A variable-point FFT operation method based on FPGA with butterfly operation hardware multiplexing
By employing a butterfly operation hardware reuse method on FPGA, the problems of resource waste and slow operation speed in FPGA hardware FFT architecture are solved, realizing variable-point FFT operation with low resource consumption and high real-time performance, which is suitable for FFT calculations with a wide range of points.
Patent Information
- Application Number
- CN202611269002.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-20
- Publication Date
- 2026-11-10
AI Technical Summary
Existing FPGA hardware FFT architectures struggle to balance resource utilization and real-time performance. Fixed-point fully parallel pipelined FFT architectures suffer from severe resource waste, while traditional variable-point serial FFT architectures are slow and have redundant computations, failing to meet real-time processing requirements.
An FPGA-based butterfly operation hardware multiplexing method is adopted. Through series segmentation, time-division multiplexing and operation truncation mechanism, FFT operation in the range of 128 to 262144 points is realized using a 128-point FFT hardware unit. Combined with address compensation and rearrangement, hardware resource utilization and operation speed are optimized.
It achieves efficient FFT operations with low resource consumption, supports a wide range of point counts, meets the real-time processing requirements of radar and high-speed communication, and optimizes the data output order.
Smart Images

Figure CN123068765A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of digital signal processing hardware acceleration technology, specifically to a variable-point FFT operation method based on FPGA-based butterfly operation hardware multiplexing. Background Technology
[0002] Fast Fourier Transform (FFT) is a core algorithm in digital signal processing, widely used in real-time systems such as communication modulation and demodulation, radar echo spectrum analysis, and image signal processing. With the continuous improvement of real-time performance in signal processing, FPGA-based hardware FFT accelerators have become the mainstream implementation solution. However, existing FPGA FFT architectures suffer from two significant drawbacks:
[0003] Fixed-point fully parallel pipelining FFT architecture: The single-point operation latency is extremely low, but the hardware logic resources are consumed in huge quantities. If the design supports fully parallel hardware that supports a maximum of 262,144 points of FFT, it will occupy most of the logic units and storage resources of the FPGA. When the system only needs low-point (128 / 1024 / 8192 points) FFT operations, most of the hardware butterfly units are idle for a long time, the hardware resource utilization rate is less than 10%, and the chip cost and power consumption are seriously wasted.
[0004] Traditional variable-point serial FFT architecture relies on single butterfly units for cyclic calculation, which consumes little hardware resources but involves a large number of iterations and has a large overall processing latency, making it unable to meet the microsecond / millisecond-level real-time processing requirements of radar and high-speed communication systems.
[0005] Existing improved schemes mostly adopt the block operation approach, but they do not design a standardized series splitting and operation truncation mechanism. After segmentation, there are a lot of redundant butterfly operations, resulting in low operation efficiency. In addition, the range of point number adaptation is limited, and it cannot simultaneously cover the full range of 128~262144 for power-2 point FFT calculation.
[0006] In summary, existing technologies struggle to balance FPGA hardware resource consumption with real-time FFT computation speed. The industry urgently needs a variable-point FFT hardware implementation architecture that features low resource consumption, high real-time performance, and a wide point number adaptation range. Summary of the Invention
[0007] To address the technical shortcomings of existing fixed-point FFTs, such as resource waste, slow operation speed of traditional variable-point serial FFTs, and redundant computation in segmented operations, this invention provides a variable-point FFT operation method based on FPGA butterfly operation hardware multiplexing. Using a 128-point FFT hardware unit as a foundation, it achieves high-speed operation of variable-point FFTs across the entire range of 128 to 262144 points with extremely low hardware resource overhead through series segmentation, time-division multiplexing, operation truncation, and address compensation rearrangement mechanisms, balancing hardware resource consumption and real-time processing performance.
[0008] Technical solution
[0009] Option 1: A variable-point FFT operation method based on FPGA-based butterfly operation hardware multiplexing
[0010] This method is deployed inside the FPGA chip and includes the following steps:
[0011] Step S1: Obtain system configuration information
[0012] It receives FFT configuration parameters from an external master controller. The core parameter is the length N of the transform to be processed. N is strictly an integer power of 2, and the value range is limited to 128≤N≤262144, covering the needs of spectrum analysis from small to large scale.
[0013] Step S2: Butterfly-shaped series decomposition
[0014] The total number of butterfly levels is calculated as m = log2N based on the transform length N. A 128-point FFT corresponds to a complete 7-level butterfly operation. Therefore, with 7 levels as the upper limit of the segment, the total m-level operation is divided into k independent sub-operations. The segmentation rule is: the first k-1 segments all perform the complete 7-level butterfly operation, and the last segment performs the remaining butterfly operation with less than 7 levels.
[0015] Step S3: Time-division multiplexing hardware unit segmented operation
[0016] All k segments of computation are performed in a time-sharing manner using the same 128-point FFT basic operation unit:
[0017] (1) First k-1 segment operation: control the hardware unit to execute the complete 7-level butterfly operation, making full use of the hardware parallel processing capability;
[0018] (2) The last sub-operation: Enable hardware truncation control logic, only execute the remaining level butterfly operation, and end the calculation of this segment directly after completion, skip the redundant butterfly operation within 7 levels, and eliminate invalid calculation overhead.
[0019] The 128-point FFT basic operation unit has a built-in 64-way parallel radix-2 butterfly processing unit, which can complete all 64 groups of butterfly operations at the first level in a single clock cycle, maximizing the single-level operation throughput.
[0020] Step S4: Address Compensation and Data Output
[0021] After all segment operations are completed, the cyclic left shift compensation bits corresponding to the last segment operation are calculated. Based on the compensation bits, the storage data addresses are reordered to correct the data disorder problem caused by the segment operation, and the standard N-point FFT transformation results in natural order are output.
[0022] Option 2: A variable-point FFT operation structure based on 128-point hardware multiplexing
[0023] To implement the above calculation methods, it is integrated into an FPGA chip and includes four main functional modules:
[0024] Acquisition module: Connects to the external master controller, parses and caches FFT configuration information, and outputs the transform length N to the splitting module;
[0025] The splitting module has built-in series calculation and segmentation logic. It receives N to solve for the total series m and outputs the series configuration instructions for each segment according to the 7-level segmentation rule.
[0026] The core hardware unit is a 128-point FFT basic operation unit, equipped with segmented control and operation truncation logic; it receives segmented series instructions, and the time-division multiplexed hardware unit performs butterfly calculations segment by segment, with the last segment automatically shielding redundant operations.
[0027] Output module: Built-in address rearrangement logic, calculates the number of bits for cyclic left shift compensation, corrects the address of the out-of-order data after segmented operation, and outputs ordered FFT spectrum data.
[0028] Option 3: Storage Media and Electronic Structure
[0029] Computer-readable storage medium: stores program instructions that implement the operation method of the present invention, which are executed by the processor after being loaded by the processor;
[0030] FPGA electronic structure: It consists of an FPGA chip and off-chip memory. The memory stores the operation control instructions, and the FPGA loads the instructions to complete the hardware-accelerated calculation of variable-point FFT.
[0031] Beneficial effects
[0032] Extremely low hardware resource consumption: Using a 128-point FFT hardware unit as the sole core computing hardware, it achieves a maximum FFT computation of 262,144 points through time-division multiplexing. Compared with a fully parallel pipelined FFT architecture with the same maximum number of points, it saves more than 90% of FPGA logic resources. There is no hardware idleness during low-point operations, and the resource utilization rate is greatly improved. In actual tests on a Xilinx XC7K325T FPGA with a 200MHz main frequency, the maximum 262,144-point FFT only occupies 15% of the chip's hardware resources.
[0033] Strong real-time performance: It integrates 64 parallel butterfly units, completing a full butterfly operation in a single clock cycle; at the same time, it sets up a terminal operation truncation mechanism to eliminate redundant butterfly calculations and reduce the latency of invalid operations; in actual tests at a 200MHz main frequency, the complete operation of 262,144-point FFT only takes 22.45ms, which meets the real-time processing requirements of radar and high-speed communication.
[0034] Wide range of point number adaptation: It uniformly supports FFT of all integer powers of 2 between 128 and 262144, and a single hardware architecture is compatible with small, medium and large scale spectrum analysis scenarios, making it highly versatile.
[0035] Output data regularization: Added circular left shift address compensation logic to automatically correct the data disorder problem caused by segmented operation, eliminating the need for secondary sorting by backend software and reducing the processing overhead of the upper layer of the system.
[0036] Obviously, based on the above description of the present invention, and according to common technical knowledge and conventional methods in the field, various other modifications, substitutions or alterations can be made without departing from the basic technical concept of the present invention.
[0037] The following detailed embodiments further illustrate the above-described content of the present invention. However, this should not be construed as limiting the scope of the present invention to the following examples. All technologies implemented based on the above-described content of the present invention fall within the scope of the present invention. Attached Figure Description
[0038] Figure 1 This is a schematic diagram of the series decomposition of the 32-point FFT of the present invention;
[0039] Figure 2 This is a schematic diagram of the hardware framework structure of the 128-point FFT basic operation unit of the present invention;
[0040] Figure 3 This is a flowchart illustrating the variable-point FFT operation method based on 128-point hardware multiplexing according to the present invention.
[0041] Figure 4 This is an exemplary split-and-compare diagram of the present invention. Detailed Implementation
[0042] The present invention is illustrated below with specific embodiments, which are not intended to limit the scope of the invention.
[0043] like Figure 1-4 As shown, a variable-point FFT operation method based on FPGA butterfly operation hardware multiplexing is presented. Combined with two typical engineering implementation cases, the implementation process of the operation method and hardware device of this invention is fully explained. The implementation cases uniformly use Xilinx XC7K325T FPGA chip, with a stable operating frequency of 200MHz. The hardware-level 128-point FFT basic unit has a built-in 64-way parallel radix-2 butterfly operation unit and a 256-bit wide FIFO cache array, which can complete zero-filling of time-domain data and segmented data interaction. The complete process is shown in the schematic diagram of the scheme.
[0044] Example 1: Implementation of FFT operation with a massive number of 262,144 points (wideband radar echo spectrum analysis scenario)
[0045] Configuration parameter acquisition and verification: The external radar main control processor sends the FFT transform length parameter N=262144. The FPGA internal acquisition module verifies that N is an integer power of 2 and 128≤262144≤262144. If the parameter is valid, the value of N is locked and transmitted to the splitting module.
[0046] Total butterfly series calculation and segmentation: The segmentation module calculates m=log2262144=18, performs segmentation operation with 7 levels as the upper limit of a single segment, 18 / 7=2 remainder 4, the total number of segmentation k=3, and the three segments are assigned levels of 7, 7 and 4 levels respectively; the segmentation module generates a three-segment control instruction sequence, which is sent to the calculation module in sequence;
[0047] Time-division multiplexing segmented butterfly operation:
[0048] (1) First stage of operation: The operation module disables the truncation control enable, fully reuses the 128-point FFT basic unit to perform 7-level standard butterfly operation, and all 64 parallel butterfly units synchronously iterate 7 levels. After the operation is completed, the intermediate data is written back to the on-chip FIFO cache and the read and write pointers are refreshed.
[0049] (2) Second stage operation: Without resetting the hardware unit, the same 128-point FFT kernel is reused directly to perform the 7-level butterfly operation again, and the operation result updates the intermediate data of the cache array;
[0050] (3) Third segment truncation operation: Pull up the truncation control enable, the hardware unit only performs the allocated 4-level butterfly operation, and immediately triggers the truncation signal after completing the 4th level iteration, skips the remaining 3 levels of redundant butterfly operation, terminates the calculation process of this segment, and the cache array stores all 262144 points of frequency domain intermediate data.
[0051] Address compensation rearrangement and data output: The output module reads the remaining number of the last segment r=4, calculates the corresponding cyclic left shift compensation bits, traverses all frequency domain data in the on-chip buffer to complete the address reverse correction, and restores the disordered spectrum data to a natural ascending order; reads the spectrum data sequentially according to the address 0~262143, and transmits it to the radar back-end signal processing unit through the LVDS high-speed interface;
[0052] Actual performance indicators: The total time for the entire 262,144-point FFT segmentation operation and address rearrangement process is 22.45ms; the overall hardware resource utilization rate of the FPGA chip is 15%, and the LUT, DSP multiplier, and BRAM resources are sufficient, which can simultaneously superimpose radar algorithms such as pulse compression and constant false alarm detection.
[0053] Example 2: Implementation of medium-scale FFT operation with 8192 points (5G communication narrowband baseband demodulation scenario)
[0054] Parameter configuration phase: The communication baseband master controller sends out the transformation length N=8192, and after obtaining the module verification parameters and confirming compliance, it outputs them to the split module;
[0055] Series decomposition calculation: m=log28192=13, 13÷7=1 remainder 6, total number of segments k=2, the segment combination is the first segment with 7 levels and the second segment with 6 levels;
[0056] Segmented reuse operations:
[0057] (1) The first complete 7-level butterfly operation fully utilizes the computing power of 64 parallel butterfly units, and the intermediate results are stored in the FIFO cache;
[0058] (2) The second segment enables truncation control, performs only 6 levels of butterfly iteration, terminates the operation directly after completion, and skips the remaining 1 level of meaningless butterfly calculation;
[0059] Address compensation output: The address compensation bits are calculated based on the remaining number of the last segment r=6. The hardware automatically rearranges the cached data addresses and outputs an ordered 8192-point frequency domain spectrum, which can be directly used for communication subcarrier amplitude and phase demodulation calculation.
[0060] Advantages of engineering implementation: Compared with the same number of fully parallel FFT architectures, hardware resource consumption is reduced by 92% and the overall power consumption of the baseband board is reduced by 35%, meeting the miniaturization and low power consumption design requirements of communication equipment.
[0061] Supplementary Implementation Instructions for FFT Operation with a Minimum of 128 Points
[0062] When the main controller is configured with N=128, the total number of butterfly stages m=log2128=7. After splitting, there is only one sub-operation, with segment k=1. No truncation control is required. The hardware unit fully executes all 7 levels of butterfly operations. The address compensation bit is 0, and the cached data is naturally ordered. No additional address rearrangement operation is required. The operation process is the simplest, and the latency is only in the microsecond range, making it suitable for real-time spectrum monitoring of narrowband low-speed signals.
[0063] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or structure that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or structure.
[0064] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural transformations made using the contents of the present invention's specification and drawings under the inventive concept of the present invention, or direct / indirect applications in other related technical fields, are included within the patent protection scope of the present invention.
[0065] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A variable-point FFT operation method based on FPGA-based butterfly operation hardware multiplexing, characterized in that, Includes the following steps: Obtain configuration information, which includes the length N of the FFT transform to be processed, where N is an integer power of 2 and the value range satisfies 128≤N≤262144; The total number of butterfly operation levels m of FFT is calculated based on the transform length N. The m-level butterfly operation is divided into k sub-operations. A single sub-operation can carry a maximum of 7 levels of butterfly operation. The 7-level butterfly operation corresponds to the hardware processing limit of 128-point FFT basic operation unit. All k sub-operations are executed sequentially in a loop, and the butterfly calculation corresponding to each sub-operation is completed by time-division multiplexing the same set of 128-point FFT basic operation units. After all suboperations are completed, the complete N-point FFT transformation result is output.
2. The variable-point FFT operation method based on FPGA-based butterfly operation hardware multiplexing according to claim 1, characterized in that, The rule for splitting an m-level butterfly operation into k sub-operations is as follows: the first k-1 sub-operations are all complete 7-level butterfly operations, and the last sub-operation executes the remaining butterfly operations that are less than 7 levels.
3. The variable-point FFT operation method based on FPGA with butterfly operation hardware multiplexing according to claim 2, characterized in that, For the last suboperation, the operation is truncated for the 128-point FFT basic operation unit. Only the butterfly operation corresponding to the remaining level is executed. After the operation is completed, the calculation process is terminated directly, skipping the subsequent meaningless and redundant butterfly operation steps.
4. The variable-point FFT operation method based on FPGA with butterfly operation hardware multiplexing according to any one of claims 1-3, characterized in that, It also includes an address rearrangement step: calculating the number of cyclic left shift compensation bits required after the last suboperation is completed, reordering the output data addresses based on the compensation bits, and restoring the disordered operation results to natural order data.
5. The variable-point FFT operation method based on FPGA with butterfly operation hardware multiplexing according to any one of claims 1-3, characterized in that, The 128-point FFT basic operation unit has a built-in 64-channel parallel radix-2 butterfly operation processing unit, which can synchronously complete all 64 groups of butterfly operations in a single clock cycle.
6. A variable-point FFT operation structure based on 128-point hardware multiplexing, characterized in that, Deployed inside the FPGA, including: The acquisition module is used to receive configuration information sent by the host computer. The configuration information includes the length N of the FFT transform to be processed, where N is an integer power of 2 and the value range is 128≤N≤262144. The splitting module is signal-connected to the acquisition module and is used to solve the total number of butterfly operation levels m according to the transformation length N, and to divide the m-level operation into k sub-operations, with each sub-operation processing a maximum of 7 levels of butterfly operation; The computation module is connected to the output of the splitting module and has a built-in 128-point FFT basic computation unit. The basic unit is time-division multiplexed to perform butterfly calculations on each segment in turn. The output module is connected to the computation module. After all k sub-operations have been completed, it outputs standard N-point FFT spectrum data.
7. The variable-point FFT operation structure based on 128-point hardware multiplexing according to claim 6, characterized in that, The level splitting logic executed by the splitting module is as follows: the first k-1 segment operations are all complete 7-level butterfly operations, and the last segment operation is executed with the remaining less than 7-level butterfly operations.
8. The variable-point FFT operation structure based on 128-point hardware multiplexing according to claim 7, characterized in that, The computation module has built-in truncation control logic. For sub-operations with fewer than 7 levels at the end, the 128-point FFT basic computation unit is controlled to stop after completing the corresponding level calculation, thus shielding subsequent redundant computation processes.
9. A computer-readable storage medium, characterized in that, The storage medium stores at least one instruction, a program, a code set, or an instruction set. When the instruction, program, code set, or instruction set is loaded and run by the FPGA-compatible processor, it implements the variable-point FFT operation method based on FPGA-based butterfly operation hardware multiplexing as described in any one of claims 1 to 5.
10. An FPGA electronic structure, characterized in that, It includes an FPGA chip and a memory, the memory storing arithmetic instructions that can be loaded and executed by the FPGA processor. When the instructions are executed, they perform the variable-point FFT arithmetic method based on FPGA butterfly arithmetic hardware multiplexing as described in any one of claims 1 to 5.
