Design Of Adaptive Filter Based On Approximate Computing

Due to its mature theoretical foundation and simple hardware implementation methods,adaptive filtering has been applied in many fields of engineering practice,such as adaptive noise cancellation,adaptive spectral line enhancement,and system identification.The realization of the adaptive filtering system relies on arithmetic units such as multipliers and adders,and the operation of the system is accompanied by a large amount of data calculations.As the scale of adaptive filter hardware becomes larger and larger,the power consumption problem it faces becomes more and more serious.Approximate calculation circuits are gradually introduced into such arithmetic units to solve the problems of power consumption density and computing power limitations in the circuit.The basic idea is to simplify the circuit hardware complexity by generating inaccurate but acceptable calculation results.So as to achieve the purpose of improving the electrical performance of the circuit.In order to explore the feasibility and specific optimization effects of introducing approximate calculations in applications such as adaptive filters.This article firstly expounds the principle and analyzes the characteristics of the important arithmetic units that constitute the system,and then approximates the design of the above-mentioned units.Then build accurate and multiple approximate adaptive filters.In the compressor unit,8-bit,16-bit multiplier unit and adaptive filter level,a large number of comparative experiments and analyses have been carried out.Finally,under the application of adaptive noise cancellation,the feasibility and necessity of approximate calculation in the application of adaptive filtering are demonstrated.Based on the probabilistic analysis of partial product arrays,this thesis proposes a new type of approximate 4-2 compressor and an adaptive error compensation unit.Thanks to the compensation feature of addition,the approximate 4-2 compressor innovatively retains only one output port.Compared with other approximate designs,the multiplier built with the compressor and the error compensation unit achieves a significant improvement in electrical performance while ensuring better accuracy.Using this 8-bit unsigned multiplier for image synthesis applications can achieve a PSNR performance of more than 60 d B.At the same time,this article proposes a theoretical analysis method for the compressor's error.When using it to build a multiplier,it can quickly estimate the error performance of a single approximate compressor link without verilog encoding and simulation.In the application of adaptive noise cancellation,the approximate adaptive filter with multiple different parameters composed of the above-mentioned units has an43.99% reduction in area than the full-precision design under the condition of insignificant accuracy degradation,the power consumption is reduced by 58.55%,and the delay is reduced by 18.40%.The insignificant decrease in accuracy indicates that it is feasible to introduce approximate calculations in the adaptive filter design.The significant electrical performance improvement illustrates the necessity of introducing approximate design in such arithmetic units at the time when low-power design is becoming more and more important.