Clock synchronization method based on subspace decomposition

Abstract

The invention relates to the technical field of communication, and provides a clock synchronization method based on subspace decomposition in order to reduce communication overhead of a master end and a slave end and improve clock synchronization precision at the same time. Initial phases of signals with different wavelengths are calculated by using subspace orthogonality, and the clock difference between the master end and the slave end is calculated according to a steering vector, so that the slave clock time is corrected, and the anti-noise performance is improved; one-way comparison is adopted, the method does not depend on the symmetry of a two-way signal propagation path, the clock synchronization precision is improved, the consumption of system communication resources is reduced, and the efficiency is higher.

Description

Technical Field

[0001] The invention relates to the field of communication technology, and in particular to a clock synchronization method based on subspace decomposition. Background Art

[0002] Traditional wireless LANs have problems such as small coverage, severe signal attenuation, and signal blind spots or dead spots. To address this problem, the industry has proposed a variety of networking methods to solve these problems, among which FTTR (Fiber To The Room) and EasyMesh networking are considered to be feasible full coverage solutions.

[0003] FTTR is a technology that extends fiber optic cabling to every room in the home, aiming to provide high-speed, full coverage, low latency and high stability network connections. FTTR ensures that every room has a stable network connection by directly introducing optical fiber into each room, eliminating signal blind spots and weak signal areas.

[0004] The EasyMesh architecture is a mesh network architecture based on Wi-Fi technology, which achieves seamless coverage and intelligent management through multiple wireless access points (APs). The introduction of EasyMesh technology can effectively solve the shortcomings of traditional wireless LANs, mainly in the following aspects: EasyMesh achieves seamless Wi-Fi coverage through multiple Mesh nodes; dynamically selects the best path for data transmission to improve network stability; the EasyMesh architecture supports multiple wireless channels to reduce channel congestion; the Mesh network can intelligently allocate bandwidth according to device and application requirements to ensure that each device can obtain the best connection quality and speed.

[0005] Time synchronization plays a fundamental role in networking technology, and is particularly critical for FTTR and EasyMesh networking technologies. In FTTR wired networking, multiple network devices (such as fiber terminal equipment, optical splitters, etc.) need to work together. FTTR fiber networks require high-precision time synchronization to ensure that data packets can be sent and received on time. The EasyMesh architecture requires seamless switching between different Mesh nodes, and time synchronization can ensure that there will be no network interruption or delay due to switching nodes.

[0006] The existing technology mostly uses a two-way time synchronization method to obtain clock error when performing time synchronization. The two-way time synchronization method relies on the symmetry of the two-way signal propagation path. The asymmetric propagation delay of the link (such as the delay difference introduced by different transmission media, routes or devices) will lead to a decrease in time synchronization accuracy; two-way communication requires at least two comparisons, and the time synchronization frame is embedded in the signal frame, which increases the system bandwidth and energy consumption requirements, especially in high-frequency synchronization operations; the two-way comparison depends on the communication quality. The noise and interference in the communication link will affect the signal quality. If the one-way comparison fails, it will affect the time synchronization quality of the entire cycle. Summary of the invention

[0007] In order to reduce the communication overhead between the master and slave ends while improving the clock synchronization accuracy, the present application provides a clock synchronization method based on subspace decomposition.

[0008] The technical solution adopted by the present invention to solve the above problems is: The clock synchronization method based on subspace decomposition includes: Step 1: The slave end sends multiple light waves of different wavelengths to the master end; Step 2: The master end obtains a steering vector based on the received light waves of different wavelengths and constructs a covariance matrix; Step 3, performing eigenvalue decomposition on the covariance matrix to decompose the covariance matrix into a signal subspace and a noise subspace; Step 4: construct a spectral function based on the orthogonality between the noise subspace and the steering vector; Step 5, obtaining the clock error based on the spectral function; Step 6: The master sends the clock difference to the slave, and the slave completes time synchronization based on the clock difference.

[0009] Furthermore, after receiving light waves of different wavelengths, step 2 further includes: selecting a wavelength, and constructing a covariance matrix based on the selected light wave.

[0010] Furthermore, step 2 selects the wavelength according to the signal-to-noise ratio.

[0011] Furthermore, step 2 selects the wavelength according to a weighted function of the signal-to-noise ratio and the wavelength.

[0012] Furthermore, in step 2, an ant colony algorithm is used to obtain the optimal solution of the weighted function.

[0013] Furthermore, step 5 is specifically as follows: determine whether there is a relationship model between the clock error and the environmental information; if not, obtain the environmental information and the corresponding clock error; when the amount of environmental information and the clock error meets the threshold, establish a relationship model between the clock error and the environmental information; if it exists, obtain the current environmental information, bring the current environmental information into the relationship model between the clock error and the environmental information to obtain the current predicted clock error, and obtain the clock error based on the predicted clock error and the spectral function.

[0014] Furthermore, a relationship model between the clock error and the environmental information is established based on the change in the environmental information and the corresponding change in the clock error.

[0015] Furthermore, the environmental information is temperature.

[0016] Furthermore, the specific steps for obtaining the clock error based on the predicted clock error and the spectral function are: divide the theoretical maximum clock error into N grids on average, estimate the grid where the real clock error is located based on the predicted clock error, and traverse the grid and several adjacent grids to obtain the clock error.

[0017] Compared with the prior art, the present invention has the following beneficial effects: by using a plurality of optical signals of different wavelengths to construct a covariance matrix and adopting a subspace decomposition method, the initial phase of signals of different wavelengths is calculated using the subspace orthogonality, and the clock difference between the master and slave ends is calculated according to the steering vector, thereby correcting the slave clock time and improving the anti-noise performance; by adopting a one-way comparison, which does not rely on the symmetry of the two-way signal propagation path, the clock synchronization accuracy is improved, the system communication resource consumption is reduced, the risk of communication failure is reduced, the anti-noise performance is improved, and the efficiency is higher. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flow chart of the clock synchronization method based on subspace decomposition; Figure 2 This is the flow chart of the ant colony algorithm. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0020] like Figure 1 As shown, the clock synchronization method based on subspace decomposition includes: Step 1: Send multiple light waves of different wavelengths from the slave end to the master end. In the EasyMesh networking architecture, the master end is connected to the main Controller router of the EasyMesh architecture, and the slave end is connected to each Mesh node of the network; in the FTTR architecture, the optical distribution gateway ODN is the master end of time synchronization, and the ONT is the slave end of time synchronization.

[0021] Step 2: The master end obtains the steering vector based on the received light waves of different wavelengths and constructs a covariance matrix.

[0022] The mixed signals of different wavelengths received by the master end at a certain moment are expressed as: ,in It is the change of different wavelength signals transmitted in optical fiber. It is a complex number with amplitude and phase. is the frequency of signals of different wavelengths, is the initial phase of signals of different wavelengths. Theoretically, the slave end sets all initial phases to 0. However, due to the time asynchrony between the master and slave ends and the influence of optical dispersion effect, is a function of the clock difference between the master and slave ends, It is the noise in the process of optical fiber transmission and signal demodulation, including thermal noise, electromagnetic noise, etc. K is the total number of wavelengths. is a complex exponential, j is a complex unit, and t is the synchronization period (since the synchronization period is very short, the clock error does not change significantly and can be regarded as a moment). , i.e., the steering vector, reflects the impact of the optical fiber channel on signals of different wavelengths, and T represents transposition.

[0023] The fiber channel is relatively stable, and the clocks at both ends of the master and slave are relatively stable, that is, the clock difference cannot jump. In a synchronization cycle, multiple snapshots are taken to construct the covariance matrix of the signal: , where x represents a random vector (signal sample), represents the conjugate transpose of x, and E[] is the expected operation. If there are N samples, the time average can be used instead of the expectation: , where N is the number of snapshots of the signal sampling, and n represents the nth sampling point. Represents the value of signal x at the nth sampling point, and H is the conjugate transpose.

[0024] Step 3, performing eigenvalue decomposition on the covariance matrix to decompose the covariance matrix into a signal subspace and a noise subspace; ,in, is the signal subspace, corresponding to the largest eigenvalue (signal component), is the noise subspace, corresponding to the remaining smaller eigenvalues ​​(noise components), is the diagonal matrix of eigenvalues ​​of the signal subspace, is the diagonal matrix of eigenvalues ​​of the noise subspace.

[0025] Step 4: construct a spectral function based on the orthogonality between the noise subspace and the steering vector; According to the subspace decomposition theory, the noise subspace The signal vector Orthogonal, the signal's steering vector Indicates that when the scanning clock error is The corresponding calculated steering vector is used to construct the spectral function .

[0026] Step 5: Obtain the clock error based on the spectral function. Scan different clock errors Value, when the scanning clock difference Equal to the true clock error hour, A peak will appear. By finding the peak of the spectrum function, the clock difference between the master and slave can be estimated.

[0027] Step 6: The master sends the clock difference to the slave, and the slave completes time synchronization based on the clock difference.

[0028] Light waves of different wavelengths are affected by the dispersion effect in the optical fiber, and their phase changes are different, thus forming a steering vector; the master end receives signals of different wavelengths sent by the slave end at a predetermined time, and demultiplexes the signals of different wavelengths to form a covariance matrix; the covariance matrix is ​​subjected to eigenvalue decomposition and divided into a signal subspace and a noise subspace; the steering vector is determined using the orthogonality of the subspace, and the initial phases of signals of different wavelengths are calculated based on the dispersion effect, and the clock difference between the two ends is calculated accordingly; the master end sends the clock difference to the slave end, and the slave end compensates to complete time synchronization.

[0029] When the number of snapshots is too large and the matrix dimension is too large (determined by the number of signals of different wavelengths), the computational complexity and storage requirements of the subspace decomposition method increase rapidly. Due to factors such as uneven gain of optical amplifiers, poor performance of wavelength selection components, and loss characteristics of the optical fiber itself, the signal noise of a certain wavelength may be large in the optical fiber, while the signal noise of other wavelengths is small. The subspace algorithm separates the signal subspace and the noise subspace by decomposing the eigenvalues ​​of the covariance matrix. Under the condition of uneven noise, if the strong noise level is close to the weak signal level, the gap between the eigenvalues ​​of the covariance matrix becomes smaller, and the signal subspace and the noise subspace are difficult to separate accurately, which in turn affects the estimation accuracy. In addition, the wavelength difference also affects the performance of the subspace decomposition method. The performance of the wavelength selection component, the characteristics of the optical fiber itself, and the nonlinear effect may affect the system's resolution of the wavelength, thereby affecting the estimation accuracy. Therefore, when executing the subspace decomposition method, in step 2, the signals of different wavelengths currently involved in the calculation are first selected.

[0030] If the combination is selected based only on the estimated signal-to-noise ratio (with prior information, calculated based on the signal power of a specific frequency or format on the spectrum and the total power of the received signal, or other calculation methods can be used), the calculated accuracy may not be the optimal solution. Therefore, the signal-to-noise ratio and the wavelength weighting function are combined to determine whether the currently selected signal combination is the optimal combination. However, when there are many wavelengths, the number of subsets of wavelength signal combinations is a combination number, and its value increases exponentially with the increase in the number of wavelengths. The weighting function of all subsets cannot be calculated. At this time, the ant colony algorithm can be used to estimate the wavelength combination corresponding to the optimal solution, such as Figure 2 As shown, specifically including: 1. Initialization phase. In the initialization phase, pheromones are first assigned to each wavelength signal and the objective function (fitness) is evaluated. The pheromones are evenly distributed and the objective function is a weighted function of wavelength and signal-to-noise ratio. Then, n wavelength combinations, i.e., paths, are randomly generated according to the probability function, and each path corresponds to a certain pheromone and fitness.

[0031] 2. Evaluate the quality of each path by calculating the objective function consisting of the wavelength and the estimated signal-to-noise ratio weight, and filter out obviously abnormal wavelength combinations (such as wavelengths with excessive signal-to-noise ratio). Reduce pheromones for the filtered wavelength combinations.

[0032] 3. Update pheromones. Reduce the pheromones of all old paths to avoid falling into local optimality, and add pheromones to the wavelength combinations with higher evaluated fitness to strengthen their attractiveness in subsequent iterations.

[0033] 4. Generate new paths based on pheromones. Regenerate n paths after updating pheromones.

[0034] 5. Check the convergence condition. If the fitness change of the current optimal path is less than the threshold or exceeds the maximum number of iterations, stop the algorithm.

[0035] 6. Output the optimal path and its corresponding wavelength combination.

[0036] Since the subspace calculation needs to calculate the covariance matrix of the received signal and its decomposition, when the number of snapshots is too large and the matrix dimension is too large, its computational complexity and storage requirements increase rapidly. In addition, in order to determine the clock error, step 5 needs to scan different clock errors, and each time the scan is performed, the value of the pseudo spectrum at the scan value needs to be calculated. As the scanning accuracy increases, its computational complexity and storage requirements also continue to increase.

[0037] Since the optical fiber channel is relatively stable, and the channel and the clock conditions at both ends of the master and slave are subject to external factors of the use environment, including temperature and humidity, the clock difference between the master and slave must change slowly and with a limited range. Therefore, physical sensors such as temperature and humidity can be set at both ends of the master and slave to measure the environment.

[0038] At the same time, the theoretical maximum clock error of the synchronization device (determined by the hardware device, limited by the physical properties of the two clock crystals, and relatively stable) is evenly divided into N grids. The clock error of the master and slave ends is recorded and its grid position is determined each time the time synchronization is completed. In the next time synchronization cycle, the clock error at both ends can be roughly estimated based on the clock error statistics of the previous cycle and the changes in the environmental measurement values ​​of the sensor information (the clock error-temperature drift curve of the clocks at both ends can be tested and obtained), and the direction of the clock error change can be determined. The grids near the clock error measured in the previous cycle can be scanned. Through gridding and prediction based on environmental information, the amount of calculation is simplified to one-Nth of the original, which significantly reduces the amount of calculation.

[0039] Specifically include: 1. Obtain environmental information The method of obtaining environmental information includes steps such as sensor selection, deployment, and data collection. First, select appropriate sensor models such as temperature and humidity according to environmental conditions. According to the environment in which the time synchronization device is located, the interaction cycle between the master and slave ends, and the speed of actual environmental parameter changes, estimate the measurement range, accuracy, response time, and environmental tolerance, and select a suitable sensor.

[0040] Secondly, ensure that the sensor layout is reasonable and covers the target area. Considering the heat source in the time synchronization device and the slow change of environmental parameters, multiple sensors should be set up in the optical fiber channel and the time synchronization device, as well as near the device, to accurately capture environmental parameters, while paying attention to not occupying too many resources and saving costs.

[0041] When collecting environmental data, a reasonable sampling frequency should be designed to balance real-time performance and energy consumption. Its frequency should at least exceed the clock error estimation frequency, but it should not be too large to occupy computing resources. After data collection, data preprocessing should be performed, including denoising, calibration, time synchronization and other steps to improve data quality and consistency.

[0042] 2. Establish a relationship model between clock error and environmental information Since the optical fiber channel is relatively stable and has relatively uniform properties, a certain type of optical fiber has the same rules. Therefore, firstly, the test is carried out in a specific environment to obtain actual data including clock error and environmental information, and then the model is built based on the actual data, and the model parameters are estimated using statistical or machine learning methods.

[0043] In order to make full use of prior information, use physical laws to improve model efficiency and weaken the impact of noise, this embodiment establishes a differential model, that is, instead of using the current value of the measured environmental information, the differential result of the current environmental information and the previous environmental information is used to establish a model with the pre-recorded clock difference information.

[0044] 3. Using the model to estimate clock error In this embodiment, a differential model is established. Therefore, after differential data is input, the trend of the clock difference is estimated. The clock difference information at the previous moment should be combined to obtain the estimated value of the clock difference at the current moment.

[0045] 4. Use estimated clock error to simplify calculations First, according to the parameter settings and internal principles of the time synchronization device, the maximum range of the time synchronization error is determined as the grid to be divided. Secondly, according to the amplitude of the clock error fluctuation within a fixed period and the accuracy requirements of the time synchronization device, the resolution of the clock error grid is defined, and the size and boundary of each clock error grid unit are determined.

[0046] Next, the maximum range of time synchronization error is divided into uniform grid cells, each of which represents a clock error range.

[0047] Finally, in the current time synchronization period, the grid where the real clock error is located is estimated in combination with the pre-estimated clock error, and the grid and several adjacent grid cells are traversed or calculated to complete the comprehensive coverage within the estimated clock error range, complete the clock error estimation, and improve the estimation efficiency.

Claims

1. A clock synchronization method based on subspace decomposition, characterized in that: include: Step 1: The slave end sends multiple light waves of different wavelengths to the master end; Step 2: The master end obtains a steering vector based on the received light waves of different wavelengths and constructs a covariance matrix; Step 3, performing eigenvalue decomposition on the covariance matrix to decompose the covariance matrix into a signal subspace and a noise subspace; Step 4: construct a spectral function based on the orthogonality between the noise subspace and the steering vector; Step 5, obtaining the clock error based on the spectral function; Step 6: The master sends the clock difference to the slave, and the slave completes time synchronization based on the clock difference.

2. The clock synchronization method based on subspace decomposition according to claim 1, characterized in that: After receiving light waves of different wavelengths, step 2 also includes: selecting a wavelength and constructing a covariance matrix based on the selected light wave.

3. The clock synchronization method based on subspace decomposition according to claim 2, characterized in that: Step 2: Select the wavelength based on the signal-to-noise ratio.

4. The clock synchronization method based on subspace decomposition according to claim 2, characterized in that: Step 2 selects the wavelength based on a weighting function of signal-to-noise ratio and wavelength.

5. The clock synchronization method based on subspace decomposition according to claim 4, characterized in that: Step 2 uses the ant colony algorithm to obtain the optimal solution of the weighted function.

6. The clock synchronization method based on subspace decomposition according to any one of claims 1 to 5, characterized in that: Step 5 is specifically as follows: determine whether there is a relationship model between the clock error and the environmental information; if not, obtain the environmental information and the corresponding clock error; when the amount of environmental information and the clock error meets the threshold, establish a relationship model between the clock error and the environmental information; if it exists, obtain the current environmental information, bring the current environmental information into the relationship model between the clock error and the environmental information to obtain the current predicted clock error, and obtain the clock error based on the predicted clock error and the spectral function.

7. The clock synchronization method based on subspace decomposition according to claim 6, characterized in that: A relationship model between clock error and environmental information is established based on the change in environmental information and the corresponding change in clock error.

8. The clock synchronization method based on subspace decomposition according to claim 7, characterized in that: The environmental information is temperature.

9. The clock synchronization method based on subspace decomposition according to claim 7, characterized in that: The specific steps for obtaining the clock error based on the predicted clock error and the spectral function are as follows: divide the theoretical maximum clock error into N grids on average, estimate the grid where the actual clock error is located based on the predicted clock error, and traverse the grid and several adjacent grids to obtain the clock error.

Images