We present an anomaly-based drone classification scheme. High dimensional spectrum data is encoded using a convolutional neural network autoencoder. This is trained on data generated from a generic mathematical drone model. Once encoded, we use quadratic discriminant analysis for non-drone classes and define anomalies in terms of the log likelihood and prior knowledge from the drone model. When integrating ten samples, we can discriminate drones from non-drone samples such as birds, with an average accuracy of 98% at 20 dB signal to noise ratio. This corresponds to an effective observation time of 90 ms.

We present two neural network solutions for data driven track before detect applications. The detected tracks may be used to estimate good initial states for traditional trackers such as Kalman filters. We evaluate the method on different scenarios with multiple targets, non-linear trajectories, and different signal to noise ratio (SNR) values. Depending on scenario, the presented method achieves 99% detection probability on Swerling 3 and 4 targets at 5 - 13 dB SNR, with 0.04 - 0.001 false tracks per frame. The presented method is compared to a theoretically optimal detector.

Multipath interference while tracking sea-skimming targets can significantly distort the estimated height of the target. If accounted for however, this interference can be used to obtain more accurate estimates. In this study, we accomplish this with a convolutional neural network (CNN) used as a parameter estimator. The performance of this network is compared with maximum likelihood and least-squares methods. We found that the CNN performs well in comparison to these methods with only a fraction of the computations.

Model selection is an indispensable part of data analysis dealing very frequently with fitting and prediction purposes. In this paper, we tackle the problem of model selection in a general linear regression where the parameter matrix possesses a block-sparse structure, i.e., the non-zero entries occur in clusters or blocks and the number of such non-zero blocks is very small compared to the parameter dimension. Furthermore, a high-dimensional setting is considered where the parameter dimension is quite large compared to the number of available measurements. To perform model selection in this setting, we present an information criterion that is a generalization of the Extended Bayesian Information Criterion-Robust (EBIC-R) and it takes into account both the block structure and the high-dimensionality scenario. We name it Generalized EBIC-R (GEBIC-R). The analytical steps for deriving the GEBIC-R are provided. Simulation results show that the proposed method performs considerably better than the existing state-of-the-art methods and achieves empirical consistency at large sample sizes and/or at high-SNR.

Model selection in linear regression models is a major challenge when dealing with high-dimensional data where the number of available measurements (sample size) is much smaller than the dimension of the parameter space. Traditional methods for model selection such as Akaike information criterion, Bayesian information criterion (BIC), and minimum description length are heavily prone to overfitting in the high-dimensional setting. In this regard, extended BIC (EBIC), which is an extended version of the original BIC, and extended Fisher information criterion (EFIC), which is a combination of EBIC and Fisher information criterion, are consistent estimators of the true model as the number of measurements grows very large. However, EBIC is not consistent in high signal-to-noise-ratio (SNR) scenarios where the sample size is fixed and EFIC is not invariant to data scaling resulting in unstable behaviour. In this article, we propose a new form of the EBIC criterion called EBIC-Robust, which is invariant to data scaling and consistent in both large sample sizes and high-SNR scenarios. Analytical proofs are presented to guarantee its consistency. Simulation results indicate that the performance of EBIC-Robust is quite superior to that of both EBIC and EFIC.

We consider a finite-state Discrete-Time Markov Chain (DTMC) source that can be sampled for detecting the events when the DTMC transits to a new state. Our goal is to study the trade-off between sampling frequency and staleness in detecting the events. We argue that, for the problem at hand, using Age of Information (AoI) for quantifying the staleness of a sample is conservative and therefore, study another freshness metric age penalty, which is defined as the time elapsed since the first transition out of the most recently observed state. We study two optimization problems: minimize average age penalty subject to an average sampling frequency constraint, and minimize average sampling frequency subject to an average age penalty constraint; both are Constrained Markov Decision Problems. We solve them using the Lagrangian MDP approach, where we also provide structural results that reduce the search space. Our numerical results demonstrate that the computed Markov policies not only outperform optimal periodic sampling policies, but also achieve sampling frequencies close to or lower than that of an optimal clairvoyant (non-causal) sampling policy, if a small age penalty is allowed.

Classifiers using convolutional neural networks (CNNs) often yield high accuracies on samples that come from the same distribution as the training data. In this study we evaluate a CNN classifier's ability to discriminate drones from non-drone targets, such as birds, when they are not represented in the training data. We found that the mean accuracy on such out-of-distribution drones was 78%. By introducing a synthetic drone class, generated from a mathematical model, the out-of-distribution drone accuracy was improved to 86%. When trained on all drone types the mean accuracy over all classes was 90%. The data was collected with a 77 GHz mechanically scanning radar with only 9 ms dwell time.

Extended Bayesian information criterion (EBIC) and extended Fisher information criterion (EFIC) are two popular criteria for model selection in sparse high-dimensional linear regression models. However, EBIC is inconsistent in scenarios when the signal-to-noise-ratio (SNR) is high but the sample size is small, and EFIC is not invariant to data scaling, which affects its performance under different signal and noise statistics. In this paper, we present a refined criterion called EBIC R where the ‘R’ stands for robust. EBIC R is an improved version of EBIC and EFIC. It is scale-invariant and a consistent estimator of the true model as the sample size grows large and/or when the SNR tends to infinity. The performance of EBIC R is compared to existing methods such as EBIC, EFIC and multi-beta-test (MBT). Simulation results indicate that the performance of EBIC R in identifying the true model is either at par or superior to that of the other considered methods.

The Bayesian information criterion (BIC) is one of the most well-known criterion used for model order estimation in linear regression models. However, in its popular form, BIC is inconsistent as the noise variance tends to zero given that the sample size is small and fixed. Several modifications of the original BIC have been proposed that takes into account the high-SNR consistency, but it has been recently observed that the performance of the high-SNR forms of BIC highly depends on the scaling of the data. This scaling problem is a byproduct of the data dependent penalty design, which generates irregular penalties when the data is scaled and often leads to greater underfitting or overfitting losses in some scenarios when the noise variance is too small or large. In this paper, we present a new form of the BIC for order selection in linear regression models where the parameter vector dimension is small compared to the sample size. The proposed criterion eliminates the scaling problem and at the same time is consistent for both large sample sizes and high-SNR scenarios.

A deep neural network (DNN) is used for achieving subpulse resolution in non-coherent stepped frequency waveform radar. The trade-off between high resolution and long range in radar systems is often addressed using pulse compression, allowing both long pulses and high resolution by increasing the pulse bandwidth. This typically requires a coherent radar. In this study we present a deep learning based solution for achieving subpulse resolution with a non-coherent radar. Our results for such a system are comparable to an equivalent coherent system for SNRs greater than 10 dB. All results are based on simulated data.