Low-power digital filtering using approximate processing with variable canonic signed digit coefficients

An approximate processing reduces the computational complexity and the power consumption of a digital filter by adjusting the order to input signal statistics. We present an alternate approximate processing method which dynamically adjusts filter coefficients represented by canonic signed digit (CSD) coefficients. Proposed method adjusts the precision of the filter coefficients by controllingthe length of the CSD coefficients. By adjusting the length of the CSD coefficients in each approximation level, the proposed method can be used to reduce the computational complexity. This paper reports a procedure to select the proper length of CSD coefficients to the input signal statistics. Our experimental results with a digital filter for Audio Codec '97 demonstrated that the numbers of addition operations can be reduced down to 29.4%, 11.8%, and 3.9% for each approximation level in computing the filter coefficients.