Machine learning and arrays in modern radars.

This report presents some techniques and results from the OPRA II project. The presented methods investigate the use of machine learning and neural networks in a radar context. These techniques are applied for the cases of sparse reconstruction and target detection. Data collected through compressed sensing methods will have gaps within and be incoherent. To reconstruct full data, sparse reconstruction methods need to be employed. These techniques rely on the use of numerical iterative methods. In this work, we show how one can instead use a neural network for this purpose. This opens up the possibilities for an efficient execution of reconstruction methods using e.g. dedicated graphical processing units. Target detection is an important task for radar systems and becomes complicated in settings with weak targets, multiple closely spaced targets and the presence of clutter. This can make it difficult to sustain a high probability of detection with a low false alarm rate. To improve upon traditional detectors, the detection process is herein proposed implemented via machine learning based techniques. The results show that it is possible to sustain a good detection capability while the neural networks can significantly reduce the false alarm rate. Array processing is another important and integral part of modern radar systems. The general theory behind many of the concepts applied in array processing is well developed. Array antennas with multiple receiver channels provide better directional information than single channel antennas, and array processing exploits this information. However, this places great demands on the antenna hardware solutions and on computational capabilities for the signal processing. This report gives a summary on various simulations and trials carried out during the project related to array processing. The results indicate array processing works well for groundbased radars operating in complex signal environments. Adaptive beamforming provides better direction of arrival estimates and detection capability than conventional beamforming during test measurements with array antennas exposed to interference.