Compressive sensing has been identified as a significant technique to reduce the volume of data collected in sensing applications to a minimum. Prior art has empirically demonstrated the effectiveness of a spinning disk for reconstruction of TeraHertZ (THZ) images. Prior empirical data has demonstrated reconstruction artifacts that are associated, in part, with the statistical Probability Density Function (PDF) of the randomly distributed transmission holes in the rotating plate. Empirical demonstration at other wavelengths such as the InfRared (IR) has also been suggested. This document summarizes the statistical requirements for artifact minimization for the previously reported spinning disk system. Consideration is given to the impact of operation at non-THZ wavelengths such as the IR.
Spectroscopic detection and classification techniques suffer from the collection of excessive data and utilize only a fraction of the information collected for classification. Compressed Sensing (CS) techniques have been utilized in optical, photonic, electronic and controls applications. This limits data collection to the essentials and reduces the hardware, software, and computational requirements. Applying CS to just the general computational system results in the collection of data which is ultimately discarded. The result is excessive power consumption, mass, physical sizes, and complexity. Compressive Sensing requires, at a minimum, a non-uniform encoding system with a non-linear decompression system for total reconstruction. Pseudorandom encoding is frequently preferred. Total reconstruction of a compressed signal has been shown to be very computational intensive and other optical-based techniques have been demonstrated to accelerate the result. Prior work has demonstrated that total reconstruction is not necessary for effective classification via PCA and other spectroscopic relevant techniques. Prior work revised the system design and modified the signal processing, both electronic and computational, to reduce system requirements. To propagate this savings back into the photonics and optical chain, it is necessary to further develop alternative techniques. In particular, a modification to the traditional LDA allows the contraction of primary optics. In this presentation an optical detector scheme is detailed. A number of configurations are considered with the most savings achieved by a spatial integrating version that allows the maintenance of optical and photonic SNR by collecting a number of photons greater than or equal to the traditional LDA. Since primary optical diameter is largely specified by the need to subtend an angle sufficient to overcome system noise, optical diameters can be reduced by up to an order of magnitude. This also mitigates optical diameter driven resolution at the detector plane. Some third order and higher issues exist and are addressed. Theoretical development with limited empirical support is to be presented.
An engineering thermodynamic approach to the plasma description associated with Laser Induced Breakdown Spectroscopy (LIBS) has been previously published. In the prior work, a non-traditional modeling approach was made to reduce the modeling system to a configuration compatible with incorporation into a TI6701. This modeling technique was necessitated by the extreme limitations that portability and robustness place on the physical size and power consumption of the computer for data processing and classification. The new modeling approach was previously reported. This presentation reports on the finalization of this and the validation of the result through comparison to more established models such as detailed balance, via the solution to a system of Boltzmann Equations. The emphasis is on the engineering modeling and its’ system implications - not a physics tutorial. Implications of the modeling approximations for the accuracy and repeatability of the complete sensor system will be presented. Possible utilization of newer, larger scale processors and the impact that would have on the model and associate sensor performance is addressed.
Proc. SPIE. 8710, Chemical, Biological, Radiological, Nuclear, and Explosives (CBRNE) Sensing XIV
KEYWORDS: Digital signal processing, Principal component analysis, Statistical analysis, Sensors, Control systems, Distance measurement, Laser induced breakdown spectroscopy, Control systems design, Performance modeling, Binary data
Portable LIBS sensor communication bandwidth limitations favor local material classification for low power consumption. Partial Least Squares - Discriminant Analysis (PLS-DA) and Principle Component Analysis (PCA) have been implementation via general purpose computers and are accepted for some Department of Defense applications. Prior publications address the creation of a low mass, low power, robust hardware spectra classifier for a limited set of predetermined materials in an atmospheric matrix. The incorporation of a PCA or a PLS-DA classifier into a predictorcorrector implementation on a TI6701 has been developed. The performance modeling of the control system with an emphasis on further optimization needs addressing. This paper characterizes, from a control system standpoint, the predictor-corrector architecture applied to LIBS data collection. In addition, the application of this as a material classifier is presented. Updates in the model implemented on a low power multi-core DSP will be presented as well. Performance comparisons to alternative control system structures will be considered.
Cavity Ring Down Spectroscopy (CRDS) has been identified as having significant potential for Department of Defense security and sensing applications. Significant factors in the development of new sensor architectures are portability, robustness and economy. A significant factor in new CRDS sensor architectures is cavity length. Prior publication has examined the role of cavity length in sensing modality both from the standpoint of the system’s design and the identification of potential difficulties presented by novel approaches. Two of interest here are new noise terms that have been designated turbulence-like and speckle-like in prior publication. In the prior publication the theoretical and some empirical data was presented. This presentation addresses the automation of the experimental apparatus, new data analysis, and implications regarding the significance of the two noise terms. This is accomplished through an Analog-to- Digital Conversion (ADC) from the output of a custom designed optical correlator. Details of the unique application of the developed instrument and implications for short cavity (portable) CRDS applications are presented.
Proc. SPIE. 8359, Sensors, and Command, Control, Communications, and Intelligence (C3I) Technologies for Homeland Security and Homeland Defense XI
KEYWORDS: Digital signal processing, Principal component analysis, Sensors, Spectroscopy, Signal processing, Distance measurement, Laser induced breakdown spectroscopy, Sensor technology, Binary data, Data analysis
There are many accepted sensor technologies for generating spectra for material classification. Once the spectra are
generated, communication bandwidth limitations favor local material classification with its attendant reduction in data
transfer rates and power consumption. Transferring sensor technologies such as Cavity Ring-Down Spectroscopy
(CRDS) and Laser Induced Breakdown Spectroscopy (LIBS) require effective material classifiers. A result of recent
efforts has been emphasis on Partial Least Squares - Discriminant Analysis (PLS-DA) and Principle Component
Analysis (PCA). Implementation of these via general purpose computers is difficult in small portable sensor
configurations. This paper addresses the creation of a low mass, low power, robust hardware spectra classifier for a
limited set of predetermined materials in an atmospheric matrix. Crucial to this is the incorporation of PCA or PLS-DA
classifiers into a predictor-corrector style implementation. The system configuration guarantees rapid convergence.
Software running on multi-core Digital Signal Processor (DSPs) simulates a stream-lined plasma physics model
estimator, reducing Analog-to-Digital (ADC) power requirements. This paper presents the results of a predictorcorrector
model implemented on a low power multi-core DSP to perform substance classification. This configuration
emphasizes the hardware system and software design via a predictor corrector model that simultaneously decreases the
sample rate while performing the classification.
A standard spectroscopic sensor technique for classification of materials is Laser Induced Breakdown Spectroscopy
(LIBS). Though LIBS, as an Atomic Emission Spectroscopy (AES) technique, is generally separated from signal
processing based classification techniques, they strongly interact in the design of sensor systems. Strict disciplinary
separation results in approaches that inadequately address the mass, power consumption and other portability parameters
of the ultimate sensor. Modifications in the sensor design approach and of the classification processing techniques
reduce redundancies in the system, resulting in more compact overall systems. An engineering thermodynamic approach
to the plasma description, as part of a predictor-corrector style classification loop, is used to reduce system requirements
for material classification. This paper presents results for the compaction of the model system. In this work, a nontraditional
approach is made to reduce the modeling system to a configuration compatible with the incorporation of the
model onto a compact DSP structure. Calculation of partition function tables allows heuristic adjustments to a
thermodynamic description of the LIBS plasma. Once the plasma environment is established, rate equation descriptions
can establish detailed balance and predict the emission properties of the sample. The resulting model must be
compatible with compact, low power, computation schemes, such as multi-core DSPs as part of a predictor-corrector
Laser Induced Breakdown Spectroscopy (LIBS) utilizes a diversity of standard spectroscopic techniques for
classification of materials present in the sample. Pre-excitation processing sometimes limits the analyte to a short list of
candidates. Prior art demonstrates that sparsity is present in the data. This is sometimes characterized as identification
by components. Traditionally, spectroscopic identification has been accomplished by an expert reader in a manner
typical for MRI images in the medicine. In an effort to automate this process, more recent art has emphasized the use of
customized variations to standard classification algorithms. In addition, formal mathematical proofs for compressive
sensing have been advanced. Recently the University of Memphis has been contracted by the Spectroscopic Materials
Identification Center to advance and characterize the sensor research and development related to LIBS. Applications
include portable standoff sensing for improvised explosive device detection and related law enforcement and military
applications. Reduction of the mass, power consumption and other portability parameters is seen as dependent on
classification choices for a LIBS system. This paper presents results for the comparison of standard LIBS classification
techniques to those implied by Compressive Sensing mathematics. Optimization results and implications for portable
LIBS design are presented.
Laser Induced Breakdown Spectroscopy (LIBS) is dependent on the interaction between the initiating Laser sequence,
the sampled material and the intermediate plasma states. Pulse shaping and timing have been empirically demonstrated
to have significant impact on the signal available for active/passive detection and identification. The transient nature of
empirical LIBS work makes data collection for optimization an expensive process. Guidance from effective computer
simulation represents an alternative. This computational method for CBRNE sensing applications models the Laser,
material and plasma interaction for the purpose of performance prediction and enhancement. This paper emphasizes the
aspects of light, plasma, and material interaction relevant to portable sensor development for LIBS. The modeling
structure emphasizes energy balances and empirical fit descriptions with limited detailed-balance and finite element
approaches where required. Dusty plasma from partially decomposed material sample interaction with pulse dynamics
is considered. This heuristic is used to reduce run times and computer loads. Computer simulations and some data for
validation are presented. A new University of Memphis HPC/super-computer (~15 TFLOPS) is used to enhance
simulation. Results coordinated with related effort at Arkansas State University. Implications for ongoing empirical
work are presented with special attention paid to the application of compressive sensing for signal processing, feature
extraction, and classification.
Laser Induced Spectroscopy (LIBS) and Cavity Ring Down Spectroscopy (CRDS) are promising chemical sensor
technologies for small mobile robotic platforms. LIBS leverages the natural surface adsorption from the atmosphere to
interior and exterior surfaces for signal enhancement. In this work, a material or class-specific adsorption surface is
combined with a miniature version of a CRDS ring cavity to achieve a similar signal enhancement for CRDS. The
combination of LIBS and CRDS allow the analysis of both classes of materials - those with long adsorption times to
permanent surfaces such as walls and those that require real time sampling of ambient concentrations. This paper
emphasizes issues related to package miniaturization, power budget limitations and ruggedness as well as basic
performance modeling of the instruments. Comprehensive sensing issues for material specific micro-detectors will be
addressed. Computer simulations and some data are presented. Applications considered include the determination of
need for remediation and the determination of the effectiveness of remediation techniques as well as the detection of
hazards and intelligence gathering.
Optical systems for THz imaging frequently consist of readily available components - typically a single spherical or
parabolic surface. For THz imaging applications, the range is short, due to limitations associated with atmospheric
attenuation. For these applications, single spherical and parabolic surfaces are neither stigmatic, aplanatic nor Herschel.
As a result, many THz imaging systems exhibit significant image degradation caused by primary aberrations. Further,
the short range limitations frequently result in image degradation due to near violation of the paraxial assumption. For
improved imaging, an aplanatic system is required. To achieve aplanatism, a minimum of two aspheric surfaces is
required. Aplanatism requires stigmatic performance which dictates surfaces that are conic sections of revolution. A
minimum of two are required to exhibit stigmatism and meet the sine condition. An improvement of receiver form
factor allows for a decrease in optical image distance and an increase in system magnification factor. This significantly
improves a number of THz imaging characteristics such as depth-of-field while maintaining the diffraction-limit
resolution and reducing the primary objective diameter. Reduction of objective diameter reduces signal strength - principally at the expense of specular reflections. This paper summarizes the results of the optical system design and its
incorporation into THz imagers containing THz receivers with improved form factors. Efforts at incorporation of optical
zoom will be presented.
The U.S. Army Night Vision and Electronic Sensors Directorate (NVESD) and the U.S. Army Research Laboratory
(ARL) have developed a terahertz-band imaging system performance model for detection and identification of
concealed weaponry. The details of this MATLAB-based model which accounts for the effects of all critical sensor and
display components, and for the effects of atmospheric attenuation, concealment material attenuation, and active
illumination, were reported on at the 2005 SPIE Europe Security and Defence Symposium. The focus of this paper is to
report on recent advances to the base model which have been designed to more realistically account for the dramatic
impact that target and background orientation can have on target observability as related to specular and Lambertian
reflections captured by an active-illumination-based imaging system. The advanced terahertz-band imaging system
performance model now also accounts for target and background thermal emission, and has been recast into a user-friendly,
Windows-executable tool. This advanced THz model has been developed in support of the Defense Advanced
Research Project Agency's (DARPA) Terahertz Imaging Focal-Plane Technology (TIFT) program. This paper will
describe the advanced THz model and its new radiometric sub-model in detail, and provide modeling and experimental
results on target observability as a function of target and background orientation.
This paper describes the design and performance of the U.S. Army RDECOM CERDEC Night Vision and Electronic
Sensors Directorate's (NVESD), active 0.640-THz imaging testbed, developed in support of the Defense Advanced
Research Project Agency's (DARPA) Terahertz Imaging Focal-Plane Technology (TIFT) program. The laboratory
measurements and standoff images were acquired during the development of a NVESD and Army Research Laboratory
terahertz imaging performance model. The imaging testbed is based on a 12-inch-diameter Off-Axis Elliptical (OAE)
mirror designed with one focal length at 1 m and the other at 10 m. This paper will describe the design considerations of
the OAE-mirror, dual-capability, active imaging testbed, as well as measurement/imaging results used to further develop
The dynamic range of the signal return from metals is a significant source of image interpretation difficulty. Techniques such as logarithmic image compression have been used to improve the recognition. Alternative techniques for improvement may be developed. This development depends in part on the ability to accurately model the surface reflective behavior including phase shifts introduced by the reflection. This work presents the results of an enhanced model development. Models of high frequency behavior in materials divide into regions such as non-relaxation region, relaxation region, optical absorption and plasma frequencies. In traditional infrared and longer wavelength imaging systems, optical absorption may play a role and it is generally assumed that the system operates in or very near the relaxation region defined as frequencies significantly greater than the reciprocal of the Boltzmann relaxation time. Though typical THz frequencies are below the relaxation time, they are not far enough below to be considered completely in the non-relaxation region. This introduces a number of issues atypical of imaging in either the RF or IR regime. Further realism is gained from the incorporation of plastic into the reflectivity and emissivity model. Empirical model validation is accomplished for selected materials.
A primary source of "clutter" in sub-millimeter wave and terahertz imagery used in security applications is the random
reflections from clothing. In this paper, techniques for modeling and characterizing these reflections are described. This
work is motivated and, in part, based on previous work done in support of imaging radar for remote sensing. A first
order model of the response of a cloth covered object is described along with a method for performing measurements on
draped cloth. The measurement method involves the simultaneous measurement of the sub-millimeter wave response of
the cloth and the underlying drape of the cloth. A rigorous model of the scattering from draped cloth is developed and
compared with results from the first order model. Conclusions regarding the suitability of the first order model for
image simulation and performance predictions are stated.
We have developed several millimeter/submillimeter/terahertz systems to study active and passive imaging and associated phenomenology. For measuring the transmission and scattering properties of materials, we have developed a dual rotary stage scattering system with active illumination and a Fourier Transform spectrometer. For imaging studies, we have developed a system based on a 12-inch diameter raster-scanned mirror. By interchange of active sources and both heterodyne and bolometric detectors, this system can be used in a variety of active and passive configurations. The laboratory measurements are used as inputs for, and model calibration and validation of, a terahertz imaging system performance model used to evaluate different imaging modalities for concealed weapon identification. In this paper, we will present examples of transmission and scattering measurements for common clothing as well as active imaging results that used a 640 GHz source and receiver.
The U.S. Army Night Vision and Electronic Sensors Directorate and the U.S. Army Research Laboratory have developed a terahertz-band imaging system performance model for detection and identification of concealed weaponry. The MATLAB-based model accounts for the effects of all critical sensor and display components, and for the effects of atmospheric attenuation, concealment material attenuation, and active illumination. The model is based on recent U.S. Army NVESD sensor performance models that couple system design parameters to observer-sensor field performance using the acquire methodology for weapon identification performance predictions. This THz model has been developed in support of the Defense Advanced Research Project Agencies' Terahertz Imaging Focal-Plane-Array Technology (TIFT) program and is presently being used to guide the design and development of a 0.650 THz active/passive imaging system. This paper will describe the THz model in detail, provide and discuss initial modeling results for a prototype THz imaging system, and outline plans to validate and calibrate the model through human perception testing.
A novel imaging technique in which frequency-modulated retides encode different pixel locations by light modulation is presented. In this technique a reticle modulates different pixel locations at different frequencies, and photodetectors collect the resulting signals. Filters decode these signals to recreate the image on a display. The technique allows multiplexing many pixels onto a fewer number of detectors by utilizing the bandwidth of the detectors more effectively. Since frequency modulation creates an additional dimension for the detector, a single detector can function as a linear array, a linear array can function as a staring array, or the additional dimension can be used to convey spectral or other information. At wavelengths requiring expensive focal plane components, costs can be greatly reduced.