A novel forward-backward predictor based low-power DSP system

In this paper, we present an algorithmic noise tolerance (ANT) technique for low-power digital signal processing systems. The proposed technique employs a low-complexity forward-backward predictor to correct errors in a main DSP (MDSP) block due to voltage overscaling, which is an ultra low-voltage operating condition. For a frequency selective FIR filtering, it is shown that the proposed technique achieves up to 43% power savings over an optimally voltage scaled MDSP with a 5% area overhead.