Design of multiplierless FIR filters with an adder depth versus filter order trade-off

An Algorithm for Jointly Optimizing Quantization and Multiple Constant Multiplication

“…In particular, for vectors of arbitrary constants, one could achieve a 20%-60% reduction with less than 10% vector approximation error for both frameworks, whereas for vectors of low-pass filter coefficients, a 15%-30% reduction is possible without exceeding 10% error in frequency response. We stress that the results of our joint framework are not directly comparable with those of the algorithms in Yli-Kaakinen and Saramäki [2001], Kang and Park [2001], Yeary et al [2006], Maskell [2007], Xu et al [2007], Yu and Lim [2007], Johansson et al [2007Johansson et al [ , 2009, Aktan et al [2008], and Shi and Yu [2011], as the joint framework assumes general constants, whereas the latter assume FIR filter coefficients.…”

“…We note that while JOINT SOLVE may be used in the case of FIR filters, we are unable to compare against the joint optimization algorithms from Yli-Kaakinen and Saramäki [2001], Kang and Park [2001], Yeary et al [2006], Maskell [2007], Xu et al [2007], Yu and Lim [2007], Johansson et al [2007Johansson et al [ , 2009, Aktan et al [2008], and Shi and Yu [2011] because they formulate the problem differently starting with some given frequency response specifications, rather than with an ideal real coefficient vector to approximate. For this reason, when designing FIR filters, jointly optimal solutions in the sense of Problem 3 might have higher adder tree costs than solutions obtained by algorithms in Yli-Kaakinen and Saramäki [2001], Kang and Park [2001], Yeary et al [2006], Maskell [2007], Xu et al [2007], Yu and Lim [2007], Johansson et al [2007Johansson et al [ , 2009, Aktan et al [2008], and Shi and Yu [2011], which are tailored specifically to FIR filters. With that said, JOINT SOLVE still achieves 15%-30% improvement over a disjointed approach in the case of low-pass filter coefficients, and an even higher improvement of 20%-60% in the general case.…”

Bit-Level Multiplierless FIR Filter Optimization Incorporating Sparse Filter Technique

“…in the design of low power consumption FIR filters. Based on the power models proposed in [42], [43], the MCM algorithms in [44]- [46] explicitly constrain the adder depth when the number of adders or FAs is minimized. Low power consumption is achieved in such algorithms.…”

“…The AAD-NFA product potentially may be used as an objective function in the tree search optimizations. Alternatively, instead of minimizing NFA with implicit AAD constraint as in this paper, minimization of the adder depth of each coefficient as in [12], [44]- [46] with proper constraints on the use of intermediate fundamentals may also achieve the lower power consumption. However, in both of the above suggestions, the search problem including the estimation of node cost and cutoff scheme needs to be redefined, and similar low power designs are expected to be obtained.…”