Reconfigurable biological signal co-processor for feature extraction dedicated to implantable biomedical microsystems

This paper reports the design of a fully reconfigurable biological signal co-processor, designed to enhance processing capabilities of a generic controller proposed by the authors. The co-processor is capable of performing feature extraction on 32 channels in both time and mathematical transform domains. In the time domain, level crossings and local maxima and minima are the features the co-processor is capable of recognizing. In the transform domain, the discrete wavelet transform (DWT) is calculated using a reconfigurable DWT core. Designed and simulated in a 180-nm standard CMOS process, the co-processor dissipates a maximum power of 427 μW when calculating the DWT coefficients with 2 levels for the Daubechies3 basis function.