Recent advances in low-cost design and fabrication enable the potential application of high-accuracy millimeter wave (mmWave) radar sensors to a variety of commercial sectors, including automotive, drones and robotics. The large bandwidth available at mmWave band enable high range resolution, while the small wavelength enhances Doppler resolution. In addition, the small wavelength allows for reduced antenna size that can be used to synthesize large aperture antenna arrays which provide narrow beams for high angular resolution. However, such antennas are expensive in terms of both cost and power consumption. In addition, individual sensors are vulnerable to blockage by larger objects in vicinity of the sensor. Therefore, an array of widely separated radar sensors is used to improve localization accuracy while combating blockage at individual sensors. In this dissertation, we discuss efficient methods for solving the large aperture antenna design and multi-sensor localization problem by exploiting intrinsic geometric properties.
We first consider the problem of designing large effective aperture antenna in 2D for accurate Direction of Arrival estimation. Conventionally, a large effective aperture antenna is constructed by filling the aperture with patch elements spaced at half the carrier wavelength or less. However, such dense array designs do not scale well with increasing aperture area in terms of cost, complexity and power consumption. On the other hand, compact antenna arrays with a moderately large number of elements can be realized at relatively low cost, especially as the carrier frequency increases. We propose a cost-effective synthesis of large apertures (and hence sharp beams) is via sparse placement of a number of such compact arrays, henceforth termed "subarrays", optimizing the placement (and controlling the phases) so as to reduce unwanted grating lobes. We assess the performance of our designs for the fundamental problem of bearing estimation for one or more sources which provides a useful tradeoff comparison of the beamwidth reduction and increase in sidelobe level.
Although mm-wave sensors provide high accuracy measurements, individual sensors are vulnerable to blockage due to the mm-wave propagation characteristics. For safety critical applications a network of sensors is required to avoid detection issues in case some sensors suffer from blockage. We study the fundamental limits on localization accuracy using a network of mmWave radar sensors. We show that super-resolution algorithms can be used to achieve good localization accuracy using low cost mmWave sensors.
Finally, we examine the spatial association of observations collected from multiple sensors in the single snapshot setting. Since the observations collected at each sensor are unordered, they need to be associated with a common target before they can be combined for location estimation. We consider the general data association problem where sensor observation contain range, Doppler information from a single snapshot only. Without any prior association information, this problem has exponential complexity. However, we show that inherent geometric relations between sensor measurements and their locations can be used to drastically reduce this association complexity. Our proposed association framework provides robustness to detection anomalies caused by blockage and achieves significant computational savings when large number of sensor are used.
