Dual-station time synchronization method for mobile scenario

Abstract

The application belongs to the technical field of communication and sensing integration, and particularly relates to a mobile scene communication and sensing integrated two-station time synchronization method, which comprises the following steps: step 1, inputting a received signal y(t, n); step 2, obtaining s(t, n) through distance compression; step 3, performing iterative constant false alarm rate detection; step 4, performing preliminary association on the detection result; step 5, initializing a reference target and an association probability; step 6, estimating a time deviation by using an EM algorithm; and step 7, compensating the time deviation based on the estimation result of the EM algorithm. The mobile scene communication and sensing integrated two-station time synchronization method can realize time synchronization of a mobile user and a fixed base station in a two-station sensing process, and solve the problem of the influence of time deviation on sensing accuracy in a communication system. The method uses target information in an environment, does not depend on stable line-of-sight link signals or other fixed reference paths, is suitable for a mobile scene, and has better flexibility.

Description

Technical Field

[0001] This invention belongs to the field of integrated communication and sensing technology, specifically relating to a method for time synchronization between two stations in a mobile scenario. Background Technology

[0002] Integrated Sensing and Communication (ISAC), or simply sensing integration, is one of the key technologies for next-generation wireless networks (6G). Through spectrum sharing, hardware reuse, and algorithm fusion, ISAC endows wireless communication networks with powerful sensing capabilities. Base stations (BS) and user equipment (UE) in the communication network can simultaneously communicate and, with the assistance of ISAC, perceive their surrounding physical environment using a dual-site sensing approach based on communication signals. This structure eliminates the need for dedicated radar transmitters, significantly reducing system hardware costs and enabling wide-area, high-precision sensing services, making it an important direction for the future evolution of communication networks.

[0003] When users possess mobility capabilities (such as drones or vehicle-mounted terminals), they can leverage these capabilities to accumulate observation angles along a trajectory, thereby overcoming the limitations of physical antenna size to improve perception accuracy and even obtain high-resolution two-dimensional or three-dimensional radar images. Integrated sensing and communication systems based on mobile platforms can serve scenarios such as wide-area geographic mapping, infrastructure deformation monitoring, disaster assessment, and low-altitude target reconnaissance, offering advantages such as all-weather operation, high precision, and low cost.

[0004] Precise echo delay and phase information are required in signal processing based on a bi-station communication architecture. Therefore, precise synchronization of signal transmission and reception times at both ends is essential during system operation. However, in separate transmitters and receivers, time offsets are unavoidable due to the use of independent local clock sources and oscillators. Furthermore, the mobile nature of the user platform and the dynamic nature of the reference path information in the environment, coupled with the inability to guarantee a stable line-of-sight link signal, make it difficult to ensure a stable and usable synchronization link. Simultaneously, the long-term coherent accumulation required for multi-view coherent processing places extremely high demands on synchronization accuracy, requiring the measurement of the absolute value of time deviation, exceeding the synchronization capabilities of traditional communication systems. Therefore, a precise time synchronization method needs to be designed to meet the synchronization accuracy requirements of bi-station sensing on mobile platforms. Summary of the Invention

[0005] The purpose of this invention is to provide a dual-station time synchronization method for integrated sensing in mobile scenarios, so as to solve the problem of insufficient time synchronization accuracy in dual-station sensing that includes a mobile platform under the existing communication system.

[0006] This invention proposes a sensor-integrated dual-station time synchronization method for mobile scenarios, the specific steps of which are as follows: The first step is for the receiver to sample according to the given azimuth rate. and observation time Acquire the received signal containing the target echo after propagation through the channel. And record the sampling times in the slow time domain. Simultaneously record the user's location Base station location .

[0007] Among them, the received signal This is the result after the transmitted signal has propagated through the channel. The transmitting end transmits the communication signal as follows: , To record the propagation process of each frame of the signal as quickly as possible, Slow time is used to distinguish signals from different frames and record node motion trajectories. Parameters Indicates the sampling time number for slow time.

[0008] The second step is that the receiving end is based on the transmitted signal. For the collected received signals Range compression is performed to obtain the processed range signal. .

[0009] The distance compression method is as follows:

[0010] in, yes The complex conjugate of . In the integral here , They are respectively and The result after variable substitution; This is a variable used in the integration process, applied over time.

[0011] The third step is to analyze the range signal. Iterative constant false alarm rate (CFAR) detection is performed to obtain target peak information, including peak delay. etc. Among them, subscripts Representing the A slow-time sampling moment signal The The peak value of the target is detected in the fast time domain.

[0012] The detailed process of the constant false alarm rate detection method used in this method is as follows.

[0013] 3.1 Calculate the threshold factor. Based on the given false alarm rate... Calculate the threshold factor .in, The solution to the following equation is:

[0014] in, It is the reference window length. It is the false alarm rate. It is the combination coefficient. These are auxiliary parameters used for summation. Unless otherwise specified, the parameters... It will then be used only as an auxiliary parameter for purposes such as summation.

[0015] 3.2 Extract detection points. For the range signal, sample it in the fast time domain at a given sampling rate. Sampling, wherein the sampling rate satisfies , This represents the bandwidth of the signal. For each sampling point... The system determines whether the signal is a peak value, i.e., whether it exceeds the signal strength of the sampling points on either side in the fast time domain. If it is a peak value, it is extracted as the point to be detected. subscript These are the sampling point numbers for the fast time domain and the slow time domain, respectively; for Pick The sampled values; where Represents the fast time domain. The time corresponding to each numbered sampling point Represents the slow time domain. The time corresponding to each numbered sampling point.

[0016] 3.3 Noise Power Estimation. For a given reference window length... and the length of the protective window Extract the points to be detected The detection windows on both the left and right sides are used, and the power average is calculated separately. The smaller value is taken as the noise estimate. The calculation formula is as follows:

[0017] Specifically, only available points near the boundary are used to calculate the average; if no available points are available on one side, the power average of that side is used as the noise estimate. (Parameters here...) In order to cooperate with the testing point The corresponding noise estimate; The sampling points in the fast time domain and the slow time domain are numbered as follows: of Sampled value; here It remains a summation parameter.

[0018] 3.4 Target discrimination. Calculate the detection threshold. .like Then record it as the target and record the index combination. Otherwise, it is judged as no target.

[0019] 3.5 Main Lobe Removal. After detecting all peak points, for each detected target point, the width is reduced to the peak value. The value of the points within the range is set to .in, The estimated value of the noise floor can be obtained by analyzing the area outside the target region. The value is obtained by averaging the energy.

[0020] 3.6 Iterative testing. Repeat steps 3.2-3.5 until the predetermined number of iterations is reached. Or, the current iteration did not detect any new targets.

[0021] 3.7 Record the target peak detection results. After completing the constant false alarm rate (CFAR) detection, set the target detection index for all targets. Record the corresponding peak value latency And measure peak power according to actual needs. and peak phase Targets with the same sampling time in the slow time domain in the detection results are then sorted according to time delay in the fast time domain. Reorder by size and The peak delay of the output is obtained by sequentially numbering the parameters. and test number Waiting for the results.

[0022] The fourth step is to perform preliminary association of the peak detection results of CFAR. This method uses the Global Nearest Neighbor (GNN) method for preliminary target association. The specific steps are as follows.

[0023] 4.1 Initialize the target trajectory. Use the first slow-time sample... All detections are considered initial observations of the new target. A number is assigned to each new target. Record the time of the most recent observation. Historical delay sequence and observation time series The target state (initially defaulted to active, recorded as disappeared if the target disappears). Record the slow-time sampling moment to be detected. .

[0024] 4.2 Predict target delay. For each activity target... The predicted delay for the current slow time is calculated based on its historical observations. The calculation method typically uses linear extrapolation of the last two observations, as shown below:

[0025] in, and ,as well as and They are respectively and The second to last and second to last elements. For predictions with only one observation, directly take... .

[0026] 4.3 Calculate the correlation cost matrix. For the current slow time... Assuming there exists Activity goals , Each detection value is used to construct the dimension according to the following formula: Association cost matrix :

[0027] in For the corresponding slow time sampling time The correlation cost matrix, Representation matrix The Line number Column elements; For peak latency, The goal The predicted time delay variance can be obtained based on the measured value of system noise or the noise preset in the model. It is the threshold for determining when the target disappears. As of Last time the target was observed slow time, and These are the delay error weight and the disappearance time weight.

[0028] 4.4 Solving the Two-Dimensional Assignment Problem. This method uses the Hungarian algorithm (Munkres algorithm) to find the optimal match with the minimum total cost, obtaining successfully matched target-measure pairs, unmatched targets (predicted without corresponding measurements), and unmatched measurements (new target candidates). The Hungarian algorithm is a commonly used algorithm for assignment problems, and its specific steps are described below.

[0029] a) For input dimension matrix If it is a square matrix, then take Otherwise, by supplementing... Transform it into a square matrix in the following way: Then in Add on the basis All of them are The row, take the result as ;like Then in Add on the basis All of them are The column, take the result as .remember The dimension is .

[0030] b) Row reduction: For the matrix For each row, find the minimum value in that row, and then subtract the minimum value from each element in that row, so that each row contains at least one zero element.

[0031] c) Column Reduction: In the row-reduced matrix, for each column, find the minimum value of that column, and then subtract the minimum value from each element of that column, ensuring that each column contains at least one zero element. This yields the reduced matrix. .

[0032] d) Mark the independent zero elements in the matrix.

[0033] i) Row scan: Check each row one by one. If there is only one unmarked and unerased zero element in the row, mark the zero element as an independent zero element and erase all other zero elements in the column containing the zero element (the erased zero elements do not participate in subsequent checks and marking).

[0034] ii) Column scan: Check each column one by one. If there is only one unmarked and unerased zero element in the column, mark the zero element as an independent zero element and erase all other zero elements in the row containing the zero element (the erased zero elements do not participate in subsequent checks and marking).

[0035] iii) Repeat steps 4.4.di and 4.4.d.ii until no new independent zero elements can be marked.

[0036] iv) Check the number of independent zero elements .like If the position of all independent zeros is optimal, the algorithm ends and proceeds to step 4.4.g; otherwise, proceed to step 4.4.e to continue adjusting.

[0037] e) Based on Kőnig's Theorem in graph theory, cover all zero elements with the fewest possible straight lines and record the corresponding set of straight lines.

[0038] i) Mark all rows that do not contain independent zeros.

[0039] ii) For all zero elements (whether independent or not) that appear in the marked rows, mark the columns in which they are located.

[0040] iii) For all independent zero elements that appear in the marked columns, mark the rows in which they are located.

[0041] iv) Repeat steps 4.4.e.ii and 4.4.e.iii until no new rows or columns are marked.

[0042] v) The minimum set of lines covering all zero elements is the sum of all unlabeled rows and labeled columns. Check the number of lines. ,like If the result is positive, the optimal solution is obtained, and proceed to step 4.4.g; otherwise, proceed to step 4.4.f to continue adjusting.

[0043] f) For the matrix Make adjustments to create new zero elements.

[0044] i) Among all elements not covered by any straight line, find the minimum value, denoted as . .

[0045] ii) Subtract each element of all rows not covered by a line .

[0046] iii) Add to each element of all columns that are already covered by lines .

[0047] iv) Update the matrix according to the results Then return to step 4.4.d.

[0048] g) Output the optimal assignment. The row and column numbers corresponding to the positions of all independent zero elements in the final matrix are the target and observation numbers of the corresponding matching results. If rows or columns with all zeros were padded in step 4.4.a, the matching corresponding to these rows or columns is ignored. The row numbers that do not appear in the matching results are the unmatched target numbers, and the column numbers that do not appear in the matching results are the unmatched measurement numbers.

[0049] 4.5 Update targets. Update each target based on the association results: If the target Successfully matched measurement Then the measurement delay Added to historical delay sequence and will Added to the observation time series And update the most recent observation time .

[0050] If the target If no measurement is matched, calculate the target disappearance time. If greater than The target status is then updated to disappear, and it will no longer participate in subsequent associations.

[0051] If measurement If no active target is matched, refer to step 4.1 to initialize a new active target.

[0052] 4.6 Take Repeat steps 4.2-4.5 until all cases have been traversed. A slow sampling time.

[0053] After GNN association processing, the target set and the corresponding detection association results can be obtained.

[0054] Step 5: Initialize reference targets and association probabilities. After completing the association of all slow-time samples, filter reference targets from the established target trajectories based on their existence time, where the existence time of the selected targets is the duration from the first observation to the last observation. The selected reference targets will be renumbered. To facilitate subsequent discussion, It is referred to as noise.

[0055] Then for each measurement Initialize its association probability with the reference target or clutter. Wherein, if the measurement It belongs to a certain reference target in the association results. Then record , when .like If no reference target is found, it is recorded as being associated with clutter, and the value is taken as... , when Here These are latent variables that indicate the association between the measured data and the target. The parameters here... As an auxiliary parameter, take Integer values ​​between these ranges are used to iterate through the target numbers.

[0056] The sixth step involves estimating the time bias based on the initial reference target and observation records using the Expectation Maximization (EM) algorithm.

[0057] The parameter set to be estimated in the EM algorithm Including: the location of each reference target , where parameters For the goal exist Axis coordinate values; time deviation model The parameter in the text, namely the initial time offset rate of change Prior association probability .

[0058] The input data for the EM algorithm includes: the time delay of all measurements. (Measurement results); Initial association probability (GNN initialization results); initial target probability (Given by system parameters or based on the number of observations of the reference target); Measurement noise variance (Derived from system noise measurements and signal bandwidth, or given by model parameters); Clutter probability density (Prior information, inversely proportional to the measurement range); Convergence threshold (Algorithm parameters); High probability threshold (Algorithm parameters); Normalized scaling vector (and (Same dimension).

[0059] The iterative process of the EM algorithm used in this method is as follows: 6.1 Algorithm Initialization: Initialize algorithm parameters based on input data, such as initial target probability. Scaling vector .

[0060] 6.2 Initial Parameter Estimation: In the first iteration of the EM algorithm, since the initial probabilities have already been obtained in the fifth step... Therefore, skipping the E-step probability estimation, we directly perform the M-step parameter estimation to obtain the initial values ​​of the parameters to be estimated. Specifically, we let... Perform step 6.4.b (target and bias parameter update in M-step) to obtain the initial target and bias parameter estimates. , and the initial target probability estimate By merging, we obtain the initial parameter estimates. After that, he ordered Proceed to step 6.3.

[0061] 6.3 E-Step: The first In the next iteration, the current parameters are calculated according to the following formula. Calculate each measurement Belonging to various targets or clutter Posterior probability:

[0062] Parameters here Still an auxiliary parameter, used to iterate through the target numbers during summation. Probability value. Prior association probability In the The estimated value obtained after one iteration. The likelihood function expression here is:

[0063] in, It is based on the parameter set Calculated target In the Slow time sampling moments The estimated echo delay at a given location is calculated using the following formula:

[0064] Parameters here The base station location and user location recorded in the first step, It is the speed of light.

[0065] 6.4 M-step: the first In this iteration, after completing the E-step posterior probability update, the lower bound of the maximum likelihood function is updated. The update process consists of two parts: prior probability update and target and bias parameter update.

[0066] a) Prior probability update: Update the prior probability according to the following formula. :

[0067] in, For measurement The total number.

[0068] b) Updating the target and bias parameters: Establishing a nonlinear least squares problem:

[0069] in, It is the parameter set of the optimization problem, containing the set of parameters to be estimated. In addition to prior association probability All other parameters besides the residual vector. The components are:

[0070] index set Here As described in step 6.3 The expressions are the same; let them be rewritten as... The function is because It already includes all the variables in its expression, and this formulation reflects its relationship with the variables in the optimization problem. The connection is established. In this method, the Levenberg-Marquardt (LM) algorithm is used to iteratively solve the nonlinear least squares problem, yielding the following results. That is, the updated target position, and the parameters in the time bias model.

[0071] 6.5 Convergence Judgment and Iteration: Calculating Parameter Changes:

[0072] Here Represents a set of variables All parameters in The vector formed by the changes of each step after normalization and scaling. If That is, the change in all parameters after normalization scaling is less than 1. If convergence is achieved, the iteration is terminated; otherwise, let... Repeat steps 6.3 and 6.4 until the algorithm converges.

[0073] Step 7: Compensate for time deviations based on the results obtained from the EM algorithm. This is done using the optimal parameter set output by the EM algorithm. Time offset correction is performed on the original signal:

[0074] Finally, a time synchronization signal is obtained. and time deviation estimate Here and The optimal parameter set The corresponding and The optimal value.

[0075] The advantages and beneficial effects of this invention are as follows: The integrated sensing dual-station time synchronization method for mobile scenarios proposed by this invention can realize the time synchronization between mobile users and fixed base stations in the dual-station sensing process, and solve the problem of time deviation affecting sensing accuracy in communication systems; This method uses target information in the environment, does not rely on stable line-of-sight link signals or other fixed reference paths, is suitable for mobile scenarios, and has better flexibility. Attached Figure Description

[0076] Figure 1 This is a scene diagram of a dual-station sensing system for mobile scenarios, which is the subject of this invention.

[0077] Figure 2 This is a flowchart of the sensor-integrated dual-station time synchronization method for mobile scenarios proposed in this invention.

[0078] Figure 3This is a flowchart of the target detection method based on iterative CFAR proposed in this invention.

[0079] Figure 4 This is a flowchart of the initial detection association method proposed in this invention.

[0080] Figure 5 This is a flowchart of the time deviation estimation method based on the EM algorithm proposed in this invention. Detailed Implementation

[0081] The following will be discussed in conjunction with the accompanying drawings. Figures 1-5 The present invention will be further described in detail with reference to the embodiments.

[0082] The specific implementation methods of the invention are described in detail below, and the overall process flow is as follows: Figure 1 As shown. The parameters used are the carrier frequency. GHz, MHz; MHz, Hz, s, detection range km; base station location m, user's initial position m, user movement speed m / s, velocity direction along the y-axis; measurement noise ns, CFAR parameters , , , Trajectory association parameters , , s, s; EM algorithm parameters ns, , , when , , , .

[0083] Step 1.1: The receiving end acquires the received signal. According to the system parameter settings, a total of [number] samples were collected. Each direction is sampled.

[0084] Step 2.1: Based on the transmitted signal ,calculate and The cross-correlation was used to obtain the distance-compressed signal. .

[0085] Step 3.1: Calculate the threshold factor of CFAR by solving the equation. approximation .

[0086] Step 3.2: Using the sampling rate MHz Sampling along the fast time path yields sampling points. Based on the detection range, each slow time sampling moment has a total of Total sampling points, The system uses 100 sampling points and extracts all peak values ​​from these sampling points as detection points. Taking the first slow-time sampling moment as an example, 308 peak values ​​were detected.

[0087] Step 3.3: Calculate the noise power estimate for each sampling point.

[0088] Step 3.4: Calculate the detection threshold for each sampling point and perform target discrimination. Taking the first slow-time sampling moment as an example, 3 targets were detected.

[0089] Step 3.5: Eliminate the main lobe of each target point and take the noise floor estimate as the average energy of the signal outside the target region after range compression. .

[0090] Step 3.6: Due to the number of iterations up to this round Furthermore, since there are new detection targets, steps 3.2-3.6 are repeated until the cutoff condition is met.

[0091] Step 3.7: After completing the CFAR detection, a total of [number] results were obtained. Take one target, rearrange and number it, and output the peak delay. .

[0092] Step 4.1: Initialize the target trajectory, The six detected targets are considered as new target initialization, and the detection time is recorded as 1. .

[0093] Step 4.2: Predict the six moving targets at the time to be detected. Since there is only one historical observation, the predicted delay is equal to the initial delay.

[0094] Step 4.3: There are 14 detection points in the next time step. Calculate... The correlation cost matrix of bits .

[0095] Step 4.4: Use the Hungarian algorithm to solve the two-dimensional assignment problem. At the current time, the 6 active targets are associated with 6 new detection points, and 8 new detection points are not matched with existing targets.

[0096] Step 4.5: Update targets: Update the measurements of the 6 successfully matched active targets; set the disappearance time of the 6 existing active targets to 0, and keep them active; initialize the 8 unmatched new observations as new active targets, resulting in a total of 14 active targets.

[0097] Step 4.6: Update Still satisfied This leads back to step 4.2 to continue iterating until all instances have been traversed. A slow sampling time.

[0098] Step 5.1: After completing the detection association, a total of 910 targets were obtained. It was calculated that 5 targets satisfied the observation duration being greater than or equal to... Select them as reference targets and number them as targets 1-5, and initialize their association probabilities.

[0099] Step 6.1: Initialize the EM algorithm, with initial target probabilities numbered as follows: Scaling vector The target position parameter is scaled by 1, and the time delay related parameters are scaled according to the speed of light. Scale to match the size of the distance parameter.

[0100] Step 6.2: Based on the initial probabilities, perform initial parameter estimation to obtain the initial target and bias parameter estimates. Combined with the initial target probability estimate, the initial parameter estimate is obtained. Update parameters ,Pick .

[0101] Step 6.3: Perform the E-step to calculate whether each measurement belongs to each target or clutter. The posterior probability. The first detection value at the first slow time sampling point. For example, ,other All values ​​are 0.

[0102] Step 6.4: Perform the M-step to update the prior probabilities, as well as the target and bias parameters. The updated prior probability values ​​are... According to the number For the target and deviation parameters, the index set is first calculated. .by For example, due to ,therefore ,at the same time Then, a nonlinear least squares problem is established based on the index set. The LM algorithm is used to solve this problem, and the updated result is obtained. .

[0103] Step 6.5: Calculate the parameter change, and obtain... greater than Therefore, the algorithm did not converge. Update the parameters. ,Pick Return to step 6.3 and continue iterating until convergence.

[0104] Step 7.1: After the EM algorithm converges, the optimal parameters are obtained. .in ns, ns / s (ppb). Therefore, the estimated time deviation is... ns, where The unit is seconds (s). The original signal is time-shifted and corrected based on the deviation parameter to obtain a time-synchronized signal. .

Claims

1. A method for synesthetic dual-station time synchronization in mobile scenarios, characterized in that, The steps are as follows: Step 1: The receiving end samples according to the given azimuth direction sampling rate. and observation time Acquire the received signal containing the target echo after propagation through the channel. And record the sampling times in the slow time domain. Simultaneously record the user's location Base station location Among them, the received signal The transmitted signal is the result after propagation through the channel; the transmitting end transmits the communication signal as follows: , To record the propagation process of each frame of the signal as quickly as possible, Slow time is used to distinguish signals from different frames and record node motion trajectories. Parameters Indicates the sampling time number for the slow time; Step 2: The receiver bases the signal transmitted at the transmitting end. For the collected received signals Range compression is performed to obtain the processed range signal. ; The distance compression method is as follows: ； in, yes The complex conjugate of; the integral here , They are respectively and The result after variable substitution; This is a variable used in the integration process, applied over time. Step 3: Range signal Iterative constant false alarm rate (CFAR) detection is performed to obtain target peak information, including peak delay. ; where subscript Representing the A slow-time sampling moment signal The The peak value of the target is detected in a fast time domain; Step 4: Perform preliminary correlation of the peak detection results of CFAR; use the Global Nearest Neighbor (GNN) method to perform preliminary correlation of targets; Step 5: Initialize reference targets and association probabilities; after completing the association of all slow-time samples, filter reference targets from the established target trajectories based on the target's existence time. The existence time of the filtered targets... The selected reference targets will be renumbered. To facilitate subsequent discussion, This is denoted as clutter; then for each measurement Initialize the association probability with the reference target or clutter; where, if the measurement It belongs to a certain reference target in the association results. Then record , when ;like If no reference target is found, it is recorded as being associated with clutter, and the value is taken as... , when ; here This refers to the latent variables that indicate the association between the measurement data and the target; the parameters here... As an auxiliary parameter, take Integer values ​​between these ranges are used to iterate through the target numbers; Step 6: Based on the initial reference target and observation records, estimate the time bias using the Expectation-Maximization (EM) algorithm; The parameter set to be estimated in the EM algorithm Including: the location of each reference target , where the parameters For the goal exist Axis coordinate values; time deviation model The parameter in the text, namely the initial time offset rate of change Prior association probability ; The input data for the EM algorithm includes: the time delay of all measurements. Initial association probability Initial target probability ; Measurement noise variance clutter probability density Convergence threshold High probability threshold Normalized scaling vector ; Step 7: Compensate for time deviations based on the results obtained from the EM algorithm; based on the optimal parameter set output by the EM algorithm. Time offset correction is performed on the original signal: ； Finally, a time synchronization signal is obtained. and time deviation estimate ; here and The optimal parameter set The corresponding and The optimal value.

2. The sensor-integrated dual-station time synchronization method for mobile scenarios according to claim 1, characterized in that: Step 3 further includes: calculating a threshold factor; and adjusting the threshold factor based on a given false alarm rate. Calculate the threshold factor ;in, The solution to the following equation is: ； in, It is the reference window length. It is the false alarm rate. It is the combination coefficient. These are auxiliary parameters used for summation; unless otherwise specified, the parameters... It will then be used only as an auxiliary parameter for summation purposes.

3. The sensor-integrated dual-station time synchronization method for mobile scenarios according to claim 2, characterized in that: Step 3 further includes: extracting detection points; sampling the range signal in the fast time domain at a given sampling rate. Sampling, wherein the sampling rate satisfies , The bandwidth of the signal; for each sampling point The system determines whether a point is a peak value, specifically whether it exceeds the signal strength of the sampling points on either side in the fast time domain. If it is a peak value, it is extracted as the point to be detected. subscript These are the sampling point numbers for the fast time domain and the slow time domain, respectively; for Pick The sampled values; where, Represents the fast time domain. The time corresponding to each numbered sampling point Indicates the slow time domain The time corresponding to each numbered sampling point.

4. The sensor-integrated dual-station time synchronization method for mobile scenarios according to claim 3, characterized in that: Step 3 further includes: noise power estimation; for a given reference window length and the length of the protective window Extract the points to be detected The detection windows on both the left and right sides are used, and the average power is calculated separately. The smaller value is taken as the noise estimate. The calculation formula is as follows: ； In this approach, only available points near the boundary are used to calculate the average; if no points are available on one side, the power average of that side is used as the noise estimate. Here, the parameters... In order to cooperate with the testing point The corresponding noise estimate; The sampling points in the fast time domain and the slow time domain are numbered as follows: of Sampled value; here It remains a summation parameter.

5. The sensor-integrated dual-station time synchronization method for mobile scenarios according to claim 4, characterized in that: Step 3 further includes: target discrimination; and calculation of the detection threshold. ;like Then record it as the target and record the index combination. Otherwise, it is judged as no target; main lobe elimination; after completing the detection of all peak points, for each detected target point, the width is centered on the peak point. The value of the points within the range is set to ;in, This is an estimate of the noise floor, obtained by analyzing the area outside the target region. The value is obtained by averaging the energy.

6. The sensor-integrated dual-station time synchronization method for mobile scenarios according to claim 5, characterized in that: Step 3 further includes: cyclic detection; until a predetermined number of iterations is reached. Or, if no new targets are detected in the current iteration; record the peak target detection result; after completing the constant false alarm rate (CFAR) detection, set the index for all target detections. Record the corresponding peak value latency And measure peak power according to actual needs. and peak phase ;Targets with the same sampling time in the slow time domain in the detection results are sorted according to time delay in the fast time domain. Reorder by size and The peak delay of the output is obtained by sequentially numbering the parameters. and test number result.

7. The sensor-integrated dual-station time synchronization method for mobile scenarios according to claim 1, characterized in that: Step 4 further includes the following steps: 4.1 Initialize the target trajectory; use the first slow-time sample All detections are considered initial observations of the new target; each new target is assigned a number. Record the time of the most recent observation. Historical delay sequence and observation time series And the target state; record the slow-time sampling moment to be detected. ; 4.2 Predict target delay; for each activity target The predicted delay for the current slow time is calculated based on historical observations; the calculation method typically uses linear extrapolation of the last two observations, as shown below: ； in, and ,as well as and They are respectively and The second to last and second to last elements; for predictions with only one observation, directly take... ; 4.3 Calculate the correlation cost matrix; for the current slow time Suppose that there exists Activity goals , Each detection value is used to construct the dimension according to the following formula: Association cost matrix : ； in, For the corresponding slow time sampling time The correlation cost matrix, Representation matrix The Line 1 Column elements; For peak latency, The goal The variance of prediction delay, It is the threshold for determining when the target disappears. As of Last time the target was observed Slow time, and These are the delay error weight and the vanishing time weight; 4.4 Solving the two-dimensional assignment problem; using the Hungarian algorithm to find the optimal matching with the minimum total cost, obtaining successfully matched target-measure pairs, unmatched targets, and unmatched measurements; 4.5 Update targets; update each target based on the association results: 4.6 Take Repeat steps 4.2-4.5 until all cases have been traversed. A slow-time sampling moment; After GNN association processing, the target set and the corresponding detection association results are obtained.

8. The sensor-integrated dual-station time synchronization method for mobile scenarios according to claim 7, characterized in that: Step 4.5 further includes: If the target Successfully matched measurement Then the measurement delay Added to historical delay sequence and will Added to the observation time series And update the most recent observation time ; If the target If no measurement is matched, calculate the target disappearance time. If greater than The target status is then updated to disappear, and it will no longer participate in subsequent associations; If measurement If no active target is matched, refer to step 4.1 to initialize a new active target.

9. The sensor-integrated dual-station time synchronization method for mobile scenarios according to claim 1, characterized in that: Step 6 further includes the following steps: 6.1 Algorithm Initialization: Initialize algorithm parameters based on input data, such as initial target probability. Scaling vector ; 6.2 Initial Parameter Estimation: In the first iteration of the EM algorithm, since the initial probabilities have already been obtained in step 5... Therefore, skipping the E-step probability estimation, we directly perform the M-step parameter estimation to obtain the initial values ​​of the parameters to be estimated; the specific operation is to let The initial target and bias parameter estimates are obtained. , and the initial target probability estimate By merging, we obtain the initial parameter estimates. ; after that, he ordered Proceed to step 6.3; 6.3 E-Step: The first In the next iteration, the current parameters are calculated according to the following formula. Calculate each measurement Belonging to various targets or clutter Posterior probability: ； Parameters here Still an auxiliary parameter, used to iterate through the target numbers during summation; probability value Prior association probability In the The estimated value obtained after one iteration; the likelihood function expression here is: ； in, It is based on the parameter set Calculated target In the Slow time sampling moments The estimated echo delay at a given location is calculated using the following formula: ； Parameters here The base station location and user location recorded in the first step, The speed of light; 6.4 M-step: the first In each iteration, after completing the E-step posterior probability update, the lower bound of the maximum likelihood function is updated. 6.5 Convergence Judgment and Iteration: Calculating Parameter Changes: ； Here Represents a set of variables All parameters in The vector formed by the changes of each step after normalization and scaling; if That is, the change in all parameters after normalization scaling is less than 1. If convergence is achieved, the iteration is terminated; otherwise, let... Repeat steps 6.3 and 6.4 until the algorithm converges.

10. The sensor-integrated dual-station time synchronization method for mobile scenarios according to claim 9, characterized in that: Step 6.4 further includes: the update process is divided into two parts: prior probability update and target and bias parameter update; a) Prior probability update: Update the prior probability according to the following formula. : ； in, For measurement The total number; b) Updating the target and bias parameters: Establishing a nonlinear least squares problem: ； in, It is the parameter set of the optimization problem, containing the set of parameters to be estimated. In addition to prior association probability All other parameters besides the residual vector. The components are: ； Index set ; here With step 6.3 With the same expression, the LM algorithm is used to iteratively solve this nonlinear least squares problem, yielding the following result. That is, the updated target position, and the parameters in the time bias model.

Images