Reliable detection and characterisation of dim target via track-before-detect
Detection of manoeuvring and small objects is a challenging task in radar surveillance applications. Small objects in high noise background induce low signal to noise ratio (SNR) reflections. Conventional methods detect such objects by integrating multiple reflections in the same range-bearing and doppler bins in sampled versions of received signals. When the objects manoeuvre, however, these methods are likely to fail to detect them because the integration is performed without taking into account the possibility of the object movements across resolution bins. Furthermore, slowly manoeuvring objects create detection difficulties in discriminating them from radar clutter. Reflections of such objects contain micro-Doppler shifts generated by their propulsion devices. These shifts can characterise specific types of objects. In this case, estimation of these shifts is a challenging task because the front-end signals at the receiver are low SNR reflections and are the superposition of all reflections from the entire object and the noise background. Conventional estimators for this purpose only use reflections collected in a coherent processing interval (CPI) and produce poor estimate outputs. In order to achieve the desired accuracy, one requires more reflections than those collected in a CPI. This thesis mainly considers the aforementioned two difficulties and aims to develop efficient algorithms, which can detect low SNR and manoeuvring objects by incorporating long-time pulse integration and micro-doppler estimation. Main contributions in this thesis are based on the following two algorithms. The first work considers the detection of manoeuvring and small objects with radars. The radar systems are considered both co-located and separated transmitter/receiver pairs, i.e., monostatic and bistatic configurations, respectively, as well as multistatic settings involving both types. The proposed detection algorithm is capable of coherently integrating reflected signals within a CPI in all these configurations and continuing integration for an arbitrarily long time across consecutive CPIs. This approach estimates the complex value of the reflection coefficients for the integration while simultaneously estimating the object trajectory. Compounded with this simultaneous tracking and reflection coefficient estimation is the estimation of the unknown time reference shift of the separated transmitters necessary for coherent processing. The detection is made by using the resulting integration value in a Neyman-Pearson test against a constant false alarm rate threshold. The second work focuses on micro-Doppler signature estimation of manoeuvring and small rotor based unmanned aerial vehicle (UAV) systems with a monostatic radar. The micro-Doppler signature is considered rotation frequencies generated by rotating rotor blades of the UAVs. This estimation uses a maximum likelihood (ML) approach that finds rotation frequencies to maximise a likelihood function conditioned on an object trajectory, complex reflection coefficients, and rotation frequencies. In particular, the proposed algorithm uses an expectation-maximisation (EM) approach such that the expectation of the likelihood mentioned above is approximated by using the state distributions generated from Bayesian recursive filtering for the trajectory estimation. The reflection coefficients and the rotation frequencies are estimated by maximising this approximated expectation. As a result, this algorithm is capable of simultaneously tracking the trajectory and estimating the reflection coefficients and the rotation frequencies of the UAVs before the decision on the object presence is made.