A sharing method for cascaded CSD complex multipliers of a multi-path FFT processor
By sharing the CSD complex multiplier in a multipath FFT processor, splitting the twitch factor coefficients into two sub-stages and staggering their calculations, the problem of excessive hardware resources and power consumption is solved, achieving optimization of hardware cost and energy consumption, and making it suitable for 6G wireless communication systems.
Patent Information
- Application Number
- CN202611304854.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-26
- Publication Date
- 2026-11-17
AI Technical Summary
In traditional multi-path FFT processors, the increase in the number of paths leads to a significant increase in hardware resource consumption, chip power consumption, and hardware cost, becoming a bottleneck restricting the engineering implementation of long-point, high-real-time FFT processors.
A serial CSD complex multiplier sharing method for multi-path FFT processors is adopted. By splitting the coefficients of the twitch factor into two sub-stages and sharing the CSD complex multiplier between adjacent paths, the calculation of the same twitch factor coefficient group is staggered, thereby reducing hardware resource consumption.
It significantly reduces the hardware overhead and power consumption of the FFT processor, adapting to the stringent constraints of power consumption and hardware resources in 6G wireless communication systems.
Smart Images

Figure CN123114411A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of digital integrated circuit design technology, and in particular relates to a sharing method for a serialized CSD complex multiplier for a multi-path FFT processor. Background Technology
[0002] The FFT processor is a core building block in the field of digital signal processing, and it has been widely used in various engineering scenarios such as spectrum analysis, image and speech signal processing, biomedical detection, radar detection, digital filtering, and wired and wireless communication. In 6G wireless communication systems, the FFT processor, as the core data processing module of the physical layer, directly determines the transmission efficiency and cost-effectiveness of the entire communication system in terms of its computing performance and resource consumption.
[0003] For future communication applications driven by 6G high speed and ultra-wideband, long-point FFT processors have become a core necessity for physical layer signal processing, leading to a significant increase in the scale of multi-coefficient twiddle factor operations. 6G communication systems place exponential demands on the real-time processing speed and bandwidth of FFT operations, and multi-path parallel architecture is the mainstream and efficient technical solution for improving the real-time processing capabilities of FFT processors. However, in traditional multi-path architectures, each path is independently configured with a dedicated cascaded CSD complex multiplier. The increase in the number of paths directly leads to a significant increase in hardware resource overhead, chip power consumption, and hardware cost, becoming a key bottleneck restricting the engineering implementation of long-point, high-real-time FFT processors. Summary of the Invention
[0004] In view of this, the purpose of this application is to provide a sharing method for serial CSD complex multipliers for multi-path FFT processors, which can significantly reduce the hardware overhead and power consumption of FFT processors and adapt to various application scenarios in 6G wireless communication systems with strict constraints on power consumption and hardware resource overhead.
[0005] This application provides a shared method for serialized CSD complex multipliers for multipath FFT processors. The method is applied to complex multiplication stages with 128, 256, and 512 twitch factors. The method includes: The coefficients of the rotation factor of the corresponding specification to be calculated are split according to the corresponding decomposition rules; Based on the specification of this rotation factor, the rotation factor of a specific coefficient is divided into two sub-stages; The first sub-stage complex multiplication operation is performed by serially connecting CSD complex multipliers that do not overlap between adjacent paths. The feedback registers of this operation stage are split on the two adjacent paths respectively, so as to avoid the calculation of the same twitch factor coefficient group that needs to be performed simultaneously on the two paths, thereby achieving the purpose of sharing the serially connected CSD complex multipliers. The second sub-stage complex multiplication operation is performed by deploying CSD complex multipliers on their respective paths; Based on the symmetry of the twitch factor with non-specific coefficients, the results of complex multiplication are derived.
[0006] Furthermore, the decomposition rule for the 128 sets of rotation factor coefficients is as follows: ; The decomposition rules for the 256 sets of twitch factor coefficients are as follows: ; The decomposition rule for the 512 sets of twitch factor coefficients is as follows: .
[0007] Furthermore, the specification based on this rotation factor divides the rotation factor of a specific coefficient into two sub-stages, including: Rotation factors with coefficients of 4, 8, 12, and 16 Divided into the first sub-stage, the rotation factors with coefficients of 1, 2, and 3 are... Divided into the second sub-stage; Alternatively, use rotation factors with coefficients of 4, 8, 12, 16, 20, 24, 28, and 32. Divided into the first sub-stage, the rotation factors with coefficients of 1, 2, and 3 are... Divided into the second sub-stage; Alternatively, use rotation factors with coefficients of 8, 16, 24, 32, 40, 48, 56, and 64. Divided into the first sub-stage, the rotation factors with coefficients of 1, 2, 3, 4, 5, 6, and 7 are... It is divided into the second sub-stage.
[0008] Furthermore, the step of splitting the feedback register of the operation stage on two adjacent paths includes: When the feedback register capacity is 4, it is split into 1 and 3, and 3 and 1 on the two paths respectively; When the feedback register capacity is 8, it is divided into 5 and 3, and 3 and 5 on the two paths respectively; When the feedback register capacity is 16, it is split into 10 and 6, and 6 and 10 on the two paths respectively; When the feedback register capacity is 32, it is split into 20 and 12, and 12 and 20 on the two paths respectively; When the feedback register capacity is 64, it is split into 40 and 24, and 24 and 40 on the two paths respectively.
[0009] The sharing method for serialized CSD complex multipliers for multipath FFT processors provided in this application can significantly reduce the hardware overhead and power consumption of FFT processors, and is suitable for various application scenarios in 6G wireless communication systems with strict constraints on power consumption and hardware resource overhead. Attached Figure Description
[0010] Figure 1 The schematic diagram of the FIFO after splitting according to an embodiment of this application is shown; Figure 2 The diagram illustrates the schematic of a dual-path cascaded CSD complex multiplier shared structure for 128-point rotation factor complex multiplication operations provided in this application embodiment. Detailed Implementation
[0011] To make the objectives, technical solutions, and advantages of this technical solution clearer, the following detailed description, in conjunction with specific embodiments, further illustrates this technical solution. It should be understood that these descriptions are merely exemplary and not intended to limit the scope of this technical solution.
[0012] Example 1: Serial CSD Complex Multiplier Multiplexing and Sharing Technology To meet the high-speed, real-time processing requirements of 6G wireless communication systems, multi-path architecture has become the mainstream implementation scheme for long-point FFT processors. However, the hardware overhead of multi-path FFT processors increases dramatically with the number of parallel paths. Simultaneously, 6G communication systems place higher demands on the number of FFT operation points, with a standard minimum of 4096 points, necessitating a large number of complex multiplication operations with multiple rotation factors during FFT computation. Compared to traditional CSD complex multipliers, cascaded CSD complex multipliers can efficiently adapt to scenarios involving more complex multiplications with multiple coefficients, making them more suitable for long-point FFT architectures. Against this backdrop, effectively reducing the hardware resource consumption of cascaded CSD complex multipliers in multi-path long-point FFT processors has become a core critical issue in the design of high-speed 6G FFT processors.
[0013] To address this issue, this application proposes a serial CSD complex multiplier reuse and sharing technology, specifically adapted to complex multiplication calculation scenarios with multiple coefficient rotation factors in FFT operations, which can effectively solve the hardware overhead problem of long-point, multi-path FFT architecture.
[0014] This reuse-sharing technology effectively reduces the overall hardware resource overhead of multi-path FFT processors by reusing a single serialized CSD complex multiplier through a dual-path architecture. This serialized structure supports complex multiplication operations with 128, 256, and 512 twitch factor coefficients. In long-point FFT processor designs, scenarios involving complex multiplication with twitch factors of these specifications can all achieve hardware cost optimization through this reuse-sharing technology.
[0015] The hardware complexity of traditional CSD complex multipliers is entirely determined by the number of rotation factor coefficients. Conventional structures can only support operations on 64 sets of coefficients or less, limiting their computational adaptability. In contrast, the cascaded CSD complex multiplier employs a two-stage complex multiplication architecture, splitting multiple sets of rotation factor coefficients and performing the operations in segments. The specific coefficient set splitting method is shown in Table 1. For example, when performing operations on 128-point rotation factors… When performing complex multiplication, first perform an AND operation on the input data. Complex number multiplication is performed, and the resulting data is then multiplied by... This significantly reduces the logical complexity and hardware cost associated with using a standard CSD complex multiplier. It's worth noting that for the rotation factor... Other The value can be calculated based on the symmetry of the rotation factor.
[0016] Table 1. Decomposition method of coefficient group of cascaded CSD complex multiplier
[0017] Furthermore, Table 2 provides the base-2 6 -2 5 The FFT algorithm implements the rotation factor distribution for an 8192-point FFT. As shown in Table 2, the 11th stage of the rotation factor distribution requires calculating the rotation factor. In this case, a cascaded CSD complex multiplier can be used to complete the calculation. In the context of multiple paths, it allows two paths to share a single cascaded CSD complex multiplier.
[0018] By reallocating the capacity of the FIFO feedback register and staggering the calculations of the same twiddle factor coefficient sets that need to be performed simultaneously on both paths, the goal of sharing the cascaded CSD complex multiplier is achieved. Twiddle factor At stage 11 of the rotation factor distribution table, the FIFO capacity is 4. The FIFO of the 2-path is decomposed, with the FIFO of one path split into 1 and 3, and the FIFO of the other path split into 3 and 1. Multi-path FFT processors generally have an even number of paths, so it can be set that the FIFO of odd-numbered paths is split into 1 and 3, and the FIFO of even-numbered paths is split into 3 and 1. Figure 1 The diagram shown is a schematic of the FIFO after it has been split. Figure 1 (a) shows the traditional dataflow computing structure, while Figure 1 (b) and Figure 1 (c) are the data flow computation structures after splitting the FIFO in odd-numbered and even-numbered paths, respectively, where BF is the basic butterfly unit. "To implement a CSD complex multiplier with 128 sets of coefficients in series."
[0019] Table 2, Base-26 -2 5 The algorithm's 8192-point FFT rotation factor distribution
[0020] Employing a shared multiplier architecture reduces the hardware footprint of cascaded CSD complex multipliers in multipath FFT processors. The core advantage of using cascaded CSD complex multipliers is that, compared to the Booth multipliers commonly used in traditional solutions, this architecture has lower hardware overhead and eliminates the need for additional read-only memory (ROM) to store the rotation factor coefficients.
[0021] like Figure 2 The diagram shows the shared architecture of a dual-path cascaded CSD complex multiplier for 128-point rotation factor complex multiplication. (Modules shown in the diagram...) For data selectors; to shorten the critical path of the circuit and improve real-time data processing performance, a pipelined register is inserted between the two stages of shared cascaded CSD complex multipliers. The mapping module relies on the symmetry property of the twiddle factor to derive... The results of the calculations for the remaining coefficients are used to complete the calculation of all rotation factors.
[0022] In summary, conventional multi-path FFT processors all employ an even number of parallel paths. This application proposes an innovative architecture that allows two paths to share a single set of cascaded CSD complex multipliers, halving the total number of cascaded CSD complex multipliers within the FFT processor.
[0023] Example 2: A sharing method for serialized CSD complex multipliers for multipath FFT processors: The method is applied to the complex multiplication stage of 128 groups, 256 groups, and 512 groups of rotation factors. The method includes: Step 1: Decompose the coefficients of the rotation factor of the corresponding specification to be calculated according to the corresponding decomposition rules.
[0024] In practical implementation, the following decomposition rules are used for splitting: The decomposition rules for the 128 sets of twitch factor coefficients are as follows: ; The decomposition rules for the 256 sets of twitch factor coefficients are as follows: ; The decomposition rule for the 512 sets of twitch factor coefficients is as follows: .
[0025] Step 2: Based on the specification of the rotation factor, divide the rotation factor of the specific coefficient into two sub-stages.
[0026] In practice, the implementation is divided into two sub-phases as follows: Rotation factors with coefficients of 4, 8, 12, and 16 Divided into the first sub-stage, the rotation factors with coefficients of 1, 2, and 3 are... Divided into the second sub-stage; Alternatively, use rotation factors with coefficients of 4, 8, 12, 16, 20, 24, 28, and 32. Divided into the first sub-stage, the rotation factors with coefficients of 1, 2, and 3 are... Divided into the second sub-stage; Alternatively, use rotation factors with coefficients of 8, 16, 24, 32, 40, 48, 56, and 64. Divided into the first sub-stage, the rotation factors with coefficients of 1, 2, 3, 4, 5, 6, and 7 are... It is divided into the second sub-stage.
[0027] Step 3: Perform the first sub-stage complex multiplication operation by serially connecting CSD complex multipliers that do not overlap between adjacent paths. Split the feedback register of this operation stage on the two adjacent paths respectively, so as to avoid the calculation of the same twitch factor coefficient group that needs to be performed simultaneously on the two paths, thereby achieving the purpose of sharing the serially connected CSD complex multipliers.
[0028] In practical implementation, the capacity of the feedback register is allocated in the following way: When the feedback register capacity is 4, it is split into 1 and 3, and 3 and 1 on the two paths respectively; When the feedback register capacity is 8, it is divided into 5 and 3, and 3 and 5 on the two paths respectively; When the feedback register capacity is 16, it is split into 10 and 6, and 6 and 10 on the two paths respectively; When the feedback register capacity is 32, it is split into 20 and 12, and 12 and 20 on the two paths respectively; When the feedback register capacity is 64, it is split into 40 and 24, and 24 and 40 on the two paths respectively.
[0029] Step 4: Perform the second sub-stage complex multiplication operation by deploying CSD complex multipliers on their respective paths.
[0030] Step 5: Based on the symmetry of the rotation factor with non-specific coefficients, derive the result of its complex multiplication operation.
[0031] When this method is applied to the complex multiplication stage with 128 sets of rotation factors, the steps are as follows: According to the decomposition rules, the 128 sets of twisting factor coefficients are... Split; For the rotation factors with coefficients of 4, 8, 12, and 16 obtained from the decomposition. The first sub-stage complex multiplication operation is performed using a CSD complex multiplier connected in series between the two paths. The feedback register of this operation stage is split on the two paths to stagger the calculation of the same twitch factor coefficient group that needs to be performed simultaneously on the two paths, so as to achieve the purpose of sharing the series CSD complex multiplier. For the rotation factors with coefficients of 1, 2, and 3 obtained from the split The second sub-stage complex multiplication operation is performed using CSD complex multipliers deployed on their respective paths; Rotation factor for other coefficients obtained from the split Based on the symmetry of the rotation factor itself, the result of its complex multiplication operation is derived.
[0032] When this method is applied to the complex multiplication operation stage with 256 sets of rotation factors, the steps are as follows: For the 256 sets of twisting factor coefficients, according to the decomposition rules Split; For the rotation factors with coefficients of 4, 8, 12, 16, 20, 24, 28, and 32 obtained from the decomposition. The first sub-stage complex multiplication operation is performed using a CSD complex multiplier connected in series between the two paths; For the rotation factors with coefficients of 1, 2, and 3 obtained from the split The second sub-stage complex multiplication operation is performed using CSD complex multipliers deployed on their respective paths; Rotation factor for other coefficients obtained from the split Based on the symmetry of the rotation factor itself, the result of its complex multiplication operation is derived.
[0033] When this method is applied to the complex multiplication stage with 512 sets of rotation factors, the steps are as follows: According to the decomposition rules, the 512 sets of twisting factor coefficients are... Split; For the rotation factors with coefficients of 8, 16, 24, 32, 40, 48, 56, and 64 obtained from the decomposition. The first sub-stage complex multiplication operation is performed using a CSD complex multiplier connected in series between the two paths; For the rotation factors with coefficients of 1, 2, 3, 4, 5, 6, and 7 obtained from the decomposition. The second sub-stage complex multiplication operation is performed using CSD complex multipliers deployed on their respective paths; Rotation factor for other coefficients obtained from the split Based on the symmetry of the rotation factor itself, the result of its complex multiplication operation is derived.
[0034] Example 3: Verification Experiment Verification was completed using the Quartus Prime development platform, and a base-2 architecture was built. 6 -2 5 A hybrid algorithm-based 8192-point FFT processor was compared with and without the shared cascaded CSD complex multiplier scheme proposed in this application in terms of logic resource consumption and dynamic power consumption. Test results show that after adopting the shared architecture, the overall logic resource consumption is reduced by 38%, and the dynamic power consumption is reduced by 28%.
[0035] In addition, Table 3 compares the hardware resource consumption of the CSD complex multiplier and the Booth complex multiplier under different numbers of coefficient groups.
[0036] Table 3. Comparison of Hardware Resource Consumption between Booth Complex Multiplier and Serial CSD Complex Multiplier
[0037] As shown in Table 3, for complex multiplication scenarios with 128, 256, and 512 rotation factors, compared to the Booth multiplier, the cascaded CSD complex multiplier reduces logic resource usage by 41%, 34%, and 25%, respectively, while dynamic power consumption decreases by 57%, 51%, and 42%, respectively. All simulation tests were performed on the Quartus Prime development platform, and the target device was a Cyclone LP series FPGA.
[0038] The above content is only a preferred embodiment of the present invention. For those skilled in the art, many changes can be made in the specific implementation and application scope based on the ideas of the present invention. As long as these changes do not depart from the concept of the present invention, they all fall within the protection scope of the present invention.
Claims
1. A method for sharing a serialized CSD complex multiplier for a multipath FFT processor, characterized in that, The method is applied to the complex multiplication stage of 128 groups, 256 groups, and 512 groups of rotation factors. The method includes: The coefficients of the rotation factor of the corresponding specification to be calculated are split according to the corresponding decomposition rules; Based on the specification of this rotation factor, the rotation factor of a specific coefficient is divided into two sub-stages; The first sub-stage complex multiplication operation is performed by serially connecting CSD complex multipliers that do not overlap between adjacent paths. The feedback registers of this operation stage are split on the two adjacent paths respectively, so as to avoid the calculation of the same twitch factor coefficient group that needs to be performed simultaneously on the two paths, thereby achieving the purpose of sharing the serially connected CSD complex multipliers. The second sub-stage complex multiplication operation is performed by deploying CSD complex multipliers on their respective paths; Based on the symmetry of the twitch factor with non-specific coefficients, the results of complex multiplication are derived.
2. The method as described in claim 1, characterized in that, The decomposition rules for the 128 sets of twitch factor coefficients are as follows: ; The decomposition rules for the 256 sets of twitch factor coefficients are as follows: ; The decomposition rule for the 512 sets of twitch factor coefficients is as follows: .
3. The method as described in claim 1, characterized in that, The specification based on this rotation factor divides the rotation factor of a specific coefficient into two sub-stages, including: Rotation factors with coefficients of 4, 8, 12, and 16 Divided into the first sub-stage, the rotation factors with coefficients of 1, 2, and 3 are... Divided into the second sub-stage; Alternatively, use rotation factors with coefficients of 4, 8, 12, 16, 20, 24, 28, and 32. Divided into the first sub-stage, the rotation factors with coefficients of 1, 2, and 3 are... Divided into the second sub-stage; Alternatively, use rotation factors with coefficients of 8, 16, 24, 32, 40, 48, 56, and 64. Divided into the first sub-stage, the rotation factors with coefficients of 1, 2, 3, 4, 5, 6, and 7 are... It is divided into the second sub-stage.
4. The method as described in claim 1, characterized in that, The step of splitting the feedback registers for this operation stage on two adjacent paths includes: When the feedback register capacity is 4, it is split into 1 and 3, and 3 and 1 on the two paths respectively; When the feedback register capacity is 8, it is divided into 5 and 3, and 3 and 5 on the two paths respectively; When the feedback register capacity is 16, it is split into 10 and 6, and 6 and 10 on the two paths respectively; When the feedback register capacity is 32, it is split into 20 and 12, and 12 and 20 on the two paths respectively; When the feedback register capacity is 64, it is split into 40 and 24, and 24 and 40 on the two paths respectively.
