Analog Signal Processing Elements for Energy-Constrained Platforms

Energy constrained processing poses a number of challenges that have resulted in tremendous innovations over the past decade. Shrinking supply voltages and limited clock speeds have placed an emphasis on processing efficiency over the raw throughput of a processor. One of the approaches to increase processing efficiency is to use parallel processing with slower, lower resolution processing elements. By utilizing this parallel approach, power consumption can be decreased while maintaining data throughput relative to other more power-hungry architectures.;This low resolution / parallel architecture has direct application in the analog as well as the digital domain. Indeed, research shows that as the resolution of a signal processor falls below a system-dependent threshold, it is almost always more efficient to preform the processing in the analog domain. These continuous-time circuits have long been used in the most energy-constrained applications, ranging from pacemakers and cochlear implants to wireless sensor motes designed to run autonomously for months in the field.;Most audio processing techniques utilize spectral decomposition as the first step of their algorithms, whether by a FFT/DFT in the digital domain or a bank of bandpass filters in the analog domain. The work presented here is designed to function within the parallel, array-based environment of a bank of bandpass filters. Work to improve the simulation of programmable analog storage elements (Floating-Gate transistors) in typical SPICE-based simulators is presented, along with a novel method of harnessing the unique properties of these Floating-Gate (FG) transistors to extend the linear range of a differential pair. These improvements in simulation and linearity are demonstrated in a Variable-Gain Amplfier (VGA) to compress large differential inputs into small single-ended outputs suitable for processing by other analog elements. Finally, a novel circuit composed of only six transistors is proposed to compute the continuous-time derivative of a signal within the sub-banded architecture of the bandpass filter bank