ACPA: Exploiting Approximate Computing for High-Level Imprecision Optimization of Fixed-point LTI systems

The high-level imprecision optimization (HLIO) is a design paradigm in a wide variety of error-resilient digital signal processing (DSP) systems to decrease the hardware cost within an acceptable deviation from the nominal output values. In these systems, Approximate computing (AC) has emerged as the new design paradigm, which has the common goal of designing efficient systems by trading accuracy for hardware cost with HLIO. In this paper, we make use of AC as a complementary paradigm in HLIOs to improve the utilization of the error budget compared to existing HLIOs with the aim of further reduction in the area and delay. So, we improve HLIO using AC on fixed-point implementation of a feed-forward linear-time-invariant (LTI) system, when the error measure budget is specified as maximum mismatch (MM), or maximum absolute error. The results show that our method can improve the area by 21% and delay by 9.3% on average compared to the state-of-the-art techniques.