Fractional Fourier transform high-precision rapid implementation method, system and equipment for digital signal processing
Through a new form of fractional Fourier transform decomposition and anti-aliasing filtering downsampling, the calculation error and efficiency problems in broadband signal processing in the prior art are solved, and high-precision and rapid calculation are achieved.
Patent Information
- Application Number
- CN202411092509.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-09
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-08-09
AI Technical Summary
The existing fractional Fourier transform fast computing technology has calculation errors and efficiency losses when processing broadband or ultra-wideband signals, and is insufficient in flexibility.
Through the new form of fractional Fourier transform decomposition method, the final result can be obtained with only one FFT operation, and combined with anti-aliasing filtering and downsampling processing, the calculation efficiency and accuracy are significantly improved.
High-precision calculation of broadband or ultra-wideband signals is realized, which reduces the amount of data, improves the calculation speed and flexibility, and avoids calculation errors.
Smart Images

Figure CN118916593A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of radar and underwater acoustic signal processing, and in particular relates to a high-precision and fast implementation method, system and equipment for fractional Fourier transform for digital signal processing. Background Art
[0002] Fractional Fourier transform (FrFT) is a popular signal time-frequency domain analysis method that is widely used in the field of signal processing. It is particularly suitable for analyzing linear frequency modulation signals. By performing a fractional Fourier transform on the signal, the signal is projected into an α-u two-dimensional transform domain, where α represents the projection angle, which directly matches the frequency modulation slope of the signal, and u has a certain correlation with the signal's time delay. Through the fractional Fourier transform, the time-frequency distribution of the signal can be easily analyzed, which is convenient for subsequent processing such as signal detection and recognition.
[0003] If the fractional Fourier transform is calculated directly according to the definition, it will have a large complexity, which is not conducive to practical engineering applications. In order to facilitate the practical application of the fractional Fourier transform, a variety of fast calculation technologies and systems of this method have been derived. The main principle of these fast calculation technologies is: for each angle α, the corresponding fractional Fourier transform is decomposed into a linear convolution form of two or more signals, and then this linear convolution operation is quickly calculated using the fast Fourier transform (FFT) technology, thereby finally realizing the fast calculation of the fractional Fourier transform. The main difference between the above-mentioned different technologies is mainly reflected in the different ways of decomposing the fractional Fourier transform, but the similarities are mainly reflected in the decomposition into a linear convolution form of different signal components.
[0004] In the FFT fast calculation process of the linear convolution, multiple FFT and IFFT (fast inverse Fourier transform) operations are used. Specifically, each convolution term is generally transformed into the discrete Fourier frequency domain by FFT, each transformation term is multiplied in the discrete Fourier frequency domain, and then the multiplication result is inversely transformed back to the time domain by IFFT, so as to obtain the fast calculation result of the linear convolution. Although the FFT and IFFT technologies have high computational efficiency, the multiple applications of FFT and IFFT obviously increase the amount of calculation exponentially compared to a single application. On the other hand, in the existing FrFT fast calculation technology, the two or more signals (or functions) decomposed into linear convolution are mainly in the form of linear frequency modulation. For the field of broadband or ultra-wideband signal processing, due to the characteristics of wide bandwidth and long signal sequence, the decomposed signals or functions and the operations between them will inevitably bring about non-negligible calculation errors, which will have a significant impact on the results and lead to loss of calculation accuracy.
[0005] The Chinese invention patent application with publication number CN 107644004A discloses a digital signal processing method and device based on a fast calculation method of discrete fractional Fourier transform. However, since this method needs to calculate the eigenvalues and eigenvectors of the discrete Fourier transform kernel matrix, the calculation complexity is increased and the calculation efficiency is lost.
[0006] The Chinese invention patent application with publication number CN105783974A discloses a method and system for detecting and estimating parameters of linear frequency modulation signals. The method realizes the calculation of fractional Fourier transform by rotation transformation and fast Fourier transform (FFT). However, the resolution accuracy and transform domain range of the transform domain calculation result corresponding to the method are strictly limited to the number of sampling points N and the sampling frequency fs of the signal, which loses the flexibility of design. In addition, it performs FFT over the entire sampling frequency range fs, which increases the calculation redundancy and also loses the calculation efficiency. Summary of the invention
[0007] In order to overcome the shortcomings of the above-mentioned prior art, the purpose of the present invention is to provide a method, system and device for high-precision and fast implementation of fractional Fourier transform for digital signal processing. The method achieves faster calculation than the prior art by performing a new form of decomposition on the fractional Fourier transform. Specifically, for each angle α, after some processing steps, only one FFT is needed to obtain the final transformation result, which greatly improves the calculation efficiency. On the other hand, in the process of processing the signal, the signal components formed by the combination of the present method tend to be narrowband or single-frequency. From the perspective of computer or processor implementation, the calculation can be realized with higher precision, which significantly improves the calculation accuracy.
[0008] In order to achieve the above object, the technical solution adopted by the present invention is:
[0009] A high-precision and fast implementation method of fractional Fourier transform for digital signal processing comprises the following steps:
[0010] S1, based on the actual discretized input signal x s (n) Situation and actual needs, determine the analysis range of the fractional Fourier transform angle α, discretize the input signal x s The total delay analysis range of (n) is T m , and the delay analysis step Δt m ;
[0011] S2, according to the angle α range and the total size T of the delay analysis range obtained in step S1 m Time delay analysis step Δt m , calculate the number of extended analysis points N corresponding to the maximum frequency domain range sAnd the point expansion multiple p, and the discretized input signal x s (n) performing zero padding processing;
[0012] S3, calculate the digital angular frequency resolution interval and the index value k0 of the starting position of the u-axis of the fractional Fourier transform;
[0013] S4, according to the digital angular frequency resolution interval obtained in step S3 The index value k0 and the discretized input signal x obtained in step S2 s (n) Calculate the synthetic signal x' s (n);
[0014] S5, the synthesized signal x' obtained in step S4 s (n) Perform anti-aliasing filtering and downsampling to obtain the downsampled signal x' sLD (n);
[0015] S6, the down-sampled signal x' obtained in step S5 sLD (n) Perform a fast Fourier transform (FFT) operation to obtain the preliminary result X' of the fractional Fourier transform α (ku s );
[0016] S7, the preliminary result X' of the fractional Fourier transform obtained in step S6 α (ku s ) performs inversion, downsampling, and truncation operations to obtain the final fractional Fourier transform result X α (ku s ).
[0017] The step S1 is specifically as follows:
[0018] For the discretized input signal x s (n) is the case where the instantaneous frequency decreases linearly with time. The analysis range of angle α should be in the interval Select from the discretized input signal x s (n) is the case where the instantaneous frequency increases linearly with time. The analysis range of angle α should be in the interval Select from; set the discretized input signal x s The delay analysis range of (n) is [0,T m ], where T m Indicates the total size of the delay analysis range, and the delay analysis step is Δt m , the total number of delay analysis points can be obtained as:
[0019]
[0020] Where int(·) indicates rounding operation; for the case where the starting time of the delay analysis range is not zero, the discretized input signal x s (n) Perform advance or delay processing to align the analysis start time to time zero.
[0021] The step S2 is specifically as follows:
[0022] First, find the delay analysis step length Δt m The corresponding frequency analysis step Δf m , and the total size of the delay analysis range T m The corresponding frequency domain analysis length f m , solve according to the following formula:
[0023] Δf m =Δt m |cotα| (1-7)
[0024] f m =T m |cotα| (1-8)
[0025] Determine the frequency domain analysis length f m Is it less than or equal to the maximum frequency domain analysis length f? s , that is, the sampling rate. If yes, continue to execute below. If no, return to equation (1-7) and equation (1-8) to reset the total size of the delay analysis range T m or angle α;
[0026] The number of extended analysis points N corresponding to the maximum frequency domain range is calculated according to the following formula: s :
[0027]
[0028] Among them, N s ≥N m , N m is the total number of delay analysis points, and the point expansion factor p represents the total number of input signal points N compared to the number of analysis points N in the maximum frequency domain range. s The multiple of is calculated as follows:
[0029]
[0030] Among them, the symbol Indicates rounding up, N is the discretized input signal x s (n) The total number of sampling points; after obtaining the point expansion factor p, the total number of analysis points corresponding to the maximum frequency domain range is pN s ;
[0031] Then, according to the total number of analysis points pN s, for the discretized input signal x s (n) is filled with zeros, that is, in x s (n) Add pN at the end s -N zeros, so that the total number of points of the signal after zero padding is just pN s To simplify the expression, let the discretized input signal after zero padding still be represented as x s (n), the time domain sequence index n takes values of 0, 1, 2, ..., pN s -1.
[0032] The step S3 is specifically as follows:
[0033] Digital angular frequency resolution Directly obtained by the following formula:
[0034]
[0035] Among them, pN s is the total analysis points;
[0036] When the angle α is in the first quadrant of the time-frequency domain, the index value k0 is calculated as follows:
[0037]
[0038] Among them, f bm Represents the signal x to be analyzed s (n) The intermediate frequency starting frequency, f bm Satisfy the relationship: f bm <f s , f s Indicates the maximum frequency domain analysis length;
[0039] When the angle α is in the fourth quadrant of the time-frequency domain, the index value k0 can be calculated as follows:
[0040]
[0041] Among them, p represents the point expansion multiple, Δf m Represents the frequency analysis step size.
[0042] The step S4 is specifically as follows:
[0043] When angle α is in the first quadrant, x' s (n) is calculated as follows:
[0044]
[0045] When angle α is in the fourth quadrant, x' s (n) is calculated as follows:
[0046]
[0047] Where j represents the imaginary unit, n represents the time domain sequence index, T s Indicates the sampling interval.
[0048] The step S5 is specifically as follows:
[0049] First, calculate the cutoff frequency parameter ω of the anti-aliasing filter according to the following formula: c :
[0050]
[0051] Among them, f m Indicates the frequency domain analysis length, f s Indicates the maximum frequency domain analysis length;
[0052] Then calculate the downsampling factor D according to the following formula:
[0053]
[0054] Among them, the symbol means rounding down, max(·) means taking the maximum value of a set of numbers;
[0055] Determine whether the downsampling multiple D is equal to 1. If yes, skip the anti-aliasing filtering and downsampling process and directly proceed to step S6. If no, continue to perform the anti-aliasing filtering and downsampling process.
[0056] Assume the length of the anti-aliasing filter is N l , N l is an odd number, and the cutoff angular frequency is ω c , it is necessary to perform advance shift processing on the anti-aliasing filter so that the symmetry center of the filter is placed at time 0, that is, at the corresponding sequence index n=0, to ensure that the anti-aliasing filter does not cause additional delay to the signal. Let the anti-aliasing filter after the above adjustment be L(n);
[0057] Next, the synthetic signal x' s (n) Perform anti-aliasing filtering to obtain the filtered signal x' sL (n) is shown in the following formula:
[0058]
[0059] in, Represents circular convolution, and the corresponding cycle period is the total number of analysis points pN s , the filtered signal x' sL The length of (n) remains the total number of analysis points pN s unchanged, time domain sequence index n=0,1,2,...,pN s -1;
[0060] Then the filtered signal x' sL (n) Downsampling is performed, and the downsampled signal x' sLD (n) is expressed as:
[0061] x' sLD (n) = x' sL (nD) (1-17)
[0062] Where D represents the downsampling multiple. After downsampling, the value range of the time domain sequence index n is shortened and its value becomes
[0063] The step S6 is specifically as follows:
[0064] The preliminary result of fractional Fourier transform is obtained as follows:
[0065]
[0066] Where j represents the imaginary unit, u s represents the discretization step size of the u axis, T s represents the sampling interval, D represents the downsampling multiple, k' and k are the transform domain sequence indexes before and after correction, respectively. When the angle α is in the first quadrant, k = k'-k0, and when the angle α is in the fourth quadrant, k = k'+k0. Its value is The u-axis discretization step length u in the formula s Obtained by the following formula:
[0067]
[0068] Among them, p represents the point expansion multiple, Δf m represents the frequency analysis step. Note that if the S5 step is not performed, the signal x' in equation (1-18) sLD (n) is x' sL (n).
[0069] The step S7 is specifically as follows:
[0070] First, determine the preliminary result X' of the fractional Fourier transform according to the angle α α (ku s ) whether it needs to be negated. If the angle α is in the first quadrant, it does not need to be negated. If the angle α is in the fourth quadrant, the preliminary result X' of the fractional Fourier transform needs to be α (ku s ) performs the inversion operation along the u axis, as follows:
[0071]
[0072] Among them, pNs is the total number of analysis points, and D is the downsampling multiple;
[0073] Then for X″ α (ku s ) Continue to do downsampling processing, the downsampling multiple is the point expansion multiple p, which is performed as follows:
[0074] X α (ku s )=X″ α (kpu s ) (1-21)
[0075] Finally, the result of equation (1-21) is X α (ku s ) to perform truncation and extract the first N m points, so far, the final fractional Fourier transform result X is obtained α (ku s ), where the modified transform domain sequence index k is: k = 0, 1, 2, ..., N m -1.
[0076] A high-precision and fast implementation system of fractional Fourier transform for digital signal processing, the processing principle of which is based on the high-precision and fast implementation method of fractional Fourier transform for digital signal processing, including:
[0077] The synthetic signal generation module determines the analysis range of the angle α of the fractional Fourier transform and the total size T of the delay analysis range of the signal to be analyzed according to the actual signal to be processed and the actual needs. m , and the delay analysis step Δt m , and then according to the angle α range and the total size of the delay analysis range T m Time delay analysis step Δt m , calculate the number of extended analysis points N corresponding to the maximum frequency domain range s And the point expansion multiple p, and the discretized input signal x s (n) Perform zero padding and then calculate the digital angular frequency resolution interval And the index value k0 of the starting position of the u axis of the fractional Fourier transform; finally, the interval is resolved according to the digital angular frequency Index value k0, and discretized input signal x s (n) Calculate the synthetic signal x' s (n);
[0078] Anti-aliasing filtering and downsampling module, for the synthetic signal x' s (n) Perform anti-aliasing filtering and downsampling to obtain the downsampled signal x' sLD (n);
[0079] Fast Fourier Transform (FFT) module, for the downsampled signal x' sLD (n) Perform a fast Fourier transform (FFT) operation to obtain the preliminary result X' of the fractional Fourier transform α (ku s );
[0080] Fractional Fourier transform post-processing module, the preliminary result X' of fractional Fourier transform α (ku s ) performs inversion, downsampling, and truncation operations to obtain the final fractional Fourier transform result X α (ku s ).
[0081] A high-precision and fast implementation device of fractional Fourier transform for digital signal processing, comprising:
[0082] Memory: used for storing a computer program for implementing the method for high-precision and fast implementation of fractional Fourier transform for digital signal processing;
[0083] Processor: used to implement the high-precision and fast implementation method of fractional Fourier transform for digital signal processing when executing the computer program.
[0084] Compared with the prior art, the beneficial effects of the present invention are:
[0085] 1. In the process of calculating the fractional Fourier transform, the present invention is based on the discretized input signal x to be processed s (n) Calculate the synthetic signal x' s (n), compared to directly using x s (n) This processing method realizes the narrowband or single-frequency processing of the input signal. For the field of broadband or ultra-wideband signal processing, this processing method avoids large number calculations in the digital calculation process, making the fractional Fourier transform result more accurate.
[0086] 2. The present invention achieves a significant reduction in data volume without loss of data information through a series of operations including anti-aliasing filtering, downsampling and fast Fourier transform (FFT), which has the effect of faster calculation.
[0087] 3. The present invention uses the modified transform domain sequence index k as the final fractional Fourier transform result X α (ku s ) so that the starting point index of the fractional Fourier transform just corresponds to the starting point of the delay analysis range of the input signal to be analyzed, which has the characteristic of consistent corresponding indexes before and after the signal transformation, and is convenient for practical use.
[0088] 4. The present invention calculates parameters such as the number of expansion points in the frequency domain range and the angular frequency resolution interval according to the input signal delay analysis range and delay analysis step size of the design requirements. The design process and results strictly meet the design requirements without the need for additional subsequent processing, thereby increasing the flexibility of the design.
[0089] In summary, the present invention has the characteristics of high calculation accuracy, fast calculation, and convenient and flexible use. BRIEF DESCRIPTION OF THE DRAWINGS
[0090] Figure 1 This is a flow chart of a method for high-precision and rapid implementation of fractional Fourier transform for digital signal processing described in the present invention.
[0091] Figure 2 This is a diagram of the fractional Fourier transform simulation results when the search angle of the present invention is the first.
[0092] Figure 3 This is a diagram of the fractional Fourier transform simulation results when the search angle of the present invention is the 21st.
[0093] Figure 4 This is a diagram of the simulation results of the α-u domain two-dimensional distribution of the fractional Fourier transform at all search angles of the prior art.
[0094] Figure 5 A diagram of simulation results of the α-u domain two-dimensional distribution of fractional Fourier transform at all search angles of the present invention. DETAILED DESCRIPTION
[0095] The following is a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0096] The following is a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0097] The definition of the fractional Fourier transform (FrFT) targeted by the present invention is as follows:
[0098]
[0099] Among them, x(t) is the signal to be analyzed, t is the time variable, α is the angle, and u can be called the displacement factor. α and u together form a two-dimensional transformation domain, K α (t,u) is the transformation kernel of FrFT, and its expression is:
[0100]
[0101] Where δ(t) represents a unit impulse signal, l is an integer, and j is an imaginary unit. Note that in formula (1-2), the cases of α = 2lπ and α = (2l+1)π are only to ensure the rigor of the fractional Fourier transform expression. In practical applications, the value range of α is usually set to: That is, α is in the first or fourth quadrant of the two-dimensional time-frequency domain. The case where α is in the second or third quadrant of the time-frequency domain can be easily equivalent to the first or fourth quadrant, so it will not be discussed here.
[0102] The discretization form of formula (1-1) is:
[0103]
[0104] Where n is the index of the discrete sequence in the time domain, and its value is n = 0, 1, 2, ..., N-1, where N is the discretized input signal x s (n) The total number of sampling points; u s is the discretization step size of the displacement factor u, then u can be discretized as k'u s , where k' is the transform domain sequence index number, and its value is related to the actual analysis range of u; the transform kernel K α (nT s ,ku s ) is expressed as:
[0105]
[0106] The cases of α=2lπ and α=(2l+1)π are ignored. The discretized signal x in (1-3) is s (n) is the sampling form of the signal x(t). Without loss of generality, let x(t) be the intermediate frequency signal after preprocessing such as down-conversion, and its starting time is time 0, then x s (n) can be expressed as:
[0107]
[0108] Where, T s is the sampling interval, f s is the sampling rate, and the two are reciprocals of each other.
[0109] like Figure 1As shown, a high-precision and fast implementation method of fractional Fourier transform for digital signal processing comprises the following steps:
[0110] S1, based on the actual discretized input signal x s (n) Situation and actual needs, determine the analysis range of the fractional Fourier transform angle α, discretize the input signal x s The total delay analysis range of (n) is T m , and the delay analysis step Δt m ;
[0111] S2, according to the angle α range and the total size T of the delay analysis range obtained in step S1 m Time delay analysis step Δt m , calculate the number of extended analysis points N corresponding to the maximum frequency domain range s And the point expansion multiple p, and the discretized input signal x s (n) performing zero padding processing;
[0112] S3, calculate the digital angular frequency resolution interval and the index value k0 of the starting position of the u-axis of the fractional Fourier transform;
[0113] S4, according to the digital angular frequency resolution interval obtained in step S3 The index value k0 and the discretized input signal x obtained in step S2 s (n) Calculate the synthetic signal x' s (n);
[0114] S5, the synthesized signal x' obtained in step S4 s (n) Perform anti-aliasing filtering and downsampling to obtain the downsampled signal x' sLD (n);
[0115] S6, the down-sampled signal x' obtained in step S5 sLD (n) Perform a fast Fourier transform (FFT) operation to obtain the preliminary result X' of the fractional Fourier transform α (ku s );
[0116] S7, the preliminary result X' of the fractional Fourier transform obtained in step S6 α (ku s ) performs inversion, downsampling, and truncation operations to obtain the final fractional Fourier transform result X α (ku s ).
[0117] The step S1 is specifically as follows:
[0118] For the discretized input signal xs (n) is the case where the instantaneous frequency decreases linearly with time. The analysis range of angle α should be in the interval Select from the discretized input signal x s (n) is the case where the instantaneous frequency increases linearly with time. The analysis range of angle α should be in the interval Select from; set the discretized input signal x s The delay analysis range of (n) is [0,T m ], where T m Indicates the total size of the delay analysis range, and the delay analysis step is Δt m , the total number of delay analysis points can be obtained as:
[0119]
[0120] Where int(·) indicates rounding operation; for the case where the starting time of the delay analysis range is not zero, the discretized input signal x s (n) Perform advance or delay processing to align the analysis start time to time zero.
[0121] The step S2 is specifically as follows:
[0122] First, find the delay analysis step length Δt m The corresponding frequency analysis step Δf m , and the total size of the delay analysis range T m The corresponding frequency domain analysis length f m , solve according to the following formula:
[0123] Δf m =Δt m |cotα| (1-7)
[0124] f m =T m |cotα| (1-8)
[0125] Determine the frequency domain analysis length f m Is it less than or equal to the maximum frequency domain analysis length f? s , that is, the sampling rate. If yes, continue to execute below. If no, return to equation (1-7) and equation (1-8) to reset the total size of the delay analysis range T m or angle α;
[0126] The number of extended analysis points N corresponding to the maximum frequency domain range is calculated according to the following formula: s :
[0127]
[0128] Among them, N s ≥Nm , N m is the total number of delay analysis points, and the point expansion factor p represents the total number of input signal points N compared to the number of analysis points N in the maximum frequency domain range. s The multiple of is calculated as follows:
[0129]
[0130] Among them, the symbol Indicates rounding up, N is the discretized input signal x s (n) The total number of sampling points; after obtaining the point expansion factor p, the total number of analysis points corresponding to the maximum frequency domain range is pN s ;
[0131] Then, according to the total number of analysis points pN s , for the discretized input signal x s (n) is filled with zeros, that is, in x s (n) Add pN at the end s -N zeros, so that the total number of points of the signal after zero padding is just pN s To simplify the expression, let the discretized input signal after zero padding still be represented as x s (n), the time domain sequence index n takes values of 0, 1, 2, ..., pN s -1.
[0132] The step S3 is specifically as follows:
[0133] Digital angular frequency resolution Directly obtained by the following formula:
[0134]
[0135] Among them, pN s is the total analysis points;
[0136] For the calculation of the index value k0 of the starting position of the u-axis of the fractional Fourier transform, it is necessary to discuss the situation according to the time-frequency domain quadrant where the angle α is located:
[0137] When the angle α is in the first quadrant of the time-frequency domain, the index value k0 is calculated as follows:
[0138]
[0139] Among them, f bm Represents the signal x to be analyzed s (n) The intermediate frequency starting frequency, f bm Satisfy the relationship: f bm <f s , f s Indicates the maximum frequency domain analysis length;
[0140] When the angle α is in the fourth quadrant of the time-frequency domain, the index value k0 can be calculated as follows:
[0141]
[0142] Among them, p represents the point expansion multiple, Δf m Represents the frequency analysis step size.
[0143] The step S4 is specifically as follows:
[0144] When angle α is in the first quadrant, x' s (n) is calculated as follows:
[0145]
[0146] When angle α is in the fourth quadrant, x' s (n) is calculated as follows:
[0147]
[0148] Where j represents the imaginary unit, n represents the time domain sequence index, T s Indicates the sampling interval.
[0149] The step S5 is specifically as follows:
[0150] First, calculate the cutoff frequency parameter ω of the anti-aliasing filter according to the following formula: c :
[0151]
[0152] Among them, f m Indicates the frequency domain analysis length, f s Indicates the maximum frequency domain analysis length;
[0153] Then calculate the downsampling factor D according to the following formula:
[0154]
[0155] Among them, the symbol means rounding down, max(·) means taking the maximum value of a set of numbers;
[0156] Determine whether the downsampling multiple D is equal to 1. If yes, skip the following anti-aliasing filtering and downsampling processing and directly proceed to step S6. If no, continue to perform the following anti-aliasing filtering and downsampling processing;
[0157] The anti-aliasing filter should be a linear phase FIR low-pass filter designed according to the known technology, and the filter length should be set to an odd number, and the filter cutoff angular frequency should be set to ω of the formula (1-14)c ; In this embodiment, a linear phase window function low-pass filter is used as an anti-aliasing filter;
[0158] Assume the length of the designed anti-aliasing filter is N l , N l is an odd number, and the cutoff angular frequency is ω c , it is necessary to perform advance shift processing on the anti-aliasing filter so that the symmetry center of the filter is placed at time 0, that is, at the corresponding sequence index n=0, to ensure that the anti-aliasing filter does not cause additional delay to the signal. Let the anti-aliasing filter after the above adjustment be L(n);
[0159] Next, the synthetic signal x' s (n) Perform anti-aliasing filtering to obtain the filtered signal x' sL (n) is shown in the following formula:
[0160]
[0161] in, Represents circular convolution, and the corresponding cycle period is the total number of analysis points pN s , the filtered signal x' sL The length of (n) remains the total number of analysis points pN s unchanged, time domain sequence index n=0,1,2,...,pN s -1;
[0162] Then the filtered signal x' sL (n) Downsampling is performed, and the downsampled signal x' sLD (n) is expressed as:
[0163] x' sLD (n) = x' sL (nD) (1-17)
[0164] Where D represents the downsampling multiple. After downsampling, the value range of the time domain sequence index n is shortened and its value becomes
[0165] The step S6 is specifically as follows:
[0166] The preliminary result of fractional Fourier transform is obtained as follows:
[0167]
[0168] Where j represents the imaginary unit, u s represents the discretization step size of the u-axis, T srepresents the sampling interval, D represents the downsampling multiple, k' and k are the transform domain sequence indexes before and after correction, respectively. When the angle α is in the first quadrant, k = k'-k0, and when the angle α is in the fourth quadrant, k = k'+k0. Its value is The u-axis discretization step length u in the formula s Obtained by the following formula:
[0169]
[0170] Among them, p represents the point expansion multiple, Δf m represents the frequency analysis step length. Note that if step S5 is not performed (corresponding to D=1), the signal x' in equation (1-18) is sLD (n) is x' sL (n).
[0171] The step S7 is specifically as follows:
[0172] First, determine the preliminary result X' of the fractional Fourier transform according to the angle α α (ku s ) whether it needs to be negated. If the angle α is in the first quadrant, it does not need to be negated. If the angle α is in the fourth quadrant, the preliminary result X' of the fractional Fourier transform needs to be α (ku s ) performs the inversion operation along the u axis, as follows:
[0173]
[0174] Among them, pN s is the total number of analysis points, and D is the downsampling multiple;
[0175] Then for X″ α (ku s ) Continue to do downsampling processing, the downsampling multiple is the point expansion multiple p, which is performed as follows:
[0176] X α (ku s )=X″ α (kpu s ) (1-21)
[0177] Finally, the result of equation (1-21) is X α (ku s ) to perform truncation and extract the first N m points, so far, the final fractional Fourier transform result X is obtained α (ku s ), where the modified transform domain sequence index k is: k = 0, 1, 2, ..., N m -1.
[0178] In principle, the present invention decomposes the fractional Fourier transform expression in a completely new form, and obtains a new decomposition form different from the prior art:
[0179]
[0180] From this expression, it can be seen that the superposition and partial Only one fast Fourier transform (FFT) operation is needed to obtain the fractional Fourier transform result of a signal at an angle α, without the need to perform multiple fast Fourier transform (FFT) and inverse fast Fourier transform (IFFT) operations as in the prior art. However, if the FFT operation is directly performed on the superposition and part, the number of points of the operation is pN s , too many points, not efficient. Therefore, the present invention continues to synthesize the signal x' s (n) Anti-aliasing filtering and downsampling are performed to further reduce the number of signal points. Then, FFT operation is performed on the processed synthetic signal, which greatly improves the calculation efficiency.
[0181] On the other hand, it can be analyzed that the synthetic signal x' in the present invention is s (n) (its form is detailed in equations (1-13a) and (1-13b)) relative to the original signal x s (n) It realizes narrowband or single frequency, and can achieve higher precision calculation in actual processing. In contrast, the prior art usually involves mutual operation between two or more synthetic signals or constructed signals in the form of broadband linear frequency modulation, which will inevitably produce errors in the actual numerical calculation process, and the wider the frequency band and the longer the duration of the signal to be processed, the greater the processing error. This reflects the advantage of the present invention in the field of broadband or ultra-wideband signal processing with high calculation precision.
[0182] A high-precision and fast implementation system of fractional Fourier transform for digital signal processing, the processing principle of which is based on the high-precision and fast implementation method of fractional Fourier transform for digital signal processing, including:
[0183] The synthetic signal generation module determines the analysis range of the angle α of the fractional Fourier transform and the total size T of the delay analysis range of the signal to be analyzed according to the actual signal to be processed and the actual needs. m , and the delay analysis step Δt m , and then according to the angle α range and the total size of the delay analysis range T m Time delay analysis step Δt m , calculate the number of extended analysis points N corresponding to the maximum frequency domain range s And the point expansion multiple p, and the discretized input signal x s(n) Perform zero padding and then calculate the digital angular frequency resolution interval And the index value k0 of the starting position of the u axis of the fractional Fourier transform; finally, the interval is resolved according to the digital angular frequency Index value k0, and discretized input signal x s (n) Calculate the synthetic signal x' s (n), steps S1 to S4 for implementing a high-precision and fast implementation method of fractional Fourier transform for digital signal processing;
[0184] Anti-aliasing filtering and downsampling module, for the synthetic signal x' s (n) Perform anti-aliasing filtering and downsampling to obtain the downsampled signal x' sLD (n), step S5 for implementing a high-precision and fast implementation method of fractional Fourier transform for digital signal processing;
[0185] Fast Fourier Transform (FFT) module, for the downsampled signal x' sLD (n) Perform a fast Fourier transform (FFT) operation to obtain the preliminary result X' of the fractional Fourier transform α (ku s ), used to implement step S6 of a high-precision and fast implementation method of fractional Fourier transform for digital signal processing;
[0186] Fractional Fourier transform post-processing module, the preliminary result X' of fractional Fourier transform α (ku s ) performs inversion, downsampling, and truncation operations to obtain the final fractional Fourier transform result X α (ku s ), used to implement step S7 of a high-precision and fast implementation method of fractional Fourier transform for digital signal processing.
[0187] A high-precision and fast implementation device of fractional Fourier transform for digital signal processing, comprising:
[0188] Memory: used to store a computer program for implementing a high-precision and fast implementation method of fractional Fourier transform for digital signal processing as claimed in any one of claims 1 to 8;
[0189] Processor: used to implement a high-precision and fast method for implementing fractional Fourier transform for digital signal processing as described in any one of claims 1 to 8 when executing the computer program.
[0190] The present invention provides a high-precision and fast calculation method for fractional Fourier transform, which is most commonly used in signal processing fields or systems, specifically involving radar signal processing, radio frequency signal processing, digital signal processing and other specific fields.
[0191] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.
[0192] The effect of the present invention can be further illustrated by the following simulation:
[0193] 1. Simulation conditions
[0194] Assume that the signal to be analyzed is a broadband linear frequency modulation pulse signal with a pulse width of 0.5 ms and a bandwidth of 500 MHz. The modulation method in which the frequency increases linearly with time is adopted. The angle α in the time-frequency domain is in the fourth quadrant. The input signal-to-noise ratio is -20 dB, the intermediate frequency sampling rate is set to 1.2 GHz, and the total number of sampling points of the signal to be analyzed is 8×10 5 , the total delay analysis size is 66.7us, the delay analysis step is 1.5ns, and the total number of search angles is set to 60, among which the real angle α corresponding to the signal to be analyzed is the 21st in the search angle sequence. Simulation computer configuration: CPU 3.8GHz / 24 cores, memory 16GB. Other parameters that have no substantial impact on the simulation results are not listed here.
[0195] 2. Simulation content
[0196] Under the above simulation conditions, the high-precision and fast implementation method of fractional Fourier transform for digital signal processing described in the present invention and the prior art are used to perform fractional Fourier transform processing on the analyzed signal, and the fractional Fourier transform results are recorded as follows: Figure 2-5 The simulation running time results are shown in Table 1.
[0197] Table 1. Comparison of operation time of the prior art and the method of the present invention
[0198] category Prior art Method of the present invention Calculation time (seconds) 52.37 12.28
[0199] Figure 2 and Figure 3 The fractional Fourier transform results when the angle is not completely matched and when the angle is completely matched are given respectively. From the results, it can be seen that the peak energy corresponding to the high-precision and fast implementation method of fractional Fourier transform for digital signal processing described in the present invention is more concentrated and the amplitude is higher, while the peak energy corresponding to the prior art is more divergent and the amplitude is slightly lower. This shows that the calculation accuracy of the high-precision and fast implementation method of fractional Fourier transform for digital signal processing described in the present invention is higher and a higher matching correlation peak can be obtained. Figure 4 and Figure 5The two-dimensional distribution result of the fractional Fourier transform further reflects the peak point energy divergence effect caused by the loss of calculation accuracy in the prior art, while the high-precision and fast implementation method of the fractional Fourier transform for digital signal processing described in the present invention realizes high-precision calculation and obtains a correlation peak with more concentrated energy.
[0200] Table 1 gives the total time consumption of fractional Fourier transform under 60 search angles. It can be clearly seen that the time consumption of the high-precision and fast implementation method of fractional Fourier transform for digital signal processing described in the present invention is 12.28s, which is much less than 52.37s in the prior art, reflecting the high efficiency of the high-precision and fast implementation method of fractional Fourier transform for digital signal processing described in the present invention in realizing fast calculation of fractional Fourier transform.
[0201] It should be noted that the signal to be analyzed in this simulation example is a high-bandwidth, high-pulse-width signal, and the existing technology will have obvious calculation accuracy loss and high time consumption, which is also the main application scenario in which the advantages of the high-precision and fast implementation method of fractional Fourier transform for digital signal processing described in the present invention are reflected. For other application scenarios, the quantitative conclusions of this simulation example are not applicable.
Claims
1. A high-precision and fast implementation method of fractional Fourier transform for digital signal processing, characterized in that: The following steps are involved: S1, based on the actual discretized input signal x s (n) Situation and actual needs, determine the analysis range of the fractional Fourier transform angle α, discretize the input signal x s The total delay analysis range of (n) is T m , and the delay analysis step Δt m ; S2, according to the angle α range and the total size T of the delay analysis range obtained in step S1 m Time delay analysis step Δt m , calculate the number of extended analysis points N corresponding to the maximum frequency domain range s And the point expansion multiple p, and the discretized input signal x s (n) performing zero padding processing; S3, calculate the digital angular frequency resolution interval and the index value k0 of the starting position of the u-axis of the fractional Fourier transform; S4, according to the digital angular frequency resolution interval obtained in step S3 The index value k0 and the discretized input signal x obtained in step S2 s (n) Calculate the synthetic signal x' s (n); S5, the synthesized signal x' obtained in step S4 s (n) Perform anti-aliasing filtering and downsampling to obtain the downsampled signal x' sLD (n); S6, the down-sampled signal x' obtained in step S5 sLD (n) Perform a fast Fourier transform (FFT) operation to obtain the preliminary result X' of the fractional Fourier transform α (ku s ); S7, the preliminary result X' of the fractional Fourier transform obtained in step S6 α (ku s ) performs inversion, downsampling, and truncation operations to obtain the final fractional Fourier transform result X α (ku s ).
2. The high-precision and fast implementation method of fractional Fourier transform for digital signal processing according to claim 1, characterized in that: The step S1 is specifically as follows: For the discretized input signal x s (n) is the case where the instantaneous frequency decreases linearly with time. The analysis range of angle α should be in the interval Select from the discretized input signal x s (n) is the case where the instantaneous frequency increases linearly with time. The analysis range of angle α should be in the interval Select from; set the discretized input signal x s The delay analysis range of (n) is [0,T m ], where T m Indicates the total size of the delay analysis range, and the delay analysis step is Δt m , the total number of delay analysis points can be obtained as: Where int(·) indicates rounding operation; for the case where the starting time of the delay analysis range is not zero, the discretized input signal x s (n) Perform advance or delay processing to align the analysis start time to time zero.
3. The high-precision and fast implementation method of fractional Fourier transform for digital signal processing according to claim 1, characterized in that: The step S2 is specifically as follows: First, find the delay analysis step length Δt m The corresponding frequency analysis step Δf m , and the total size of the delay analysis range T m The corresponding frequency domain analysis length f m , solve according to the following formula: Δf m =Δt m |cotα| (1-7) f m =T m |cotα| (1-8) Determine the frequency domain analysis length f m Is it less than or equal to the maximum frequency domain analysis length f? s , that is, the sampling rate. If yes, continue to execute below. If no, return to equation (1-7) and equation (1-8) to reset the total size of the delay analysis range T m or angle α; The number of extended analysis points N corresponding to the maximum frequency domain range is calculated according to the following formula: s : Among them, N s ≥N m , N m is the total number of delay analysis points, and the point expansion factor p represents the total number of input signal points N compared to the number of analysis points N in the maximum frequency domain range. s The multiple of is calculated as follows: Among them, the symbol Indicates rounding up, N is the discretized input signal x s (n) The total number of sampling points; after obtaining the point expansion factor p, the total number of analysis points corresponding to the maximum frequency domain range is pN s ; Then, according to the total number of analysis points pN s , for the discretized input signal x s (n) is filled with zeros, that is, in x s (n) Add pN at the end s -N zeros, so that the total number of points of the signal after zero padding is just pN s To simplify the expression, let the discretized input signal after zero padding still be represented as x s (n), the time domain sequence index n takes values of 0, 1, 2, ..., pN s -1.
4. The high-precision and fast implementation method of fractional Fourier transform for digital signal processing according to claim 1, characterized in that: The step S3 is specifically as follows: Digital angular frequency resolution Directly obtained by the following formula: Among them, pN s is the total analysis points; When the angle α is in the first quadrant of the time-frequency domain, the index value k0 is calculated as follows: Among them, f bm Represents the signal x to be analyzed s (n) The intermediate frequency starting frequency, f bm Satisfy the relationship: f bm <f s , f s Indicates the maximum frequency domain analysis length; When the angle α is in the fourth quadrant of the time-frequency domain, the index value k0 can be calculated as follows: Among them, p represents the point expansion multiple, Δf m Represents the frequency analysis step size.
5. The high-precision and fast implementation method of fractional Fourier transform for digital signal processing according to claim 1, characterized in that: The step S4 is specifically as follows: When angle α is in the first quadrant, x' s (n) is calculated as follows: When angle α is in the fourth quadrant, x' s (n) is calculated as follows: Where j represents the imaginary unit, n represents the time domain sequence index, T s Indicates the sampling interval.
6. The high-precision and fast implementation method of fractional Fourier transform for digital signal processing according to claim 1, characterized in that: The step S5 is specifically as follows: First, calculate the cutoff frequency parameter ω of the anti-aliasing filter according to the following formula: c : Among them, f m Indicates the frequency domain analysis length, f s Indicates the maximum frequency domain analysis length; Then calculate the downsampling factor D according to the following formula: Among them, the symbol means rounding down, max(·) means taking the maximum value of a set of numbers; Determine whether the downsampling multiple D is equal to 1. If yes, skip the anti-aliasing filtering and downsampling process and directly proceed to step S6. If no, continue to perform the anti-aliasing filtering and downsampling process. Assume the length of the anti-aliasing filter is N l , N l is an odd number, and the cutoff angular frequency is ω c , it is necessary to perform advance shift processing on the anti-aliasing filter so that the symmetry center of the filter is placed at time 0, that is, at the corresponding sequence index n=0, to ensure that the anti-aliasing filter does not cause additional delay to the signal. Let the anti-aliasing filter after the above adjustment be L(n); Next, the synthetic signal x' s (n) Perform anti-aliasing filtering to obtain the filtered signal x' sL (n) is shown in the following formula: in, Represents circular convolution, and the corresponding cycle period is the total number of analysis points pN s , the filtered signal x' sL The length of (n) remains the total number of analysis points pN s unchanged, time domain sequence index n=0,1,2,...,pN s -1; Then the filtered signal x' sL (n) Downsampling is performed, and the downsampled signal x' sLD (n) is expressed as: x' sLD (n)=x' sL (nD) (1-17) Where D represents the downsampling multiple. After downsampling, the value range of the time domain sequence index n is shortened and its value becomes 7. The high-precision and fast implementation method of fractional Fourier transform for digital signal processing according to claim 1, characterized in that: The step S6 is specifically as follows: The preliminary result of fractional Fourier transform is obtained as follows: Where j represents the imaginary unit, u s represents the discretization step size of the u axis, T s represents the sampling interval, D represents the downsampling multiple, k' and k are the transform domain sequence indexes before and after correction, respectively. When the angle α is in the first quadrant, k = k'-k0, and when the angle α is in the fourth quadrant, k = k'+k0. Its value is The u-axis discretization step length u in the formula s Obtained by the following formula: Among them, p represents the point expansion multiple, Δf m represents the frequency analysis step. Note that if the S5 step is not performed, the signal x' in equation (1-18) sLD (n) is x' sL (n).
8. The high-precision and fast implementation method of fractional Fourier transform for digital signal processing according to claim 1, characterized in that: The step S7 is specifically as follows: First, determine the preliminary result X' of the fractional Fourier transform according to the angle α α (ku s ) whether it needs to be negated. If the angle α is in the first quadrant, it does not need to be negated. If the angle α is in the fourth quadrant, the preliminary result X' of the fractional Fourier transform needs to be α (ku s ) performs the inversion operation along the u axis, as follows: Among them, pN s is the total number of analysis points, and D is the downsampling multiple; Then to X" α (ku s ) Continue to do downsampling processing, the downsampling multiple is the point expansion multiple p, which is performed as follows: X α (stand s )=X” α (cap s ) (1-21) Finally, the result of equation (1-21) is X α (ku s ) to perform truncation and extract the first N m points, so far, the final fractional Fourier transform result X is obtained α (ku s ), where the modified transform domain sequence index k is: k = 0, 1, 2, ..., N m -1.
9. A high-precision and fast implementation system of fractional Fourier transform for digital signal processing, the processing principle of which adopts a high-precision and fast implementation method of fractional Fourier transform for digital signal processing according to any one of claims 1 to 8, characterized in that: include: The synthetic signal generation module determines the analysis range of the angle α of the fractional Fourier transform and the total size T of the delay analysis range of the signal to be analyzed according to the actual signal to be processed and the actual needs. m , and the delay analysis step Δt m , and then according to the angle α range and the total size of the delay analysis range T m Time delay analysis step Δt m , calculate the number of extended analysis points N corresponding to the maximum frequency domain range s And the point expansion multiple p, and discretize the input signal x s (n) Perform zero padding and then calculate the digital angular frequency resolution interval And the index value k0 of the starting position of the u axis of the fractional Fourier transform; finally, the interval is resolved according to the digital angular frequency Index value k0, and discretized input signal x s (n) Calculate the synthetic signal x' s (n); Anti-aliasing filtering and downsampling module, for the synthetic signal x' s (n) Perform anti-aliasing filtering and downsampling to obtain the downsampled signal x' sLD (n); Fast Fourier Transform (FFT) module, for the downsampled signal x' sLD (n) Perform a fast Fourier transform (FFT) operation to obtain the preliminary result X' of the fractional Fourier transform α (ku s ); Fractional Fourier transform post-processing module, the preliminary result X' of fractional Fourier transform α (ku s ) performs inversion, downsampling, and truncation operations to obtain the final fractional Fourier transform result X α (ku s ).
10. A high-precision and fast implementation device of fractional Fourier transform for digital signal processing, characterized in that: include: Memory: used to store a computer program for implementing a high-precision and fast implementation method of fractional Fourier transform for digital signal processing as claimed in any one of claims 1 to 8; Processor: used to implement a high-precision and fast method for implementing fractional Fourier transform for digital signal processing as described in any one of claims 1 to 8 when executing the computer program.
Citation Information
Patent Citations
Anti-aliasing filtering method and device, and programmable logic device
CN108199998A
Adaptive time-frequency analysis method combining short-time Fourier transform and WVD
CN114564682A
High-precision time-varying spectrum analysis method based on fractional order S transformation
CN117110709A
Synchronous channelized extraction method, device and equipment for multiple different signals and medium
CN117251717A
Frequency analysis device and method
JP2988914B1
