The performance of GaAs/AlGaAs multiple quantum well long-wavelength infrared detectors is evaluated for potential applications in focal plane arrays. A number of GaAs/AlGaAs quantum well infrared detectors with absorption between 7 to 12 micrometers have been fabricated and characterized. In these samples, the quantum well width and barrier height were held approximately constant, while the AlGaAs barrier thickness was varied from 300 to 500 angstrom. These detectors were characterized by FTIR absorption, dark current, responsivity, spectral noise density, and thermal activation energy measurements at temperatures ranging from 6 to 80 K. A maximum detectivity of 4 X 1013 cm(root)Hz/W at 6 K is obtained at (lambda) equals 8.4 micrometers for the 500 angstrom barrier sample.
We demonstrate the first long wavelength quantum well infrared photodetectors (QWIPs) using lattice matched n-doped In0.47As/InP and n-doped 1.3 micrometers InGaAsP/InP materials systems. The responsivity of In0.52Ga0.47As/InP detectors has been found to be larger than that for similar GaAs/AlxGa1 - xAs detectors. In addition we demonstrate the first p-doped In0.53Ga0.47As/InP QWIPs. This detector has the shortest wavelength response, (lambda) p equals 2.7 micrometers , ever achieved in a QWIP and operates at normal incidence.
In the last five years, photoconductive infrared detectors using intersubband transitions in AlGaAs/GaAs multiple-quantum-well structures have attained more than adequate single-pixel detectivity for thermal imaging. Because AlGaAs/GaAs materials technology can potentially deliver a high yield of extremely uniform large-area detector arrays, lower-cost LWIR thermal imagers may indeed be possible. In this paper, we ask whether arrays of these detectors are suitable for staring LWIR thermal imagers. At first look they do not appear particularly attractive because of excessive dark-current pattern noise. For practical operating temperatures above 68 K (approximately the lowest temperature for single-stage cryocoolers), total dark- current variation must be reduced by more than an order of magnitude to attain a practical noise-equivalent temperature difference (NETD). Reducing dark current is therefore the most effective way to improve staring array performance. Several other measures can also improve NETD. Optimizing bias voltage for minimum NETD rather than for maximum detectivity can improve NETD by a factor of 2. Integrated micro-optics can reduce detector area, reducing pattern noise by reducing dark current. A multiple-quantum-well chopper can be used to improve correction for dark-current nonuniformity. We conclude that a practical thermal imager can probably be made with AlGaAs/GaAs MQW technology, but success will require careful modeling of all relevant factors.
We report optical modulation results on InGaAs/GaAs coupled multiple quantum well spatial light modulators. The structures consist of an n+ GaAs buffer, an undoped 250 period coupled multiple quantum well layer, and a p+ GaAs/InAs cap. The samples are probed at room temperature using photoabsorption spectroscopy. An absorption peak is observed at 969 nm, and this shifts to 982 nm as the field is increased from 0 to 67 kV/cm, in good agreement with theory. For a single pass through the structure, this results in a contrast ratio > 8:1 at 969 nm. A second modulator exhibits a contrast ratio > 8:1 at 1.04 micrometers .
Metal-Semiconductor-Metal (MSM) photodetectors on semi-insulating gallium arsenide (SIGaAs) have found widespread application as front-end detectors in receivers for optical fiber communications. Their major attributes are (a) simplicity of design and construction, (b) high responsivity, and (c) high speed of response. In this paper the performance of MSM detectors at short wavelengths is examined and considerations are given for their use in GaAs charge-coupled-device (CCD) imagers. The simplicity of their structure allows MSM detectors to be easily integrated with two-phase meander channel CCD registers in GaAs for production of high speed light scanners. Test results show that GaAs MSM detectors have high sensitivity in the wavelength range from 200 to 800 nm and good linearity over three orders of magnitude variation of the incident intensity. These features make MSM detectors potential candidates for both linear and 2-D light scanners for ultraviolet (UV) spectrophotometry and astronomy.
The potential of focal-plane signal processing for space-borne scientific imagers is discussed. Significant improvement in image quality and consequent scientific return may be enabled through the utilization of focal-plane signal processing techniques. The possible application of focal-plane signal processing to readout noise reduction, cosmic ray circumvention, non-uniformity correction, and throughput enhancement is described. On-focal-plane analog-to-digital (A/D) conversion and micromotion stabilization are also discussed. It is the intention of this paper to stimulate further thought and efforts in this field.
Digital technology has made dramatic strides in providing higher signal processing throughput in smaller packages. However, meeting processor throughput requirements for future imagers will be difficult using current or near-term available digital technology. As fabrication technology allows for more dense focal planes, typical image processing throughput requirements per frame increase with the number of pixels; for example, the number of pixel gain and offset computations quadruple when the focal plane size increases from a 128 X 128 to a 256 X 256 array. Several analog technologies are answering the driving throughput requirements for image processing applications; one such technology is the movement of charge packets through a piezoelectric GaAs channel, called acoustic charge transport (ACT). Analog solutions such as ACT offer orders of magnitude higher throughput imagers than A/D conversions and digital processing approaches. This paper includes (1) an overview of the theory of ACT device operation, (2) currently available ACT technology and devices, and (3) computational benefits. Detailed proprietary architectures of current ACt image processor designs are not discussed here.
ISC has completed test on an IC which has 32 channels of amplifiers, low pass anti-aliasing filters, 13-bit analog-to-digital (A/D) converters with non-uniformity correction per channel and a digital multiplexer. The single slope class of A/D conversion is described, as are the unique variations required for incorporation of this technique for use with on-focal plane detector readout electronics. This paper describes the architecture used to implement the digital on-focal plane signal processing functions. Results from measured data on a test IC are presented for a circuit containing these functions operating at a sensor frame rate of 1000 hertz.
GaAs MESFET technology is ideal for use in lightwave receiver applications. FET devices have a fundamental advantage over BJT transistors in low noise applications because of their inherent high input impedance. Another advantage comes from the fact that FETs are majority carrier devices and can be easily used as feedback elements in automatic gain control applications.
We designed and tested a two-dimensional silicon receptor array constructed from pixels that temporally high-pass filter the incident image. There are no surround interactions in the array; all pixels operate independently except for their correlation due to the input image. The high- pass output signal is computed by sampling the output of an adaptive, high-gain, logarithmic photoreceptor during the scanout of the array. After a pixel is sampled, the output of the pixel is reset to a fixed value. An interesting capacitive coupling mechanism results in a controllable high-pass filtering operation. The resulting array has very low offsets. The computation that the array performs may be useful for time-domain image processing, for example, motion computation.
With rapid advancements in infrared focal plane array (IRFPA) technology, greater demands are being placed on nonuniformity correction (NUC) techniques to provide near-BLIP performance over a wide dynamic range. Standard NUC techniques involve calibrating each detector using reference temperature sources before imaging the IRFPA. Usually the correction needs to be re-calibrated after a short period of time due to IRFPA drift or to adjust for changes in the level of background flux. Adaptive NUC techniques eliminate the need for calibration by continuously updating the correction coefficients based on radiance levels of the scene being viewed. In this manner, continuous compensation can be applied adaptively for individual detector non-idealities and background changes. Two adaptive NUC techniques are discussed; one is a temporal highpass filter and the other involves a neural network with lateral interconnects to nearest neighbor pixels. Both have similarities to biological retinal processing. Questions of implementation and stability are discussed and performance results are given for several test image sequences which were obtained from an MWIR HgCdTe array and a HIDAD uncooled array. We conclude that adaptive techniques will be very useful in future IRFPA sensors, primarily because of their ability to adapt over a wide range of background flux without calibration sources, but also because they can offer improved sensitivity under most operating conditions.
The time and expense involved in producing infrared hybrid focal plane arrays (IRFPAs) dictates that optimal designs be used. The use of accurate signal and noise models, which identify the critical focal plane design parameters, are essential in obtaining these optimal designs. We describe the methodology behind building a signal and noise model for a hybrid infrared focal plane array consisting of photodetectors coupled to silicon complementary- symmetry metal-oxide-semiconductor (CMOS) readouts. The CMOS readout is essential in transducing the infrared photosignal to a usable voltage domain form, but introduces front-end noise in the hybrid FPA. A signal and noise model for a photodetector-readout combination that is currently used in many applications is examined in order to identify the parameters that affect performance.
The low background performance of most infrared systems has been limited by the read noise of the focal plane array (FPA). In many cases the cause has been attributed to the use of MOSFETS or to the detector itself. In this paper we describe a multiple correlated sampling technique and demonstrate noise improvements by a factor of 9 over conventional readout techniques in use today. The FPA under discussion is a `Source Follower per Detector' or SFD type readout structure; specifically, the Santa Barbara Research Center 58 X 62 InSb FPA and the Hughes Technology Center (HTC) 256 X 256 PtSi FPA. The quoted read noise for the SBRC FPA (typically 400 electrons rms) has been reduced to 45 electrons rms using this technique.
This paper addresses the use of an adaptive noise canceling technique to eliminate the coherent noise generated in scanner data. The technique is based on a Finite Impulse Response (FIR) adaptive noise canceler. A two-weight FIR filter is used to adaptively learn the characteristics of a sinusoid. This sinusoid is then removed from the data. The least Mean Squares (LMS) algorithm is used to converge to the coefficients of the adaptive filter during the learning process. An image corrupted with a single frequency periodic noise is used for investigating the algorithm. It is observed that the efficiency of the algorithm is dependent on the convergence gains and the initial positioning of the weights of the FIR filter. Because of the computational simplicity of the algorithm, it is possible to implement this in real-time mode.
3-D focal plane array read-out structures offer the means for a fully interconnected, many level neural network with the potential for biological recognition capabilities. In this, the first of a two part paper, the 3DANN concept is presented and described. The second part discusses and compares the hardware implementation with more conventional approaches.
During the 1930s and 1940s Norbert Wiener and others invented the core concepts of linear signal processing. These ideas quickly became popular and played a significant role in the Allies' victory in World War II. During and after the war, linear signal processing theory was greatly expanded and began to take on the character of an imposing monolith. By the mid- 1940s, Wiener (and others, such as Dennis Gabor) came to recognize that linear signal processing theory, while interesting and very useful, was only a piece of a much larger picture. In 1946 and 1958 Gabor and Wiener, respectively, attempted to address the whole picture. While they were not completely successful, they did implicitly set an agenda for a more general approach to signal processing. Although a few others have, from time to time, addressed this agenda; in terms of the signal processing community as a whole it still remains lost in the shadow of the ever-growing monolith of linear signal processing theory. The thesis of this paper is that it is now time to get on with the Wiener and Gabor agenda. It is time to make general signal processing the mainstream focus of the subject. It is argued here that the best way to do this is to abandon the transfer function/Fourier analysis/z-transform approach of the current linear signal processing regime and replace it with a much more natural intellectual framework for general signal processing--the framework offered by neurocomputing. A potential benefit of this refocusing of the field is that the detailed engineering might soon be left to machines, while human technologists will be able to concentrate on the art of signal sculpting.
This paper estimates the upper and lower bounds of the frame noise of a linear detector array that uses a one-dimensional scan pattern. Using chi-square distribution, it is analytically shown why it is necessary to use the average of the variances and not the average of the standard deviations to estimate these bounds. Also, a criteria for determining whether any excessively noisy lines exist among the detectors is derived from these bounds. Using a Gaussian standard random variable generator, these bounds are demonstrated to be accurate within the specified confidence interval. A silicon detector array is then used for actual dark current measurements. The criterion developed for determination of noisy detectors is checked on the experimentally obtained data.
This paper discusses a novel technique for reducing the effects of microphony in pyroelectric infrared arrays. The method is an enhancement of the existing compensation techniques and uses the digital signal processing methods. Specifically adaptive noise cancellation is employed to perform the compensation. The paper discusses the choice of the system parameters, type of algorithm to be used and performance of the technique when applied to array elements both close and distant in the array. A comparison with conventional compensation is also made.
A focal-plane processing array for laser speckle interferometry is described. The imager consists of a 32 X 32 array of analog processing elements which determine the average phase shift of a holographic fringe pattern. The phase computation algorithm requires only the addition, subtraction and rectification (absolute value) of signals from adjacent pixels in the imager. Charge-coupled device structures are used to implement these operations in parallel and eliminate the need to read the pixel data off the array. This allows imager frame rates in excess of 104 frames/sec to be achieved. The imager sensitivity is currently limited by a 20% fill factor and non ideal performance of the rectifier. A higher resolution array is currently being designed which utilizes amorphous silicon photodiodes to improve the fill factor. This design also includes an integrated amplifier at each pixel to obtain photon shot noise limited sensitivity.
Outer-product learning (also referred to as Hebbian learning) has been used as a very simple training algorithm for neural networks. Outer-product learning also has been proposed as a method for training a network with higher-order interconnections. It is shown in this paper that outer-product learning is inappropriate for higher-order networks because it does not have the ability to perform a non-monotonic mapping.
Hyperspectral imagery, spatial imagery with associated wavelength data for every pixel, offers a significant potential for improved detection and identification of certain classes of targets. The ability to make spectral identifications of objects which only partially fill a single pixel (due to range or small size) is of considerable interest. Multiband imagery such as Landsat's 5 and 7 band imagery has demonstrated significant utility in the past. Hyperspectral imaging systems with hundreds of spectral bands offer improved performance. To explore the application of different sub pixel spectral detection algorithms a synthesized set of hyperspectral image data (hypercubes) was generated utilizing NASA earth resources and other spectral data. The data was modified using LOWTRAN 7 to model the illumination, atmospheric contributions, attenuations and viewing geometry to represent a nadir view from 10,000 ft. altitude. The base hypercube (HC) represented 16 by 21 spatial pixels with 101 wavelength samples from 0.5 to 2.5 micrometers for each pixel. Insertions were made into the base data to provide random location, random pixel percentage, and random material. Fifteen different hypercubes were generated for blind testing of candidate algorithms. An algorithm utilizing a matched filter in the spectral dimension proved surprisingly good yielding 100% detections for pixels filled greater than 40% with a standard camouflage paint, and a 50% probability of detection for pixels filled 20% with the paint, with no false alarms. The false alarm rate as a function of the number of spectral bands in the range from 101 to 12 bands was measured and found to increase from zero to 50% illustrating the value of a large number of spectral bands. This test was on imagery without system noise; the next step is to incorporate typical system noise sources.
Performance testing of large area mosaic focal planes requires high speed and high volume test methods because of the large number of individual elements to be tested. Extensive use of parallel activities is required to avoid the sequential delays involved in loading, evacuating, cooling, testing, warming, pressurizing, and unloading the test articles. A parallel path production test system designed for automated testing of over twelve thousand detector channels per hour has been built and placed into operation. Operation of the test system will be described and early performance data will be presented.
The marriage of superconducting electronics with Z-plane FPA readout structures offer the potential for high speed, low power parallel digital processing on-focal plane. This paper reports on some early research into this marriage of two technologies conducted by Irvine Sensors Corporation (ISC) and TRW. Progress is reviewed for both low and high temperature superconducting technologies.
3-D focal plane array readout structures offer the means for a fully interconnected, many level neural network with the potential for biological recognition capabilities. In this, the second of a two part paper, the 3DANN hardware design is discussed and its neural processing capabilities compared with other technologies.
This paper presents conceptual designs for high density packaging of parallel processing systems. The systems fall into two categories: global memory systems where many processors are packaged into a stack, and distributed memory systems where a single processor and many memory chips are packaged into a stack. Thermal behavior and performance are discussed.
Detailed radiometric characterization data must be reviewed at the module level(2048 JR detectors) prior to integrating
the hardware into higher level assemblies. Software has been developed that highly automates the process of reducing test
data for noise, responsivity, uniformity, output offset, saturation, linearity, filter pole location, and crosstalk. Output
consists of both text and graphics at different levels of detail in order to accommodate the needs of engineering, test,
manufacturing, quality assurance, and program management. All the results are placed on a LAN so that the necessary
reviews can occur in essentially a paperless environment.
This paper discusses the relationship of focal plane architectures and signal processing functions currently used in infrared sensors. It then discusses the development of an algorithm derived from the models developed by biologists to explain the functions of insect eyes and the hardware realization of this algorithm using commercially available silicon chips. The conclusion of this study is that there are important lessons to be learned from the architecture of biological sensors, which may lead to new techniques in electro-optic sensor design.