Multi-channel single target joint detection and tracking method based on FPGA pipeline design
Through the multi-channel single-object joint detection and tracking method designed by FPGA flow, the Gaussian random number generator and particle filtering algorithm are used to solve the problem of detection and tracking of radar system under low signal-to-noise ratio conditions, and efficient target detection and Parameter estimation meets the real-time processing requirements of high-frequency radars.
Patent Information
- Application Number
- CN202211537827.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-12-02
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2042-12-02
AI Technical Summary
Existing radar systems are difficult to effectively detect and track stealth targets under low signal-to-noise ratio conditions, and the single-target joint detection and tracking process implemented by traditional DSP or ARM is difficult to meet the processing time requirements of high-frequency radars, especially in high-precision parameter estimation. in the system.
Using FPGA-based flow design, the initial particle system is generated through a three-way Gaussian random number generator, and noise normalization processing and particle filtering are performed. The real-time performance of data processing is improved by parallel processing and flow design, and the multi-channel single-target joint is realized. Detection tracking.
Under low signal-to-noise ratio conditions, the target tracking and detection performance and parameter estimation accuracy are greatly improved, and the high-speed processing time requirements are met to ensure real-time data processing.
Smart Images

Figure CN115932832B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of radar detection technology, and in particular to a multi-channel single-target joint detection and tracking method based on FPGA pipeline design. Background Art
[0002] The radar cross-section of stealth targets decreases by orders of magnitude, posing a huge challenge to existing terminal guidance radar systems. In order to improve the interception probability and parameter estimation accuracy of stealth targets, it is urgent to improve the detection performance of existing or newly researched radars. Therefore, target detection and parameter estimation under low signal-to-noise ratio conditions have become important research directions.
[0003] In the traditional single target tracking and detection system, target tracking and detection are performed independently, and the detection-first-then-tracking method is adopted. The main principle is: firstly, the target is judged based on a certain threshold criterion, and then the tracking operation is determined according to the judgment result, and finally the target parameters are estimated, which is suitable for high signal-to-noise ratio. In the case of extremely low signal-to-noise ratio, the information loss caused by the detection process leads to poor performance.
[0004] In addition, the realization of single-target joint detection and tracking involves complex operations such as exponential, logarithmic, and matrix operations. Traditional solutions use DSP (Digital Signal Processing) or ARM (processor) to achieve this. This solution is a simple method that is easy to program for systems with a small number of particles and low real-time requirements. However, for high-repetition-rate radars, the processing time requirement is microseconds, and the high-precision parameter estimation system requires thousands of particles. It is difficult to meet this processing time requirement using a multi-core DSP architecture. Summary of the invention
[0005] Based on this, it is necessary to provide a multi-channel single target joint detection and tracking method based on FPGA pipeline design that can be suitable for high-repetition-rate PD (pulse Doppler) radar in complex backgrounds to detect and track weak single targets and improve the real-time performance of data processing in response to the above technical problems.
[0006] A multi-channel single target joint detection and tracking method based on FPGA pipeline design, the method comprising:
[0007] A three-way Gaussian random number generator is used to generate three-way Gaussian distribution random numbers in parallel, and an initial particle system is generated according to the three-way Gaussian distribution random numbers; wherein the initial particle system includes Doppler frequency particles, misalignment angle particles and signal-to-noise ratio particles and particle weights corresponding to each particle;
[0008] The multi-channel spectrum data observed by the radar in the current frame is obtained and noise normalized to obtain the spectrum data after noise normalization. The weight vector of each particle is obtained by calculating the observed value in the spectrum data after noise normalization and the particle value in the initial particle system.
[0009] The particle weights are updated and normalized according to the weight vectors of each particle to obtain updated weights;
[0010] The particles are resampled according to the updated weights of each particle to obtain the resampled particles, parameter estimation and particle update are performed based on the resampled particles to obtain particle filtering results and predicted particles, and target detection and tracking are performed on the spectrum data of the next frame according to the predicted particles until the particle filtering results of all spectrum data are obtained, thus completing target detection and tracking of all spectrum data.
[0011] In one embodiment, the method is applied in FPGA and adopts serial-parallel pipeline processing to serially process the particle X1 before resampling and the particle X2 after resampling, and to parallelly process the particles in X1 and X2.
[0012] In one of the embodiments, the inter-frame particle data in the method is stored in a dedicated storage unit of the FPGA, and the intra-frame process data is stored in a shared storage unit of the FPGA.
[0013] In one embodiment, a three-way Gaussian random number generator includes a linear shift feedback register, a phase generator, an amplitude lookup table, and an IP core multiplier;
[0014] A three-way Gaussian random number generator is used to generate three-way Gaussian distribution random numbers in parallel, and an initial particle system is generated according to the three-way Gaussian distribution random numbers, including:
[0015] A linear shift feedback register with preset fixed parameters is used to generate two uniformly distributed random numbers. The first uniformly distributed random number is input into a phase generator for processing according to the Box-Muller algorithm to obtain a phase value corresponding to the first uniformly distributed random number. The second uniformly distributed random number is input into an amplitude query table for query to obtain an amplitude value corresponding to the second uniformly distributed random number.
[0016] Input the phase value and the amplitude value into the first IP core multiplier and the second IP core multiplier in the FPGA for multiplication to obtain the first Gaussian distribution random number and the second Gaussian distribution random number respectively, input the first Gaussian distribution random number and the second Gaussian distribution random number into the third IP core multiplier for multiplication to obtain the third Gaussian distribution random number;
[0017] Generate the initial particle system {xi ,ω i}, where x i represents the value of the i-th particle in the initial particle system, ω i Represents the particle weight corresponding to the i-th particle.
[0018] In one embodiment, obtaining multi-channel spectrum data observed by the radar in the current frame and performing noise normalization processing to obtain noise-normalized spectrum data includes:
[0019] The sum channel spectrum data and difference channel spectrum data of the current frame of radar target detection are obtained, and the spectrum data is normalized according to the noise power of the sum channel to obtain the spectrum data after noise normalization, which is expressed as Among them, f i represents the ith spectrum data before noise normalization, n o Represents the noise power of the sum channel.
[0020] In one embodiment, the weight vector of each particle is obtained by calculating according to the observed value in the noise-normalized spectrum data and the particle value in the initial particle system, including:
[0021] For the Doppler frequency particles in the initial particle system, it is determined whether the Doppler frequency particles meet the preset Doppler frequency range requirements. If it is determined that the requirements are met, the weight vector of the Doppler frequency particles is obtained by calculation based on the Doppler frequency particle observation value in the noise-normalized spectrum data and the Doppler frequency particle value in the initial particle system; otherwise, the Doppler frequency particle observation value is discarded, and the weight vector of the Doppler frequency particle is obtained by calculation based on the Doppler frequency particle value;
[0022] For the misaligned angle particles and signal-to-noise ratio particles in the initial particle system, the weight vectors corresponding to the misaligned angle particles and signal-to-noise ratio particles are calculated according to the particle observation values in the noise-normalized spectrum data and the corresponding particle values in the initial particle system;
[0023] Among them, the weight vector of each particle in the initial particle system is expressed as
[0024] Ln=ss′*Cov_ss -1 *ss-log(det(Cov_ss))
[0025] Where ss represents the difference between the observed value of a particle and the corresponding particle value in the initial system, ss′ represents the transpose of ss, Cov_ss represents the particle covariance matrix, and Cov_ss -1 Represents the inverse matrix of Cov_ss.
[0026] In one embodiment, the particle weights are updated and normalized according to the weight vectors of each particle to obtain updated weights, including:
[0027] The particle weight is updated according to the weight vector of each particle to obtain a new weight, which is expressed as
[0028] The new weights are normalized to obtain the updated weights of each particle, expressed as
[0029] In one of the embodiments, in the calculation process of the weight vector and the updated weight, a data processing method of mixed operations of fixed-point numbers and floating-point numbers is adopted.
[0030] In one embodiment, in the calculation process of the weight vector and the updated weight, a data processing method of mixed fixed-point and floating-point operations is adopted, including:
[0031] In the calculation process of the weight vector, the integer b in the exponential expression Ln=Exp(a+b) of the weight vector is input into the ROM memory for table lookup to obtain an integer calculation result, the decimal a in the exponential expression of the weight vector is input into the CordicIP core for dynamic calculation to obtain a decimal calculation result, the integer calculation result and the decimal calculation result are input into the IP core multiplier and the IP core in turn for processing, and the weight vector is output;
[0032] In the calculation process of the updated weight, the particle weight of the current frame is input into the IP core for fixed-point to floating-point conversion to obtain the converted particle weight, the converted particle weight and the weight vector are input into the IP core multiplier for multiplication, and the new weight of each particle is stored in the RAM memory, the new weights of all particles are obtained from the RAM memory and the new weights of each particle are normalized, and the normalized weights are input into the IP core for floating-point to fixed-point conversion, and the updated weights of each particle are output.
[0033] In one embodiment, parameter estimation and particle updating are performed according to the resampled particles to obtain particle filtering results and predicted particles, including:
[0034] The resampled particles are screened according to the preset threshold. When the particle value corresponding to the resampled particle is not equal to the preset threshold, the resampled particle is determined to be an invalid particle; otherwise, the resampled particle is determined to be a valid particle, and the parameters of the valid particle are estimated to obtain the particle filtering result, which is expressed as Among them, x i represents the i-th valid particle, and n represents the number of valid particles; wherein the preset threshold refers to the dynamic threshold calculated based on the particles after resampling of each frame;
[0035] According to the current frame particles in the resampled particles The particles are updated with process noise to generate predicted particles, which can be expressed as Among them, k represents the correlation coefficient, s i represents the process noise of the ith particle.
[0036] The above-mentioned multi-channel single-target joint detection and tracking method based on FPGA pipeline design first uses a three-way Gaussian random number generator to generate three-way Gaussian distribution random numbers in parallel, and generates an initial particle system according to the three-way Gaussian distribution random numbers; then obtains the multi-channel spectrum data observed by the radar in the current frame and performs noise normalization processing to obtain the noise-normalized spectrum data, and calculates according to the observation value in the noise-normalized spectrum data and the particle value in the initial particle system to obtain the weight vector of each particle; then updates the particle weight according to the weight vector of each particle and performs normalization processing to obtain the updated weight; finally, resamples the particles according to the updated weights of each particle to obtain the resampled particles, and performs parameter estimation and particle update according to the resampled particles to obtain the particle filtering results and predicted particles, and performs target detection and tracking on the spectrum data of the next frame according to the predicted particles until the particle filtering results of all spectrum data are obtained, and the target detection and tracking of all spectrum data is completed. This method can make full use of the effective information in the observation data and greatly improve the target tracking detection performance and parameter estimation accuracy under low signal-to-noise ratio conditions. At the same time, this method is implemented using a high-speed pipelined FPGA design to ensure real-time data processing requirements. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1 A schematic diagram of a flow chart of a multi-channel single target joint detection and tracking method based on FPGA pipeline design in one embodiment;
[0038] Figure 2 A schematic diagram of a flow chart of serial-parallel pipeline processing steps in an FPGA in one embodiment;
[0039] Figure 3 A schematic diagram of a process of generating a three-way Gaussian distribution random number in one embodiment;
[0040] Figure 4 Schematic diagram of the calculation process of the weight vector and the updated weight in one embodiment. DETAILED DESCRIPTION
[0041] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0042] In one embodiment, Figure 1 As shown, a multi-channel single target joint detection and tracking method based on FPGA pipeline design is provided, including the following steps:
[0043] Step S1, particle initialization: a three-way Gaussian random number generator is used to generate three-way Gaussian distribution random numbers in parallel, and an initial particle system is generated according to the three-way Gaussian distribution random numbers; wherein the initial particle system includes Doppler frequency particles, misalignment angle particles and signal-to-noise ratio particles and particle weights corresponding to each particle. wherein the generated initial particles are stored in a storage unit (RAM).
[0044] It can be understood that the target detection and tracking method of the present application is based on a particle filtering algorithm, which regards the parameters of the target signal in the high-repetition-rate PD mode as the state vector of the particle, takes the data input of the radar and difference channels as the observation values, and uses the optimal Bayesian filter to realize joint target detection and estimation.
[0045] Step S2, noise normalization: obtaining multi-channel spectrum data observed by the radar in the current frame and performing noise normalization processing to obtain spectrum data after noise normalization.
[0046] It can be understood that the multi-channel Fourier (FFT) spectrum data observed by the radar also needs to be subjected to noise normalization processing before it can be used for the calculation of particle weights, and the noise normalization processing is performed once for each frame of data.
[0047] Step S3, weight vector calculation: Calculate the weight vector of each particle based on the observed value in the noise-normalized spectrum data and the particle value in the initial particle system.
[0048] Step S4, weight update: update the particle weight according to the weight vector of each particle and perform normalization processing to obtain the updated weight. The updated weight is stored in RAM.
[0049] It can be understood that in order to improve the estimation accuracy and take into account resource overhead, a mixed data processing method of "fixed-point + floating-point" is adopted in the data calculation process of weight vector and weight update, in which the fixed-point number to floating-point number calls the Floating-point IP (floating-point IP) core in FPGA (programmable array logic). For the weight vector exponent calculation, the integer and decimal of the exponent are calculated independently, that is, the integer is completed by a lookup table, and the decimal is dynamically calculated by the Cordic IP (coordinate rotation digital calculation IP) core in FPGA, which effectively improves the weight calculation accuracy and takes into account the FPGA logic unit resources.
[0050] Step S5, particle resampling: resample the particles according to the updated weights of the particles to obtain resampled particles.
[0051] It can be understood that according to the system resampling method, the large weight particles are copied and retained, and the small weight particles are discarded, that is, the weight values are read from the RAM for accumulation. When the weight accumulation value is greater than the reference value of random sampling, the particle is copied and retained, otherwise, the particle is discarded. The output resampled particles are mainly used for parameter estimation and particle update, and all resampled particles have equal weights.
[0052] Step S6, parameter estimation: perform parameter estimation based on the resampled particles to obtain the particle filtering result.
[0053] It can be understood that since the resampled particle value is closer to the posterior density, parameter estimation refers to estimating the target existence probability and target echo measurement information (including parameters such as Doppler frequency, misalignment angle and signal-to-noise ratio) based on the resampled particles.
[0054] Step S7, particle update: perform particle update according to the resampled particles to obtain predicted particles, wherein the predicted particles are stored in RAM.
[0055] Step S8, looping through steps S2 to S7, performing target detection and tracking on the spectrum data of the next frame according to the predicted particles until the particle filtering results of all the spectrum data are obtained, thereby completing target detection and tracking of all the spectrum data.
[0056] The above-mentioned multi-channel single-target joint detection and tracking method based on FPGA pipeline design fully utilizes the effective information in the observation data by performing particle initialization, noise normalization, weight vector calculation, weight update, particle resampling, parameter estimation and particle update in high-speed FPGA, and greatly improves the target tracking detection performance and parameter estimation accuracy under low signal-to-noise ratio conditions. At the same time, it is implemented using high-speed FPGA to ensure the real-time requirements of data processing.
[0057] In one embodiment, if Figure 2 As shown in FIG. 1 , the multi-channel single target joint detection and tracking method based on FPGA pipeline design is applied in FPGA and adopts serial-parallel pipeline processing mode, and the particle X1 before resampling and the particle X2 after resampling are processed serially, and the particles in X1 and X2 are processed in parallel. In addition, when processing the particle X1 before resampling, the adjacent processing time interval is one particle.
[0058] It can be understood that the use of serial-parallel pipeline processing, that is, a two-level buffer request pipeline structure with a "ping-pong operation" for particle access, can reduce pipeline pause bubbles of particle requests, improve processing efficiency, reduce data waiting time, and greatly reduce the time for single target tracking and detection.
[0059] In one embodiment, if Figure 1 As shown in the figure, the inter-frame particle data in the multi-channel single-target joint detection and tracking method based on FPGA pipeline design is stored in the dedicated storage unit of FPGA, and the intra-frame process data is stored in the shared storage unit of FPGA. By classifying the data RAM, the resource overhead of FPGA is effectively reduced.
[0060] In one embodiment, the three-way Gaussian distribution random number generation method is shown as follows Figure 3 As shown, the three-way Gaussian random number generator includes a linear shift feedback register, a phase generator, an amplitude lookup table and an IP core multiplier. The three-way Gaussian random number generator can quickly generate three-way approximate Gaussian distribution random numbers in parallel. The specific steps are:
[0061] Firstly, a linear shift feedback register with preset fixed parameters is used to generate two uniformly distributed random numbers. According to the Box-Muller (Box-Muller transform) algorithm, the first uniformly distributed random number is input into a phase generator based on a DDS (direct digital frequency synthesizer algorithm) for processing to obtain a phase value corresponding to the first uniformly distributed random number. The second uniformly distributed random number is input into an amplitude query table based on a ROM (read-only memory) for query to obtain an amplitude value corresponding to the second uniformly distributed random number.
[0062] Then the phase value and amplitude value are input into the first IP core multiplier and the second IP core multiplier in the FPGA for multiplication to obtain the first Gaussian distribution random number G1 and the second Gaussian distribution random number G2 respectively. The first Gaussian distribution random number G1 and the second Gaussian distribution random number G2 are input into the third IP core multiplier for multiplication to obtain the third Gaussian distribution random number G3.
[0063] Further, the initial particle system {x i ,ωi}, where x i represents the value of the i-th particle in the initial particle system, ω i Represents the particle weight corresponding to the i-th particle.
[0064] In one embodiment, obtaining multi-channel spectrum data observed by the radar in the current frame and performing noise normalization processing to obtain noise-normalized spectrum data includes:
[0065] The sum channel spectrum data and difference channel spectrum data of the current frame of radar target detection are obtained, and the spectrum data is normalized according to the noise power of the sum channel to obtain the spectrum data after noise normalization, which is expressed as Among them, f i represents the ith spectrum data before noise normalization, n o Represents the noise power of the sum channel.
[0066] In one embodiment, the weight vector of each particle is obtained by calculating according to the observed value in the noise-normalized spectrum data and the particle value in the initial particle system, including:
[0067] For the Doppler frequency particles in the initial particle system, first determine whether the Doppler frequency particles meet the preset Doppler frequency range requirements. If it is determined that the requirements are met, calculate the Doppler frequency particle observation value in the noise-normalized spectrum data and the Doppler frequency particle value in the initial particle system to obtain the weight vector of the Doppler frequency particle; otherwise, discard the Doppler frequency particle observation value, calculate according to the Doppler frequency particle value, and obtain the weight vector of the Doppler frequency particle; specifically, the user-preset Doppler frequency range requirement is [-20kHz, 20kHz];
[0068] For the misaligned angle particles and signal-to-noise ratio particles in the initial particle system, the weight vectors corresponding to the misaligned angle particles and signal-to-noise ratio particles are calculated according to the particle observation values in the noise-normalized spectrum data and the corresponding particle values in the initial particle system;
[0069] Among them, the weight vector of each particle in the initial particle system is expressed as
[0070] Ln=ss′*Cov_ss -1 *ss-log(det(Cov_ss))
[0071] Where ss represents the difference between the observed value of a particle and the corresponding particle value in the initial system, ss′ represents the transpose of ss, Cov_ss represents the particle covariance matrix, and Cov_ss -1 Represents the inverse matrix of Cov_ss.
[0072] In one embodiment, the particle weights are updated and normalized according to the weight vectors of each particle to obtain updated weights, including:
[0073] The particle weight is updated according to the weight vector of each particle to obtain a new weight, which is expressed as
[0074] The new weights are normalized to obtain the updated weights of each particle, expressed as
[0075] In order to improve the weight calculation accuracy and take into account resource consumption, further, such as Figure 4 As shown, in the calculation process of the weight vector and the updated weight, a data processing method of mixed fixed-point and floating-point operations is adopted. The specific steps are:
[0076] In the calculation process of the weight vector, the integer b in the exponential expression Ln=Exp(a+b) of the weight vector is input into the ROM memory for table lookup to obtain an integer calculation result, the decimal a in the exponential expression of the weight vector is input into the CordicIP core for dynamic calculation to obtain a decimal calculation result, the integer calculation result and the decimal calculation result are input into the IP core multiplier and the IP core in turn for processing, and the weight vector is output;
[0077] In the calculation process of the updated weight, the particle weight of the current frame is input into the IP core for fixed-point to floating-point conversion to obtain the converted particle weight, the converted particle weight and the weight vector are input into the IP core multiplier for multiplication, and the new weight of each particle is stored in the RAM memory, the new weights of all particles are obtained from the RAM memory and the new weights of each particle are normalized, and the normalized weights are input into the IP core for floating-point to fixed-point conversion, and the updated weights of each particle are output.
[0078] In one embodiment, parameter estimation and particle updating are performed according to the resampled particles to obtain particle filtering results and predicted particles, including:
[0079] The resampled particles are screened according to the preset threshold. When the particle value corresponding to the resampled particle is not equal to the preset threshold γ or γ+1, the resampled particle is determined to be an invalid particle; otherwise, the resampled particle is determined to be a valid particle, and the parameters of the valid particle are estimated to obtain the particle filtering result, which is expressed as Among them, x i represents the i-th valid particle, and n represents the number of valid particles; wherein the preset threshold refers to the dynamic threshold calculated based on the particles after resampling of each frame;
[0080] According to the current frame particles in the resampled particles The particles are updated with process noise to generate predicted particles, which can be expressed as Among them, k represents the correlation coefficient, s i represents the process noise of the ith particle.
[0081] It should be understood that although Figure 1 , Figure 3 and Figure 4 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 , Figure 3 and Figure 4 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.
[0082] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0083] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.
Claims
1. A multi-channel single target joint detection and tracking method based on FPGA pipeline design, characterized in that: The method comprises: A three-way Gaussian random number generator is used to generate three-way Gaussian distribution random numbers in parallel, and an initial particle system is generated according to the three-way Gaussian distribution random numbers; wherein the initial particle system includes Doppler frequency particles, misalignment angle particles and signal-to-noise ratio particles and particle weights corresponding to each particle; Acquire multi-channel spectrum data observed in the current frame of the radar and perform noise normalization processing to obtain noise-normalized spectrum data, and calculate according to the observation value in the noise-normalized spectrum data and the particle value in the initial particle system to obtain the weight vector of each particle; The particle weights are updated and normalized according to the weight vectors of each particle to obtain updated weights; The particles are resampled according to the updated weights of each particle to obtain the resampled particles, parameter estimation and particle update are performed according to the resampled particles to obtain particle filtering results and predicted particles, and target detection and tracking are performed on the spectrum data of the next frame according to the predicted particles until the particle filtering results of all the spectrum data are obtained, thereby completing the target detection and tracking of all the spectrum data.
2. The method according to claim 1, characterized in that The method is applied in FPGA and adopts a serial-parallel pipeline processing mode, performs serial processing on the particle X1 before resampling and the particle X2 after resampling, and performs parallel processing on the particles in X1 and X2.
3. The method according to claim 2, characterized in that In the method, the inter-frame particle data is stored in a dedicated storage unit of the FPGA, and the intra-frame process data is stored in a shared storage unit of the FPGA.
4. The method according to claim 1, characterized in that: The three-way Gaussian random number generator includes a linear shift feedback register, a phase generator, an amplitude lookup table and an IP core multiplier; A three-way Gaussian random number generator is used to generate three-way Gaussian distribution random numbers in parallel, and an initial particle system is generated according to the three-way Gaussian distribution random numbers, including: A linear shift feedback register with preset fixed parameters is used to generate two uniformly distributed random numbers. The first uniformly distributed random number is input into a phase generator for processing according to the Box-Muller algorithm to obtain a phase value corresponding to the first uniformly distributed random number. The second uniformly distributed random number is input into an amplitude query table for query to obtain an amplitude value corresponding to the second uniformly distributed random number. Input the phase value and the amplitude value into the first IP core multiplier and the second IP core multiplier in the FPGA for multiplication to obtain a first Gaussian distribution random number and a second Gaussian distribution random number respectively, input the first Gaussian distribution random number and the second Gaussian distribution random number into the third IP core multiplier for multiplication to obtain a third Gaussian distribution random number; Generate the initial particle system {x i ,ω i }, where x i represents the value of the i-th particle in the initial particle system, ω i Represents the particle weight corresponding to the i-th particle.
5. The method according to claim 4, characterized in that Obtain the multi-channel spectrum data observed by the radar in the current frame and perform noise normalization processing to obtain the spectrum data after noise normalization, including: The sum channel spectrum data and difference channel spectrum data of the current frame of radar target detection are obtained, and the spectrum data is normalized according to the noise power of the sum channel to obtain the spectrum data after noise normalization, which is expressed as Among them, f i represents the ith spectrum data before noise normalization, n o Represents the noise power of the sum channel.
6. The method according to claim 5, characterized in that The weight vector of each particle is obtained by calculating the observed value in the noise-normalized spectrum data and the particle value in the initial particle system, including: For the Doppler frequency particles in the initial particle system, it is determined whether the Doppler frequency particles meet the preset Doppler frequency range requirements. When it is determined that the requirements are met, a weight vector of the Doppler frequency particles is obtained by calculation according to the Doppler frequency particle observation value in the noise-normalized spectrum data and the Doppler frequency particle value in the initial particle system; otherwise, the Doppler frequency particle observation value is discarded, and a calculation is performed according to the Doppler frequency particle value to obtain the weight vector of the Doppler frequency particle; For the misaligned angle particles and the signal-to-noise ratio particles in the initial particle system, the particle observation values in the noise-normalized spectrum data and the corresponding particle values in the initial particle system are calculated to obtain weight vectors corresponding to the misaligned angle particles and the signal-to-noise ratio particles; Among them, the weight vector of each particle in the initial particle system is expressed as Ln=ss′*Cov_ss -1 *ss-log(det(Cov_ss)) Where ss represents the difference between the observed value of a particle and the corresponding particle value in the initial system, ss′ represents the transpose of ss, Cov_ss represents the particle covariance matrix, and Cov_ss -1 Represents the inverse matrix of Cov_ss.
7. The method according to claim 6, characterized in that The particle weights are updated and normalized according to the weight vectors of each particle to obtain the updated weights, including: The particle weight is updated according to the weight vector of each particle to obtain a new weight, which is expressed as The new weights are normalized to obtain the updated weights of each particle, expressed as 8. The method according to claim 6 or 7, characterized in that: In the calculation process of the weight vector and the updated weight, a data processing method of mixed operations of fixed-point numbers and floating-point numbers is adopted.
9. The method according to claim 8, characterized in that In the calculation process of the weight vector and the updated weight, a data processing method of mixed fixed-point and floating-point operations is adopted, including: In the calculation process of the weight vector, the integer b in the exponential expression Ln=Exp(a+b) of the weight vector is input into the ROM memory for table lookup to obtain an integer calculation result, the decimal a in the exponential expression of the weight vector is input into the Cordic IP core for dynamic calculation to obtain a decimal calculation result, the integer calculation result and the decimal calculation result are sequentially input into the IP core multiplier and the IP core for processing, and the weight vector is output; In the calculation process of the updated weight, the particle weight of the current frame is input into the IP core for fixed-point to floating-point conversion processing to obtain the converted particle weight, the converted particle weight and the weight vector are input into the IP core multiplier for multiplication, and the obtained new weight of each particle is stored in the RAM memory, the new weights of all particles are obtained from the RAM memory and the new weights of each particle are normalized, and the obtained normalized weights are input into the IP core for floating-point to fixed-point conversion processing, and the updated weights of each particle are output.
10. The method according to claim 9, characterized in that Parameter estimation and particle update are performed based on the resampled particles to obtain particle filtering results and predicted particles, including: The resampled particles are screened according to a preset threshold. When the particle value corresponding to the resampled particle is not equal to the preset threshold, the resampled particle is determined to be an invalid particle; otherwise, the resampled particle is determined to be a valid particle, and the parameters of the valid particle are estimated to obtain a particle filtering result, which is expressed as Among them, x i represents the i-th valid particle, and n represents the number of valid particles; wherein the preset threshold refers to a dynamic threshold calculated based on the particles after resampling of each frame; According to the current frame particles in the resampled particles The particles are updated with process noise to generate predicted particles, which can be expressed as Among them, k represents the correlation coefficient, s i represents the process noise of the ith particle.
