Parallel Sphere Decoding with Sub-Tree Search and Pruning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The complexity of sphere decoding increases exponentially with the number of interfering streams, making it difficult to achieve maximum-likelihood performance for large numbers of interfering streams, and simple parallelization approaches can result in increased processing latency and inefficiency.
Innovation Solution
A parallel-processing method that partitions a tree graph into sub-trees, identifies nodes likely to be part of the solution using metrics of promise, and processes these sub-trees in parallel using multiple processing elements, allowing for efficient tree pruning and solution determination.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If sphere decoding is used to optimally demultiplex mutually interfering signals, then maximum-likelihood performance is achieved, but complexity increases exponentially with the number of interfering streams
Solution Approach 1:
The patent divides the sphere decoding tree into multiple sub-trees that can be processed in parallel. Each processing element handles a specific sub-tree, breaking down the exponentially complex single-tree search into manageable parallel segments. This segmentation maintains maximum-likelihood performance while reducing the complexity burden on individual processing elements.
2Productivity
If traditional processors are used for sphere decoding, then implementation is straightforward, but processing speed has reached a plateau and cannot handle large numbers of interfering streams
Solution Approach 1:
The decoding tree is segmented into parallel sub-trees assigned to different processing elements, enabling simultaneous processing that increases overall productivity. This parallel segmentation allows the system to handle larger numbers of interfering streams by distributing the computational load across multiple elements rather than overloading a single processor.
Solution Approach 2:
The patent performs preliminary identification of metric-of-promise nodes before parallel processing begins. This preliminary action allows processing elements to focus only on promising sub-trees, reducing the actual processing workload and enabling faster convergence to the maximum-likelihood solution without exhaustive search of the entire tree.
3Productivity
If simple parallelization approaches are used, then processing load is distributed, but tree pruning efficiency decreases and processing latency increases
Solution Approach 1:
The patent calculates metric-of-promise values for nodes before parallel processing to identify which sub-trees are most likely to contain the optimal solution. This preliminary ranking allows processing elements to prioritize high-value sub-trees, maintaining pruning efficiency and reducing latency by avoiding exhaustive search of low-probability regions even in parallel mode.
Solution Approach 2:
Different processing elements can apply different pruning strategies or thresholds based on their assigned sub-trees' characteristics. The metric-of-promise calculation allows each processing element to optimize its local search strategy, maintaining overall pruning efficiency while distributing the workload across parallel elements.
Data Source
AI summary
Apparatus and methods for finding a vector solution to a tree search problem are disclosed. In some embodiments, the apparatus and methods can be used for sphere decoding. The tree search is performed by: obtaining a tree graph; identifying a plurality of nodes in the tree graph that are likely to be part of the solution to the tree graph; partitioning the tree graph into a plurality of sub-trees, each sub-tree including one or more of the identified nodes; processing the plurality of sub-trees in parallel by allocating one or more of the processing elements to each of the sub-trees; and determining the solution based on results of the sub-tree processing.