A pulse TOA estimation method based on CE-Adam combined algorithm

By using the CE-Adam combination algorithm in pulsar signal processing, and using a combination of cross entropy algorithm and Adam algorithm, the problem of large amount of pulse TOA estimation in the existing technology is solved, and fast and high-precision pulse TOA estimation is achieved. , supports the rapid implementation of pulsar navigation algorithm.

CN111814335BActive Publication Date: 2024-11-26NAT UNIV OF DEFENSE TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202010662717.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-07-10
Publication Date
2024-11-26
Estimated Expiration
2040-07-10

AI Technical Summary

Technical Problem

The existing pulse TOA estimation method increases dramatically when the observation time increases, resulting in too long algorithms and hinders the rapid implementation of pulsar navigation algorithms.

Method used

Using the method based on the CE-Adam combination algorithm, the global optimal solution is calculated through the cross entropy algorithm and the Adam algorithm is used for precise optimization, which reduces the calculation amount and improves the solution speed.

Benefits of technology

On the premise of ensuring reliability, by processing a small amount of observation data through cross-entropy algorithm, combined with the Adam algorithm, the calculation amount is significantly reduced, the speed of pulse TOA estimation is increased, and the rapid implementation of pulsar navigation algorithm is provided.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN111814335B_ABST
    Figure CN111814335B_ABST
Patent Text Reader

Abstract

The present invention provides a pulse TOA estimation method based on the CE-Adam combined algorithm, comprising the following steps: step 1, determining the target function, the estimated range of the initial phase and the estimated range of the frequency according to prior information; step 2, using the cross entropy algorithm CE to calculate a rough global optimal solution; step 3, using the rough global optimal solution obtained in step 2 as a starting point, applying the Adam algorithm, and outputting an accurate global optimal solution. The present invention determines the rough global optimal solution by applying the cross entropy algorithm to a small amount of observation data, and then uses the Adam algorithm for precise optimization, which can reduce the amount of calculation and improve the solution speed while ensuring reliability, and provides strong support for the rapid implementation of the pulsar navigation algorithm.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of pulsar signal processing, and in particular to a pulse TOA estimation method based on a CE-Adam combined algorithm. Background Art

[0002] The estimation of pulse TOA is one of the important contents of pulsar signal processing. The pulse TOA estimation method that directly uses the photon TOA is an asymptotically effective pulse TOA estimation method. This method transforms the estimation problem of the initial phase and frequency of the pulsar signal into a two-dimensional optimization problem by constructing a maximum likelihood estimation problem. By solving the optimization problem, the estimated values ​​of the initial phase and frequency can be obtained. This type of method uses the photon TOA almost losslessly and has high calculation accuracy.

[0003] At the same time, optimization problems often have multiple extreme values, and the traditional optimization method based on gradient descent is prone to fall into local optimality.

[0004] In order to overcome this defect, there are two main methods currently adopted. One is to determine the search range through prior information and use grid search within the search range; the other is to use Newton iteration method, select multiple starting points, and solve the optimization problem.

[0005] However, in order to obtain a high-precision pulse TOA estimation result, it is necessary to use observation data for a longer period of time for pulse TOA estimation. As the observation time increases, the amount of calculation of the two existing methods increases sharply, and the algorithm takes too long, which hinders the rapid implementation of the pulsar navigation algorithm.

[0006] Therefore, it is of great significance to design a new pulse TOA estimation method with fast calculation speed and high accuracy. Summary of the invention

[0007] The present invention provides a pulse TOA estimation method based on the CE-Adam combined algorithm, and the specific technical solution is as follows:

[0008] A pulse TOA estimation method based on CE-Adam combined algorithm comprises the following steps:

[0009] Step 1: Determine the target function, the estimated range of the initial phase and the estimated range of the frequency according to the prior information;

[0010] Step 2: Use the cross entropy algorithm CE to calculate the approximate global optimal solution;

[0011] Step 3: Using the approximate global optimal solution obtained in step 2 as the starting point, apply the Adam algorithm to output the precise global optimal solution.

[0012] Preferably, in the above technical solution, in step 1:

[0013] The objective function using expression 1) is:

[0014]

[0015] λ(t k )=λ(φ(t k ))=β+αh(φ(t k ))

[0016] φ(t k )=q+f(t k -t1) 1);

[0017] Where: J represents the objective function; λ represents the rate function; t k represents the arrival time of the kth photon, t1 represents the arrival time of the first photon; Q is the total number of photons; φ represents the pulsar phase; α represents the average flow rate of the pulsar signal; β represents the background noise flow rate; h represents the profile function of the pulsar; q represents the initial phase of the pulsar signal, f represents the frequency of the pulsar signal, and the estimated values ​​of q and f are expressed as

[0018] The estimated range of the initial phase q of the pulsar signal is [0,1];

[0019] The estimated range of the frequency f of the pulsar signal is [f'-Δf, f'+Δf], where: f' is the estimated value of the frequency of the pulsar signal at the current moment, f0 is the frequency of the pulsar signal at time t0, is the frequency change rate of the pulsar signal at time t0, Δt represents the time difference between the current time and time t0; Δf is an empirical value;

[0020] Preferably, in the above technical solution, the cross entropy algorithm in step 2 specifically includes the following steps:

[0021] Step 2.1, m is the number of cycles, take m = 0; give the initial value of the distribution parameter μ q0 , σ q0 , μ f0 and σ f0 , where: μ q0 and σ q0 represents the initial expectation and initial standard deviation of parameter q, μ f0 and σ f0 represents the initial expectation and initial standard deviation of the parameter f;

[0022] Step 2.2: Make a judgment. If the termination condition σ is met qm <σ qend And σ fm <σfend , then go to step 2.5; otherwise go to step 2.3; where: σ qend and σ fend represent the standard deviation thresholds of parameters q and f respectively;

[0023] Step 2.3: Generate N independent samples based on the probability density function and calculate the objective function value corresponding to the samples; sort the samples in order from small to large according to the objective function value, and select the N samples with the largest objective function value. e Elite samples, N e =ρN, ρ is the quantile parameter;

[0024] Use expression 3) to obtain the distribution parameters:

[0025]

[0026]

[0027]

[0028]

[0029] Where: μ qm represents the expectation of parameter q in the mth cycle (i.e. the expected value); μ fm represents the expectation of parameter f in the mth cycle (i.e. the expected value); σ qm represents the standard deviation of parameter q in the mth cycle; σ fm represents the standard deviation of the parameter f in the mth cycle; N represents the sample size; J i represents the objective function value corresponding to the i-th sample; q i and f i represents the values ​​of parameters q and f corresponding to the i-th sample; γ represents the threshold; I{P(J i ≥γ)} represents the indicator function,

[0030] Step 2.4: According to N e The elite samples update the distribution parameters, taking m=m+1, and return to step 2.2;

[0031] Step 2.5: Output μ qm and μ fm As a rough global optimal solution.

[0032] Preferably, in the above technical solution, in step 2.3:

[0033] The probability density function is expressed as:

[0034]

[0035] Among them, μ q0 , σ q0 , μ f0 and σ f0 is the initial distribution parameter, μ q0 and σ q0 represents the initial expectation and initial standard deviation of parameter q, μ f0 and σ f0 represents the initial expectation and initial standard deviation of the parameter f.

[0036] In the above technical solution, the μ q0 The value is 0.5, σ q0 The value is 0.5, μ f0 Take the value of f', σ f0 The value is Δf.

[0037] Preferably, in the above technical solution, the Adam algorithm in step 3 specifically includes the following steps:

[0038] Step 3.1, take the number of iterations n = 0;

[0039] Step 3.2, calculate the objective function at point θ n =(q n ,f n ) is the gradient g n ;Record the number of iterations n=n+1;

[0040] Step 3.3: Use expression 4) to calculate the weighted average m of the gradient at the nth iteration: n and momentum v n :

[0041] m n =β1m n-1 +(1-β1)g n

[0042]

[0043] Among them: β1 and β2 are constants close to 1; m0 = v0 = 0;

[0044] Step 3.4: Calculate m using expression 5) n and v n Correction value and

[0045]

[0046] Step 3.5: Update the value of parameter θ using expression 6):

[0047]

[0048] Among them, η represents the step size, ò is a very small constant, is the learning rate;

[0049] Step 3.6: Determine whether the iteration termination condition is met. If so, output the values ​​of parameters q and f as the exact global optimal solution; otherwise, return to step 3.2;

[0050] The iteration termination condition is: n -q n-1 <q end And f n -f n-1 <f end , where: q n ,q n-1 、f n 、f n-1 Indicates the values ​​of parameters q and f obtained by the nth and n-1th iterations, q end and f end Indicates the change threshold of parameters q and f.

[0051] The pulse TOA estimation method based on the CE-Adam combined algorithm of the present invention determines the global optimal approximate solution by implementing the cross entropy algorithm on a small amount of observation data, and then uses the Adam algorithm for precise optimization. It can reduce the amount of calculation and improve the solution speed while ensuring reliability, and provide strong support for the rapid implementation of the pulsar navigation algorithm.

[0052] In addition to the above-described purposes, features and advantages, the present invention has other purposes, features and advantages. The present invention will be further described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The drawings constituting a part of this application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0054] Figure 1 It is a flow chart of a pulse TOA estimation method based on a CE-Adam combination algorithm according to a preferred embodiment of the present invention;

[0055] Figure 2a Schematic diagram showing the first iteration of the cross entropy algorithm;

[0056] Figure 2b Schematic diagram showing the second iteration of the cross entropy algorithm;

[0057] Figure 2c Schematic diagram showing the third iteration of the cross entropy algorithm. DETAILED DESCRIPTION

[0058] The embodiments of the present invention are described in detail below with reference to the accompanying drawings, but the present invention can be implemented in many different ways as defined and covered by the claims.

[0059] Example:

[0060] A pulse TOA estimation method based on the CE-Adam combined algorithm, see Figure 1 , specifically including the following steps:

[0061] Step 1: Based on the prior information, determine the target function, the estimated range of the initial phase, and the estimated range of the frequency. The details are:

[0062] The objective function using expression 1) is:

[0063]

[0064] λ(t k )=λ(φ(t k ))=β+αh(φ(t k ))

[0065] φ(t k )=q+f(t k -t1) 1);

[0066] Where: J represents the objective function; λ represents the rate function; t k represents the arrival time (TOA) of the kth photon, t1 represents the arrival time of the first photon; Q is the total number of photons; φ represents the pulsar phase; α represents the average flow rate of the pulsar signal; β represents the background noise flow rate; h represents the profile function of the pulsar; q represents the initial phase of the pulsar signal, f represents the frequency of the pulsar signal, and the estimated values ​​of q and f are expressed as

[0067] The estimated range of the initial phase q of the pulsar signal (i.e. The value range of q is [0,1]. The estimated range of q can also be appropriately narrowed according to actual needs, such as [0,0.5].

[0068] The estimated range of the frequency f of the pulsar signal (i.e. The value range of is [f'-Δf, f'+Δf], where: f' is the frequency estimate of the pulsar signal at the current moment, f0 is the frequency of the pulsar signal at time t0, is the frequency change rate of the pulsar signal at time t0, Δf is an empirical value; Δt represents the time difference between the current time and time t0.

[0069] Step 2: Based on the estimated range in step 1, use the cross entropy algorithm (CE) to calculate the approximate global optimal solution, specifically:

[0070] ①. According to the objective function and the estimated range (the estimated range of the initial phase and the estimated range of the frequency), determine the termination condition, set the sample size N, set the quantile parameter ρ, and set the distribution parameter of the initial sampling probability density function;

[0071] The termination condition is: qm <σ qend And σ fm <σ fend , the value of m is 0, 1, 2,,,,M (maximum number of cycles), σ qend and σ fend Represent the standard deviation thresholds of parameters q and f, respectively, generally σ qend Take 1e-03 and σ fend Take 1e-05; μ q0 , σ q0 , μ f0 and σ f0 is the initial distribution parameter, μ q0 and σ q0 represents the initial expectation and initial standard deviation of parameter q, μ f0 and σ f0 represents the initial expectation and initial standard deviation of the parameter f; preferably: q0 The value is 0.5, σ q0 The value is 0.5, μ f0 Take the value of f', σ f0 The value is Δf;

[0072] ②, make a judgment, if the termination condition σ is met qm <σ qend And σ fm <σ fend , then go to step ⑤; otherwise go to step ③; σ qend and σ fend Respectively represent the standard deviation thresholds of parameters q and f; σ qm represents the standard deviation of parameter q in the mth cycle; σ fm represents the standard deviation of the parameter f in the mth cycle;

[0073] ③. Generate N independent samples based on the probability density function and calculate the objective function value corresponding to the samples; sort the samples in order from small to large according to the objective function value, and select the N samples with the largest objective function value. e Elite samples, N e =ρN, ρ is the quantile parameter; the sampling probability density function is expressed by expression 2):

[0074] Here, expression 3) is used to calculate the distribution parameters and update them:

[0075]

[0076]

[0077]

[0078]

[0079] Where: N represents the sample size; J i represents the objective function value corresponding to the i-th sample; q i and f i represents the values ​​of parameters q and f corresponding to the i-th sample; γ represents the threshold, which is not fixed. Its value is equal to the smallest value of the objective function corresponding to the elite sample screened each time; I{P(X i ≥γ)} represents the indicator function,

[0080] ④ According to N e Elite samples update the distribution parameters, taking m=m+1, and return to step ②;

[0081] ⑤、Output μ qm and μ fm As a rough global optimal solution;

[0082] Step 3: Using the rough global optimal solution obtained in step 2 as the starting point, apply the Adam algorithm to obtain the precise global optimal solution, specifically:

[0083] Step 3.1: Set the iteration termination condition, which is that the change in the parameters between the two calculations is less than a certain threshold, i.e., q n -q n-1 <q end And f n -f n-1 <f end , where q n ,q n-1 、f n 、f n-1 represents the values ​​of q and f obtained at the nth and n-1th iterations, q end and f end It represents the threshold of change of q and f, which is related to many factors such as the number of photons and the selected pulsar. Generally, q end The value is 1e-04 and f end The value is 1e-08;

[0084] Take the number of iterations n = 0;

[0085] Step 3.2, calculate the objective function at point θ n =(q n ,f n ) is the gradient g n ;Record the number of iterations n=n+1;

[0086] Step 3.3: Use expression 4) to calculate the weighted average m of the gradient at the nth iteration: n and momentum v n :

[0087] m n =β1m n-1 +(1-β1)g n

[0088]

[0089] Among them: β1 and β2 are constants close to 1; m0 = v0 = 0;

[0090] Step 3.4: Calculate m using expression 5) n and v n Correction value and

[0091]

[0092] Step 3.5: Update the value of parameter θ using expression 6):

[0093]

[0094] Among them, η represents the step size, ò is a very small constant, is the learning rate;

[0095] Step 3.6: Determine whether the iteration termination condition is met. If so, output the values ​​of parameters q and f as the exact global optimal solution; otherwise, return to step 3.2.

[0096] Step 4: End.

[0097] Taking the Crab (PSRB0531+21) pulsar as an example, the combined algorithm is used to process a simulation data with an initial phase of 0.3, a period of 33.4ms (frequency 29.9401202HZ), and an observation time of 600s. Figure 2a -c indicates the convergence process of the cross entropy algorithm. Figure 2a The first iteration diagram of the cross entropy algorithm can be represented as follows. Figure 2b Schematic diagram of the second iteration of the cross entropy algorithm, attached Figure 2cSchematic diagram of the third iteration of the cross entropy algorithm. Figure 2a -c shows that the distribution of sample points during the iteration of the cross entropy algorithm is characterized by the sample points continuously converging to the location of the optimal solution. In 2a-c, the points represent sample points and the plane represents the threshold plane.

[0098] The combined algorithm of the present invention is used, and the simulation environment is CPU: Intel CORE I5-6300HQ@2.3GHz, and the compilation environment is: Matlab R2018a. The combined algorithm only takes 51 seconds to obtain the estimated result, and the frequency error of the estimated result is 5.20e-07Hz, and the initial phase error is 6.20e-04 phase. However, if the grid search method is used, it takes about 52 hours to obtain the result of the same accuracy when there is little prior information. Therefore, the scheme of the present invention can greatly improve the speed of calculation.

[0099] The pulse TOA estimation method based on the CE-Adam combined algorithm of the present invention determines the global optimal approximate solution by implementing the cross entropy algorithm on a small amount of observation data, and then uses the Adam algorithm for precise optimization. It can reduce the amount of calculation and improve the solution speed while ensuring reliability, and provides strong support for the rapid implementation of the pulsar navigation algorithm.

[0100] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A pulse TOA estimation method based on CE-Adam combined algorithm, characterized in that: The following steps are involved: Step 1: Determine the target function, the estimated range of the initial phase and the estimated range of the frequency according to the prior information; Step 2: Use the cross entropy algorithm CE to calculate the approximate global optimal solution; Step 3: Using the rough global optimal solution obtained in step 2 as the starting point, apply the Adam algorithm to output the precise global optimal solution; In the step 1: The objective function using expression 1) is: Where: J represents the objective function; λ represents the rate function; t k represents the arrival time of the kth photon, t1 represents the arrival time of the first photon; Q is the total number of photons; φ represents the pulsar phase; α represents the average flow rate of the pulsar signal; β represents the background noise flow rate; h represents the profile function of the pulsar; q represents the initial phase of the pulsar signal, f represents the frequency of the pulsar signal, and the estimated values ​​of q and f are expressed as The estimated range of the initial phase q of the pulsar signal is [0,1]; The estimated range of the frequency f of the pulsar signal is [f'-Δf, f'+Δf], where: f' is the estimated value of the frequency of the pulsar signal at the current moment, f0 is the frequency of the pulsar signal at time t0, is the frequency change rate of the pulsar signal at time t0, Δt represents the time difference between the current time and time t0; Δf is an empirical value.

2. The pulse TOA estimation method based on the CE-Adam combined algorithm according to claim 1, characterized in that: The cross entropy algorithm in step 2 specifically includes the following steps: Step 2.1, m is the number of cycles, take m = 0; give the initial value of the distribution parameter μ q0 , σ q0 , μ f0 and σ f0 , where: μ q0 and σ q0 represents the initial expectation and initial standard deviation of parameter q, μ f0 and σ f0 represents the initial expectation and initial standard deviation of the parameter f; Step 2.2: Make a judgment. If the termination condition σ is met qm <σ qend And σ fm <σ fend , then go to step 2.5; otherwise go to step 2.3; where: σ qend and σ fend Respectively represent the standard deviation thresholds of parameters q and f; σ qm represents the standard deviation of parameter q in the mth cycle; σ fm represents the standard deviation of the parameter f in the mth cycle; Step 2.3: Generate N independent samples based on the probability density function and calculate the objective function value corresponding to the samples; sort the samples in order from small to large according to the objective function value, and select the N samples with the largest objective function value. e elite samples, Ne = ρN, ρ is the quantile parameter; Use expression 3) to obtain the distribution parameters: Where: μ qm represents the expectation of parameter q in the mth cycle; μ fm represents the expectation of the parameter f in the mth cycle; N represents the capacity of the sample; J i represents the objective function value corresponding to the i-th sample; q i and f i represents the values ​​of parameters q and f corresponding to the i-th sample; γ represents the threshold; I{P(J i ≥γ)} represents the indicator function, Step 2.4: According to N e The elite samples update the distribution parameters, taking m=m+1, and return to step 2.2; Step 2.5: Output μ qm and μ fm As a rough global optimal solution.

3. The pulse TOA estimation method based on the CE-Adam combined algorithm according to claim 2 is characterized in that: In step 2.3: The probability density function is expressed as: Among them, μ q0 , σ q0 , μ f0 and σ f0 is the initial distribution parameter, μ q0 and σ q0 represents the initial expectation and initial standard deviation of parameter q, μ f0 and σ f0 represents the initial expectation and initial standard deviation of the parameter f.

4. The pulse TOA estimation method based on the CE-Adam combined algorithm according to claim 3 is characterized in that: The μ q0 The value is 0.5, σ q0 The value is 0.5, μ f0 Take the value of f', σ f0 The value is Δf.

5. The pulse TOA estimation method based on the CE-Adam combined algorithm according to claim 2, characterized in that: The Adam algorithm in step 3 specifically includes the following steps: Step 3.1, take the number of iterations n = 0; Step 3.2, calculate the objective function at point θ n =(q n ,f n ) is the gradient g n ;Record the number of iterations n=n+1; Step 3.3: Use expression 4) to calculate the weighted average m of the gradient at the nth iteration: n and momentum v n : m n =β1m n-1 +(1-β1)g n Among them: β1 and β2 are constants close to 1; m0 = v0 = 0; Step 3.4: Calculate m using expression 5) n and v n Correction value and Step 3.5: Update the value of parameter θ using expression 6): Among them, η represents the step size, ò is a very small constant, is the learning rate; Step 3.6: Determine whether the iteration termination condition is met. If so, output the values ​​of parameters q and f as the exact global optimal solution; otherwise, return to step 3.2; The iteration termination condition is: n -q n-1 <q end And f n -f n-1 <f end , where: q n ,q n-1 、f n 、f n-1 Indicates the values ​​of parameters q and f obtained by the nth and n-1th iterations, q end and f end Indicates the change threshold of parameters q and f.