David C. Hendry

IP core implementation of a self-organizing neural network

This paper reports on the design issues and subsequent performance of a soft intellectual property (IP) core implementation of a self-organizing neural network. The design is a development of a previous 0.65-/spl mu/m single silicon chip providing an array of 256 neurons, where each neuron stores a 16 element reference vector. Migrating the design to a soft IP core presents challenges in achieving the required performance as regards area, power, and clock speed. This same migration, however, offers opportunities for parameterizing the design in a manner which permits a single soft core to meet the requirements of many end users. Thus, the number of neurons within the single instruction multiple data (SIMD) array, the number of elements per reference vector, and the number of bits of each such element are defined by synthesis time parameters. The construction of the SIMD array of neurons is presented including performance results as regards power, area, and classifications per second . For typical parameters (256 neurons with 16 elements per reference vector) the design provides over 2 000 000 classifications per second using a mainstream 0.18-/spl mu/m digital process. A RISC processor, the array controller (AC), provides both the instruction stream and data to the SIMD array of neurons and an interface to a host processor. The design of this processor is discussed with emphasis on the control aspects which permit supply of a continuous instruction stream to the SIMD array and a flexible interface with the host processor.

In Situ Underwater Electronic Holographic Camera for Studies of Plankton

In this paper, we describe an underwater electronic holographic camera (eHoloCam) that has been developed for in situ studies of the distribution and dynamics of plankton and other marine organisms and particles. The eHoloCam uses an Nd-YAG pulsed laser to freeze-frame fast moving particles and a complementary metal-oxide-semiconductor (CMOS) sensor for high-resolution image capture. Digital holograms and holographic videos are recorded at rates from 5 to 25 Hz over a period of several hours. Data is stored locally on an embedded computer. The eHoloCam is capable of recording all organisms and particles located in a water volume of 36.8 cm3 in a single hologram frame. The recorded holographic videos may subsequently be reconstructed numerically at a desired image plane. The main optical and mechanical specifications for eHoloCam are also described. To record electronic holographic videos of marine organisms, the eHoloCam was deployed from a towed sampling frame [autorecording instrumented environmental sampler (ARIES)] on the research vessel RV Scotia at speeds up to 4 kn (about 2 mldrs-1) in the North Sea off the Shetland Isles. Various images of marine organisms obtained from this deployment are shown, together with preliminary measurements on the distribution of Calanus copepods.

Underwater digital holography for studies of marine plankton

Conventional and digital holographies are proving to be increasingly important for studies of marine zooplankton and other underwater biological applications. This paper reports on the use of a subsea digital holographic camera (eHoloCam) for the analysis and identification of marine organisms and other subsea particles. Unlike recording on a photographic film, a digital hologram (e-hologram) is recorded on an electronic sensor and reconstructed numerically in a computer by simulating the propagation of the optical field in space. By comparison with other imaging techniques, an e-hologram has several advantages such as three-dimensional spatial reconstruction, non-intrusive and non-destructive interrogation of the recording sampling volume and the ability to record holographic videos. The basis of much work in optics lies in Maxwells electromagnetic theory and holography is no exception: we report here on two of the numerical reconstruction algorithms we have used to reconstruct holograms obtained using eHoloCam and how their starting point lies in Maxwells equations. Derivation of the angular spectrum algorithm for plane waves is provided as an exact method for the in-line numerical reconstruction of digital holograms. The Fresnel numerical reconstruction algorithm is derived from the angular spectrum method. In-line holograms are numerically processed before and after reconstruction to remove periodic noise from captured images and to increase image contrast. The ability of the Fresnel integration reconstruction algorithm to extend the reconstructed volume beyond the recording sensor dimensions is also shown with a 50% extension of the reconstruction area. Finally, we present some images obtained from recent deployments of eHoloCam in the North Sea and Faeroes Channel.

An overview of the COBRA-ABS high level synthesis system for multi-FPGA systems

This paper presents an overview of the COBRA-ABS behavioural high-level synthesis tool. COBRA-ABS has been designed to synthesise custom architectures for arithmetic intensive algorithms, specified in C, for implementation on multi-FPGA Custom Computing Machine (FCCM) platforms. It performs globally optimising high level synthesis using simulated annealing, integrating all partitioning, scheduling, binding and allocation operations in one optimisation step, and has been designed to be retargetable to different board architectures. COBRA-ABS synthesises a custom Very Long Instruction Word (VLIW) architecture for the given algorithm for implementation on the specified FCCM. The paper gives details of the architectural issues which have influenced the design of the tool, looks at how it fits into the overall design flow and reviews the fundamental concepts and implementation of the globally optimising synthesis methodology. To illustrate the operation of the tool, a number of results for synthesis of a Fast Fourier Transform algorithm are presented.

The COBRA-ABS high-level synthesis system for multi-FPGA custom computing machines

This paper describes the column oriented butted regular architecture-algorithmic behavioral synthesis (COBRA-ABS) high-level synthesis tool which has been designed to synthesize DSP algorithms, specified in C, onto multi-field programmable gate array (FPGA) custom computing machines (FCCMs). COBRA-ABS performs synthesis using a new simulated annealing-based methodology, which maps the specified behavior into a four-dimensional (4-D) space and then optimizes the implied architecture. COBRA-ABS synthesizes custom very long instruction word (VLIW) style architectures partitioned across the FPGAs of the FCCM and has been used to compile C algorithms down to FPGA configuration bit-streams. This paper describes the tool and synthesis concepts and presents simulation results from a number of synthesized fast Fourier transform (FFT) related algorithms.

Comparator trees for winner-take-all circuits

This paper presents architectures for comparator trees capable of finding the minimum value of a large number of inputs. Such circuits are of general applicability although the intended application for which the circuits were designed is the winner-take-all function of a digital implementation of a neural network based on the self organising map. Mechanisms for reducing delay based on look-ahead logic within individual comparators and mechanisms based on multiplexor architectures of a comparator are compared for both propagation delay and area.

HoloCam: a subsea holographic camera for recording marine organisms and particles

The HoloCam system is a major component of a multi-national multi- discipline project known as HoloMar (funded by the European Commission under the MAST III initiative). The project is concerned with the development of pulsed laser holography to analyse and monitor the populations of living organisms and inanimate particles within the worlds oceans. We describe here the development, construction and evaluation of a prototype underwater camera, the purpose of which is to record marine organisms and particles, in-situ. Recording using holography provides several advantages over conventional sampling methods in that it allows non-intrusive, non-destructive, high- resolution imaging of large volumes (up to 105 cm3) in three dimensions. The camera incorporates both in-line and off-axis holographic techniques, which allows particles from a few micrometres to tens of centimetres to be captured. In tandem with development of the HoloCam, a dedicated holographic replay system and an automated data extraction and image processing facility are being developed. These will allow, optimisation of the images recorded by the camera, identification of species and particle concentration plotting.

eHoloCam - an electronic holographic camera for subsea analysis

Recently we have developed an underwater holographic camera for the analysis of plankton and other marine organisms. This camera (HoloMar) was unique in that it was able to record simultaneous in-line and off-axis holograms to cover a range of size of marine organisms from a few microns to tens of millimetres and at concentrations from a few particles per cubic centimetre to dense aggregates. However, HoloMar suffered for being physically large and heavy and difficult to deploy. It also was based on the use of photographic emulsions to record the holograms. To overcome some of these difficulties we have started to develop a new holographic camera (eHoloCam) based on digital holography. In electronic or digital holography (eHolography) an electronic hologram is directly recorded onto a CCD or CMOS sensor and then numerically reconstructed by simulation of the optical hologram reconstruction. In this paper, we discuss some of the possible optical layouts and algorithms under consideration. We present some eHolograms produced in the laboratory prototypes of the eHoloCam

DSP datapath synthesis eliminating global interconnect

CASS (column architecture synthesis system) is a behavioral high level synthesis system for DSP applications. It uses a column based target architecture and in-the-cell routing to produce compact layout which eliminates the need for global wiring. This is achieved using bit-sliced cells which butt together to produce the data path. A description of the architecture and algorithms which produce the datapath is given. It is also shown that this approach gives large area savings when compared to a conventional system.<<ETX>>

Ocean Plankton Imaging Using an Electronic Holographic Camera

In this paper, we describe an electronic holographic camera that has been developed for in situ underwater studies of the distribution and dynamics of plankton and other marine organisms and particles. Holographic data are stored on an embedded computer in the camera for later data extraction and analysis. We describe the main optical and mechanical specifications and outline the design, development and operation of eHoloCam. We summarise the eHoloCam performance in four in situ deployments in North Sea and Faeroe Channels at water depths ranging from about 10 m to 450 m with research vessel Scotia. eHoloCam is capable of capturing opaque and transparent organisms in the size range from about 50 mum up to a 10 millimetres. A clear advantage of eHoloCam over other imaging and counting techniques is the ability to capture high-resolution images without destroying the organism. This feature is extremely valuable in facilitating new in-situ studies of plankton dynamics. The recorded holographic videos are reconstructed numerically using one of our reconstruction algorithms at various planes through the light path. The overall system resolution for the recorded images is 8 mum and 36 mum at a distance of 100 mm and 470 mm, respectively. We show various images of marine organisms recorded on these 4 cruises, and preliminary data on size distributions of Calanus are also presented and discussed.