J.T. Taylor

A scalable high-speed current-mode winner-take-all network for VLSI neural applications

This paper describes a very flexible, high-speed current-mode winner-take-all network (WTA) for use in artificial neural systems. Since the WTA is based on a network of identical current switching cells requiring only adjacent transistor matching, it is particularly suited to use in large systems. In addition, since the network is based on a tree structure, very little space is occupied by interconnect, thus ensuring a very compact layout. This WTA has current (matching score) inputs and provides a buffered, binary encoded voltage output. >

A 100-Mb/s 2.8-V CMOS current-mode analog Viterbi decoder

We describe a K = 3, rate-1/2 analogue Viterbi decoder fabricated in 0.8µm CMOS technology, intended for convolutional decoding applications. The decoder is the first of its kind to employ current-mode analogue circuit techniques. It operates at data rates of at least 100Mb/s and consumes 40mW at that rate from a single 2.8V power supply. The chip contains about 6K transistors of which less than 1K are used in the analogue sections of the system and has a core area of approximately 1mm2.

High speed vision-based quality grading of oranges

We describe a novel system for grading oranges into three quality bands, according to their surface characteristics. This processing operation is currently the only non-automated step in citrus packing houses. The system must handle fruit with a wide range of size (55-100 mm), shape (spherical to highly eccentric), surface coloration and defect markings. Furthermore, the point of stem attachment (the calyx) must be recognised in order to distinguish it from defects. A neural network classifier on rotation invariant transformations (Zernike moments) is used to recognise radial colour variation, that is shown to be a reliable signature of the stem region. This application requires both high throughput (5-10 oranges per second) and complex pattern recognition. Three separate algorithmic components are used to achieve this, together with state-of-the-art processing hardware and novel mechanical design. The grading is achieved by simultaneously imaging the fruit from six orthogonal directions as they are propelled through an inspection chamber. In the first stage processing colour histograms from each view of an orange are analysed using a neural network based classifier. Views that may contain defects are further analysed in the second stage using five independent masks and a neural network classifier. The computationally expensive stem detection process is then applied to a small fraction of the collected images. The succession of oranges constitute a pipeline, and, time saved in the processing of defect free oranges is used to provide additional time for other oranges. Initial results are presented from a performance analysis of this system.

An implant for chronic selective stimulation of nerves

An implantable stimulator system has been developed for nerve stimulation. The system is capable of stimulating selectively, either by fibre position, fibre size or by sending action potentials in one direction only, based on the use of nerve cuffs. The stimulator produces either quasi-trapezoidal current pulses, to allow anodal blocking, or conventional rectangular-shaped current pulses, of amplitude 20 microA to 5 mA (in 20 microA steps) with duration of 16 micros to 1 ms (in 8 micros steps). For safety, both active and passive charge balancing is used. The amplitude of the active charge-balancing phase can be varied between 1/7 and 1/47 of the pulse amplitude. During manufacture, each implant is customised so as to drive either 6 quasi-tripolar (dipolar), 4 tripolar or 2 pentapolar cuffs. Possible applications of the device are: improved defaecation and bladder voiding after spinal cord injury, by stimulation of the sacral motor roots; neuromodulation to reduce hyperreflexia without concomitant muscle contractions; in stroke patients, to enable balanced inversion-eversion while dorsiflexing the ankle by stimulating the peroneal nerve. It may also be used in chronic animal experiments.This paper describes the implant system, its hardware and communication protocol, and shows results from in vitro tests of the device and the first acute anodal-blocking experiments in pigs.

Vision and neural control for an orange harvesting robot

We describe the system control architecture of a large orange harvesting robot. This robot has two independent electrically driven telescopic arms mounted on a common platform which is itself held by a large hydraulic arm. This arm, in turn, is mounted on a tracked vehicle. The telescopic arms have cameras within the end-effecters, which are used to detect and measure the position and distance of the fruit within the canopy of a tree. Most of the development and control software was implemented using the matrix-based Virtual Machine Language (VML). This language was designed to implement neural networks, and has been extended and enhanced for robotic applications and the particular low level control requirements of the hardware. The device drivers provide the interface to frame grabbers, motor drivers, digital interface electronics, proximity detectors, and file handling. The same interface is used to implement interprocess communications with display and monitoring tools.

High-speed electro-optic analogue to digital converters

Interleaved architectures can substantially increase the speed of analogue-to-digital converters (ADCs). Problems of timing jitter have limited the maximum speed of all-electronic ADCs to about 1 Gs/s. An optoelectronic solution is proposed which will remove this restriction and fully exploit the potential of the interleaved architecture. Fully integrated optically triggered sample-and-hold circuits are fabricated for the first time, using GaAs MESFET technology. Preliminary test results are presented.<<ETX>>

A new architecture for low power analogue convolutional decoders

A new architecture for convolutional decoding, the modified feedback decoding algorithm (MFDA), is presented. For the specific codes considered, its error-correction performance compares very favourably to the truncated Viterbi algorithm (VA) but it requires less hardware. The realisation of its primary subsystem, an add-compare-select (ACS) circuit, is presented which can also be employed in analogue/digital realisations of the VA. Preliminary results indicate the switched-current ACS circuit described is faster, more compact, and uses less power than a digital ACS unit designed for the same level of resolution.

Design issues for a switched-capacitor filter using GaAs technology

Various aspects of the design of a high-speed GaAs switched-capacitor filter are discussed. The circuit is based on a formerly unexploited finite-region insensitive integrator-pair, taking advantage of a recently recognized property of its inverting integrator. Device modeling and the influence of transistor characteristics on subcircuit design are discussed. Details of the filter-optimization procedures are given. The circuit has been fabricated and is now being mounted for testing.<<ETX>>

Low-power CMOS and BiCMOS circuits for analog convolutional decoders

Analog techniques have been proposed as a method of realizing high-speed low-power Viterbi decoders, mostly for use in digital magnetic recording applications. However, such methods can also be used in more general applications of the Viterbi algorithm, such as convolutional decoding. We describe two current-mode analog circuits for use in such applications. These circuits operate in excess of 100 MHz, and consume less than 4 mW per state, with 2.8-V supply rails when realized in 0.8-/spl mu/m technology.

Video Grading of Oranges in Real-Time

We describe a novel system for grading oranges into three quality bands, according to their surface characteristics. The system is designed to process fruit with a wide range of size (55–100 mm), shape (spherical to highly eccentric), surface coloration and defect markings. This application requires both high throughput (5–10 oranges per second) and complex pattern recognition. The grading is achieved by simultaneously imaging each item of fruit from six orthogonal directions as it is propelled through an inspection chamber. In order to achieve the required throughput, the system contains state-of-the-art processing hardware, a novel mechanical design, and three separate algorithmic components. One of the key improvements in this system is a method for recognising the point of stem attachment (the calyx) so that it can be distinguished from defects. A neural network classifier on rotation invariant transformations (Zernike moments) is used to recognise the radial colour variation that is shown to be a reliable signature of the stem region. The succession of oranges processed by the machine constitute a pipeline, so time saved in the processing of defect free oranges is used to provide additional time for other oranges. Initial results are presented from a performance analysis of this system.