Mixed-mode correlator for micropower acoustic transient classification

Abstract

A mixed-mode very large scale integration (VLSI) processor for acoustic transient classification performs a running correlation between a time-frequency decomposed analog input signal and a corresponding template. Differential encoding of the inputs allows simplification of the multiply-and-accumulate operations, operating on binary templates and positive-valued inputs, implemented in current mode with eight MOS transistors per cell including SRAM template storage. The use of a bucketbrigade device (BBD) instead of a charge-coupled device (CCD) for shift-and-accumulate operations allows long integration times necessary for acoustic transients spanning several hundred milliseconds. Correlated double sampling at the output compensates for dark current losses in the BBD and other common-mode effects. The 64 2 2 2 16 time-frequency correlator measures 700 m 2 2 2 1170 m in a 1.2-m CMOS process and dissipates an average of 30 W in continuous operation.

Evaluation of Gunshot Detection Algorithms

“…The search window, nonetheless, must be limited in length, as this structure quickly degrades the signal [31]. A better option in terms of signal conservation is the use of a delay chain, as proposed in [32] and shown in Fig.…”

“…Detection and classification methods based on correlation matching are common in plenty of fields, from brain machine interfaces [15] to ultrawideband receivers [36]. Digital simplified detectors based on correlation with very low power dissipation have been successfully built for particular applications [37], and mixed-signal general classifiers have also been proposed [31]. Here, a full-scale method is initially proposed (16-bit resolution and 48-kHz sampling rate) as a top metric for the evaluation of the method's efficiency.…”

“…Subquantization was applied in order to search for simpler digital implementations but just going from 16-to 8-bit degraded performance, with the FPR increasing to almost 0.3 to 0.4 in most cases. A plausible alternative is Edwards and Cauwenberghs' proposal of a mixed-mode implementation of this algorithm [31], with a reported consumption of 50 to 30 W. Here, the computation is not done directly on the signal itself but on the features provided by a wavelet, cochlear or any other kind of preprocessing algorithm. The sampling vector used is a BBD structure of 64 taps, and the correlation is done by an analog current multiplication of the input with a binary pattern.…”

Neuromorphic processor for real-time biosonar object detection

“…An array of four Kerneltron chips are embedded on a printed circuit card interfacing with a host PC and the front-end processor, for a total of 1,024 input units and 128 hidden neurons (support vectors). Since inputs represent time-frequency features, each output implements a cortical cell with particular time-frequency response as observed in the mammal auditory system [9] and implemented in silicon models of acoustic transient classification [10,11]. These cortical units are combined in the output layer for object recognition.…”

Low-Power and Programmable Analog Circuitry for Wireless Sensors

“…Further improvements on this acoustic processor could be gained by using more sophisticated classifiers (e.g. [82][83][84]) that have memory, and also by using signal features other than the spectrum, such as the cepstrum, which has been shown to provide better separation between acoustic classes for both speech and vehicle applications, and has been previously implemented in analog ICs [31]. If the application is changed to something that is not suitable for frequency analysis (e.g.…”