Zigzag low-complexity approximate DCT using frequency upscaling technique

Abstract

Approximate computing is widely used in image processing field for reducing the energy dissipation, increasing the throughput, and decreasing the amount of computation by sacrificing some accuracy in the output results. Thus, this paper presents a runtime-based approximate computing approach by using frequency upscaling (FUS) technique into zigzag low-complexity approximate DCT (ZLCADCT). The presented technique allows to change the degree of approximation without changing the circuit of ZLCADCT while providing increased throughput by applying the input vectors of circuits at a rate beyond the maximum operating frequency. The results show that when FUS are implemented into ZLCADCT, approximate full adder can sustain significant higher input frequency (around 19.2GHZ using 32nm adders) when compared with an exact full adder (at around 15.4GHZ) without having significant decreases in PSNR value. The number of completed DCT operations for ZLCADCT (2.95 to 3.53 for approximate full adder at 16.6GHZ) is higher than 2.3 to 2.77 for ADCT by using FUS technique.