of the two masking filters remain unchanged. To this end, a new prefilter-equalizer filter is proposed to replace the edge-shaping filter in the frequency response masking approach. 

We considered two types of prefilter-equalizer filters. The first type utilizes the Interpolated Finite-Impulse Response technique to realize the edge-shaping filter. This type of filter works very well for a narrow band filter. The second type uses three sub-filters. The first sub-filter is a multiplier-free first-order even length filter. When connected in parallel with an odd length filter, a pre-filter pair is formed. This pair is cascaded with an equalizer to achieve a similar frequency response as an edge-shaping filter. The sampling rate of the pre-filter pair is double that of the rate used in the frequency response masking approach while the band edges of the two masking filters remain unchanged. An example of this type of pre-filter is shown in Figure 1. This type of filter can be used to obtain an arbitrary bandwidth edge-shape filter, and leads to a reduction of the number of multipliers compared with the first type. With this new approach about 20% reduction in the number of multipliers is achieved compared to the conventional frequency-response masking approach. Figure 2 shows the frequency response of a filter designed using the proposed method.

 
Figure 1: An example of the proposed prefilter-equalizer filter.   


Figure 2: The overall frequency-response of the filter with passband and stopband edges at 0.3 and 0.301, respectively.

The proposed FIR filter structure was designed and implemented using the 0.35mm CMOS technology. The designed filter is equivalent to a 1550 tap conventional FIR filter. The coefficients in all sub-filters were converted to power-of-two terms to eliminate the multipliers. The overall filter was able to operate at a clock rate of 100MHz. 

This work was performed by Assoc Prof Y Lian , Assoc Prof C C Ko, and graduate student L Zhang.