An Experimental Evaluation of Error Spectrum Shaping Applied to Mixed-Signal Image Convolutions

In this experimental research paper, analog circuits are manufactured and their performance evaluated for suitability in performing FIR image filtering with the convolution kernel and data altered via error spectrum shaping and oversampling so as to preclude corruption by circuit imperfections. Similar to the one-dimensional (1-D) binary signals output by Sigma Delta analog-to-digital converters, the representational noise caused by the circuit's resolution inaccuracy is pushed into an unused portion of the spectrum in these analog signals, permitting the inband portion of the oversampled signal to be more effectively represented and processed by imperfect circuits. An analysis of image convolutions performed using the circuits' data establishes that ESS is successful at reducing the computational error in certain analog image convolutions.

Duke Scholars

Author Martin A. Brooke Electrical and Computer Engineering