CPU Load Distribution in Block-Based Computation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for distributing computing load in data processing devices during block-based computing operations, such as audio and video signal processing, lead to uneven utilization of processor resources and significant delays, making real-time processing challenging due to the separation of input, output, and calculation phases.
Innovation Solution
The method subdivides the calculation rule into multiple calculation steps processed sequentially within a block, ensuring each step generates a consistent computing load, allowing for even distribution of processing capacity across time windows, thereby eliminating the separation of input and output phases and optimizing CPU utilization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If fast convolution with FFT is used to reduce calculation effort, then computational efficiency is improved, but real-time processing capability deteriorates because the entire audio signal must be available before transformation can begin
Solution Approach 1:
The patent divides the impulse response into multiple blocks and processes convolution in sections rather than requiring the entire signal. This segmentation allows incremental processing where each block can be transformed and combined progressively, enabling real-time output generation without waiting for complete signal availability.
Solution Approach 2:
The patent performs preliminary FFT transformation on impulse response blocks in advance and stores the transformed data. When processing is needed, the pre-transformed blocks are readily available for multiplication and combination, eliminating the need to wait for complete signal availability before starting transformation operations.
2Loss of time
If section convolution is used to enable real-time processing, then real-time capability is improved, but computing load distribution becomes uneven with alternating input/output phases and calculation phases
Solution Approach 1:
The patent segments both the impulse response into blocks and the FFT operation into multiple stages. This dual segmentation allows the calculation work to be distributed across multiple processing steps rather than concentrated in single phases, creating a more uniform computing load distribution over time.
Solution Approach 2:
The patent implements dynamic scheduling where multiple blocks are processed in an interleaved manner rather than sequentially completing one block entirely before starting the next. This dynamic approach overlaps input, transformation, and output operations across different blocks, smoothing out the computing load distribution.
3Productivity
If asynchronous section convolution is used to separate input/output phases from calculation phases, then load distribution is improved, but synchronization overhead increases and real-time capability deteriorates
Solution Approach 1:
The patent maintains continuous processing by ensuring that while one block is being transformed, another block's results are being multiplied and combined, and a third block is being prepared for transformation. This continuous overlapping of operations eliminates idle synchronization periods while maintaining proper data dependencies.
Solution Approach 2:
The patent uses dynamic block selection and processing scheduling that adapts to current system state and data availability. This dynamic approach allows flexible coordination between different processing stages without rigid synchronization points, reducing synchronization overhead while maintaining real-time capability.
Data Source
Figure 1~2
Figure 3~4
Figure 5~6
AI summary
The method involves structuring the input values in block operation of the algorithm in two computational steps (20). The computational steps are processed one after the other within a block. The computational steps generate the same computational load. The computational steps are processed in the time interval of the input values. The block needs a blockwise predetermined number of input values running in a chronological sequence. Independent claims are also included for the following: (1) an arrangement for executing the method; (2) a computer program for executing the method; and (3) a computer-readable storage medium for executing the method.