Show all abstracts
View Session
- Pattern Recognition I
- Poster Presentations
- Pattern Recognition II
- Pattern Recognition III
- Pattern Recognition IV
- Information Processing I
- Information Processing II
- Poster Presentations
- Optoelectronic Neural Networks Session 8
- Optoelectronic Neural Networks Session 9
- Optoelectronic Neural Networks Session 11
- Optoelectronic Neural Networks Session 10
- Optoelectronic Neural Networks Session 11
- Optoelectronic Neural Networks Session 9
- Optoelectronic Neural Networks Session 11
- Memory Architectures
- Poster Presentations
- Optical Memory Materials
- Memory Components
- Information Processing I
- Information Processing II
- Poster Presentations
- Information Processing II
- Pattern Recognition I
- Pattern Recognition IV
- Poster Presentations
- Information Processing II
Pattern Recognition I
Correlation filter fusion for detection: morphological, wavelet, and Gabor methods
David P. Casasent,
John Scott Smokelin,
Anqi Ye,
et al.
Show abstract
We consider the detection of candidate objects (regions of interest) in a scene containing high clutter, multiple objects in different classes, independent of aspect view, with hot/cold/bimodal/partial object variations, and with low contrast targets. We use three different filters with each designed to produce high probability of detection (PD). We fuse the results from different outputs to reduce false alarms (PFA). All filters are realizable on a correlator.
Reduced classification error probability with ternary correlation filters
Show abstract
Much of the filter design work that has been performed to date for filter SLMs with both constrained and unconstrained modulation characteristics has been concerned with optimizing the design for certain performance criteria associated only with the correlation function of the target image. However, in most likely application scenarios there will be multiple objects that may populate the field of view, and the most important correlation performance criterion is ultimately the probability of correct classification of a given object as either belonging to the in-class set or the out-of-class set. In this work, we study the problem of designing ternary phase and amplitude filters (TPAFs) that reduce the probability of image misclassification. We use the Fisher ratio as a measure of the correct classification rate, and we attempt to maximize this quantity in our filter designs. Given the nonanalytical nature of the design problem, we employ a simulated annealing optimization technique. We present computer simulation results for several cases including single in-class and out-of-class image sets and multiple image sets corresponding to the design of synthetic discriminant function filters. We find significant reductions in expected rates of classification error in comparison to BPOFs and other TPAF designs.
Optimum receiver design for pattern recognition with spatially disjoint target and scene noise
Show abstract
The design of an optimum receiver for pattern recognition problems with input scene noise that is spatially disjoint (non-overlapping) with the target is described. The processor is designed based on multiple alternative hypothesis testing.
Poster Presentations
Spatially multiplexed composite filters for optical pattern recognition
Show abstract
A new class of composite filters is proposed. This class of filters can be used in optical correlator based recognition systems. It introduces a novel approach for rotation and scale invariant systems. Also it can be used to improve power throughput without sacrificing the signal-to-ratio in optical correlators based on complex matched filters.
Pattern Recognition II
Multichannel disk-based optical correlator
Show abstract
Angularly multiplexed transmission holograms were recorded at multiple spots on a photopolymer coated disk. By imaging in the `along track' direction and Fourier transforming in the radial direction when recording and presenting images, disk rotation can be used to generate the 2D correlation functions between stored templates and presented images. The correlation functions for holograms multiplexed at a given spot are generated line by line and detected in parallel. A theoretical model of operation was developed and experimentally verified.
Evaluation of optimal synthetic discriminant filters on an application to angle determination
Show abstract
The performances of the OTSDF filters are illustrated on a particular application which consists in determining the angle and position of an airplane. For this purpose, different numerical experiments are presented. In particular, the influence of the noise characteristics is analyzed. The stability of the filters is investigated and simple regularization techniques are proposed. The robustness of correlation to different kinds of hidden parts is also studied. A real time numerical experimental system has been realized using different airplane scale models. We shall show the performances of the system with different kinds of perturbations. The correlation system has a hierarchical structure. At the first level, a rough estimation is performed, at the second level a fine angular determination is achieved and it is only at the third level that the type of the airplane is determined. This system allows to obtain 0.1 degree of mean precision in the absence of noise. This mean precision is still better than 1 degree with an input signal to noise ratio of about -22 dB. These results are obtained for 128 X 128 pixels input images.
Tracking and pattern recognition performance of a breadboard system using a nonlinear joint-transform correlator
R. A. Dobson,
M. Gregg Roe,
Kevin L. Schehrer
Show abstract
The experimental performance of a breadboard system that included a nonlinear joint- transform correlator is described. This system was capable of rapid tracking and time- multiplexed optical pattern recognition.
Optical continuous phase-only correlator using liquid crystal television
Show abstract
We use a commercially available liquid crystal television (LCTV) for encoding real-time on- axis continuous phase-only filter (POF) in a Vander Lugt type optical correlator. We show by theory and experiments that filter coded on the LCTV with phase mismatching and coupled amplitude modulation maintains a term that is the true POF with a diffraction efficiency to about 70%, plus a zero order spot. The correlation output is a superposition of the POF correlation peak with an image of the input. The filter, referred as to the phase dominant filter, behaves like the POF with advantages of on-axis correlation: high light efficiency and utilization of all the available space bandwidth product of the LCTV. Optical experimental results are shown.
Pattern Recognition III
Nonlinear decision boundaries from the use of complex constraints in synthetic discriminant function filters
Show abstract
Conventional Synthetic Discriminant Function (SDF) filter formulations employ the same real correlation output (at origin) for all input training images from the same class, but this leads to poor discrimination. However, this can be overcome by using complex correlation constraints in the SDF filter design. In this paper, we characterize the recognition sets for such SDF filters.
Illumination-invariant pattern recognition with a binary nonlinear joint transform correlator using spatial frequency dependent threshold function
Show abstract
The correlation performance of binary joint transform correlators with unknown input light illumination is investigated for different thresholding methods. It is shown that a binary joint transform correlator that uses a spatial frequency dependent threshold function for binarization of the joint power spectrum is illumination invariant. The performance of the binary joint transform correlator with unknown input light illumination using a variety of thresholding methods is investigated.
Examination of metrics and assumptions used in correlation filter design
Show abstract
This paper examines some of the metrics that are commonly used to design correlation filter's for optical pattern recognition, including: the Fisher ratio, the signal-to-noise ratio, the equal correlation peak constraint, and normalized correlation. Attention is given to the underlying assumptions that are required to move the Bayesian decision theory to a particular metric or design principle. Since a Bayes classifier is statistically optimum, this provides a means for assessing the merit of a particular approach. Although we only examine a few metrics in this paper, the approach is general and should be useful for assessing the merit and applicability of any of the numerous filter designs that have been proposed in the optical pattern recognition community.
Hard-clipped two-beam coupling correlators
Show abstract
An all optical nonlinear joint transform correlator is proposed and demonstrated. A hard- limiting quadratic nonlinearity is used in the Fourier plane to allow the first implementation of a phase only filter using photorefractive materials.
In-line acousto-optic interferometer as a correlator and spectrum analyzer
Nabeel A. Riza
Show abstract
A novel, efficient, stable, in-line, two Bragg cell, acousto-optic architecture is introduced for signal correlation and spectrum analysis, and experimental results are described. The processor employs an image inversion technique to produce a correlation output that does not have to be generated on a spatial carrier. The system is particularly suited for wide instantaneous bandwidth signal processing.
Pattern Recognition IV
Optical fractal image processor for noise-embedded targets detection
Show abstract
A new fractal dimensionality measurement and analysis-based automatic target recognition (ATR) is presented. ATR is a multi-step procedure, such as preprocessing, interest locating, segmenting, feature extracting, and classifying. Physical Optics Corporation has developed a new ATR system, based on fractal image processing, which can perform interest locating, segmenting and feature extraction simultaneously. Experimental results of the proof-of- principle is also presented.
Performance of the optical wavelet matched filter
Show abstract
Optical wavelet matched filter performs the wavelet transforms for edge enhancement and correlations between the wavelet transform coefficients in a single step for shift invariant pattern recognition. The performance of the wavelet matched filter depends on the scale factor s. A systematic study and comparison on the light efficiency, discrimination capability and robustness to noise is made for the wavelet matched filter, the matched spatial filter and the phase-only matched filter.
Applying time modulation to optical joint transform correlators to improve target detection in high-clutter/multiple-target environments
Show abstract
A major limitation on the optical joint transform correlation (JTC) is that the output plane is dominated by unwanted self correlation products. Temporal encoding can be applied to the JTC to separate the correlation components in the output plane. This results in SNR/PSR improvement, removal of input plane location constraints, and elimination of the detection problems which result from multiple targets. A general implementation of this is the superheterodyne image mixer. The reference and scene input images are modulated at two different frequencies. Demodulation of the transform plane output at the sum or difference of the input frequencies results in the total separation of the correlation signals. This technique can be combined with spatial filtering to further improve system performance. For the case of square wave modulation, it is shown that both modulation and demodulation can be easily implemented. Potential performance improvement is demonstrated through computer simulation.
Algorithm to choose which pixels to zero out in a zero and phase filter
Robert R. Kallman
Show abstract
An algorithm to choose which pixels to zero out in an otherwise phase-only filter is presented. This algorithm involves a simple thresholding technique applied to the general spatial filter described in previous work. Having fixed those pixels which are to be set equal to 0 in the filter, one can then view the remaining phases as free parameters and design a good 0 and continuous phase filter by optimizing the filter's signal-to-clutter ratio using the author's techniques. One can the discretize this zero and phase filter into one whose entries are 0 or one of the nth roots of unity. For example, ternary (0, +1, and -1) filter are of this form as are filters whose entries are 0 and the sixteenth roots of unity. Examples of such are given. This paper together with its ancestor provides a completely automatic way to start with a training set of true and false targets of any size and automatically and without human intervention produce optimized filters for use in extant spatial light modulators.
Modified interferometric in-line time integrating correlator
Christopher W. Keefer,
Michael J. Ward
Show abstract
We built and tested a novel in-line interferometric time integrating acousto-optic correlation which has distinct advantages over existing designs. The novel design demonstrates improved vibrational stability, system light efficiency and dynamic range over the dual-path Mach- Zehnder interferometric correlator. The design also provides a tunable spatial frequency at the detector, unlike the non-interferometric in-line correlator which has a fixed spatial frequency.
Pattern recognition system by a matched spatial filter and neural networks for imperfect input characters
Shun-ichi Kamemaru,
Tsutomu Baba
Show abstract
A hybrid pattern recognition system using a matched spatial filter and neural networks is described. In the system, a multiplexed matched spatial filter synthesized by feature-extracted reference patterns is used for optical processing and an algorithm of a neural network is used for digital processing. The concept of the neural is based on an associative memory. It enables the system to recognize unknown input patterns with slightly different shapes from the reference patterns, which we could not distinguish by a conventional matched spatial filter. Using the system, imperfect input letters such as given by poor printing are fairly recognized as well as the perfect ones are recognized.
Information Processing I
Statistical analysis and modeling of analog optical processors
Show abstract
A generic three-plane optical processor is investigated from a statistical viewpoint. The means and the mutual coherence functions of the output field amplitude and intensity are derived. The photodetection process is then studied, and the mean and the autocorrelation function of the output current are found, thus establishing the functional form of the signal dependence of noise at the processor output. An integral equation and a series expression are also presented for the probability density function of the output signal. A special case is then analyzed, and the use of these expressions is demonstrated. Finally, device models to be used within this framework are summarized.
High-space bandwidth product computer-generated holography using volume holography
Show abstract
We suggest a method for coding high resolution computer-generated volume holograms. It involves splitting the computer-generated hologram into multiple holograms, each individually recorded as a volume hologram utilizing the maximal resolution available from the spatial light modulator. Our method enables their simultaneous subsequent reconstruction. We demonstrate the recording and the reconstruction of a computer-generated volume hologram with a space bandwidth product much higher than the maximal one of the spatial light modulator used as an interface. Finally, we analyze the scheduling procedure of the multiple holographic recording process in photorefractive medium in this specific application.
Three-dimensional optical time delay units for radar
Nabeel A. Riza
Show abstract
2D SLM based optical time delay units (OTDUs) are introduced for radar and wideband signal processing applications. In particular, the mature nematic liquid crystal (NLC) SLM technology is considered for the proposed architecture. A 1 X 2 optical switch is demonstrated using a parallel-rub birefringent-mode NLC cell, a bulk optics cube polarizing beamsplitter, and a sheet polarizer. Switch measurements taken at 633 nm show a > 3400 : 1 or > 35 dB output port optical isolation. The 1 X 2 NLC switch is used to build a 1-bit, 3.33 ns duration free-space OTDU using mirrors and total internal reflection corner prisms. The unit demonstrated a > 30 dB optical signal-to-noise ratio for both delay and no-delay positions.
Computer-generated holograms optimized by a global iterative coding
Show abstract
A global iterative coding method for computer-generated holograms (CGH) is introduced. The method is based on the iterative correction of a CGH with the use of standard Lee coding. The correction, i.e. the difference between the desired and the obtained reconstruction, is coded and added to the current CGH. The learning factor and the weighting factor are introduced in order to control the speed of convergence and the signal-to-noise ratio (SNR). Advantages lie in low computing time as well as improvement of the object SNR, the neighborhood SNR and the background SNR. Reducing the standard deviation of the phase of the reconstructed object also results. A slight improvement of the diffraction efficiency is also observed. The comparison is realized using the Lee interferogram method as a reference.
Optical processing at MICOM
Show abstract
The Weapons Sciences Directorate of the U.S. Army Missile Command has been directly involved in research and development of optical processing technology for several years. Extensive publications have described developments in several key areas of research. Current efforts are aimed toward military applications, for which optical processing offers great potential for improved speed and performance in automatic target detection, location, and recognition. Research is directed toward overcoming the remaining technical issues. Topics of investigation include: spatial light modulator performance and use; photorefractive materials for data storage; Vander Lugt, joint transform, and acousto-optic correlator designs; neural network implementations for morphological transforms, region and texture segmentation, rapid training techniques; and composite filter design, development, and implementation. This paper describes progress and accomplishments in some of these areas, with projections for future work.
Information Processing II
Designing fast optically controlled waveguide switches
Show abstract
Optically controlled two way switches are useful for constructing interconnection networks and for performing optical logic for parallel computing. A switch is constructed by placing two waveguides parallel and sufficiently close to each other that their fields couple transversely. The supermodes for this configuration are even and odd and travel at different phase velocities so that the output switches from one waveguide to the other repetitively with distance. The output can also be switched from one waveguide to the other by altering the optical length of the waveguide. Nonlinear materials are incorporated in the waveguide so that the optical path length may be altered by using an optical control beam to vary the intensity of light. A finite element method is described that enables the modeling of the nonlinear dual waveguide structure. Nonlinearity is handled by iteration and assuming an equivalent nonlinear relative permittivity.
Reduction of error effects in digital partitioning by error-correction coding
Show abstract
Analog optical matrix-vector multipliers (OMVMs) compute matrix-vector products rapidly due to the parallelism and high speed of optics. However, their low analog accuracy hinders their widespread application. Digital partitioning is a promising new technique for achieving high accuracy computations on analog OMVMs. One potential drawback of digital partitioning is its sensitivity to errors. The results presented here show how error-correcting codes can reduce this sensitivity to errors. Various computer simulation results are presented to show that a significant reduction of error can be obtained.
Compact electro-optic controller for microwave phased-array antennas
Nabeel A. Riza
Show abstract
A compact, low cost, phased array antenna optical beamformer with element level analog phase (0 - 2 (pi) ) and amplitude control using nematic liquid crystal display-type technology is experimentally demonstrated. Measurements indicate > 6-bit phase control and 52.6 dB of amplitude attenuation control. High quality error calibration and antenna sidelobe level control is possible with this low control power, analog antenna controller. Optical system options using rf Bragg cells or wideband Bragg cells are discussed.
Optical morphological processing of optical correlator signals
W. Michael Crowe
Show abstract
The U.S. Army Missile Command is investigating several optical correlator systems for target acquisition, recognition, and tracking. The correlation signal is often less than ideal and surrounded by a very noisy background. A morphological approach to noise rejection in the correlation plane could enhance the performance of these systems. A morphological algorithm, realized via threshold decomposition, has been tested to provide noise rejection in the correlation plane of an optical correlator system. The peak-signal-to-peak-noise ratio has been greatly enhanced, and this algorithm lends itself to implementation as a real-time (video frame rate) processor.
Poster Presentations
Optical control system for millimeter-wave phased-array antennas
Nabeel A. Riza
Show abstract
Novel compact electrooptic control systems are introduced for microwave and millimeter wave phased array antennas that use dual-laser frequency heterodyning techniques for radio frequency (rf) signal generation, and nematic liquid crystals for rf phase and amplitude control.
Implementation of arbitrary linear optical transformations with diffractive optics
Show abstract
All applications for diffractive optical elements can be seen as some type of linear optical transform implementation. Methods for the implementation of arbitrary linear optical transformations are discussed. Necessary and sufficient conditions for the implementation of linear optical transformations are considered. These are derived from the properties of a linear optical system. A taxonomy of linear optical transformations is provided. Point transforms are one of the groups of linear optical transformations. These are considered in more detail. The necessary and sufficient conditions naturally leads to the discrete phase technique of implementation. This technique comprise the weighted summation of localized diffraction gratings (or Fresnel lenses in the case of a lensless implementation). As an example of an implementation of this technique the Hough transform is considered. This well known transform is used for the processing of two dimensional images. The conventional Hough transform maps lines in an input image to points in a two dimensional output plane. The cartesian coordinates of the points in the output plane denote the orientations and locations of the lines in the input image.
Fractal dimension estimation for optical image segmentation
Show abstract
This research investigated the use of fractal dimension measure to segment regions of interest from simulated terrain imagery. The underlying assumption is: a given region of interest in a real-world image has a different fractal dimension than its surrounding terrain. Virtually illuminated, digitally simulated fractal surfaces with known fractal dimensions were investigated. The terrain types that were considered varied from bumpy to flat surfaces. An optically based image segmentation system was constructed to perform the otherwise computationally intensive Fourier transform of the image to be segmented.
Optical wavelet processor for wavelets defined in the time domain
Yan Zhang,
Emmanuel G. Kanterakis,
Al Katz,
et al.
Show abstract
In this paper, we present an optical implementation for 1D wavelet processing which is able to represent wavelets defined in the time domain and deliver complex coefficients. The system is the combination of our previously reported Smartt interferometry based wavelet processor and a modified version of VanderLugt filter. Haar wavelet, which requires complex representation in the frequency domain, was used to demonstrate the system operation.
Optoelectronic Neural Networks Session 8
Monolithic optoelectronic transistor: a new optical neuron device
Show abstract
An integrated optoelectronic device, the monolithic optoelectronic transistor (MOET), has been demonstrated. The MOET functions as an optical sum-and-threshold device with large- signal optical gain. It can be electrically biased to achieve either abrupt switching thresholds or quasi-sigmoidal optical transfer characteristics, and excitatory and inhibitory inputs can be incorporated through a simple modification of the single-input device. Initial MOET devices displayed an optical gain greater than 10 and an output contrast ratio exceeding 50. The MOET has promising characteristics as a building block of optoelectronically implemented neural networks and image preprocessing systems.
Optoelectronic feature extractor for classification of fingerprints
Show abstract
We present the design and test results of an optoelectronic feature detector array for use in a fingerprint ridge extraction system. The system comprises a microlens array for taking multiple Fourier transforms and an array of detectors. Each detector is composed of a set of wedge photodetectors with associated winner-take-all circuitry and encoders for reporting the maximally stimulated wedge element.
Large-scale neural network model for multiclass pattern recognition
Show abstract
This paper presents a large-scale neural network training model--a gray-scale interpattern association neural network model for feature extraction and fast training. A neural network based composite filter (NNCF) concept is proposed for neural network training of Fourier plane filters. The NNCF generation methods can selectively enhance features in the Fourier domain. The nonlinear combination of multiple filters through neural network training enables multi-class pattern recognition.
Optoelectronic Neural Networks Session 9
New dimensions in D-STOP neural systems
Show abstract
This paper describes a scalable, highly connected, 3D optoelectronic neural system that uses free-space optical interconnects with silicon-VLSI based hybrid optoelectronic circuits. The system design uses an efficient combination of pulse-width modulating optoelectronic neurons and pulse-amplitude modulating electronic synapses. A prototype system is built and applied to a simple classification problem. An optoelectronic testbench for evaluating learning algorithms suitable for the optoelectronic architecture is implemented. Future directions for the optoelectronic architecture are also discussed; these include limited interconnect neural systems and parallel weight loading that allow receptive fields of arbitrary sizes and connection multiplexing to be achieved.
Experimental demonstration of a 4D optical neural network
Show abstract
This paper describes the construction and operation of a 4D neural network computer. This demonstration system uses holographic interconnects recorded in a volume spectral hole burning medium. The paper provides an overview of the demonstration system and includes experimental details of components: the tunable laser, the detector arrays, the spatial light modulators, and preparation and cooling of the spectral hole burning medium. Experimental results showing association of image patterns and a bidirectional associative memory experiment are presented and discussed.
Optoelectronic Neural Networks Session 11
Feature-enhanced IPA neural network and optical realization
Wenlu Wang,
Minxian Wu,
Shutian Liu,
et al.
Show abstract
The dynamic behavior of a neural network is demonstrated by its interconnection weighted matrix. In this paper, we present a Feature Enhanced Interpattern Association (FEIPA) neural network model which is sensitive to special features of reference patterns in the reconstruction. We think of the common part of the stored patterns as redundance and discard its contributions in the associating process. It is equal to enhance the role of special features of the reference pattern in the IWM and in the reconstruction procession. Therefore the IWM of FEIPA is well-distributed and the output before threshold is a little more uniform than that of IPA model. A 2D (8 X 8) optical system is constructed using lenslet array as interconnection to realize the FEIPA model. Digital simulation and experiment results are provided.
Optoelectronic Neural Networks Session 10
Holographic neurocomputer for backpropagation based on cascaded-grating holography
Show abstract
We describe recent results for implementation of the backpropagation neural network on an experimental optical neurocomputer developed under ARPA sponsorship. Weights are stored holographically in a photorefractive crystal using a recording method which eliminates crosstalk due to Bragg degeneracy and permits full utilization of the input and output. Experimental results are discussed for handwritten digit recognition.
Optoelectronically implemented neural network with a wavelet preprocessor
Show abstract
An optoelectronic neural network based upon the Neocognitron paradigm has been implemented at JPL and successfully demonstrated for automatic target recognition for both focal plane array imageries and range-Doppler radar signatures. A novel feature of this neural network architectural design is the use of a shift-invariant multichannel Fourier optical correlation as a building block for iterative multilayer processing. An innovative bipolar neural weights holographic synthesis technique was utilized to implement both the excitatory and inhibitory neural functions and dramatically increase its discrimination capability. In order to further increase the optoelectronic Neocognitron's self-organization processing ability, a wavelet preprocessor has been developed for feature extraction preprocessing (orientation, size, location, etc.). The addition of this wavelet processor would enable the neocognitron to dynamically focus on the incoming targets based on their known features and result in higher discrimination and lower false alarm rate. The theoretical analysis of an orientation and scale selective wavelet is provided. A multichannel optoelectronic wavelet processor using an e- beam complex-valued wavelet filter is also presented. Experimental demonstrations of wavelet preprocessing for feature extraction are also provided.
Optoelectronic Neural Networks Session 11
Image processing applications of optical neural networks
Show abstract
In this paper we explore the use of 3D disks for the construction of networks with extremely large storage capacity. 3D disks can store up to 1012 weights per disk. In this paper we discuss how 3D disks are used to implement an optical neural network and then derive the capacity and speed of the resulting architecture.
Optoelectronic Neural Networks Session 9
Implementing neural nets on non-ideal analog hardware
Leonard Neiberg,
David P. Casasent
Show abstract
We consider the implementation of high capacity Ho-Kashyap associative processors on non- ideal optical and analog VLSI systems. Processor non-idealities considered include quantization, non-uniform beam illumination, and nonlinear device characteristics. New training-out techniques to overcome these non-idealities are advanced. We obtain optimal performance in the presence of stochastic noise by proper selection of the processor parameter (sigma) syn. We derive important results that allow us to a priori determine the optimal value of (sigma) syn and the expected recall accuracy P'c without having to simulate the specific processor. We present a new algorithm that allow us to achieve storage near the theoretical maximum capacity (2N, where N is the dimensionality of the input vector) with excellent recall accuracy. Optical laboratory results are included. We achieved storage of 1.5 N with recall accuracy P'c >= 95% with input noise of standard deviation (omega) 1 equals 0.02 present and with optical analog components with 5 bit input accuracy and 8 bit memory matrix accuracy. With higher accuracy analog VLSI components (10 bit input accuracy and 11 bit weight accuracy), we achieve storage of 1.75 N with P'c equals 96.43%.
Optoelectronic Neural Networks Session 11
Full simulation of optical neural networks
Walter B. Marvin,
Dhananjay S. Phatak,
Wayne P. Burleson
Show abstract
The use of optics to realize neural networks is not new. High level simulations of neural networks are common, but detailed optical simulations of the implemented machines are not. The full simulation of such optical-neural networks using advanced software simulation techniques may pave the way for future practical working models. Both physical and logical simulations are provided. The combination of both algorithmic and physical simulations of a neural network provides the opportunity to set up `thought experiments' in a very short time and at low cost. The characteristics of the network would be known at the time of physical implementation. This allows efficient use of scarce laboratory resources.
High-capacity content-addressable memory architecture
Show abstract
This paper describes a new hardware architecture for searching and accessing data. This Content Addressable Memory (CAM) can be implemented using holographic storage in spectral hole burning media. The use of laser wavelength as a fourth dimension for volume holographic recording provides an additional addressing variable which can be used to advantage in a CAM architecture. This paper consists of three parts: definition of a CAM, presentation of two CAM concepts for digital data string and analog function search, and a discussion of architecture issues.
Optical time-division multi/demultiplexer utilizing optoelectronic switches
Xianjun Mao,
Ruibo Wang
Show abstract
Bistable light-emitting Diode (BILED) circuit is a useful optical switch for optical information processing. In this paper, a new design for time-division multi/demultiplexer utilizing BILED circuits is proposed and demonstrated.
Memory Architectures
Optical coherent transient memory systems
W. Randall Babbitt
Show abstract
A performance analysis of an optical coherent transient optical memory system is presented. The simplest configuration of a coherent transient memory is assumed in which a large block of bits are stored at each laser spot by a single data pulse. The reference and recall beams are assumed to be temporally brief and a photon gated storage medium is assumed. Even with this basic system configuration, the performance predictions yield promising results. With the assumed material parameters, data rates of 5 GHz, data block sizes of 4 X 105, and storage densities up to 4.3 X 1011 bits per cm2 are predicted.
Optical and electronic error correction schemes for highly parallel access memories
Show abstract
We have fabricated and tested an optically addressed, parallel electronic Reed-Solomon decoder for use with parallel access optical memories. A comparison with various serial implementations has demonstrated that for many instances of code block size and error correction capability, the parallel approach is superior from the perspectives of VLSI layout area and decoding latency. The demonstrated Reed-Solomon parallel pipeline decoder operates on 60 bit input words and has been demonstrated at a clock rate of 5 MHz yielding a demonstrated data rate of 300 Mbps.
Holographic memory design tradeoffs and upper performance bounds
Show abstract
This paper addresses the issues associated with simultaneous achievement of high capacity, high data rate, and a short access time in volume holographic memories. We show that fundamental limits impose performance tradeoffs in any volume holographic memory system. When combined with the state of the art in compact lasers, spatial light modulators, and detector arrays, the overall performance of these memories can be bounded. Achieving greater performance will require either significant improvements in these components, or memory architectures which permit parallel storage systems to be used to increase the capacity or data rate. Conversely, component performance requirements should be evaluated within the context of an entire memory system.
Poster Presentations
Three-dimensional optical storage system based on electron-trapping thin films
Xiangyang Yang,
Charles Y. Wrigley,
Joseph Lindmayer
Show abstract
A 3D optical memory based on stacked-layer transparent electron trapping (ET) thin films is under development at Quantex Corporation. As a bit-plane oriented 3D optical storage system, 2D data pages can be accessed fully in parallel. A page of data are written into the memory by properly imaging the page composer onto the addressed ET layer with a dynamic focusing lens By suitably controlling the voltage applied to each of the ET film layers, only an addressed layer is accessible, while all the other layers are turned off. To read out a data page, a slice of infrared (IR) light is guided into the ET film from the edge. Due to the high refractive index of ET materials, the IR light is restricted within the addressed ET film and can not leak to the other layers. The emission corresponding to the written data is stimulated by the IR reading light and is detected by an array detector. The feasibility of this 3D storage system has been demonstrated by preliminary experiments.
Data transfer rates and parallelism of two-photon memories
James H. Strickler
Show abstract
Data transfer rates of recently proposed 3D two-photon memories are potentially limited by repetition rates of pulsed lasers. Herein we propose a method of spatially and temporally multiplexing pulse from a multi-Watt, 100 MHz repetition rate, sub-picosecond pulse width mode-locked laser to a linear array of parallel data channels in order to increase the two- photon writing rate. Nearby channels are desynchronized so that pulses incident on adjacent regions of the storage medium do not overlap far from the focal plane. This prevents pairs of photons directed to proximate bits from being simultaneously absorbed in regions of beam overlap which would otherwise degrade the optical sectioning property of the two-photon absorption method. This method may be applied to an earlier proposed architecture that uses 2D arrays of channels for massively parallel memories. Our modified writing method obviates the necessity for intersecting beams and thus reduces the diffraction limit to the storage density of this alternative approach.
Formation of optical disk direction paths and optical master disks with the help of inorganic resists
Show abstract
In this work, the results of investigations performed on As40S40Se20 and As40S60 layers are presented, peculiarities of optical disks direction paths and optical master disks formation with the help of laser lithography are described.
Optical Memory Materials
Spectral hole-burning holography in optical memory systems
Show abstract
Experiments verifying a new method of storing spectral hole burning holograms, which yields reduced crosstalk, have been conducted. Results demonstrating the reduced width of this type of hologram in both frequency and applied electric field are presented. Using this technique, 6000 holograms have been recorded in a single spectral hole burning sample.
Write/read performance in two-photon 3D memories
Show abstract
We describe progress in developing a 3D optical memory using 2 photon optical recording. The memory can be designed for bit, line, or page oriented data storage. We discuss advantages and drawbacks to each addressing method in terms of the capacity, storage density, and peripheral devices. We conclude with the most recent experimental results for multiple image storage using 2-photon recording.
Memory Components
Experimental demonstration of ultradense 4D holographic mass storage
Show abstract
This paper presents the results of a two-year Phase II SBIR program investigating a number of the key aspects of the use of Spectral Hole Burning media in a high capacity holographic optical digital computer memory. Factors which were experimentally examined include data longevity and unintentional erasure, and fundamental capacity issues relating to data densities and crosstalk. An experimental memory system was constructed and tested which had all the key elements of a digital memory system. Our experimental results confirm our previous analyses which indicate useful storage densities of 1012 bytes/cm3.
Large-scale holographic memory: experimental results
Show abstract
We present experimental results of a page-formatted random-access holographic memory capable of storing up to 1012 bits of information. Up to 500 holograms were angularly multiplexed at each of 8 spatially multiplexed locations, using a mechanical scanner and a segmented mirror array.
Programmable beamlet generator, dynamic lens, and optical memory using electrically switched holographic devices
Lawrence H. Domash,
Conrad M. Gozewski,
Arthur R. Nelson,
et al.
Show abstract
This report is an update on our continuing exploration of the wide range of potential device designs using electrically switchable holographic composites (ESHC) for memory applications. Here we shall describe a self contained programmable beamlet array generator, a laser focusing lens with four discrete selectable focal points, and a simple associative holographic memory showing how multiplexed data storage can be implemented with no moving parts through the use of ESHC.
Very high-density magneto-optical disk drive by using pulse erasure technology
Wai William Wang,
Tzuo-Chang Lee,
Kerry D. Rhea,
et al.
Show abstract
A pulse erasure technology is presented which allows accurate erasure of individual marks on a track. This makes it possible to obtain very high track densities without the risk of partially erasing marks on the adjacent tracks. Very large storage capacities can thus be achieved.
Motionless head for parallel readout optical disk for optoelectronic content-addressable memory system
Show abstract
The latest experimental results of the motionless head parallel readout system for optical disks are presented. The system is designed to read data blocks encoded as 1D Fourier holograms distributed radially on the disk active surface. Such systems offer several advantages: high data rates, low retrieval times and simple optical implementation.
Information Processing I
NSF's role in optical information processing
Albert B. Harvey
Show abstract
In this presentation the topic surrounding NSF's role in the growing research area of Optical Information Processing (Lightwave Technology, Photonics, Optoelectronics, Optical Communications, and Optical Computing), is discussed. In particular, the interaction of university research and industrial participation is explored. Opportunities for joint projects, information exchange, sharing of equipment, expertise, and facilities among research partners are discussed. The changing posture of NSF in terms of emphasis on education, cross- disciplinary activities, human resources and interaction with government labs and industry is introduced.
Information Processing II
Direction-dependent filtering using self-pumped phase conjugation
Show abstract
A new direction-dependent spatial filtering is presented with a self-pumped BaTiO3 crystal as an optical phase conjugator in two different geometries. In the first geometry, an input object (e.g., the letter K) is focused into the crystal by a Fourier transforming lens. As the crystal was located at the back focal plane of the lens, the vertical edges of the object orthogonal to the plane of incidence became most significantly enhanced. Other deviated from the orthogonal direction gradually fell off in brightness. Much better performance in speed and system control could be achieved with the second geometry where a reference beam along with a spatial filter forms gratings inside the crystal and the object beam focuses toward the gratings. The direction-dependency was then achieved with a long-narrow spatial filter oriented horizontally in the reference beam. The result is a high speed direction-dependent edge enhancement. The effect of the filter size on the performance is also discussed.
Poster Presentations
Minimum mean square error filter for pattern recognition with spatially disjoint signal and scene noise
Show abstract
A minimum mean square error filter for pattern recognition problems with input scene noise that is spatially disjoint (or nonoverlapping) with the target is described. The filter is designed to locate the target by producing a delta function output at the target position. The filter minimizes the mean square of the difference between the desired output delta function and the filter output in response to a noisy input data. We show that the filter output has a well defined peak and small sidelobes in the presence of spatially disjoint target and scene noise.
Information Processing II
Photorefractive phased-array-radar processor dynamics
Show abstract
We derive, and experimentally verify the dynamic and steady state behavior of a high- bandwidth, large degree-of-freedom adaptive phased-array-radar optical processor. The large number of adaptive weights necessary for processing in a complex radar signal environment with large arrays are computed in the form of dynamic three-dimensional volume holograms in a photorefractive crystal. The processor computes the angles-of-arrival of multiple interfering narrowband radar jammers and adaptively steers nulls in the antenna pattern in order to extinguish the jammers. The theoretical model developed provides analytical expressions relating system parameters such as feedback gain and phase to suppression depth and convergence rates for multiple narrowband jammers of arbitrary spatial profile. We have obtained experimental verification of the system behavior showing excellent agreement with the theoretical model and experimental jammer suppression as high as -35 dB.
Pattern Recognition I
Comparison of MINACE and smart TPAF correlation performance
Show abstract
The correlation performance of MINACE SDF filters and distortion invariant TPAFs (ternary phase-amplitude filters) on visible-wavelength images of an actual tank on a natural terrain background is compared in simulations. The comparison is performed for four test images and with three types of preprocessing: none, edge-enhancement, and binarization. Testing is realistic in that it uses only scenes not included in the training images used to construct the filters. The implications of the results for correlator design are discussed.
Pattern Recognition IV
Dual-channel solid block optical correlator
Show abstract
A new solid block correlator has been built for the U.S. Army Missile Command by OCA Applied Optics. This Advanced Solid Block Correlator (ASBC) is based on the same design approach useful for the original solid block correlator. The ASBC provides both improved performance and a vehicle for evaluating the recently available Liquid Crystal Televisions as spatial light modulators (SLMs). Performance is significantly enhanced by: (1) real-time programmable filter capability, (2) improved SLM speed, and (3) improved SLM uniformity. Furthermore, the ASBC also has the advantage of two separate filter channels to increase filter throughput or allow simultaneous monitoring of the response to two different filters. This paper will present the results obtained from initial tests performed with the ASBC.
Poster Presentations
Partial response precoding for parallel readout optical memories
Show abstract
The data density that can be resolved in optical memories is limited by the inherently bandlimited nature of optical systems. In this paper, we show how a precoding technique used in serial communications, Partial Response (PR) precoding, can be applied to a parallel readout optical system to pre-compensate for the spatial data broadening that occurs as a result of this bandlimiting. We examine the performance of 1+D PR precoding, a particular form of PR precoding, in an unapodized coherent imaging system. However, this technique can also be applied to other readout systems. A factor of 15 improvement in average worst case contrast ratio and a factor of 11 improvement in average contrast ratio were measured experimentally. Approximately 50% more area would be needed to achieve the same contrast ratio in a system without precoding. PR precoding was also found to be robust to defocusing and slight tilts, and can be used with a broad range of detector sizes.
Information Processing II
Distance classifier correlation filters for distortion tolerance, discrimination, and clutter rejection
Show abstract
A new approach to correlation filters based on quadratic distance calculations is described. The problem of distortion tolerance is addressed in terms of similarity measures. Discrimination is simultaneously addressed by optimizing the filters to maximally separate the classes. Mathematically, filter synthesis requires the inversion of diagonal matrices in the frequency domain and is a generalization of the MACE idea. The approach is shift-invariant, does not require feature extraction or image registration, and is significantly different from traditional pattern recognition techniques such as the Fisher LDF. The proposed approach is also suitable for the rejection of unknown clutter. Since recognition is based on similarity, clutter and false images which exhibit `large' distances from true target classes, are easily rejected. With improved recognition and discrimination performance, and low false alarm rates, the proposed distance classifier is a promising method for multiclass target recognition in cluttered environments.