FFT Value Generation via Index Parity Segmentation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing methods for generating fast Fourier transform (FFT) values for multi-dimensional input values face challenges in efficiency and scalability, particularly in computer systems with distributed memories, due to calculation dependencies between dimensions.

Innovation Solution

The method involves classifying multi-dimensional input values into sub-input data sets based on even and odd indices in each dimension, performing FFT operations on these sub-sets in parallel using independent pipelines, and then combining the results to generate FFT value sets.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If FFT operations are performed sequentially on multi-dimensional input values using existing methods, then calculation accuracy is maintained, but calculation time increases and productivity decreases

Engineering Contradiction:

ImproveFFT calculation speedVSAvoidCalculation time

Core Design Contradiction:

ProductivityVSLoss of time

Solution Approach 1:

The patent divides multi-dimensional input values into multiple sub-input data sets based on even and odd indices in different dimensions. For example, a 2D input is segmented into four sub-sets: (even, even), (odd, even), (even, odd), and (odd, odd) indices. This segmentation enables parallel processing of independent sub-sets, significantly reducing calculation time while maintaining accuracy.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary classification of input values into sub-input data sets before executing FFT operations. By pre-organizing data according to index parity (even/odd) in each dimension, the system prepares data structures that can be processed independently and in parallel, eliminating sequential dependencies and improving overall calculation efficiency.

Inventive Principle:

Principle #10Preliminary action

2Productivity

If FFT operations are parallelized across distributed memory systems, then productivity improves, but calculation dependencies between dimensions increase device complexity

Engineering Contradiction:

ImproveParallel processing capabilityVSAvoidCalculation dependency management

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The patent segments the calculation process into independent dimensional operations. By classifying data based on index parity in each dimension separately, it creates sub-input data sets that can be processed in parallel without inter-dependencies. This segmentation simplifies the management of calculation dependencies in distributed memory systems.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent transforms the problem by adding a classification dimension based on index parity (even/odd). This additional dimension allows the system to organize data in a way that enables parallel processing across multiple dimensions simultaneously, reducing the complexity of managing calculation dependencies in distributed systems.

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

3Adaptability or versatility

If multi-dimensional input values are processed using distributed memories, then scalability improves, but data communication overhead increases device complexity

Engineering Contradiction:

ImproveScalability in distributed memory systemsVSAvoidData communication structure

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent segments input data into sub-input data sets that can be distributed across multiple memory units. Each sub-input data set contains values with specific index parity combinations, allowing them to be processed independently in distributed memory systems. This segmentation reduces data communication overhead by minimizing the need for data exchange between memory units.

Inventive Principle:

Principle #1Segmentation

4Productivity

If classification of input values into sub-input data sets is performed, then parallel processing capability improves productivity, but data arrangement and communication increase device complexity

Engineering Contradiction:

ImproveParallel FFT processingVSAvoidData arrangement and communication protocol

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The patent performs preliminary arrangement of input values into sub-input data sets based on index parity classification before parallel processing begins. This pre-organization of data ensures that each processing unit receives pre-sorted data, eliminating the need for complex runtime data arrangement and communication protocols during parallel execution.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS20250148046A1Method of generating FFT value and electronic device for performing the method

Publication Date: 2025.05.08 SAMSUNG ELECTRONICS CO LTD

AI summary

A method of performing a fast Fourier transform (FFT) operation for a plurality of input values includes generating a plurality of sub-input data sets based on a plurality of input values, each having an index in a first dimension and an index in a second dimension, calculating a plurality of intermediate values by performing an FFT on each of the plurality of sub-input data sets, calculating partial FFT values for the plurality of input values based on the plurality of intermediate values, and generating a plurality of FFT value sets for the plurality of input values based on the partial FFT values.