Thesis Examination Committee
Prof Weijia WEN, PHYS/HKUST (Chairperson)
Prof Daniel PALOMAR, ECE/HKUST (Thesis Supervisor)
Prof Hing Cheung SO, Department of Electronic Engineering, City University of Hong Kong (External Examiner)
Prof Jun ZHANG, ECE/HKUST
Prof Wai Ho MOW, ECE/HKUST
Prof Jianfeng CAI, MATH/HKUST
Waveform design plays an important role in radar. By exploiting the waveform diversity, the performance of the radar systems can be improved significantly. The waveform design should have some nice properties so that it will be compatible with the hardware configuration and perform well in the application scenarios. Second, the design algorithm should be computationally efficient for the real-time radar applications. The focus of this dissertation is on the development of efficient optimization methods for radar waveform design.
Among many design metrics, the most important one is SINR, which decides the probability of detection. The existing methods for SINR maximization are time-consuming, and in some cases, without guarantee of monotonicity or further convergence. To deal with these issues, we propose optimization methods based on the majorization-minimization (MM) framework for the joint design of transmit waveforms and receive filter for multiple constraints in the context of MIMO radar. It can achieve the same or even better SINR than the benchmarks with much less computational cost.
Due to the ever-growing demand of spectrum resources from multiple RF services, spectral sharing is becoming a solution to tackle the issue, which makes spectral shape an important property of the designed waveform. Specifically, the transmit waveform should avoid certain frequency bands or try to minimize the spectral power on those bands. Recently proposed spectral level ratio (SLR) is interesting and simple compared with other existing approaches. However, the SLR minimization problem is very hard to solve because it is fractional, nonconvex and nondifferentiable. Thus, our goal is to develop an optimization method for the problem. The algorithm is designed based on the combination MM and the Dinkelbach’s algorithm, which achieves better performance than the benchmark.