Externally Linear Low-Power Analog Signal Processing Circuits

ABSTRACT

0209109

Columbia U

Yannis Tsividis

This research explores the possibilities of performing linear analog signal processing using systems which are internally nonlinear. The context is fully integrated systems, fabricated on a single silicon chip, containing a digital signal processor or a computer plus an analog signal processor; the latter is often a necessary part of the interface to the physical world. Conventional linear techniques for implementing analog processors often result in very large energy consumption, which rules them out in many practical cases. We are investigating the use of internally nonlinear processors to bypass those limitations. The applications of this work extend from wireless communications and computer hardware to consumer products to biomedical intrumentation.

The use of input-output linear signal processors which are permitted to be internally nonlinear allows the internal signals to have, for a large input signal range, a magnitude well above that of noise and interference, while at the same time remaining safely below overload levels. The result is an adequate signal-to-noise ratio over the entire input range, without requiring large power dissipation and chip area. A number of other related ideas are being investigated as well. One is the use of dynamic biasing, in which the bias levels of dynamical analog circuits, such as filters, are dynamically varied depending on the signal level; the other is the use of multiple signal processors, each being placed dynamically in the path of the signal as required. Since output disturbances would normally occur when the internal parameters of a dynamical system are varied, a key consideration in this research is how to eliminate such disturbances or prevent them from occurring.