Micro-Doppler models of drones, birds, and bird-like dronesNucum, Kester; Biswas, Sabyasachi; Ball, John E.
doi: 10.1117/12.3052657pmid: N/A
Counter-drone systems can confuse drones, birds, and bird-like drones. Their radar micro-Doppler (-D) effects can be used to differentiate these targets from each other. Due to the lack of availability of -D signature datasets, -D return models from quadcopters with spinning rotors, birds with flapping wings, and bird-like drones with flapping wings were implemented for classification purposes. To showcase their use, these models were used to generate a synthetic dataset of spectrograms for binary classification (Drone vs. Bird) and ternary classification (Quadcopter vs. Bird-Like Drone vs. Bird). The classifiers examined were support vector machine (SVM), k-nearest neighbors (KNN), Naïve Bayes, and a convolutional neural network (CNN). All binary classifiers produced at least 92.0% accuracy and at least 93.9% F1 score. All ternary classifiers produced at least 89.5% accuracy and 89.4% F1 score. The classification results indicate the -D models possess suitable fidelity for further research and development in drone vs. bird classification.
Impact of positioning errors in radar using distributed repeatersNusrat, Tasin; Sharma, Shivani; Vakalis, Stavros
doi: 10.1117/12.3053649pmid: N/A
Imaging at the millimeter-wave bands of the electromagnetic spectrum is prone to specular reflections. This means that targets reflect incident electromagnetic signals in specific directions, depending on the angle of incidence and target orientation. These mirror-like reflections can make radar detection very challenging, as transmitted radar energy is not necessarily directed back to the radar receiver. An approach for countering the specular responses from challenging targets by utilizing distributed millimeter-wave repeater apertures has recently been introduced. A distributed repeater can receive and retransmit the specular reflections to enhance target detection. In this paper, we review the theoretical basis of our work and analyze the impact of positioning errors in radar systems using distributed repeaters.
Discriminant analysis of radar micro-Doppler signatures for musculoskeletal injury risk assessmentFerdous, Jannatul; Ahmad, Fauzia; Onks, Cayce
doi: 10.1117/12.3055054pmid: N/A
Musculoskeletal injuries (MSKI) constitute a significant portion of all injury-related healthcare visits. Current screening methods lack the sensitivity and scalability needed to identify who is at risk so that valuable preventive and treatment resources can be directed to the most at-risk population. In this work, we investigate the feasibility of micro-Doppler (MD) radar as a cost-effective and privacy-preserving screening modality to identify those at high risk for MSKI where injury risk is previously unknown. We collected radar MD signatures of student athletes performing a drop jump at preseason screenings. This specific movement has been shown previously to potentially indicate elevated risk for MSKI when performed in an impaired manner. The injury status of the athletes was then tracked through the season to document all MSKI resulting in time loss from athletic participation. This information was used to generate labels for the preseason radar measurements, thereby enabling discriminant-analysis-based supervised classification in a retrospective manner. Specifically, we consider linear discriminant analysis (LDA) and regularized uncorrelated multilinear discriminant analysis (R-UMLDA) for feature extraction. The results demonstrate the radar’s ability to detect subtle differences in the MD signatures to identify individuals at high risk for MSKI, with R-UMLDA significantly outperforming LDA in terms of prediction accuracy, sensitivity, and specificity.
Addressing privacy and cost challenges in remote patient monitoring with streamlined 60GHz radar and edge processingChetty, Kevin; Brennan, Paul; Tang, Chong; Lok, Lai Bun; Shi, Fangzhan
doi: 10.1117/12.3052476pmid: N/A
Remote patient monitoring is critical in elderly care, particularly for detecting incidents like falls and abnormal gait, but current systems face high deployment costs, privacy concerns, and the complexities of continuous data processing. To overcome these challenges, we developed a 60GHz millimetre-wave radar system that processes data on-device, eliminating the need for constant internet access and mitigating privacy risks. This system is optimized for real-time healthcare monitoring in residential and clinical environments. Data captured by the radar are processed on an ATmega328 microcontroller using a quantized Long Short-Term Memory model. The quantization ensures efficient operation under tight resource constraints, enabling accurate classification of movement patterns. The system achieves an inference latency of approximately 300s, suitable for real-time response. A key innovation is our global confidence mechanism, which reduces false alarms by aggregating predictions over multiple detection frames. This significantly improves reliability in detecting critical events like falls, reducing false positives. The system was tested on five distinct activities: falls, normal walking, irregular walking, standing, and painful standing, achieving over 91% accuracy. Compared to conventional solutions, it provides a cost-effective, privacy-preserving alternative suitable for scalable deployment across healthcare settings. By leveraging on-device machine learning, our approach reduces computational demands and enhances real-time performance without relying on cloud-based processing.
Automatic classification of radar and communication waveforms through interpretable deep learningHicks, Bruce; Biswas, Sabyasachi; Ball, John E.; Gurbuz, Ali C.
doi: 10.1117/12.3054111pmid: N/A
In this paper, we introduce an automatic method for classifying radar and communication waveforms using complex-valued convolutional neural networks (CV-CNNs) with parameterized, learnable sinc filters. Unlike traditional approaches that depend on computationally expensive preprocessing and transformations, our system processes complex-valued raw IQ RF data directly. By integrating learnable bandpass-like sinc filters, we efficiently extract time-frequency features while reducing the number of trainable parameters, thereby improving both interpretability and robustness in dynamic RF environments. We evaluated the proposed model on a synthetic dataset of 18 types of waveform modulation with signal-to-noise ratios (SNRs) ranging from -20 dB to 20 dB. The results show that our method outperforms conventional models in all key metrics, achieving an average accuracy of 76.47%. Moreover, it demonstrates improved noise robustness and effectively discriminates among a wide variety of modulation classes, even under challenging conditions.
Millimeter-wave stepped frequency radar for high-accuracy water level monitoringSharma, Shivani; Nusrat, Tasin; Yazgan, Mehmet; Vakalis, Stavros
doi: 10.1117/12.3053647pmid: N/A
Radar monitoring of water bodies is attracting significant research interest over contact-based sensors because of its simple installation and corrosion resistance. However, the current water monitoring radar systems have limited accuracy. In this paper, the benefits of stepped-frequency continuous-wave (SFCW) radar for water monitoring are discussed and an approach to get high-accuracy sensing is presented. To verify our method, a 36-40 GHz stepped-frequency radar is built and experimental water-displacement measurements are performed. At a water level of 21 cm from the radar, a measurement standard deviation equal to 650 m is achieved.
Passive radar-based target localization through self-mixing processing and binary search minimizationHenry, Justin K. A.; Narayanan, Ram M.
doi: 10.1117/12.3053593pmid: N/A
Passive radar is a specialized form of the bistatic radar configuration in which a sensing system consists solely of appropriately designed receivers to take advantage of transmitters already operating in the target environment. These transmitters are termed illuminators of opportunity. Bistatic radar ranging requires knowledge of the time difference of arrival between the direct and target-reflected signals, the baseline length, and the direction of arrival. Furthermore, current localization methods involve matrix operations, adaptive thresholding, and triangulation. Generally, elliptical contours of constant range are formed for each range estimate, and the intersection of these ovals represents a localized solution. However, many solutions may exist – therefore introducing ghost targets. This paper presents a concise method of localization depending on the time difference of arrival resolved through self-mixing processing and range estimation via binary search minimization. This method allows for efficient target range determination without the occurrence of ghost targets. The mathematical derivation and scenario simulations are explored in this work and demonstrate the feasibility of this technique in space domain awareness applications.
Assessing permittivity dependence on inhomogeneities in materialsEllenwood, Carson; Weatherall, James C.; Barber, Jeffrey
doi: 10.1117/12.3053072pmid: N/A
Permittivity measurements are performed with the two-port filled WR-75 rectangular waveguide technique (frequency 10 - 15 GHz). Samples are 3D-printed rectangular prisms of 100% infill Polylactic acid (PLA) with cavities of differing size, position, geometry, and directional orientation to test a variety of inhomogeneities. S-parameter data are collected and fit to curves from theory with four free parameters: real permittivity, imaginary permittivity, and the two displacements of the sample from the calibration planes. The optimized complex permittivity is taken as the measure. Measured values for permittivity of solid PLA are in good agreement with literature. The measured real permittivity for samples is consistent with the Landau-Lifshitz/Looyenga (LLL) mixing formula, although there is a weak variance with void geometry. However, results of the S-parameter fit indicate that the losses within the inhomogeneous samples are sensitive to geometry and directional orientation.
Front Matter: Volume 13471doi: 10.1117/12.3073178pmid: N/A
This PDF file contains the front matter associated with SPIE Proceedings Volume 13471, including the Title Page, Copyright information, Table of Contents, and Conference Committee information.
A comparison of nearest neighbor pixel deconvolution and spatial variant apodization for image enhancement in synthetic aperture radar imageryCain, Jonathan A.; Wang, (Sam) Yu; Medl, Thomas; Kano, Patrick O.; Israeli, Yeshayahu
doi: 10.1117/12.3052275pmid: N/A
Image enhancement is a well-studied problem for Synthetic Aperture Radar (SAR) imagery, drawing on a rich history from RADAR and EO/IR signal processing. This paper introduces a novel EO/IR approach, Nearest Neighbor Pixel Deconvolution (NNPD), to SAR images. The NNPD method rewrites the Point Spread Function (PSF) in terms of the nearest neighbor pixel groups, applies the Shift Theorem of the Fourier Transform to modify the PSF in the Fourier domain, and achieves a higher spatial resolution with the inverse Fourier Transform. We compare NNPD to a classic SAR algorithm, Spatial Variant Apodization (SVA), a technique specifically for SAR images that suppresses sidelobes while maintaining mainlobe resolution. The SVA relies on a wavenumber domain weighting function, convolved in the image domain, and then choosing the weights such that they minimize the amplitude of each of the pixels. NNPD can enhance EO image resolution up to and beyond the diffraction limit for single frame images with tradeoffs for noise and ringing. Similarly, the Generalized and Robust versions of SVA, which can use both in-band or quadrature data, together or separately, also demonstrate similar tradeoffs with respect to the Nyquist sampling rate. Beyond the analytical comparison of NNPD and SVA, this paper provides quantitative examination over sample SAR imagery with figures of merit such image entropy, image contrast, signal-to-noise ratio, and peak-to-sidelobe ratio, among others.