Junhyeong Bae

Theory and Application of SNR and Mutual Information Matched Illumination Waveforms

A comprehensive theory of matched illumination waveforms for both deterministic and stochastic extended targets is presented. Design of matched waveforms based on maximization of both signal-to-noise ratio (SNR) and mutual information (MI) is considered. In addition the problem of matched waveform design in signal-dependent interference is extensively addressed. New results include SNR-based waveform design for stochastic targets, SNR-based design for a known target in signal-dependent interference, and MI-based design in signal-dependent interference. Finally we relate MI and SNR in the context of waveform design for stochastic targets.

Evaluation of modulus-constrained matched illumination waveforms for target identification

In prior work, we have applied matched illumination strategies to target identification by a closed-loop radar system. In the closed-loop system, multiple waveforms are transmitted in succession, but each is customized based on the returns from prior transmissions. In this prior work, however, the matched waveforms were not constrained to be constant modulus. This current paper evaluates the performance of closed-loop radar with constant-modulus matched illumination. We also compare the performance of non-constant-modulus illumination under a peak power constraint. Finally, we use simple target models and assume unknown orientation, rather than the deterministic or Gaussian target models used in earlier work.

Widely separated MIMO radar with adaptive waveform for target classification

In prior work, we have shown the advantage of adaptive waveforms for monostatic radar target recognition performance. In this paper, we extend our approach to the widely separated multi-input multi-output (MIMO) radar scenario. MIMO radar exploits a diversity of radar waveforms and target scattering to improve radar performance. We present an iterative information-based waveform design method that results in waveforms having narrow, disjoint bands so as not to interfere with each other. Applying a constant modulus constraint causes spectral spreading, but we study the impact of this spreading and find that the performance loss is minimal.

Automatic target recognition with unknown orientation and adaptive waveforms

In previous work, we have demonstrated the utility of a feedback loop for enabling optimized transmit pulse shaping in radar target recognition. This previous work was based on low-fidelity target models, but in this paper, we demonstrate the closed-loop, adaptive-waveform approach applied to high-fidelity target model signatures generated by commercial electromagnetic FDTD software. We also incorporate the radar equation into our models for us in the waveform design procedure. Because SNR varies with range, so do our optimized waveforms for target recognition. Constant-modulus waveform constraints are enforced, and a template-based classification strategy is used.

Adaptive PRF selection technique for multiple targets in track-before-detect

We present an adaptive pulse repetition frequency (PRF) selection technique to mitigate ambiguity, blind-zone, and clutter-zone issues in a particle-based track-before-detect (TkBD) method. A feedback loop provides information about TkBD states to the transmitter in the form of possible target locations and probabilities. The adaptive PRF selection technique is then based on a predicted entropy computation derived from the TkBD target existence probabilities. For each potential PRF, the possible target states fall on different locations in the range-Doppler map, including clutter ridges and blind zones. Thus, the information that can be expected in the next dwell varies with PRF, and the PRF that minimizes the total expected posterior entropy is selected for the next processing interval. While the TkBD enables detection of low-SNR targets that traditional radar will might miss, the adaptive PRF selection helps to minimize track loss before detections are declared. In this context, we compare the performance of adaptive PRF selection to fixed, cyclical PRF selection.

Pose estimation with non-uniform angular sectors and adaptive waveforms

In previous work, we have shown the benefit of forming feedback loop between radar transmitter and radar receiver for target recognition in a widely separated multi-input multi-output (MIMO) radar scenario as well as a monostatic radar scenario. In this work, we extend our techniques to target pose estimation for monostatic radar scenario using high-resolution range profile (HRRP). Angular sectors are required to reduce the complexity of pose estimation techniques and we apply fuzzy K-means clustering algorithm to form angular sectors. The fuzzy K-means clustering algorithm seeks for similarity between target signatures and generates non-uniform angular sectors. We compare the performance of non-uniform angular sectors to uniform angular sectors in target pose estimation.

Target recognition with high-fidelity target signatures and adaptive waveforms in MIMO radar

In previous work, we have demonstrated the advantage of a closed-loop radar system that provides a feedback loop from radar receiver to transmitter in a widely separated multi-input multi-output (MIMO) radar scenario. The results for widely separated MIMO radar showed performance benefit when exploiting spatial diversity with adaptive waveforms. However, the bistatic target signatures used in prior work were simply assumed to be impulse functions with complex-Gaussian reflectivity, which models a wide bistatic angle having poor range resolution. In this work, we use high-fidelity bistatic target signatures obtained by commercial electromagnetic FDTD software (XFdtd), and apply to the widely separated MIMO radar scenario. We optimize the waveform based on both the bistatic and monostatic target signatures and compare the performance to non-optimized (wideband) and partially optimized (optimized for monostatic target signatures only) waveforms. We also compare to results that ignore either the bistatic path (exploit monostatic only) or the monostatic path (exploit bistatic only). Constant-modulus constraints are enforced on the adaptive waveforms, and template-based classification is adopted for final decision.

Waveform Design for Target Class Discrimination with Closed-Loop Radar

The basic concepts of cognitive radar and considered the Bayesian framework as an engine for its operation has been summarized in this chapter. Clearly, two critical functions of CR are a probability update function that quantifies the systems understanding and an ability to respond by adapting subsequent transmit waveforms. We applied two waveform strategies to the problem of discriminating an unknown target realization in additive Gaussian noise. The waveforms are based on the SNR and MI criteria. The potential benefits of closed-loop operation are apparent, and we have proposed a diversity-based explanation for the observed performance.