In this paper we demonstrate that the use of a relatively low crossrange resolution may be advantageous for aircraft identification in many situations. With high range resolution but low crossrange resolution, one can obtain a remarkably high feature extraction accuracy, provided the appropriate processing techniques are used. This approach will give a better identification performance than if one attempted to form an image with high crossrange resolution when the flight of the aircraft is unsteady. The radar can make the choice adaptively, depending on the flight behavior of the aircraft to be identified.

We describe three different processing approaches for stepped-frequency high-resolution radar that allow for the estimation of target motion parameters along the line of sight. Accurate knowledge of the motion parameters is necessary to generate a focused range-doppler image of the target. The approached adopted for this purpose are based upon the inverse discrete Fourier transform (IDFT), the modified chirp z-transform, and Prony's method. Estimates produced by these approaches are compared to each other as well as to the target's actual parameters. The IDFT approach, which is the standard procedure, requires zero padding to yield accurate estimates of the actual motion parameters and tends to be computationally intensive. The modified chirp z-transform approach yields accurate estimates with high computational efficiency for a wide SNR range. Prony's method proves to be a more robust approach for higher SNR. Considering these results, an MCZT adaptive algorithm for motion compensation is formulated which guarantees a well focused image.

SAR radars create 2D maps of terrain. Each image pixel corresponding uniquely to a terrain resolution cell. Due to terrain inhomogeneity and SAR's inadequate resolution, some of the resolution cells may contain multiple elementary scatterers belonging to different classes. Radar returns from such a resolution cell are attributable to collective interaction of electromagnetic waves which are scattered by various kinds of elementary scatterers within the resolution cell. Simultaneous use of multispectral and polarimetric measurements at the pixel makes it possible to decorrelate signals from different classes of elementary scatterers within the corresponding resolution cell, facilitating the separation of different subpixel-sized elementary scatterers in each pixel. Frequency and polarization diversity offered by multifrequency and polarimetric SAR provides a method to tackle an important yet challenge task: detecting stationary manmade ground subpixel-sized targets from complex natural clutter. The objective of this investigation is to optimize frequencies and polarizations for detecting subpixel-sized target in clutter using multilook, multifrequency polarimetric SAR data for a pixel. In the general target-plus-clutter versus clutter case, a general target detection model is established first. Then, based on this model, typical polarimetric covariance matrix parameters at several bands are used to illustrate the procedures of optimizing frequencies and polarizations for small target detection in clutter.

This paper presents results of the investigation of the CLEAN technique used for the generation of 3D images from sparse frequency space data taken on curved SAR aperture paths. Preliminary images resulting from a helicoptor bourne spotlight mode SAR is encouraging. An efficient visualization technique allows improved interpretation of CLEAN images to a degree that a focusing error can be identified. A correlation technique able to accomodate edge effects and possibilities for larger 3D CLEAN images are presented.

We have previously presented the architecture and basic analytic results for a functional 1D pipelined hybrid optical/digital processing concept capable of generating a target range- doppler profile in real time. Here we address the fundamental system processing algorithm and hardware development issues in some detail. The approach to performing real-time phase correction of the individual range profiles is outlined, along with the basic system operational runtime algorithms and system processing pipeline. A description of the receiver hardware and its component functionality in terms of the presented operational theory is given as well.

A 3D formulation of inverse synthetic aperture (ISAR) is presented showing how the various geometry and motion parameters determine the image generation properties. We then show how the arbitrary formatting capability of an opto-electronic processor can be used to format the data in such a way as to focus the image. The focusing parameters are found from tracking prominent points in the radar data itself and using rigid body constraints imposed on the data. The opto-electronic processor is particularly suited for generated image data such as found with ISAR. The processor is a time-integrating architecture that uses acousto-optic scanners for arbitrary formatting and a modified Kosters interferometer for stable Fourier transformation. This research is being funded by the Office of Naval Research.

Essex has developed a hybrid acousto-optic/digital electronic processor, called Hawkeye^TM, that generates high dynamic range, high resolution range-doppler images from wideband radar returns. The processor is described and results of laboratory tests of a partially completed breadboard version are presented. Key capabilities of this processor are: high dynamic range to detect small cross section targets in a severe clutter background; large image sizes of 1024 or greater range bins by up to 128 doppler resolution bins; dynamically adjustable doppler resolution and dynamic reconfigurability of modules to switch between coarse range resolution covering a large range extent for acquisition mode to a fine resolution mode for target discrimination; the ability to accommodate time compression/dilation to eliminate blurring of moving targets in high resolution range-doppler images; and, the ability to efficiently combine multiple low bandwidth, low range resolution radar returns to obtain high range resolution range-doppler images. This is achieved in a compact, lightweight, and cost effective package.

Signal separation or editing of components in the time-frequency domain remains an important problem that is equivalent to time-varying filtering. In this paper, a method to better carry out this process is presented and is based on the use of superresolution in the short-time Fourier transform (STFT) domain. This enhancement of the STFT is accomplished by nonparametric adaptive stationary extrapolations performed on each block of time samples. Superresolution also facilitates the definition of a component's support region in the time-frequency plane before it is edited. Synthesis is performed using the overlap add method that has been previously used to reconstruct a signal from its STFT, or a modified/edited version of it. A heuristic comparison is presented using a test signal using both: conventional and superresolution STFT-domain editing. Applications to radar imaging are also discussed.

A good portion of modern signal processing deals with statistical signal processing and array signal processing, in which models of both the signal and the noise are used to attain precise estimates of the underlying signal parameters. These parameters include frequencies of sinusoids, angles of arrival of wavefronts, scatterer location of radar targets, etc. The explicit use of a signal-noise model and the use of an optimal statistical estimator leads to an accuracy and resolution which is clearly superior to the fast Fourier transform (FFT). At low signal-to- noise ratios, the estimators can be made as robust as the FFT. We present applications of superresolution signal estimators to synthetic aperture radar (SAR) and inverse SAR imaging.

In this paper we propose a newly developed wavelet shrinkage approach that achieves superior performance in clutter suppression by using discrete wavelet transforms. We utilize the relationship of the magnitudes of the wavelet coefficients and the smoothness of the signal function to address the clutter suppression problem with different thresholding methods. We conclude that in heavily cluttered scenario the wavelet shringkage can provide an excellent preformance for clutter suppression.

Small radar detection and tracking systems--in particular, radar guided missile systems--are of great utility because of their all-weather performance and their long range capabilities. A major drawback with these systems results from the relatively simple data that they collect and the difficulty in using these data for target classification and identification purposes. We are investigating a technique which employs the statistics of the tracking data used by many missile seekers and which creates a cross-range target structure map that can be expressed as a function of the target's down-range extent. These data consist of ordinary angle-of-arrival measurements collected from a small burden and may hold the potential to be used for automatic (machine-based) target classification in realistic (time and data limited) environments.

In this paper, autoregressive (AR) modeling is used to characterize the frequency responses of the radar targets. The prominent resonances of the radar target are extracted by modeling the wideband target response and are used for target identification. The advantage of this technique compared to the resonance extraction by the multiple frequency radar is the it is applicable to a variety of targets. No a-priori target resonance information is necessary. Results of this approach on real target data classification in terms of percentage correct classification are presented.

Inverse synthetic aperture radar (ISAR) is an imaging technique that can be utilized in the identification of targets such as ships and aircraft. Since these targets are free to maneuver during the time required to collect their signature, kinematic motion parameter estimates are needed to focus ISAR imagery. In order to perform this estimation, a burst derivative measure, which has global minimum coordinates that provide optimum estimates of the motion parameters, is utilized in conjuction with unconstrained optimization algorithms. It is shown that the burst derivative is a multivariate function with a strong dependence on radar parameters. Results indicate that this dependence can be exploited by the optimization algorithms to obtain efficient motion estimation, thus improving the overall processing speed.

High quality ultrawide-band measurements provide a basis for understanding the transient scattering phenomena necessary for the development of short-pulse radar target identification and detection schemes. This paper describes several techniques used at Michigan State University (MSU) for the acquisition, processing, and interpretation of ultra-wideband scattering data. By performing measurements over a sufficiently large bandwidth, the early- time specular nature of a radar target and the late-time resonant behavior can be observed simultaneously within a single target signature. Special attention has been given at MSU to enhancing the equivalent bandwidth of the measurement system through a spectral slicing and extrapolation method. Observation and interpretation of the various scattering phenomena and their dependance on target aspect are then interpreted through several visualization techniques, including scattering plots, frequency-time plots and images.

Our pattern theoretic approach to the automated understanding of complex scenes brings the traditionally separate endeavors of detection, tracking, and recongition together into a unified jump-diffusion process. Concentrating on an air-to-ground scenario, we postulate data likelihood models for a low-resolution, wide field-of-view millimeter wave radar (for detection) and a high-resolution, narrow field-of-view forward-looking infrared sensor (for recognition). The interaction between the sensors is governed by a jump-diffusion process which provides a mathematical foundation for saccadic detection and computationally efficient target hypothesis during recognition. New objects are detected and object types are recognition through discrete jump moves. Between jumps, the location and orientation of objects are estimated via continuous diffusions. The methodology outlined may be applied to any scenario involving the fusion of low-resolution and high-resolution sensor data.

Previously we presented a jump-diffusion based random sampling algorithm for generating conditional mean estimates of scene representations for the tracking and recongition of maneuvering airborne targets. These representations include target positions and orientations along their trajectories and the target type associated with each trajectory. Taking a Bayesian approach, a posterior measure is defined on the parameter space by combining sensor models with a sophisticated prior based on nonlinear airplane dynamics. The jump-diffusion algorithm constructs a Markov process which visits the elements of the parameter space with frequencies proportional to the posterior probability. It consititutes both the infinitesimal, local search via a sample path continuous diffusion transform and the larger, global steps through discrete jump moves. The jump moves involve the addition and deletion of elements from the scene configuration or changes in the target type assoviated with each target trajectory. One such move results in target detection by the addition of a track seed to the inference set. This provides initial track data for the tracking/recognition algorithm to estimate linear graph structures representing tracks using the other jump moves and the diffusion process, as described in our earlier work. Target detection ideally involves a continuous research over a continuum of the observation space. In this work we conclude that for practical implemenations the search space must be discretized with lattice granularity comparable to sensor resolution, and discuss how fast Fourier transforms are utilized for efficient calcuation of sufficient statistics given our array models. Some results are also presented from our implementation on a networked system including a massively parallel machine architecture and a silicon graphics onyx workstation.

This paper describes a technique that has been developed for autonomously constructing a low- level point model of the surface of an object. Noisy range images of the object are acquired from many different known viewpoints. Data from the range images are resampled onto a regular grid in a reference coordinate system, and then combined to form a single voxel based representation of the surface. A dense set of points lying on the surface is then extracted from the voxel representation. Results using synthetic range data are presented.

The Air Force Phillips Laboratory is in the process of demonstrating an advanced space surveillance capability with a heterodyne laser radar (ladar) system to be used, among other applications, for range-resolved imaging. Recently, image domain signal-to-noise rations (SNRs) have been derived both for the intensity projections calculated from the range-resolved reflective data and for image information obtained using linear combinations of the projections. Also, other recent results have indicated that superior image quality is obtained by first converting the heterodyne returns into intensity projections before using tomographic techniques to reconstruct an image, as intensity projections is validated using a laboratory heterodyne setup. In addition, the laboratory results are used to validate the conclusion that intensity projections provide superior image reconstructions.

Expressions for the detected signal reflected from an object using a coherent laser radar system are developed in this paper. Post-processing methods to obtain intensity projection data from the detected signals are derived. Image domain signal-to-noise ratios (SNRs) are derived both for images reconstructed using a convolution-backprojection algorithm from range- resolved reflective measurements and for the intensity projection data. The two noise sources considered are photon noise and laser speckle noise. It is shown that the SNR of an individual projection at a point is limited above by one, and the upper limit of the SNR of the reconstructed image is on the order of the square root of the number of projections used to create the image.

The way of the general dot tracking is not suitable for near distance tracking and neither is the imaging tracking because of the lower imaging rate in the laser radar. A new tracking method which is called as the staring target edge tracking (STET) is therefore discussed in the paper. The physical model and range equation of the (STET) are presented, and the demands of the STET to the optical system are analyzed. This method provides a new analyzing way for miniature laser radar.

In this paper, we introduce a neural network (NN) architecture that utilizes nonparametric as well as the conventional parametric statistics. Use of the Wilcoxon two-sample test along with the classical model (e.g. Gaussian) parameters provide a qualitative as well as a quantitative representation of the target and the background. On an ordinal scale the radar returns from the target background are ranked according to a specified order and the neural network is trained with a qualitative factor for deviation from the normal distribution. In addition, the actual background distribution also depends on the type of the sensor as well as the wavelength of operation. Accordingly, the independence of the neural network training from the background noise and the clutter distribution provides a unified design approach for the microwave and the laser radar detection systems.

We develop a nonlinear optimal signal processing algorithm for estimating the time delays and amplitudes of target reflections from a set of frequency-stepped continuous wave (FSCW) measurements in the presence of noise. The optimization algorithm solves the nonlinear problem directly in an iterative fashion after prewindowing a range for the estimation of the time delay. The optimal estimate of the corresponding amplitude is solved for each time delay in the range. A performance measure is calculated based on each of the solved amplitude and time delay pairs. The minima in the performance metric correspond to the locations of the reflections. The optimization algorithm is applicable to any set of FSCW measurements from which the target range resolution is to be maximized. The derivation is general in the number of reflections, and a uniform frequency interval between the FSCW measurements is not required. We quantify the effects of noise on the accuracy of estimation through analytic expressions and illustrate through simulation. We demonstrate the performance of this optimization algorithm using synthetic FSCW data in the presence of noise through comparison with the IFFT method. The results show this algorithm is robust in estimating target reflection ranges from noisy FSCW data.

We present a new ladar (laser radar) for the detection of objects off the line-of-sight. This is accomplished by a transceiver and a fiberoptic cable that relays an outgoing laser beam to, and a returning signal from a target. The transmission signal is a laser diode emitted beam at 1550 nm, ideal from the aspects of both eyesafety and minimum loss in a silica fiber. In our immediate application, the detection of an obstacle on the railroad track of a high-speed train, the laser pulses propagate through air and the fiberoptic cable, successively. Under a variety of simulated weather conditions and by traversing twice through a 2 km fiber, we measured a signal-to-noise of 300.

By transmitting and receiving several discrete tones, a tactical Navy radar can image a target via ISAR with little EMI on surrounding radar systems. Standard methods of stepped frequency sequentially transmit single frequency information which has the striking drawback of large amounts of contiguous transmit time. For large bandwidths and maneuvering targets, requiring several hundreds of milliseconds per frequency set could produce range smearing. FM-CW has the drawback that again, (in tactical environment) occupying several hundred MHz of contiguous bandwidth could interfere with other uses of the band. The method of ISAR image formation that will be described here utilizes a transmit waveform comprising a discrete set of tones. This waveform has many advantages over waveforms that require tactically significant amounts of either time or bandwidth. The specific system that will be discussed will be an augmented x-band CW tracking radar. A method of selecting the specific frequencies will be discussed along with a method of pairwise mixing the receive tones, known as Delta-K, that can be utilized for range profile sidelobe reduction. A method of calibrating the phase differences between channels will be discussed in detail. Results from simulations and data from a real system will be discussed.