Multi-Stream Cross Correlation for Low-Latency Spike Delay Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing cross correlation methods suffer from high latency and require buffering, which is inefficient for processing streaming data with significant stochastic content, particularly in applications like spiking neural networks.
Innovation Solution
The method employs a Cross Correlation Unit (CCU) with independent operation of each unit, utilizing Learn Delay and Produce Delay blocks that operate based on stochastic spike streams, allowing for lossy memory usage and conflict resolution techniques to minimize latency, using exponential decay curves and threshold detection for delay adjustments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If established cross correlation methods (FFT, sliding window) are used, then measurement precision is improved, but latency increases and buffering is required
Solution Approach 1:
The patent divides the cross correlation computation into multiple independent parallel CCUs, each handling a specific delay offset. This segmentation eliminates the need for sequential processing and memory buffering, allowing simultaneous computation of multiple correlation values without latency accumulation.
Solution Approach 2:
The system pre-computes and stores correlation values for positive delay offsets in advance. When a negative delay offset is needed, the pre-computed positive offset values are directly accessed and reused, eliminating the need for additional buffering and reducing latency for bidirectional search.
2Adaptability or versatility
If bidirectional search with memory buffer is implemented, then cross correlation completeness is improved, but device complexity increases
Solution Approach 1:
The patent merges the handling of positive and negative delay offsets by using a single set of CCUs that can serve both directions. Pre-computed correlation values for positive offsets are reused for negative offset queries, eliminating the need for separate buffering structures and reducing overall system complexity.
Solution Approach 2:
The system creates and stores copies of correlation computation results for positive delay offsets. These copied values are then reused when negative delay offsets are required, avoiding the need for complex bidirectional buffering while maintaining complete search capability.
3Measurement precision
If multiple comparisons are performed, then measurement precision is improved, but productivity decreases
Solution Approach 1:
The patent segments the multiple comparison operations into parallel independent CCUs, each computing a specific correlation value simultaneously. This parallelization maintains the precision benefits of multiple comparisons while dramatically improving processing throughput and eliminating sequential bottlenecks.
Solution Approach 2:
Correlation values are pre-computed and stored for positive delay offsets before they are needed for negative offset queries. This preliminary computation allows rapid retrieval and comparison operations, maintaining precision while improving overall processing productivity.
Data Source
AI summary
A multi-stream cross correlator for spiking neural networks, where each stream contains significant stochastic content. At least one event occurs, with a fixed temporal relationship across at least two streams. Each stream is treated as a Frame Of Reference (FOR), and subject to an adjustable delay based on comparison to the Other streams. For each spike of the FOR, a timing analysis, relative to the last and current FOR spikes, is completed by comparing Post and Pre accumulators. Also, a new timing analysis is begun, with the current FOR spike, by restarting the production of Post and Pre weighting functions, the values of which are accumulated, upon the occurrence of each Other spike, until a next FOR spike. A one-spike delay unit can be used, if time-neutral conflict resolution is used. The average spike rate of the FOR can be determined and used for the Post and Pre weighting functions.