A Deep Learning-Based Fiber Optic Bus Clock Synchronization Method

Abstract

The present invention discloses a fiber optic bus clock synchronization method based on deep learning, which includes obtaining a path transmission delay data group and the clock compensation value at the current moment, performing mean processing on the clock compensation value to obtain the mean value of the clock compensation value; obtaining the de-mean result of the clock compensation value according to the mean value of the clock compensation value and the clock compensation value at the current moment; using the de-mean result of the clock compensation value as the input of the deep learning network model to complete the training of the deep learning network model; inputting the de-mean result of the clock compensation value into the trained deep learning network model to obtain the output value of the deep learning network model, and further obtaining the synchronization compensation result; and realizing fiber optic bus clock synchronization by using the synchronization compensation result. The present invention realizes the autonomous synchronization of the fiber optic bus clock and improves the time synchronization accuracy.

Description

Technical Field

[0001] A fiber optic bus clock synchronization method based on deep learning according to the present invention belongs to the technical field of fiber optic buses. Background Art

[0002] Fiber optic buses include communication buses using protocols such as FC-AE bus protocol, SpaceWire bus protocol, and Ethernet protocol with optical fiber as the transmission medium. Fiber optic buses have the characteristics of high transmission rate, high reliability, low cost, light weight, and low latency, and are suitable for fields such as aviation, aerospace, industrial automation, and transportation. Fiber optic buses use optical fiber as the transmission medium. For long-distance transmission modes, single-mode optical fiber can be used to meet the requirements of long-range high-speed communication transmission; for low-cost short-distance application modes, multi-mode optical fiber can be used, which is suitable for short-range low-cost and relatively low-speed networking modes. For the FC-AE bus protocol and SpaceWire bus protocol, a time synchronization mode of time code broadcasting is adopted in the protocol, and the difference in clock drift between the master and slave nodes is not considered. As the working time increases, the clock drift of the nodes becomes larger, resulting in a deterioration of the clock synchronization accuracy. To reduce the impact of clock drift on the clock synchronization accuracy, generally, the synchronization accuracy is improved by increasing the synchronization frequency. However, with the increase of the synchronization frequency, the effective communication bandwidth will decrease, reducing the communication efficiency. Summary of the Invention

[0003] The purpose of the present invention is to overcome the above defects, provide a fiber optic bus clock synchronization method based on deep learning, solve the technical problem that the synchronization accuracy gradually deteriorates with the extension of time in the prior art, and the present invention provides a time synchronization result with higher accuracy without adding additional equipment.

[0004] To achieve the above invention purpose, the present invention provides the following technical solutions:

[0005] A fiber optic bus clock synchronization method based on deep learning includes:

[0006] The first stage:

[0007] S1.1 Obtain the path transmission delay data group of a round-trip communication between the master node and the slave node in the fiber optic bus at the current moment;

[0008] S1.2 Obtain the clock compensation value at the current moment according to the path transmission delay data group;

[0009] S1.3 Perform mean processing on the clock compensation values obtained by repeatedly performing steps S1.1 to S1.2 multiple times to obtain the mean value of the clock compensation values;

[0010] The second stage:

[0011] S2.1 Obtain the mean-removed result of the clock compensation value at the current moment based on the mean value of the clock compensation values and the clock compensation value at the current moment;

[0012] S2.2 Use the multiple mean-removed results of the clock compensation values obtained by repeatedly executing step S2.1 as the input of the deep learning network model to complete the training of the deep learning network model;

[0013] The third stage:

[0014] S3.1 Obtain the mean-removed result of the clock compensation value at the current moment based on the mean value of the clock compensation values and the clock compensation value at the current moment;

[0015] S3.2 Input the mean-removed result of the clock compensation value at the current moment into the trained deep learning network model to obtain the output value of the deep learning network model;

[0016] S3.3 Use the output value of the deep learning network model to obtain the synchronization compensation result for the next moment;

[0017] S3.4 Use the synchronization compensation result for the next moment to achieve fiber optic bus clock synchronization at the next moment.

[0018] Furthermore, the master node in the fiber optic bus is the master control node or the node with the highest clock accuracy in the fiber optic bus network, and the master node is the internal clock synchronization source of the fiber optic bus network.

[0019] Furthermore, the path transmission delay data group of one round-trip communication between the master node and the slave node in the fiber optic bus includes the counting result T0 of the master node starting from the start moment in units of 1 us at the moment when the master node sends the synchronization information, the counting result T1 of the slave node starting from the start moment in units of 1 us at the moment when the slave node receives the master node synchronization information, the counting result T2 of the slave node starting from the start moment in units of 1 us at the moment when the slave node sends the synchronization information, and the counting result T3 of the master node starting from the start moment in units of 1 us at the moment when the master node receives the synchronization information.

[0020] Furthermore, the clock compensation value Tc is calculated according to the following formula:

[0021] Tc = [(T3 - T0) - (T2 - T1)] / 2.

[0022] Furthermore, the deep learning network model is an LSTM deep learning network model;

[0023] The input layer of the LSTM deep learning network model has 1 layer, the hidden layer has 3 layers, and the transfer function of the hidden layer is the tangent sigmoid transfer function.

[0024] Furthermore, the mean-removed result Te of the clock compensation value at the current moment is calculated according to the following formula:

[0025] Te = Tc - Td;

[0026] Wherein, Tc is the clock compensation value at the current moment, and Td is the average value of the clock compensation values.

[0027] Furthermore, the synchronization compensation result Tg at the next moment is calculated according to the following formula:

[0028] Tg = (Tc - Tf);

[0029] Wherein, Tc is the clock compensation value at the current moment, and Tf is the output value of the deep learning network model.

[0030] Furthermore, the method for realizing fiber optic bus clock synchronization at the next moment by using the synchronization compensation result at the next moment is as follows:

[0031] Compensate the synchronization compensation result Tg at the next moment into the clock synchronization result. When the master and slave nodes achieve time synchronization, the count values recorded by each are supposed to be the same during synchronization. However, due to the clock deviation, there are differences in the clock synchronization results of the master and slave nodes. Through the method of the present invention, the differences in the clock synchronization results can be effectively reduced. Specifically, the slave node is compensated here, that is to say, after compensating the slave node, the count value results of the slave node and the master node are approximately the same (with errors), achieving the purpose of synchronization.

[0032] Furthermore, when the fiber optic bus network starts, it enters the first stage. After the fiber optic bus network starts for X1 time, it enters the second stage. After the fiber optic bus network starts for X2 time, it enters the third stage.

[0033] Furthermore, X2 > X1 > 0.

[0034] The present invention has the following beneficial effects compared with the prior art:

[0035] (1) Aiming at the situation that the synchronization accuracy gradually deteriorates over time in the prior art, the present invention uses mean filtering and the LSTM network to separate the synchronization jitter caused by the transmission and reception time jitter and crystal oscillator drift, realizing the compensation of errors, and providing a time synchronization result with higher accuracy without adding additional equipment;

[0036] (2) The present invention uses mean filtering to filter out the time jitter caused by the transmission and reception of the master node / slave node, including the reception and transmission trigger delays, hardware circuit delays, etc., reducing the influence brought by the time jitter;

[0037] (3) The present invention uses the LSTM network to predict the crystal oscillator clock drift of the master node / slave node, compensating for the influence brought by the time synchronization of the crystal oscillator clock;

[0038] (4) The present invention adopts an independent synchronization method and does not require the support of external operations or external devices. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 FIG. is a flowchart of a fiber optic bus clock synchronization method based on deep learning according to the present invention;

[0040] Figure 2 FIG. is a flowchart of the synchronization counting processing of the method of the present invention;

[0041] Figure 3 FIG. is a flowchart of the LSTM network estimation compensation of the method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0042] The following is a detailed description of the present invention, and the features and advantages of the present invention will become clearer and more definite with these descriptions.

[0043] As used herein, the term "exemplary" means "serving as an example, embodiment, or illustration". Any embodiment described herein as "exemplary" is not necessarily to be construed as superior or better than other embodiments. Although various aspects of the embodiments are shown in the drawings, the drawings do not have to be drawn to scale unless specifically noted.

[0044] Regarding the synchronization method of each terminal node of the fiber optic bus communication protocol, under the condition that the clock of the core main communication node is determined, in order to avoid the situation that the clock accuracy of the slave node is relatively lower than that of the master clock, as the working time increases, the synchronization accuracy will deteriorate accordingly, resulting in a reduction in the synchronization accuracy of the entire fiber optic bus network. In the process of realizing the clock synchronization of the entire fiber optic bus network, in order to improve the time synchronization accuracy of the entire fiber optic bus network and not reduce the effective communication bandwidth, it is necessary to fully consider the influence of the clock drift of each terminal node, estimate the clock drift, and use the clock of the core main node of the entire network as the synchronization basis to achieve high-precision clock synchronization of the fiber optic bus without significantly reducing the bandwidth.

[0045] The present invention provides a fiber optic bus clock synchronization method based on deep learning, which can improve the synchronization accuracy of the fiber optic bus during long-term operation and reduce the influence of the clock drift of each communication node on the synchronization accuracy.

[0046] A fiber optic bus clock synchronization method based on deep learning in the present invention uses mean filtering and Long Short-Term Memory (LSTM) networks to estimate and predict the impact of clock drift of terminal nodes on the synchronized clock. Since the access positions of each communication node in the fiber optic bus are fixed, except for the time jitter caused by clock drift and sending and receiving triggers, the path delay of each data transmission is the same. Through the synchronization operation of the master and slave terminal nodes, a set of path transmission delay data can be obtained, including the path transmission delay, sending and receiving jitter, and the delay caused by clock drift jitter, etc. Both the master and slave communication nodes include a counter, and each counter uses the clock of this communication node as the timing clock. The master node calculates the path transmission delay data of a round-trip communication for calculating the synchronized clock compensation value. After being processed by mean filtering and de-meaning, this path delay data further reduces the impact of measurement noise caused by sending and receiving jitter. The data processed by mean filtering is used as the learning sample of the LSTM deep learning network to complete the training of the network model. According to the clock accuracy of the terminal device, when the fiber optic bus network starts, it enters the first stage. After the fiber optic bus network starts for X1 time, it enters the second stage. After the fiber optic bus network starts for X2 time, it enters the third stage, and the trained LSTM deep learning network is enabled to predict the time jitter caused by clock drift. X1 time is the result of the synthesis of the clock accuracy of the master and slave nodes and the user's requirement for the final synchronization accuracy, and X2 time is determined by the CPU performance calculated by the master node model. For example: the master time accuracy is 5*10 -6 , the slave clock accuracy is 5*10 -5 , if the synchronization accuracy is 5 ms after 10 hours, X1 time is 10 min, the computer CPU has a main frequency of 2.5 GHz, and the model training iterates 50 times, X2 time is about 15 min.

[0047] On this basis, the LSTM prediction result is used as the clock drift result to compensate the clock synchronization result.

[0048] The master node is the master control node in the fiber optic network or the node with the highest clock accuracy, and is the internal clock synchronization source of the fiber optic bus network.

[0049] The set of path transmission delay data includes the counting result T0 of the master node starting from the start time in units of 1 us when the master node sends the synchronization information; the counting result T1 of the slave node starting from the start time in units of 1 us when the slave node receives the master node synchronization information; the counting result T2 of the slave node starting from the start time in units of 1 us when the slave node sends the synchronization information; the counting result T3 of the master node starting from the start time in units of 1 us when the master node receives the synchronization information. The path delay result (clock compensation value) is Tc = [(T3 - T0) - (T2 - T1)] / 2

[0050] The mean filtering is the process of accumulating the path delay results of the previous N calculations and dividing the sum by N.

[0051] The de-mean processing is to subtract the current path delay result from the result of the mean filtering.

[0052] The number of input layers of the LSTM deep learning network is 1, the number of hidden layers is 3, the number of hidden layer nodes is 20, the output value of the output layer is 1, and the transfer function of the hidden layer is the tangent sigmoid transfer function.

[0053] Embodiment:

[0054] A fiber optic bus clock synchronization method based on deep learning utilizes the transmission of synchronization information between the master and slave nodes during the synchronization process, and obtains the deviation of the clock compensation value through calculating the time difference between the receiving time and the sending time, as well as mean filtering and de-mean calculation. On this basis, the LSTM deep learning network is used to estimate the de-mean output result, estimate the clock drift of the master and slave nodes, and compensate the counting result of the slave node, so as to achieve high-precision clock synchronization between the master and slave nodes.

[0055] As Figure 1 shown, the steps are as follows:

[0056] Step 1: As Figure 2 , record the sending and receiving times T0 to T3 of the master and slave node synchronization counts respectively, and obtain the clock compensation value Tc according to the following formula. In the figure, Ta represents the delay between the sending and receiving of the master node synchronization information, and Tb represents the delay between the slave node receiving the synchronization information and forwarding the synchronization information:

[0057] Tc = [(T3 - T0) - (T2 - T1)] / 2;

[0058] Step 2: Perform a mean calculation on the clock compensation value Tc to obtain the result Td, and subtract the current clock compensation value Tc from Td to obtain the de-mean result Te;

[0059] Each moment corresponds to a Tc. Due to the jitter of the master and slave clocks (mainly including frequency random walk, phase random walk, frequency flicker noise, phase flicker noise, phase white noise), each Tc value is different. As time drifts, the de-mean reflects the superposition of these noises.

[0060] Step 3: As Figure 3 , the data Te generated within a period of time after the synchronization starts is input into the LSTM deep learning network in chronological order to start the training of the LSTM network model.

[0061] Step 4: After the model completes training, the data Te obtained at the current moment is used as the input of the LSTM deep learning network, and its output result is Tf. The synchronous compensation result Tg = (Tc - Tf) is obtained. Tf is actually the prediction of the clock drift between the master and slave nodes, and Tg is equivalent to the result of compensating for the clock drift.

[0062] Step 5: The slave node compensates Tg into its counting result to achieve high-precision synchronization between the master and slave nodes.

[0063] The present invention has been described in detail in combination with specific embodiments and exemplary examples. However, these descriptions should not be construed as limitations on the present invention. Those skilled in the art understand that without departing from the spirit and scope of the present invention, various equivalent replacements, modifications or improvements can be made to the technical solutions and their implementation manners of the present invention, and these all fall within the scope of the present invention. The protection scope of the present invention is subject to the appended claims.

[0064] The content not described in detail in the specification of the present invention belongs to the well-known technology of those skilled in the art.

Claims

1. A fiber optic bus clock synchronization method based on deep learning, characterized in that Including: The first stage: S1.1 Obtain the path transmission delay data group of one round-trip communication between the master node and the slave node in the fiber optic bus at the current moment; S1.2 Obtain the clock compensation value at the current moment according to the path transmission delay data group; S1.3 Perform mean processing on the clock compensation values obtained by repeatedly performing steps S1.1 - S1.2 multiple times to obtain the mean value of the clock compensation values; The second stage: S2.1 Obtain the result of removing the mean value of the clock compensation value at the current moment according to the mean value of the clock compensation value and the clock compensation value at the current moment; S2.2 Use the results of removing the mean value of multiple clock compensation values obtained by repeatedly performing step S2.1 as the input of the deep learning network model to complete the training of the deep learning network model; The third stage: S3.1 Obtain the result of removing the mean value of the clock compensation value at the current moment according to the mean value of the clock compensation value and the clock compensation value at the current moment; S3.2 Input the result of removing the mean value of the clock compensation value at the current moment into the trained deep learning network model to obtain the output value of the deep learning network model; S3.3 Use the output value of the deep learning network model to obtain the synchronization compensation result at the next moment; S3.4 Use the synchronization compensation result at the next moment to achieve clock synchronization of the fiber optic bus at the next moment; The path transmission delay data group of one round-trip communication between the master node and the slave node in the fiber optic bus includes the counting result T0 of the master node starting from the start time in units of 1 us at the moment when the master node sends the synchronization information, the counting result T1 of the slave node starting from the start time in units of 1 us at the moment when the slave node receives the master node synchronization information, the counting result T2 of the slave node starting from the start time in units of 1 us at the moment when the slave node sends the synchronization information, and the counting result T3 of the master node starting from the start time in units of 1 us at the moment when the master node receives the synchronization information; The result of removing the mean value of the clock compensation value Te at the current moment is calculated according to the following formula: Te = Td - Tc; Where, Tc is the clock compensation value at the current moment, and Td is the mean value of the clock compensation value; The clock compensation value Tc is calculated according to the following formula: Tc = [(T3 - T0) - (T2 - T1)] / 2.

2. The method for synchronizing the optical fiber bus clock based on deep learning according to claim 1, characterized in that The master node in the fiber optic bus is the master control node or the node with the highest clock accuracy in the fiber optic bus network, or is the internal clock synchronization source of the fiber optic bus network.

3. The method for fiber optic bus clock synchronization based on deep learning according to claim 1, characterized in that The deep learning network model is an LSTM deep learning network model; The number of input layers of the LSTM deep learning network model is 1, the number of hidden layers is 3, and the transfer function of the hidden layer is the tangent sigmoid transfer function.

4. A fiber optic bus clock synchronization method based on deep learning according to claim 1, characterized in that The synchronization compensation result Tg at the next moment is calculated according to the following formula: Tg = (Tc - Tf); Where, Tc is the clock compensation value at the current moment, and Tf is the output value of the deep learning network model.

5. A fiber optic bus clock synchronization method based on deep learning according to claim 1, characterized in that The method for achieving fiber optic bus clock synchronization at the next moment by using the synchronization compensation result at the next moment is: Compensate the synchronization compensation result Tg at the next moment to the slave node.

6. The method for fiber optic bus clock synchronization based on deep learning according to claim 1, wherein, When the fiber optic bus network starts, it enters the first stage. After the fiber optic bus network starts for X1 time, it enters the second stage. After the fiber optic bus network starts for X2 time, it enters the third stage.

7. A fiber optic bus clock synchronization method based on deep learning according to claim 6, characterized in that, X2 > X1 > 0.

Images