Revisiting FPGA Implementation of Digital Filters and Exploring Approximate Computing on Biomedical Signals

Abstract

In recent years, the approximate computing becomes popular in the era of VLSI (very large scale integration) domain to arrive better power, area, and delay outcomes at the cost of lower precision loss. Also, the human beings are not so intelligent to see/observe/listen the processed digital data; means even if some of the data loss occurs human beings are unable to notice them. This behavior set the engineers to research on approximate computing which are very useful in the multimedia data processing, data communications, high-volume data storage, etc. In this study, the experiments such as hum-noise removal, filters on QRS detection are implemented on an Altera FPGA EP4CEF29C7 device using Quartus II 13.1 synthesis software tool and the simulation results on device utilization reports, the speed and the power are obtained. Simulation results reveal that the approximate computational filters offer better power, area, and speed results than the conventional ones. Also, Matlab 9.4 (R2018a) simulation was used to carry out the functional verification of the actual and approximate filters.

On Numerical Robustness of Bi-quad Structures using Fixed-Point Approximate Multiplication

“…Since in all three bi-quad structures the BBM noise is introduced into the critical feedback path, we limit our experiments to focus on varying the pole locations, i.e., for all experiments we fix the b i coefficients and alter the a j coefficients only, Equ. (9). We opt for a double zero in z = −1, i.e., b 0 = b 2 = 0.5 and b 1 = 1, (1 − ∆ in 2's complement).…”

“…AC circuits have been applied to signal processing functions such as Finite Impulse Response (FIR) filters, Fast Fourier Transform (FFT), and Discrete Cosine Transform (DCT), e.g., [7], [8], [9], demonstrating the ability to obtain A-, T -, and/or P W -reduction at the expense of a decreased accuracy. These functions are all characterized by a feed-forward data flow.…”

A First Experimental Study of Fixed-Point Approximate Arithmetic in Recursive Lattice Filters

“…In recent years, many AxC circuits have been proposed, mainly for addition and multiplication, but circuits for other arithmetic operations also exist, see the survey in [4]. Several studies have demonstrated the applicability of AxC circuits in signal processing algorithms having feed-forward branches only, e.g., Finite Impulse Response filters, the Fast Fourier Transform, and the Discrete Cosine Transform, [5], [6], [7]. For such linear time-invariant algorithms, the inaccuracy, i.e., the numerical noise induced by the AxC circuits, is transferred via certain gain factors directly to the system output.…”

“…2 that the PPs, according to Equ. (7), are subsequently left-shifted two bit positions according to their individual numerical weighting. Furthermore, the PP word…”