Vector Cumulative Sum Circuit for High-Throughput Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vector processors face throughput issues when computing moving average filters due to the length of input samples, leading to decreased efficiency in high-throughput signal processing operations.
Innovation Solution
A vector cumulative sum circuit and associated instruction are introduced, which enable concurrent input and output operations across multiple registers, allowing for the computation of cumulative sums and facilitating efficient implementation of filtering operations by reducing the number of cycles required.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional vector processors are used to compute moving average filters, then the processing can be performed with standard architecture, but the throughput decreases due to the length of input samples requiring sequential processing
Solution Approach 1:
The filter computation is divided into two independent stages: (1) computing differences between input samples at相距 positions, and (2) computing cumulative sums of these differences. This segmentation allows parallel execution of multiple difference computations simultaneously, followed by parallel cumulative sum computations, thereby increasing throughput independent of filter length
Solution Approach 2:
The difference values between input samples are pre-computed and stored before the cumulative sum operation. This preliminary action separates the computation into independent phases where difference computation results can be reused across multiple cumulative sum calculations, reducing redundant operations and improving overall processing efficiency
2Quantity of substance
If the filter size increases, then more input samples can be processed, but the number of cycles required increases proportionally
Solution Approach 1:
By segmenting the filter computation into difference computation and cumulative sum phases, the architecture can process multiple input samples in parallel during the difference phase, then process the cumulative sums in parallel during the second phase. This allows the number of cycles to remain constant regardless of the number of input samples processed
Solution Approach 2:
The cumulative sum circuit continuously accumulates the difference values across all input samples without interruption. The carry-forward mechanism ensures that each cumulative sum computation builds on the previous one in a continuous pipeline, maximizing resource utilization and maintaining constant throughput independent of filter size
Data Source
AI summary
A vector cumulative sum circuit can include a set of input registers, a carry-forward data source, a set of output registers, and a network of adder circuits coupling the input registers to the output registers such that the output value in a given output register is the sum of a value provided by the carry-forward data source and the input values from all of the input registers (in logical order) up to (and including) the corresponding input register. The value in the last output register can be carried forward to enable cumulative summing of a larger number of input values. The vector cumulative sum circuit can be implemented in a programmable processor, and a vector cumulative sum instruction can be defined in the instruction set. Using the vector cumulative sum circuit and instruction, filtering operations can be accelerated.