MIMO Signal Detection With Threshold-Pruned Tree Search

The patent segments the exhaustive MLD search process into staged tree searching with pruning. Instead of evaluating all possible symbol combinations simultaneously, the detector processes antennas sequentially (segmentation by stage) and divides symbol candidates into subsets (segmentation by candidate groups). This segmentation transforms the exponential complexity into manageable polynomial complexity while preserving detection performance through systematic exploration of candidate symbols across multiple stages.

The patent applies partial action by evaluating only a subset of symbol candidates at each stage rather than all possible candidates. The tree searching algorithm with pruning evaluates partial candidate sets (those meeting threshold criteria) and eliminates unnecessary branches early in the process. This partial evaluation approach maintains detection accuracy for promising candidates while avoiding computation on clearly suboptimal paths, thereby reducing overall computational complexity.

The patent incorporates feedback mechanisms where detection results from previous stages inform candidate selection in subsequent stages. The accumulated branch metrics from earlier antenna detections are fed back to guide the pruning decisions and candidate extension in later stages. This feedback loop ensures that promising candidates are tracked across stages while maintaining computational efficiency, thereby preserving detection performance without requiring exhaustive search.

The patent employs dynamic threshold adjustment and adaptive candidate selection based on accumulated metrics. The pruning threshold and number of candidates to extend are dynamically adjusted based on the detection progress and metric accumulation across stages. This dynamic approach allows the algorithm to maintain high detection performance when needed while reducing complexity when confidence is high, optimizing the trade-off between performance and computational load.