Large-point FFT (Fast Fourier Transform) operation method and device, chip, equipment, storage medium and program product

Abstract

The invention provides a large-point FFT operation method and device, a chip, equipment, a storage medium and a program product, and the method comprises the steps: carrying out the zero padding operation of an obtained original signal sequence according to a preset spectrum resolution requirement, and obtaining a to-be-transformed sequence; calculating the length ratio of the original signal sequence in the sequence to be transformed; if the length ratio is not lower than a set threshold value, converting the sequence to be converted into a frequency spectrum sequence by adopting a mode of performing two rounds of FFT operation after sequence decomposition according to a Cooley-Tukey FFT algorithm; and if the length ratio is lower than a set threshold value, converting the sequence to be converted into a frequency spectrum sequence by adopting a mode of performing a round of FFT operation after sequence decomposition according to a simplified decomposition method. For the condition that the length proportion of the original signal sequence in the to-be-transformed sequence is lower than the set threshold value, only one round of FFT operation needs to be executed when the simplified decomposition method is adopted, one round of FFT operation is directly omitted compared with a Cooley-Tukey FFT algorithm, calculation complexity is reduced, calculation redundancy is reduced, and calculation resources can be greatly saved.

Description

Technical Field

[0001] This invention relates to the field of signal processing technology, and more specifically, to a large-point FFT operation method, apparatus, chip, device, storage medium, and program product. Background Technology

[0002] In the field of modern digital signal processing, the Fast Fourier Transform (FFT) is one of the core algorithms for converting time-domain signals to frequency-domain signals. It is widely used in many high-tech scenarios such as communication systems, radar signal processing, audio analysis, image processing, and sensor data parsing. Through FFT operations, the frequency components of a signal can be extracted efficiently, thereby supporting functions such as modulation recognition, channel estimation, and fault diagnosis.

[0003] To achieve higher frequency resolution, the number of FFT points N is typically increased. This depends on the frequency domain resolution. (in The signal sampling frequency, (number of sampling points), increase It can distinguish adjacent frequency components more precisely.

[0004] In practical applications, many systems possess only a finite length of valid data. However, to improve spectral resolution or meet the needs of subsequent processing (such as interpolation, filtering, and matched filtering), zero-padding is often used to extend the original short sequence into a longer input sequence before performing a large-point FFT operation. For example, data with only 4000 valid samples might be zero-padding to 16384 points before performing a 16384-point FFT. While this approach does not add new information, it provides a denser sampling point in the frequency domain, facilitating the observation and analysis of spectral details.

[0005] In practical chip design, readily available small- to medium-scale FFT IP cores (such as those supporting a maximum of 4096 points) are often used. To implement large-point FFT operations, existing technologies are mostly based on Cooley-Tukey-like decomposition algorithms. The core idea of ​​this approach is to decompose large-point FFT operations into a combination of multiple small-point FFTs multiplied by twiddle factors. For example, a 16384-point FFT can be decomposed into a butterfly operation structure consisting of four 4096-point FFTs and 4096 4-point FFTs. If using existing FFT IP cores that support a maximum of 4096 points, this decomposition can be achieved through matrix rearrangement.

[0006] However, the computational complexity after splitting is greater than before splitting, and existing technology still performs the full FFT operation on all sub-blocks, including those subsequences consisting entirely of zeros, resulting in significant computational redundancy. Summary of the Invention

[0007] The purpose of this invention is to provide a large-point FFT operation method, apparatus, chip, device, storage medium, and program product to improve the problems existing in the prior art.

[0008] The embodiments of the present invention can be implemented as follows: In a first aspect, the present invention provides a large-point FFT operation method, comprising: The acquired original signal sequence is padded with zeros according to the preset spectral resolution requirements to obtain the sequence to be transformed; Calculate the length proportion of the original signal sequence in the sequence to be transformed; If the length ratio is not lower than the set threshold, the sequence to be transformed is converted into a spectral sequence by using the Cooley-Tukey FFT algorithm, which involves sequence decomposition followed by two rounds of FFT operations. If the length ratio is lower than the set threshold, the sequence to be transformed is converted into a spectral sequence by performing a round of FFT operation after sequence decomposition according to the simplified decomposition method.

[0009] In an optional implementation, the step of converting the sequence to be transformed into a spectral sequence by performing a round of FFT operation after sequence decomposition according to the simplified decomposition method includes: Reshape the sequence to be transformed into The signal matrix; Based on the pre-stored rotation matrix data, the signal matrix is ​​converted into a first intermediate frequency domain matrix; For each column of the intermediate frequency domain matrix, perform The FFT operation on the point data yields the second intermediate frequency domain matrix; The second intermediate frequency domain matrix is ​​spliced ​​column by column, starting from the first column, to obtain the spectrum sequence.

[0010] In an optional implementation, the rotation matrix data includes The original rotation matrix; The step of converting the signal matrix into a first intermediate frequency domain matrix based on pre-stored rotation matrix data includes: Copy the first column of the signal matrix. Share, received Column replication signal; The remaining signals in the signal matrix, excluding the first column. All column signals are replaced with the replicated signals to obtain a replicated signal matrix; The first intermediate frequency domain matrix is ​​obtained by multiplying the replicated signal matrix element by element with the original rotation matrix.

[0011] In an optional implementation, the rotation matrix data includes A rotation sequence, wherein the first rotation sequence is... The second column is the same as the original rotation matrix, the third... The rotation sequence includes the first rotation sequence in the original rotation matrix. After the list All elements of a part; The step of converting the signal matrix into a first intermediate frequency domain matrix based on pre-stored rotation matrix data includes: The first column signal of the signal matrix is ​​directly used as the first column frequency domain signal of the first intermediate frequency domain matrix; The result of multiplying the second column signal of the signal matrix with the first rotation sequence element by element is taken as the second column frequency domain signal of the first intermediate frequency domain matrix; For the first signal matrix The column signal, starting from the first element of the first rotated sequence, increments by the index value for each... Extract the pattern of an element to extract the current reused rotation sequence; The first Column signal center front The part is multiplied element-wise with the current multiplexed rotation sequence, and the first part is multiplied element-wise. The last 1 in the column signal Part of the first Multiplying the rotation sequence element by element, we obtain the first intermediate frequency domain matrix. Column frequency domain signal; where, .

[0012] In an optional implementation, the step of padding the acquired original signal sequence with zeros according to a preset spectral resolution requirement to obtain the sequence to be transformed includes: Obtain the original signal sequence; When the target number of points required by the preset spectral resolution is N, the original signal sequence is padded with zeros until the sequence length reaches N, thus obtaining the sequence to be transformed.

[0013] In a second aspect, the present invention provides a large-point FFT computing device, comprising: The zero-padding module is used to pad the acquired original signal sequence with zeros according to the preset spectral resolution requirements to obtain the sequence to be transformed. A sparsity determination module is used to calculate the length ratio of the original signal sequence in the sequence to be transformed; The first computation module is used to convert the sequence to be transformed into a spectral sequence by performing two rounds of FFT operations after sequence decomposition according to the Cooley-Tukey FFT algorithm when the length ratio is not less than a set threshold. The second operation module is used to convert the sequence to be transformed into a spectral sequence by performing a round of FFT operation after sequence decomposition according to the simplified decomposition method when the length ratio is lower than the set threshold.

[0014] Thirdly, the present invention provides a chip for implementing the large-point FFT operation method as described in the first aspect above.

[0015] Fourthly, the present invention provides a signal processing device, the signal processing device comprising the chip described in the third aspect above.

[0016] Fifthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the large-point FFT operation method described in the first aspect above.

[0017] In a sixth aspect, the present invention provides a program product that, when executed by a processor, implements the large-point FFT operation method described in the first aspect above.

[0018] Compared with existing technologies, this invention provides a method, apparatus, chip, device, storage medium, and program product for large-point FFT operations. First, the acquired original signal sequence is padded with zeros according to a preset spectral resolution requirement to obtain the sequence to be transformed. Next, the length ratio of the original signal sequence in the sequence to be transformed is calculated. If the length ratio is not lower than a set threshold, the sequence to be transformed is converted into a spectral sequence by performing two rounds of FFT operations after sequence decomposition according to the Cooley-Tukey FFT algorithm. If the length ratio is lower than the set threshold, the sequence to be transformed is converted into a spectral sequence by performing one round of FFT operations after sequence decomposition according to a simplified decomposition method. When the length ratio of the original signal sequence in the sequence to be transformed is lower than the set threshold, this invention only requires one round of FFT operations when using the simplified decomposition method for FFT transformation. Compared to the Cooley-Tukey FFT algorithm, which directly omits one round of FFT operations, this not only reduces computational complexity but also reduces computational redundancy, significantly saving computational resources. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating a large-point FFT operation method provided in an embodiment of the present invention.

[0021] Figure 2 A schematic diagram illustrating the process of performing small-point FFT operations and phase correction on a row of elements of a matrix in the Cooley-Tukey FFT algorithm.

[0022] Figure 3 This is an equivalent replacement diagram illustrating the implementation of small-point FFT operation and phase correction for a row of elements in an embodiment of the present invention.

[0023] Figure 4 This is a schematic diagram of a large-point FFT computing device provided in an embodiment of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0025] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0026] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0027] Furthermore, the terms "first" and "second" are used only to distinguish descriptions and should not be interpreted as indicating or implying relative importance.

[0028] It should be noted that, where there is no conflict, the features in the embodiments of the present invention can be combined with each other.

[0029] Please see Figure 1 , Figure 1 This is a flowchart illustrating a large-point FFT operation method provided in an embodiment of the present invention. The execution subject of this method can be a chip in a signal processing device, and the method includes the following steps S101 to S104.

[0030] S101. The acquired original signal sequence is padded with zeros according to the preset spectral resolution requirements to obtain the sequence to be transformed.

[0031] To meet the requirements of high-precision frequency resolution, the system typically requires the execution of large-point FFT operations, such as 4096 points, 8192 points, or even 16384 points. However, the actual effective signal length acquired is often much smaller than the target FFT point number. Therefore, in this embodiment, zero-padding is performed on the original signal sequence to ensure that the total signal length of the resulting sequence to be transformed reaches the target FFT point number N, resulting in a finer frequency interval in the output spectrum and thus improving the frequency resolution capability.

[0032] S102. Calculate the length ratio of the original signal sequence in the sequence to be transformed.

[0033] The calculation of the effective data percentage (FFT) involves the proportion of the original signal sequence length in the sequence to be transformed. This proportion is the ratio of the effective data length to the total number of FFT points. Therefore, the effective data percentage reflects the sparsity of the effective signal in the sequence to be transformed. A smaller length percentage indicates a sparser effective signal in the sequence; conversely, a larger length percentage indicates a denser effective signal.

[0034] S103. If the length ratio is not lower than the set threshold, the sequence to be transformed is converted into a spectral sequence by performing two rounds of FFT operations after sequence decomposition according to the Cooley-Tukey FFT algorithm.

[0035] In this embodiment, if the length ratio is not lower than a set threshold, the input signal is determined to be relatively dense. In this case, the traditional Cooley-Tukey FFT algorithm is used to decompose the sequence to be transformed into rows or columns, and two rounds of FFT operations are performed sequentially: the first round performs a small-point FFT operation on the decomposed subsequences, and the second round, after introducing a rotation factor, performs another small-point FFT operation on the intermediate results, ultimately generating a complete spectral sequence. This method is applicable to most common scenarios and can guarantee computational accuracy and algorithm stability.

[0036] The threshold can be set to... , This indicates the number of columns in the matrix after sequence decomposition. For example, the threshold can be set to 25% or 12.5%.

[0037] S104. If the length ratio is lower than the set threshold, the sequence to be transformed is converted into a spectral sequence by performing a round of FFT operation after sequence decomposition according to the simplified decomposition method.

[0038] If the length ratio is lower than the set threshold, it indicates that the input signal exhibits significant sparsity, meaning that most of the data in the sequence to be transformed is zero-padded, with only a small amount of front-end data containing valid information. In this case, a simplified decomposition method is activated to avoid performing redundant row-direction FFT operations on the all-zero segment. The equivalent transformation from the sequence to be transformed to the spectral sequence can be completed by performing only one round of FFT operations on each column of the decomposed matrix.

[0039] The large-point FFT operation method provided in this invention introduces length proportion as a decision variable to dynamically adapt to different FFT implementation mechanisms. This allows for differentiated processing strategies for input signals with varying sparsity while ensuring the accuracy of the output results. When the length proportion of the original signal sequence in the sequence to be transformed is less than a set threshold, a simplified decomposition method requires only one round of FFT operation. Compared to the Cooley-Tukey FFT algorithm, which directly omits one round of FFT operation, this not only reduces computational complexity but also minimizes computational redundancy, significantly saving computational resources.

[0040] Optionally, for the above step S101, the implementation process is as follows: obtain the original signal sequence; when the target number of points required by the preset spectral resolution is N, perform zero-padding operation on the original signal sequence until the sequence length reaches N, and obtain the sequence to be transformed, where N may be 4096 points, 8192 points, 16384 points, or 32768 points, etc.

[0041] For example, the original signal sequence has a total of 4000 points (which can be regarded as a 4000-dimensional one-dimensional vector), while the target number of points required by the preset spectral resolution is N, which is 16384 points. Therefore, the original signal sequence needs to be padded with zeros to 16384 points to obtain the sequence to be transformed. It should be noted that this example is only an example and is not intended to be limiting.

[0042] Regarding step S103 above, although the Cooley-Tukey FFT algorithm is an existing technology, its implementation process will be briefly introduced here.

[0043] If the number of target points N = A point, for a Given the input signal sequence at each point (which may or may not include zero-padding), assuming n=12 and m=2, the input signal sequence is: .

[0044] Ignoring the length proportions, in existing technologies, the input signal sequence is processed according to the Cooley-Tukey FFT algorithm. The process of converting it into an output spectrum sequence is as follows: Step 1, Reshaping: Reshape the input signal sequence Remodeled as follows Matrix Y1:

[0045] Step 2: Perform a 4-point FFT operation on each row of matrix Y1 to obtain matrix Y2; Step 3, Phase Correction: Multiply each element in matrix Y2 by its corresponding rotation factor to obtain matrix Y3; where the rotation factor in the r-th row and k-th column of matrix Y3 is... , , , ; Step 4: Perform a 4096-point FFT operation on each column of matrix Y3 to obtain matrix Y4; Step 5: Concatenate the matrix Y4 column by column, starting from the first column, to obtain the output spectrum sequence.

[0046] When the input signal sequence When the length of the effective signal is less than 25%, please refer to [the relevant documentation]. Figure 2 After multiple verifications, it was found that using steps 1 to 5 above for such sparse input signal sequences... During the processing, the following special characteristics were observed: (1) Matrix The (r+1)th line in the code: ,in , The result of the 4-point FFT operation in step 2 is ; (2) In the corresponding 4 twitch factors Step 3 involves multiplying the two element-wise to obtain: .

[0047] Combination Figure 3 ,based on The data transformation characteristics during the transformation process can be considered to be obtained without the need for a 4-point FFT operation. Instead, directly from Simply copy the first element and overwrite the following three zeros to get the result. Then, based on the rotation factor, phase correction processing is performed on matrix Y1 to achieve equivalent substitution.

[0048] Based on this equivalent substitution approach, the simplified decomposition method of this invention is designed to achieve fast FFT transformation for sparse sequences to be transformed where the proportion of effective data is lower than a set threshold.

[0049] Therefore, the implementation process of "converting the sequence to be transformed into a spectral sequence by performing a round of FFT operation after sequence decomposition according to the simplified decomposition method" in step S104 above may include the following sub-steps S1041 to S1044.

[0050] S1041, Reshape the sequence to be transformed into... The signal matrix; S1042. Based on the pre-stored rotation matrix data, convert the signal matrix into a first intermediate frequency domain matrix; S1043, Perform the following for each column of the intermediate frequency domain matrix: The FFT operation on the point data yields the second intermediate frequency domain matrix; S1044. Start from the first column and concatenate the second intermediate frequency domain matrix column by column to obtain the spectrum sequence.

[0051] The core of the simplified decomposition method lies in step S1042.

[0052] In the first optional implementation method, it is possible to Figure 3 The equivalent substitution approach shown can be extended to row 4096 of matrix Y1, and then the 4096×4 rotation factors of the matrix can be pre-stored.

[0053] Therefore, the rotation matrix data in S1042 above can include The original rotation matrix is ​​obtained. In this case, step S1042 is implemented by steps a1 to a3: Step a1: Copy the signal from the first column of the signal matrix. Share, received Column replication signal; Step a2: Extract the remaining signals from the signal matrix, excluding the first column. All column signals are replaced with replicated signals to obtain a replicated signal matrix; Step a3: Multiply the copied signal matrix element by element with the original rotation matrix to obtain the first intermediate frequency domain matrix.

[0054] Through steps a1 to a3, step 2 of the Cooley-Tukey FFT algorithm can be performed... OK Point-based FFT operations are equivalent to signal copying and replacement operations, which can significantly reduce computational complexity.

[0055] For example, if the original rotation matrix is ​​a 4096×4 matrix, then the original rotation matrix... for:

[0056] By observing the original rotation matrix The following patterns can be observed: (1) Original rotation matrix All elements in the first column are 1; (2) Original rotation matrix The second column is: ; (3) Original rotation matrix The third column is: That is, the first one in the third column All of the first 2048 elements appear in the second column; (4) Original rotation matrix The fourth column is: That is, the first one in the fourth column Some of them (i.e., the first 1366 elements) all appeared in the second column.

[0057] Therefore, based on the above principles, in order to reduce storage space usage, it is not necessary to store the entire original rotation matrix. Instead, it only needs to store the original rotation matrix. In: the last column of the first column, the second column, and the third column Part and the last of column 4 In this case, you can reuse only the second column when using it.

[0058] Therefore, this kind of rotation matrix is ​​used to... Extending the idea of ​​partial storage to When dealing with the rotation matrix, an equivalent substitution can be achieved for steps 2 and 3 above using another method. The principle of substitution is explained below.

[0059] Regarding step S104 above, the rotation matrix data includes A rotation sequence, the first rotation sequence and... The second column is the same as the original rotation matrix, the third... The rotation sequence includes the first rotation sequence in the original rotation matrix. After the list All elements of a portion. Therefore, the implementation process of step S104 may include the following steps b1 to b4.

[0060] Step b1: Directly use the first column signal of the signal matrix as the first column frequency domain signal of the first intermediate frequency domain matrix.

[0061] Among them, it is precisely because of the original rotation matrix All elements in the first column are 1. Since 1 multiplied by any number has no effect, the first column of the signal matrix can be directly used as the first column of the frequency domain signal of the first intermediate frequency domain matrix.

[0062] Step b2: Multiply the second column of the signal matrix element by element with the first rotation sequence and use the result as the second column of the frequency domain signal of the first intermediate frequency domain matrix.

[0063] Step b3, for the signal matrix The column signal starts from the first element of the first rotated sequence and increments by the index value. Extract the pattern of an element to extract the current reused rotation sequence; Step b4, the first Column signal center front The part is multiplied element-wise with the current multiplexed rotation sequence, and the first element is multiplied by the second element. The last 1 in the column signal Part and the Multiplying the rotation sequences element by element, we obtain the first intermediate frequency domain matrix. Column frequency domain signal; where, .

[0064] In this embodiment, in the first rotation sequence, the index value of an element + 1 indicates the element's position in the first rotation sequence. That is, in the first rotation sequence, the index values ​​of each element are: .

[0065] For example, again assuming n=12 and m=2, then for the signal in the 3rd column of the signal matrix (i.e. ), it is necessary to extract the index value from the first rotation sequence. The total of 2048 elements yields the current reused rotation sequence; Next, for the signal in the 4th column of the signal matrix (i.e. ), it is necessary to extract the index value from the first rotation sequence. The total of 1366 elements yields the current reused rotation sequence. It should be noted that this example is for illustrative purposes only and is not intended to be limiting.

[0066] Therefore, through steps b1 to b4, not only can step 2 of the Cooley-Tukey FFT algorithm be performed... OK The point-wise FFT operation is directly omitted, and the rotation factor multiplication in the first column of step 3 is also eliminated, thus significantly reducing computational complexity. Furthermore, leveraging the pattern that some elements in each column of the original rotation matrix appear in the second column starting from the third column, only the elements mentioned above need to be stored. A rotation sequence reduces storage usage.

[0067] It should be noted that the execution order of each step in the above method embodiments is not limited to what is shown in the accompanying drawings and step numbers, and the execution order of each step shall be subject to the actual application situation.

[0068] In order to perform the corresponding steps in the above method embodiments and various possible implementations, an implementation of a large-point FFT operation device is given below.

[0069] Please see Figure 4 , Figure 4 A schematic diagram of the structure of a large-point FFT computing device provided in an embodiment of the present invention is shown. The large-point FFT computing device 200 includes: a zero-padding module 210, a sparsity judgment module 220, a first computing module 230, and a second computing module 240.

[0070] The zero-padding module 210 is used to perform zero-padding on the acquired original signal sequence according to the preset spectral resolution requirements to obtain the sequence to be transformed; The sparsity judgment module 220 is used to calculate the length ratio of the original signal sequence in the sequence to be transformed; The first operation module 230 is used to convert the sequence to be transformed into a spectrum sequence by using the Cooley-Tukey FFT algorithm, which decomposes the sequence and performs two rounds of FFT operations after sequence decomposition, when the length ratio is not lower than a set threshold. The second operation module 240 is used to convert the sequence to be transformed into a spectral sequence by performing a round of FFT operation after sequence decomposition according to the simplified decomposition method when the length ratio is lower than a set threshold.

[0071] Optionally, the zero-padding module 210 can be used to: acquire the original signal sequence; when the target number of points required by the preset spectral resolution is N, perform zero-padding operation on the original signal sequence until the sequence length reaches N, to obtain the sequence to be transformed.

[0072] Optionally, the second operation module 240 may include: a reshaping module for reshaping the sequence to be transformed into... The signal matrix; the equivalent transformation module, used to convert the signal matrix into a first intermediate frequency domain matrix based on pre-stored rotation matrix data; the column transformation module, used to transform each column of the intermediate frequency domain matrix. The FFT operation on the point data yields the second intermediate frequency domain matrix; the organization module is used to concatenate the second intermediate frequency domain matrix column by column, starting from the first column, to obtain the spectrum sequence.

[0073] Optionally, when the rotation matrix data includes When dealing with the original rotation matrix, the equivalent transformation module can specifically be used to: copy the signal in the first column of the signal matrix. Share, received Copy the signal to the column; copy the remaining signals in the signal matrix except for the first column. All column signals are replaced with replicated signals to obtain a replicated signal matrix; the replicated signal matrix is ​​multiplied element by element with the original rotation matrix to obtain the first intermediate frequency domain matrix.

[0074] Optionally, when the rotation matrix data includes When there are rotation sequences, and the first rotation sequence is... The second column is the same as the original rotation matrix, the third... The rotation sequence includes the first rotation sequence in the original rotation matrix. After the list All elements of the part. At this point, the equivalent transformation module can specifically be used to: directly use the first column of the signal matrix as the first column of the frequency domain signal of the first intermediate frequency domain matrix; use the result of element-wise multiplication of the second column of the signal matrix with the first rotation sequence as the second column of the frequency domain signal of the first intermediate frequency domain matrix; for the signal matrix... The column signal starts from the first element of the first rotated sequence and increments by the index value. Extract the pattern of an element to perform element extraction, and obtain the current reuse rotation sequence; then... Column signal center front The part is multiplied element-wise with the current multiplexed rotation sequence, and the first element is multiplied by the second element. The last 1 in the column signal Part and the Multiplying the rotation sequences element by element, we obtain the first intermediate frequency domain matrix. Column frequency domain signal; where, .

[0075] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the large-point FFT computing device 200 described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0076] This invention also provides a chip for implementing the large-point FFT operation method described above.

[0077] Embodiments of the present invention also provide a signal processing device, which includes the above-described chip.

[0078] This invention also provides a computer-readable storage medium storing a computer program. When executed by a processor, this computer program implements the large-point FFT operation method disclosed in the above embodiments. The computer-readable storage medium can be, but is not limited to, RAM (Random Access Memory), ROM (Read Only Memory), FLASH (Flash Memory), PROM (Programmable Read-Only Memory), EPROM (Erasable Programmable Read-Only Memory), EEPROM (Electric Erasable Programmable Read-Only Memory), USB flash drives, portable hard drives, FLASH disks, or optical discs, and other media capable of storing program code.

[0079] This invention also provides a computer program product that, when executed by a processor, implements the large-point FFT operation method disclosed in the above embodiments.

[0080] In summary, this invention provides a method, apparatus, chip, device, storage medium, and program product for large-point FFT operations. First, the acquired original signal sequence is padded with zeros according to a preset spectral resolution requirement to obtain the sequence to be transformed. Next, the length proportion of the original signal sequence in the sequence to be transformed is calculated. If the length proportion is not lower than a set threshold, the sequence to be transformed is converted into a spectral sequence by performing two rounds of FFT operations after sequence decomposition according to the Cooley-Tukey FFT algorithm. If the length proportion is lower than the set threshold, the sequence to be transformed is converted into a spectral sequence by performing one round of FFT operations after sequence decomposition according to a simplified decomposition method. When the length proportion of the original signal sequence in the sequence to be transformed is lower than the set threshold, this invention only requires one round of FFT operations when using the simplified decomposition method for FFT transformation. Compared to the Cooley-Tukey FFT algorithm, which directly omits one round of FFT operations, this not only reduces computational complexity but also reduces computational redundancy, significantly saving computational resources.

[0081] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for large-point FFT operation, characterized in that, include: The acquired original signal sequence is padded with zeros according to the preset spectral resolution requirements to obtain the sequence to be transformed; Calculate the length proportion of the original signal sequence in the sequence to be transformed; If the length ratio is not lower than the set threshold, the sequence to be transformed is converted into a spectral sequence by using the Cooley-Tukey FFT algorithm, which involves sequence decomposition followed by two rounds of FFT operations. If the length ratio is lower than the set threshold, the sequence to be transformed is converted into a spectral sequence by performing a round of FFT operation after sequence decomposition according to the simplified decomposition method.

2. The large-point FFT operation method according to claim 1, characterized in that, The step of converting the sequence to be transformed into a spectral sequence by performing a round of FFT operation after sequence decomposition according to the simplified decomposition method includes: Reshape the sequence to be transformed into The signal matrix; Based on the pre-stored rotation matrix data, the signal matrix is ​​converted into a first intermediate frequency domain matrix; For each column of the intermediate frequency domain matrix, perform The FFT operation on the point data yields the second intermediate frequency domain matrix; The second intermediate frequency domain matrix is ​​spliced ​​column by column, starting from the first column, to obtain the spectrum sequence.

3. The large-point FFT operation method according to claim 2, characterized in that, The rotation matrix data includes The original rotation matrix; The step of converting the signal matrix into a first intermediate frequency domain matrix based on pre-stored rotation matrix data includes: Copy the first column of the signal matrix. Share, received Column replication signal; The remaining signals in the signal matrix, excluding the first column. All column signals are replaced with the replicated signals to obtain a replicated signal matrix; The first intermediate frequency domain matrix is ​​obtained by multiplying the replicated signal matrix element by element with the original rotation matrix.

4. The large-point FFT operation method according to claim 2, characterized in that, The rotation matrix data includes A rotation sequence, wherein the first rotation sequence is... The second column is the same as the original rotation matrix, the third... The rotation sequence includes the first rotation sequence in the original rotation matrix. After the list All elements of a part; The step of converting the signal matrix into a first intermediate frequency domain matrix based on pre-stored rotation matrix data includes: The first column signal of the signal matrix is ​​directly used as the first column frequency domain signal of the first intermediate frequency domain matrix; The result of multiplying the second column signal of the signal matrix with the first rotation sequence element by element is taken as the second column frequency domain signal of the first intermediate frequency domain matrix; For the first signal matrix The column signal, starting from the first element of the first rotated sequence, increments by the index value for each... Extract the pattern of an element to extract the current reused rotation sequence; The first Column signal center front The part is multiplied element-wise with the current multiplexed rotation sequence, and the first part is multiplied element-wise. The last 1 in the column signal Part of the first Multiplying the rotation sequence element by element, we obtain the first intermediate frequency domain matrix. Column frequency domain signal; where, .

5. The large-point FFT operation method according to claim 1, characterized in that, The step of padding the acquired original signal sequence with zeros according to a preset spectral resolution requirement to obtain the sequence to be transformed includes: Obtain the original signal sequence; When the target number of points required by the preset spectral resolution is N, the original signal sequence is padded with zeros until the sequence length reaches N, thus obtaining the sequence to be transformed.

6. A large-point FFT computing device, characterized in that, include: The zero-padding module is used to pad the acquired original signal sequence with zeros according to the preset spectral resolution requirements to obtain the sequence to be transformed. A sparsity determination module is used to calculate the length ratio of the original signal sequence in the sequence to be transformed; The first computation module is used to convert the sequence to be transformed into a spectral sequence by performing two rounds of FFT operations after sequence decomposition according to the Cooley-Tukey FFT algorithm when the length ratio is not less than a set threshold. The second operation module is used to convert the sequence to be transformed into a spectral sequence by performing a round of FFT operation after sequence decomposition according to the simplified decomposition method when the length ratio is lower than the set threshold.

7. A chip, characterized in that, The chip is used to implement the large-point FFT operation method as described in any one of claims 1-5.

8. A signal processing device, characterized in that, The signal processing device includes the chip as described in claim 7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the large-point FFT operation method according to any one of claims 1-5.

10. A program product, characterized in that, When the program product is executed by the processor, it implements the large-point FFT operation method as described in any one of claims 1-5.

Images