Automotive Radar DOA Estimation With Recursive Tree Search

The patent segments the exhaustive search process into a hierarchical tree structure with multiple levels. Each level processes a subset of candidate DOA angles, allowing the system to evaluate promising directions first and prune unlikely branches early. This segmentation transforms the single-stage exhaustive search into a multi-stage progressive refinement process, reducing the number of computations required while maintaining estimation accuracy.

The patent performs preliminary actions by pre-calculating and storing path metrics and pseudo-spectra for candidate nodes before completing the full DOA estimation. These preliminary calculations provide early indicators of promising search directions, enabling the algorithm to prioritize certain paths and avoid computationally expensive evaluations of unlikely candidates. The preliminary path metrics serve as filtering criteria that reduce the search space before detailed analysis.

The patent implements a dynamic search strategy where the algorithm adapts its behavior based on the data being processed. The tree traversal dynamically adjusts which branches to explore based on path metric thresholds and pseudo-spectrum analysis. The algorithm can terminate early when confidence thresholds are met, or continue searching when additional objects are detected. This dynamic adaptation eliminates the need for predetermined object counts while maintaining estimation precision.

The patent incorporates feedback mechanisms where the results of intermediate calculations (path metrics, pseudo-spectra) are used to guide subsequent search decisions. The algorithm continuously evaluates whether current search paths are promising based on accumulated evidence and adjusts its traversal strategy accordingly. This feedback loop allows the system to adapt to the actual number of objects present in the scene without requiring prior knowledge, improving both precision and adaptability.

The patent employs a nested hierarchical tree structure where each node contains sub-nodes representing progressively refined DOA candidates. The recursive search algorithm traverses this nested structure, reusing calculations from parent nodes in child node evaluations. Path metrics and pseudo-spectra computed at higher levels inform and optimize the search at lower levels, creating a nested computation pattern that improves productivity while managing complexity through systematic organization.

The patent implements discarding and recovering by pruning search branches that fall below certain path metric thresholds, discarding computationally expensive evaluations of unlikely candidates. However, the algorithm recovers by systematically backtracking and exploring alternative branches when promising directions are identified later in the search. This selective discarding and recovering strategy optimizes processing efficiency by avoiding redundant computations while ensuring thorough exploration of promising search spaces.