Direct sequence spread spectrum underwater acoustic communication Doppler fast estimation method and system based on GPU

Abstract

The invention discloses a direct sequence spread spectrum underwater acoustic communication Doppler rapid estimation method and system based on a GPU. The method comprises the following steps: modeling a received signal of a direct sequence spread spectrum underwater acoustic communication system; performing resampling and FFT operation on a local pilot signal in sequence, and storing an FFT operation result to obtain a Doppler matching matrix based on an equipment memory preloading method of a GPU (Graphics Processing Unit); a two-step Doppler estimation process is adopted, in the rough search stage, cross ambiguity function calculation is carried out based on GPU complex number multiplication, reduced module value normalization calculation is carried out based on GPU parallel, and a rough Doppler factor to be estimated is obtained through threshold detection; in the fine search stage, a plurality of Doppler branches near the coarse Doppler factor are selected, cross ambiguity function calculation is carried out based on GPU complex number multiplication, reduced module value normalization calculation is carried out based on GPU parallel, and a fine Doppler factor to be estimated is obtained through threshold detection.

Description

Technical Field

[0001] The present invention belongs to the intersection of underwater acoustic communication physical layer technology and GPU scientific computing, and specifically relates to a GPU-based direct sequence spread spectrum underwater acoustic communication Doppler fast estimation method and system. Background Art

[0002] Underwater acoustic communication is the best choice for underwater long-distance wireless communication, but the complex characteristics of underwater acoustic channels such as time-varying, frequency-varying, and space-varying bring many challenges to achieving reliable communication. Sound waves propagate relatively slowly in water, and the fluctuations of water flow and the relative movement of the signal transmitter and receiver will cause obvious Doppler frequency deviation to the underwater acoustic signal, which has a great impact on the realization of highly reliable underwater acoustic communication. Therefore, it is particularly important to achieve fast and accurate Doppler estimation for underwater acoustic communication systems.

[0003] At present, the commonly used underwater acoustic communication Doppler estimation is mostly achieved by inserting a known waveform into the communication signal. One method is to insert a Doppler-insensitive waveform before and after the communication signal to form a leading and trailing sequence. The receiving end detects the interval change between the leading and trailing sequences to obtain the average Doppler scale estimate of the entire communication signal. For example, a linear frequency modulation signal (LFM) is used as the leading and trailing sequence of a single-carrier underwater acoustic communication signal to achieve a more accurate Doppler estimation at a lower computational cost. The method of measuring interval changes has a low estimation accuracy and regards the Doppler factor in the entire communication data packet as constant. The Doppler estimation effect is poor in the scenario of variable speed movement of the communication machine.

[0004] Another method is to insert a Doppler-sensitive waveform with a sharp ambiguity function before the communication signal as a preamble sequence. The receiver calculates the cross ambiguity function (CAF) of the preamble sequence by calculating the received signal and the Doppler scaling factor and searches for the maximum branch index of the correlation peak to obtain the Doppler scale estimate. The direct sequence spread spectrum (DSSS) underwater acoustic communication system uses a pseudo-random sequence (PN) for spectrum expansion, where the PN sequence has a relatively sharp ambiguity function. Using the PN sequence as a pilot can ensure the consistency of the DSSS communication signal. Therefore, the DSSS underwater acoustic communication system often uses the CAF method for Doppler estimation. The CAF method can achieve high estimation accuracy and can also achieve good estimation results for Doppler in variable speed motion scenarios under the condition of high-density pilot insertion. However, this method requires multiple searches in the delay-Doppler factor grid, has a relatively dense computational complexity, and will cause considerable computational overhead when implemented in hardware.

[0005] Graphics Processing Unit (GPU) is specially used for highly parallel computing and has been widely used in signal processing and data-intensive computing in recent years.

[0006] In summary, for direct sequence spread spectrum underwater acoustic communication systems, the difficulty of using the cross ambiguity function method for fast Doppler estimation lies in reducing the computational complexity and avoiding the computational overhead during hardware implementation; parallel optimization of the Doppler estimation algorithm and deployment on the GPU platform is a better solution. Summary of the invention

[0007] The purpose of the present invention is to overcome the defects of the prior art and propose a GPU-based direct sequence spread spectrum underwater acoustic communication Doppler fast estimation method and system.

[0008] In view of this, the present invention proposes a GPU-based direct sequence spread spectrum underwater acoustic communication Doppler fast estimation method, the method comprising:

[0009] Step 1) Modeling the received signal of the direct sequence spread spectrum underwater acoustic communication system;

[0010] Step 2) resampling and FFT operation are performed on the local pilot signal in sequence, and the FFT operation result is stored to obtain the Doppler matching matrix based on the device memory preloading method of the GPU;

[0011] Step 3) adopts a two-step Doppler estimation process. In the coarse search stage, the cross ambiguity function is calculated based on GPU complex multiplication, the reduced modulus normalization calculation is performed based on GPU parallelization, and the coarse Doppler factor to be estimated is obtained through threshold detection;

[0012] Step 4) In the fine search stage, multiple Doppler branches near the coarse Doppler factor are selected, the cross ambiguity function is calculated based on GPU complex multiplication, the reduced modulus normalization calculation is performed based on GPU parallelism, and the fine Doppler factor to be estimated is obtained through threshold detection.

[0013] Preferably, the received signal in step 1) Modeled as:

[0014]

[0015] in, is the transmitted signal, N pa is the number of multipath channels, A p (t) and τ p They represent the amplitude and delay of the pth path respectively, α is the Doppler factor, n(t) is the additive noise, and t represents the time domain.

[0016] Preferably, the local pilot signal of step 2) The length is N s T s , the resampling factor is the calibration Doppler factor α j ,j=1,2,...,J-1, J is the number of Doppler factors to be estimated; perform N FFT Point FFT operation, where N FFT is an integer power of 2 and is greater than the number of signal sampling points; the size of the Doppler matching matrix is ​​N FFT ×J.

[0017] Preferably, the step 3) comprises:

[0018] In the coarse search stage, the Doppler branches to be estimated are divided into K groups, each with J branches, the Doppler estimation accuracy is Δα1, and the Doppler search range of each group is greater than Twice the width of the main peak of the fuzziness function; where, It is Q-channel signal;

[0019] The cross ambiguity function is calculated based on GPU complex multiplication, and the reduced modulus normalization calculation is performed based on GPU parallelization to calculate the energy integral of the normalized cross ambiguity function of the kth group of Doppler branches. The threshold detection method is used. If the variance of the kth group is significantly higher than the average variance threshold, the branch with the smallest energy integral in the group is selected. The rough Doppler factor to be estimated is

[0020] Preferably, the step 4) comprises:

[0021] The rough Doppler factor to be estimated Select J′ Doppler branches for fine search with Doppler estimation accuracy Δα2 nearby, calculate cross ambiguity function based on GPU complex multiplication, perform reduced modulus normalization calculation based on GPU parallelization, select the minimum value of the energy integral of the normalized cross ambiguity function of J′ branches, and the branch with the minimum value generates the fine Doppler factor to be estimated.

[0022] Preferably, the method is designed based on a heterogeneous parallel algorithm of a GPU.

[0023] On the other hand, the present invention proposes a GPU-based direct sequence spread spectrum underwater acoustic communication Doppler fast estimation system, comprising:

[0024] A modeling unit, used for modeling a received signal of a direct sequence spread spectrum underwater acoustic communication system;

[0025] The operation storage unit is used to perform resampling and FFT operation on the local pilot signal in sequence, and store the FFT operation result to obtain the Doppler matching matrix based on the device memory preloading method of the GPU;

[0026] A coarse search unit, which adopts a two-step Doppler estimation process. In the coarse search stage, a cross ambiguity function is calculated based on GPU complex multiplication, a reduced modulus normalization calculation is performed based on GPU parallelization, and a coarse Doppler factor to be estimated is obtained through threshold detection; and

[0027] The fine search unit is used to select multiple Doppler branches near the coarse Doppler factor in the fine search stage, calculate the cross ambiguity function based on GPU complex multiplication, perform reduced modulus normalization calculation based on GPU parallelism, and obtain the fine Doppler factor to be estimated through threshold detection.

[0028] Compared with the prior art, the advantages of the present invention are:

[0029] The present invention is aimed at the demand for real-time processing of underwater acoustic communication signals. From the perspective of engineering implementation, the GPU is innovatively applied to the Doppler estimation task of underwater acoustic communication signals. The algorithm is optimized from three aspects: algorithm improvement, computing unit, and GPU platform deployment. The algorithm performance is tested and verified using measured data. The test and verification results show that the present invention is reliable and effective. The present invention greatly improves the processing efficiency of direct sequence spread spectrum underwater acoustic communication receivers, solves the key problem of real-time Doppler estimation of received signals, and provides support for the research and development of underwater acoustic communication systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 is a block diagram of a direct sequence spread spectrum underwater acoustic communication system used in the present invention;

[0031] Figure 2 It is a schematic diagram of the calculation flow of the mutual ambiguity function Doppler estimation method used in the present invention;

[0032] Figure 3 It is a schematic diagram of an implementation method of a GPU-based device memory preloading method proposed in the present invention;

[0033] Figure 4 It is a schematic diagram of the calculation flow of the cross ambiguity function calculation method based on GPU complex multiplication proposed by the present invention;

[0034] Figure 5 It is a schematic diagram of the calculation flow of the module normalization calculation method based on GPU parallel reduction proposed by the present invention;

[0035] Figure 6 It is a schematic diagram of the calculation flow of the GPU-based heterogeneous parallel algorithm design proposed by the present invention;

[0036] Figure 7 is the relative speed of the two communicating parties;

[0037] Figure 8 The results of Doppler estimation using five methods for measured data are shown in Figure 2.

[0038] Fig. 9 The average calculation time and speedup ratio of Doppler estimation under five methods;

[0039] Fig.10 Receiver running time and method speedup ratio under five methods. DETAILED DESCRIPTION

[0040] The present invention proposes a GPU-based direct sequence spread spectrum underwater acoustic communication Doppler fast estimation method, which realizes accurate and fast estimation of the direct sequence spread spectrum underwater acoustic communication signal Doppler.

[0041] In order to achieve the above-mentioned purpose, the present invention designs a fast Doppler estimation method for direct sequence spread spectrum underwater acoustic communication signals on a CPU+GPU heterogeneous platform, the method comprising: an improved two-step Doppler estimation method based on a cross ambiguity function, a device memory preloading method based on GPU, a cross ambiguity function calculation method based on GPU complex multiplication, a modulus normalization calculation method based on GPU parallel reduction, and a heterogeneous parallel algorithm design based on GPU.

[0042] The improved two-step Doppler estimation method based on the cross ambiguity function reduces the number of Doppler estimation branches from the algorithm level, is suitable for GPU platform deployment and improves the execution efficiency of the Doppler algorithm;

[0043] A GPU-based device memory preloading method is used to pre-store the calibration Doppler matching matrix to reduce repeated calculations, avoid frequent data interactions, and reduce the cost of computing and storage resources during hardware implementation;

[0044] The cross ambiguity function calculation method based on GPU complex multiplication is used for the Doppler cross ambiguity function calculation within the algorithm, and the GPU parallel computing advantage is used for optimization to further reduce the calculation time;

[0045] The module value normalization calculation method based on GPU parallel reduction is used to calculate the normalized module value of the cross ambiguity function, mainly using the fast reduction algorithm to improve the calculation efficiency;

[0046] A GPU-based heterogeneous parallel algorithm is designed for deployment of the Doppler fast estimation algorithm on a GPU platform.

[0047] The invention of this paper is verified using measured data, which proves that the method proposed in this paper can achieve accurate and fast estimation of the Doppler of direct sequence spread spectrum underwater acoustic communication signals.

[0048] The technical solution of the present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0049] Example 1

[0050] Embodiment 1 of the present invention proposes a GPU-based direct sequence spread spectrum underwater acoustic communication Doppler fast estimation method, the method comprising:

[0051] Step 1) Modeling the received signal of the direct sequence spread spectrum underwater acoustic communication system;

[0052] Step 2) resampling and FFT operation are performed on the local pilot signal in sequence, and the FFT operation result is stored to obtain the Doppler matching matrix based on the device memory preloading method of the GPU;

[0053] Step 3) adopts a two-step Doppler estimation process. In the coarse search stage, the cross ambiguity function is calculated based on GPU complex multiplication, the modulus normalization calculation of the reduction is performed based on GPU parallelization, and the coarse Doppler factor to be estimated is obtained through threshold detection;

[0054] Step 4) In the fine search stage, multiple Doppler branches near the coarse Doppler factor are selected, the cross ambiguity function is calculated based on GPU complex multiplication, the reduced modulus normalization calculation is performed based on GPU parallelism, and the fine Doppler factor to be estimated is obtained through threshold detection.

[0055] The present invention includes an improved two-step Doppler estimation method based on a cross ambiguity function, a device memory preloading method based on a GPU, a cross ambiguity function calculation method based on GPU complex multiplication, a modulus normalization calculation method based on GPU parallel reduction, and a heterogeneous parallel algorithm design based on GPU. First, a direct sequence spread spectrum underwater acoustic system model applicable to the invention is briefly introduced.

[0056] Direct sequence spread spectrum underwater acoustic communication system Figure 1 As shown in the figure, the signal is generated by IQ modulation, the I channel is used to transmit actual information, and the Q channel transmits a known PN sequence for frame synchronization, Doppler estimation and channel estimation. I (t) is expressed as c is the pulse width of the spread spectrum code, T s is the symbol duration,

[0057] T s =LT c , g(t) is the pulse shaping function.

[0058]

[0059] The Q path signal consists of a s ×L known PN sequence generation, baseband spread spectrum signal S Q (t) is expressed as:

[0060]

[0061] The IQ two-way baseband signal is respectively modulated by the carrier and superimposed to obtain the transmission signal, where f c is the carrier center frequency.

[0062]

[0063] The signal model after passing through the underwater acoustic channel is as follows, where N pa is the number of multipath channels, A p (t) and τ p denote the amplitude and delay of the pth path, α is the Doppler factor, n(t) is the additive noise,

[0064]

[0065] On the improved two-step Doppler estimation method based on cross ambiguity function, Figure 2 The mutual ambiguity function Doppler estimation method is firstly MF The calculation of (α, τ) is implemented in the frequency domain as follows, where and represent Fourier transform and inverse Fourier transform respectively.

[0066]

[0067] The calculation process of the improved two-step Doppler estimation method based on the cross ambiguity function proposed in the present invention is as follows: (1) In the coarse search stage, the Doppler branches to be estimated are divided into K groups, each group has J branches, the Doppler estimation accuracy is Δα1, and the Doppler search range of each group is greater than (2) Calculate the energy integral of the normalized cross ambiguity function of the kth group of Doppler branches, and use the threshold detection method. If the variance of the kth group is significantly higher than the average variance threshold, take the branch with the smallest energy integral in the group. The rough Doppler factor to be estimated is (3) Get the coarse Doppler factor After that, Select J′ Doppler branches with Doppler estimation accuracy Δα2 for fine search, select the minimum value of the energy integral of the normalized cross ambiguity function of J′ branches, and the branch with the minimum value will generate the fine Doppler factor to be estimated.

[0068] The implementation method of GPU-based device memory preloading method is as follows Figure 3 As shown. First, for a length of N s T s Local pilot signal Resample, the resampling factor is the calibration Doppler factor α j ,j=1,2,...,J-1. Among them, the Doppler factor ranges from α1 to α j The increment interval is the fine search accuracy Δα2. FFT Point FFT operation, where N FFT is an integer power of 2 and N FFT >α J ×N s T s . The FFT results are transferred from the host memory to the device memory in sequence, and the size is N FFT ×J Doppler matching matrix.

[0069] The calculation flow diagram of the cross ambiguity function calculation method based on GPU complex multiplication is as follows Figure 4 shown.

[0070] The calculation flow diagram of the module normalization calculation method based on GPU parallel reduction is as follows Figure 5 shown.

[0071] The schematic diagram of the calculation process of heterogeneous parallel algorithm design based on GPU is as follows Figure 6 shown.

[0072] Figure 6 It is mainly divided into two parts: coarse search and fine search. To receive the signal, is the resampled local pilot signal, and the resampling factor is α k,j , F[·] represents FFT operation, and J is the number of coarse search Doppler branches. "CAF normalized energy calculation" calculates the CAF normalized energy r_MF_NORM and ss_MF_NORM in the kth group, and Kernel_variance is the kernel function used to calculate the variance v of r_MF_NORM k When v k Greater than the threshold v threshold When , the kernel function Kernel_min calculates the ss_MF_NORM minimum index to obtain the rough search Doppler factor In the fine search stage, J' is the number of fine search Doppler branches, "CAF normalized energy calculation" calculates the CAF normalized energy ss_MF_NORM, and the kernel function Kernel_min calculates the minimum index of ss_MF_NORM to obtain the fine search Doppler factor

[0073] Example 2

[0074] Embodiment 2 of the present invention proposes a GPU-based direct sequence spread spectrum underwater acoustic communication Doppler fast estimation system, which is implemented based on the method of embodiment 1. The system includes:

[0075] A modeling unit, used for modeling a received signal of a direct sequence spread spectrum underwater acoustic communication system;

[0076] The operation storage unit is used to perform resampling and FFT operation on the local pilot signal in sequence, and store the FFT operation result to obtain the Doppler matching matrix based on the device memory preloading method of the GPU;

[0077] A coarse search unit is used to adopt a two-step Doppler estimation process. In the coarse search stage, a cross ambiguity function is calculated based on GPU complex multiplication, a reduced modulus normalization calculation is performed based on GPU parallelization, and a coarse Doppler factor to be estimated is obtained through threshold detection;

[0078] The fine search unit is used to select multiple Doppler branches near the coarse Doppler factor in the fine search stage, calculate the cross ambiguity function based on GPU complex multiplication, perform reduced modulus normalization calculation based on GPU parallelism, and obtain the fine Doppler factor to be estimated through threshold detection.

[0079] The heterogeneous parallel algorithm is deployed on the GPU platform: (1) The algorithm framework is an improved two-step Doppler estimation method based on the cross-ambiguity function. The estimation of the Doppler factor is completed in two stages: coarse search and fine search; (2) The spatial storage is optimized by using the GPU device memory preloading method; (3) The CAF normalized energy calculation module is optimized by using the cross-ambiguity function calculation method based on GPU complex multiplication and the modulus normalization calculation method based on GPU parallel reduction.

[0080] To verify the reliability of the present invention, the Doppler factor of the measured data is estimated using the method of the present invention. Figure 7 As shown in Figure 2, the Doppler estimation results of the five methods are as follows: Figure 8 As shown, the five methods are: ①CPU method 1; ②CPU method 2; ③GPU method 1; ④GPU method 2; ⑤the method of the present invention. Fig. 9 This shows that the method of the present invention has the same Doppler estimation accuracy as the other four methods, verifying the reliability of the present invention.

[0081] To verify the effectiveness of the present invention, the Doppler estimation time of the present invention is compared with that of the other four methods. The average calculation time of Doppler estimation and the receiver running time under different methods and the acceleration of the method are as follows: Fig. 9 , Fig.10 shown. Fig. 9 , Fig.10 The results shown verify the effectiveness of the present invention.

[0082] Summary:

[0083] The present invention provides a method for fast Doppler estimation of direct sequence spread spectrum underwater acoustic communication signals based on a CPU+GPU heterogeneous platform, the method comprising: an improved two-step Doppler estimation method based on a cross ambiguity function, for reducing the number of Doppler estimation branches; a device memory preloading method based on a GPU, for pre-storing a calibration Doppler matching matrix; a cross ambiguity function calculation method based on GPU complex multiplication, for Doppler cross ambiguity function calculation; a modulus normalization calculation method based on GPU parallel reduction, for calculating the normalized modulus of the cross ambiguity function; and a heterogeneous parallel algorithm design based on a GPU, for deployment of the fast Doppler estimation algorithm on a GPU platform. The present invention utilizes the advantages of high efficiency of GPU parallel computing to provide an accurate and efficient method for Doppler estimation of direct sequence spread spectrum underwater acoustic communication signals, the method can accurately estimate the Doppler of measured underwater acoustic communication signals, and can greatly reduce the time of Doppler estimation, improve the processing efficiency of the receiver, and provide strong support for the real-time processing of underwater acoustic communication receiving signals and the research and development of underwater acoustic communication system equipment.

[0084] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention is described in detail with reference to the embodiments, it should be understood by those skilled in the art that any modification or equivalent replacement of the technical solutions of the present invention does not depart from the spirit and scope of the technical solutions of the present invention and should be included in the scope of the claims of the present invention.

Claims

1. A GPU-based direct sequence spread spectrum underwater acoustic communication Doppler fast estimation method, the method comprising: Step 1) Modeling the received signal of the direct sequence spread spectrum underwater acoustic communication system; Step 2) resampling and FFT operation are performed on the local pilot signal in sequence, and the FFT operation result is stored to obtain the Doppler matching matrix based on the device memory preloading method of the GPU; Step 3) adopts a two-step Doppler estimation process. In the coarse search stage, the cross ambiguity function is calculated based on GPU complex multiplication, the reduced modulus normalization calculation is performed based on GPU parallelization, and the coarse Doppler factor to be estimated is obtained through threshold detection; Step 4) In the fine search stage, multiple Doppler branches near the coarse Doppler factor are selected, the cross ambiguity function is calculated based on GPU complex multiplication, the reduced modulus normalization calculation is performed based on GPU parallelism, and the fine Doppler factor to be estimated is obtained through threshold detection.

2. The GPU-based direct sequence spread spectrum underwater acoustic communication Doppler fast estimation method according to claim 1 is characterized in that: The received signal of step 1) Modeled as: in, is the transmitted signal, N pa is the number of multipath channels, A p (t) and τ p They represent the amplitude and delay of the pth path respectively, α is the Doppler factor, n(t) is the additive noise, and t represents the time domain.

3. The GPU-based direct sequence spread spectrum underwater acoustic communication Doppler fast estimation method according to claim 2 is characterized in that: The local pilot signal of step 2) The length is N s T s , the resampling factor is the calibration Doppler factor α j ,j=1,2,...,J-1, J is the number of Doppler factors to be estimated; perform N FFT Point FFT operation, where N FFT is an integer power of 2 and is greater than the number of signal sampling points; the size of the Doppler matching matrix is ​​N FFT ×J.

4. The GPU-based direct sequence spread spectrum underwater acoustic communication Doppler fast estimation method according to claim 3 is characterized in that: The step 3) comprises: In the coarse search stage, the Doppler branches to be estimated are divided into K groups, each with J branches, the Doppler estimation accuracy is Δα1, and the Doppler search range of each group is greater than Twice the width of the main peak of the fuzziness function; where, It is Q-channel signal; The cross ambiguity function is calculated based on GPU complex multiplication, and the reduced modulus normalization calculation is performed based on GPU parallelization to calculate the energy integral of the normalized cross ambiguity function of the kth group of Doppler branches. The threshold detection method is used. If the variance of the kth group is significantly higher than the average variance threshold, the branch with the smallest energy integral in the group is selected. The rough Doppler factor to be estimated is 5. The GPU-based direct sequence spread spectrum underwater acoustic communication Doppler fast estimation method according to claim 4 is characterized in that: The step 4) comprises: The rough Doppler factor to be estimated Select J′ Doppler branches for fine search with Doppler estimation accuracy Δα2 nearby, calculate cross ambiguity function based on GPU complex multiplication, perform reduced modulus normalization calculation based on GPU parallelization, select the minimum value of the energy integral of the normalized cross ambiguity function of J′ branches, and the branch with the minimum value generates the fine Doppler factor to be estimated.

6. The GPU-based direct sequence spread spectrum underwater acoustic communication Doppler fast estimation method according to claim 1 is characterized in that: The method is based on the heterogeneous parallel algorithm design of GPU.

7. A GPU-based direct sequence spread spectrum underwater acoustic communication Doppler fast estimation system, characterized in that: include: A modeling unit, used for modeling a received signal of a direct sequence spread spectrum underwater acoustic communication system; The operation storage unit is used to perform resampling and FFT operation on the local pilot signal in sequence, and store the FFT operation result to obtain the Doppler matching matrix based on the device memory preloading method of the GPU; A coarse search unit is used to adopt a two-step Doppler estimation process. In the coarse search stage, a cross ambiguity function is calculated based on GPU complex multiplication, a reduced modulus normalization calculation is performed based on GPU parallelization, and a coarse Doppler factor to be estimated is obtained through threshold detection; and The fine search unit is used to select multiple Doppler branches near the coarse Doppler factor in the fine search stage, calculate the cross ambiguity function based on GPU complex multiplication, perform reduced modulus normalization calculation based on GPU parallelism, and obtain the fine Doppler factor to be estimated through threshold detection.

Images