Bidirectional FFT Architecture for Mixed-Order Signal Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing FFT architectures are unidirectional, inefficient, and require multiple instances to perform both forward and inverse FFT operations, leading to high resource usage and low throughput.
Innovation Solution
A bidirectional FFT architecture that can perform both forward and reverse FFT operations using the same structure, allowing for mixed-radix and optimal twiddle memory usage, and capable of processing data in natural and radix-reversed orders.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If unidirectional FFT architecture is used, then the structure is simple, but multiple instances are required for forward and inverse FFT operations leading to high resource usage
Solution Approach 1:
The patent implements a bidirectional FFT architecture where the same hardware structure can perform both forward FFT and inverse IFFT operations by controlling the data flow direction. The architecture includes configurable interfaces that can accept input data in natural order or radix-reversed order, and output results accordingly, allowing one instance to replace multiple unidirectional instances.
2Productivity
If data is reordered to radix-reversed order for FFT processing, then the FFT calculation can be performed, but the reordering process increases processing time and reduces throughput
Solution Approach 1:
The patent inverts the conventional approach by accepting input data in natural order and producing output in the required order, rather than requiring pre-reordered input. The architecture internally handles the reordering through its stage structure and twiddle factor application, eliminating the need for external pre-processing reordering operations and improving overall system throughput.
3Adaptability or versatility
If separate architectures are used for forward FFT and inverse IFFT, then each can be optimized independently, but the overall device size and resource consumption increase
Solution Approach 1:
The patent merges the forward FFT and inverse IFFT functionalities into a single unified architecture. The same computational stages, twiddle factor generators, and data paths are used for both operations, with direction control achieved through configuration of the input/output interfaces and data flow routing, thereby halving the resource requirements compared to having separate dedicated architectures.
4Adaptability or versatility
If decimation-in-time algorithm is used to process natural order input, then the algorithm can work with natural order data, but it cannot compute certain radix values such as radix-2 or mixed radix 4/2
Solution Approach 1:
The patent implements a dynamic architecture that can adapt its operation mode based on the required radix value and data order. The stage structure and twiddle factor generation are configurable to support different radix values (radix-2, radix-4, mixed radix 4/2), and the data flow can be dynamically routed to handle both natural order and radix-reversed order inputs, providing flexibility without sacrificing ease of operation.
Data Source
AI summary
Methods and systems for performing Fast Fourier Transform (FFT) operations are disclosed. A plurality of cascaded stages compute a selectable one of a forward decimation-in-frequency (DFT) FFT operation in a forward direction from an ith stage to a jth stage, and a reverse DFT FFT operation in a reverse direction from the jth stage to the ith stage. The reduction in time and resource usage from employing the disclosed bidirectional DIF FFT architectures may significantly increase performance of DIF applications, for example, ones that involve streaming FFTs.