A Master of Science thesis in Electrical Engineering by Mohammad Moufeed Sahnoon entitled, "Target Detection Using Learning Methods," submitted in June 2017. Thesis advisor is Dr. Khaled Assaleh and thesis co-advisors are Dr. Usman Tariq and Dr. Hasan Mir. Soft and hard copy available.
Adaptive beamforming is an array processing method that can be used for target detection. In the absence of clutter signals, this method uses a one-dimensional adaptive filter called the space filter in the spatial dimension using a uniformly linear array as a receiver that is made of N-channels separated by a distance d. The N-receiver channels work on collecting target-free data that can be used as training data for the radar along with collecting the target signal with all types of interferences. The training data are then used to build the covariance matrix that is used in determining the adaptive beamformer filter weights. After that, the received data are projected onto these weights to null the jamming signals, minimize noise, and amplify the target signal. Finally, the output, after projection, is compared with a measured threshold value to decide upon the presence of the target. This conventional method suffers from several problems such as target cancellation when the training data collected are not target free. Furthermore, the amount of secondary data required is usually not available in such applications. Thus, different algorithms must be found or developed to overcome or improve the problems of the conventional method. In this report, a target detection system that involves direction of arrival estimation and learning based algorithms is proposed. The proposed system is assumed to overcome the problem of the jamming signal direction of arrival variations between the training and testing stages, signal-to-interference-plus-noise-ratio variations and the necessity for target free secondary data. Another target detection system is also proposed, i.e. the cascade system. This system uses the adaptive beamforming method as an unsupervised dimensionality reduction technique in line with the learning-based method for target detection, and it shows a comparable performance as compared to the original proposed system.