Bistatic Automotive Radar Sensing: Signal Processing for Motion Parameter Estimation

Radar sensors are nowadays an integral part of a road vehicle for their ability of detecting targets and extracting information about their motion and location. However, little research has been done on the ability of a vehicle to operate such application when the transmitter is located externally, a setup commonly referred to as bistatic. This thesis incorporates the communication potential of new radio with the merits of bistatic radar to advance the existing automotive sensing technology. It proposes an application envisioned in a smart highway where a vehicle switches to an economic mode, and the long-range radar module relies on cooperative roadside transmitters to locate other road targets. Firstly, a proof of concept is developed using existing Fourier techniques to show the advantage of the bistatic setup over the popular monostatic counterpart. It is proven that a theoretical bistatic range up to twice the monostatic one may be achieved, as well as the ability of correctly locating targets with elevated noise levels (up to 3 dB). A sparse representation of the bistatic automotive radar signal model is then developed and a sparsity-based method for two-dimensional location and Doppler estimation is proposed. After that, an extension of this application to a multistatic scheme is proposed by deploying multiple cooperative roadside sensors, and adopting the group-sparsity concept for parameter estimation. Two methods for data association with varying complexity and performance are proposed with both achieving 100% pairing probability at typical noise levels in the presence of 2 targets. Finally, a rigorous signal model for the wideband case is derived and the associated artefacts are identified. A sparsity-based solution for decoupled motion parameter estimation is proposed which naturally resolves such unwanted artefacts. Extensive computer simulations were conducted to mimic a real automotive scenario and verify the capability of the proposed methods. It was shown through Monte Carlo trials, under different operational settings, that the proposed solutions can outperform the state-of-art bearing an added computational cost. In most cases the improvement in estimation accuracy is at least 1 resolution cell, and can range up to 20 resolution cells in some cases, meanwhile up to 40 folds increase in processing run-time was recorded with the proposed algorithms.