Technical report | A Review of Sparsity-Based Methods for Analysing Radar Returns from Helicopter Rotor Blades
Radar imaging of rotating blade-like objects, such as helicopter rotors, using narrowband radar has lately been of significant interest; these objects cannot be adequately described by the classic point-scatterer model. Recently, a novel tilted-wire' scatterer model has been developed that can provide an accurate and sparse representation of radar returns from such objects. Following a literature review on compressed sensing algorithms, covering both greedy and lp minimisation methods (0 < p ≤ 1), the report focuses on a comparative study of various greedy pursuit algorithms, using both simulated and real radar data, with a particular emphasis on the use of the tilted-wire scatterer model. It is observed that the greedy algorithms that select multiple atoms at the matched-filtering stage do not perform well when the atoms used in the dictionary are significantly correlated. Amongst the greedy algorithms, Orthogonal Matching Pursuit (OMP) exhibits the best performance, closely followed by Conjugate Gradient Pursuit (CGP), which has a much smaller computational complexity than OMP. In applications where the tilted-wire model requires large dictionaries and large CPI atoms, CGP is the preferred option.
Signal analysis and radar imaging of fast-rotating objects such as helicopter rotor blades are of particular research interest because these micro-Doppler signals cannot be processed by conventional range-Doppler techniques and should be separated from other non-rotating scattering components prior to further processing. In addition, the point-scatterer model, which is commonly assumed in the development of various inverse SAR (ISAR) and radar tomography approaches, is not always the most fitting model for the analysis of this type of micro-Doppler signals. The novel tilted-wire model offers new possibilities to overcome the limitations of the point-scatterer model; it however also introduces a new degree of complexity which requires the use of state-of-the-art sparsity-based techniques.
Since the tilted-wire scatterer model can be used to facilitate an accurate sparse representation of signals from rotating blades, a comprehensive review of known algorithms for sparse parameter estimation techniques is carried out, covering both greedy and lp minimisation methods (0 < p ≤ 1). The report focuses on a comparative study of various greedy pursuit algorithms using both simulated and real helicopter radar data, with an aim to accurately estimate the tilted-wire parameters associated with a rotor blade. These parameters are presented as scatter plots which show the orientation, length and tilt of the estimated wires used to represent the rotor blade.
Amongst the greedy algorithms, the so-called Orthogonal Matching Pursuit (OMP) technique exhibits the best performance, closely followed by Conjugate Gradient Pursuit (CGP), which has a much smaller computational complexity than OMP. Important improvements and exploitation of these modern techniques will be published separately in the near future.