Low Complexity Multiplier-less Modified FRM Filter Bank using MPGBP Algorithm

Abstract

The design of a low complexity multiplier-less narrow transition band filter bank for the channelizer of multi-standard software-defined radio (SDR) is investigated in this paper. To accomplish this, the modal filter and complementary filter in the upper and lower branches of the conventional Frequency Response Masking (FRM) architecture are replaced with two power-complementary and linear phase filter banks. Secondly, a new masking strategy is proposed to fully exploit the potential of the numerous spectra replicas produced by the interpolation of the modal filter, which was previously ignored in the existing FRM design. In this scheme, the two masking filters are appropriately modulated and alternately masked over the spectra replicas from 0 to 2$\pi$, to generate even and odd channels. This Alternate Masking Scheme (AMS) increases the potency of the Modified FRM (ModFRM) architecture for the design of a computationally efficient narrow transition band uniform filter bank (termed as ModFRM-FB). Finally, by combining the adjoining ModFRM-FB channels, Non-Uniform ModFRM-FB (NUModFRM-FB) for extracting different communication standards in the SDR channelizer is created. To reduce the total power consumption of the architecture, the coefficients of the proposed system are made multiplier-less using the Matching Pursuits Generalized Bit-Planes (MPGBP) algorithm. In this method, filter coefficients are successively approximated using a dictionary of vectors to give a sum-of-power-of-two (SOPOT) representation. In comparison to all other general optimization techniques, such as genetic algorithms, the suggested design method stands out for its ease of implementation, requiring no sophisticated optimization or exhaustive search schemes. Another notable feature of the suggested approach is that, in comparison to existing methods, the design time for approximation has been greatly reduced. To further bring down the complexity, adders are reused in recurrent SOPOT terms using the Common Sub-expression Elimination (CSE) technique without compromising the filter performance.