Sharif Digital Repository

To address one of the most challenging industry problems, we develop an enhanced training algorithm for anomaly detection in unlabelled sequential data such as time-series. We propose the outputs of a well-designed system are drawn from an unknown probability distribution, U, in normal conditions. We introduce a probability criterion based on the classical central limit theorem that allows evaluation of the likelihood that a data-point is drawn from U. This enables the labelling of the data on the fly. Non-anomalous data is passed to train a deep Long Short-Term Memory (LSTM) autoencoder that distinguishes anomalies when the reconstruction error exceeds a threshold. To illustrate our... 

Some periodic and quasi-periodic pulse trains are emitted by different sources in the environment and a number of sensors receive them through a single channel simultaneously. We are often interested in separating these pulse trains for source identification at sensors. This identification process is termed as deinterleaving pulse trains. Deinterleaving pulse trains has wide applications in communications, radar systems, neural systems, biomedical engineering, and so on. This paper studies the deinterleaving problem with the assumption that both sources and sensors are fixed. In this study, the problem of deinterleaving pulse trains is modeled as a blind source separation (BSS) problem. To... 

This paper concentrates on the target localization problem in a distributed multiple-input multiple-output radar system using the bistatic range (BR) measurements. By linearizing the BR measurements and considering the relationship between the nuisance parameter and the target position, a constrained weighted least squares (CWLS) problem is formulated, which is an indefinite quadratically constrained quadratic programming problem. Since the constraint is non-convex, it is a nontrivial task to find the global solution. For this purpose, an improved Newton's method is applied to the CWLS problem to estimate the target position. Numerical simulations are included to examine the algorithm's... 

In this paper, we focus on the moving target localization problem in a multiple-input multiple-output radar with widely separated antennas. By exploiting jointly different types of information including time delay, Doppler shift and azimuth and elevation angles of arrival, we develop an algebraic closed-form two-stage weighted least squares solution for the problem. The proposed algorithm is shown analytically to attain the CramerRao lower bound accuracy under the small Gaussian noise assumption. Numerical simulations are included to examine the algorithm's performance and corroborate the theoretical developments  

In this paper, the problem of locating a target in a distributed multiple-input multiple-output radar system using bistatic range measurements is addressed. An algebraic closed-form two-stage weighted least squares solution for the considered problem is developed and analyzed. In the first stage, we establish a set of linear equations by eliminating the nuisance parameters first and then we apply a weighted least squares estimator to determine the target position estimate. In the second stage, in order to improve the localization performance and refine the solution of the first stage, an estimate of the target position estimation error is obtained. The final solution is obtained by...