ACCURACY VS. PRECISION IN DIGITAL VLSI ARCHITECTURES FOR SIGNAL-PROCESSING

The paper provides a sensitivity analysis to measure the loss in accur acy induced by perturbations affecting acyclic computational flows com posed of linear convolutions and nonlinear functions. We do not assume a large number of coefficients or input independence for the convolut ion module, nor strict requirements on the nonlinear function. The ana lysis is tailored to digital VLSI implementations where perturbations, associated with data quantization. affect the device inputs, coeffici ents, internal values, and outputs. The sensitivity analysis can be us ed to measure the loss in accuracy along the computational chain, to c haracterize the tolerated perturbations, and to dimension the whole ar chitecture.