A Low-Complexity Optical Communication Carrier Phase Recovery Method
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-12-05
- Publication Date
- 2026-08-14
AI Technical Summary
[0004]为了解决相干光通信系统中载波相位恢复算法BPS复杂度较高的问题，本发明的目的是提供一种低复杂度光通信载波相位恢复方法，主要分为两阶段，第一阶段采用改进的DBSCAN算法对信号进行聚类，获得聚类中心；第二阶段对聚类中心进行盲相位搜索算法BPS处理，将BPS处理结果应用于对信号的载波相位恢复中，提高光通信接收端信号处理精度
[0039]1、本发明公开的一种低复杂度光通信载波相位恢复方法，采用先聚类再对聚类中心进行盲相位搜索BPS处理，相比于对全部数据进行BPS的方法，减少BPS处理的数据量，提高载波相位恢复的效率。
Smart Images
Figure CN117675028B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a low-complexity optical communication carrier phase recovery method, belonging to the field of optical fiber communication technology. Background Technology
[0002] Optical fiber communication, a type of optical communication, uses optical fibers as the medium for information transmission and features high communication capacity, resistance to electromagnetic interference, and low signal interference. With the rapid development of technologies such as big data and the Internet of Things, the demand for high-speed, high-capacity transmission systems is increasing. Traditional direct detection techniques can no longer meet the needs of this high-speed, large-scale information development. Coherent optical communication technology has gained widespread attention due to its high spectral efficiency, high receiver sensitivity, and support for multiple modulation formats. However, coherent optical communication systems are characterized by high cost and high power consumption, which hinders their application in cost- and power-sensitive short-distance transmission systems. Therefore, research on reducing the complexity of coherent optical communication systems is essential.
[0003] Due to spontaneous emission and various noise effects, the coherent light generated by stimulated emission in a laser is subject to random frequency and phase perturbations. The spectrum of the laser output light is an envelope with a certain width, and its half-maximum width is the laser linewidth. The existence of the laser linewidth will cause the signal constellation points to rotate, so an algorithm is needed to recover the carrier phase of the signal. Commonly used carrier phase recovery algorithms in coherent optical communication systems include blind phase search (BPS) and Viterbi-Viterbi phase estimation algorithms. The Viterbi-Viterbi phase estimation algorithm first removes the modulation information of the QPSK signal through a fourth power operation, and then averages multiple adjacent symbols to estimate the phase noise of the laser. However, this algorithm has low tolerance for laser linewidth. BPS is widely used because of its high linewidth tolerance and ability to handle signals with various modulation formats, but its complexity is very high, which may require significant resource consumption in practical applications. Therefore, research on reducing its complexity is also very necessary. Summary of the Invention
[0004] To address the high complexity of the Blind Phase Search (BPS) algorithm in coherent optical communication systems, this invention aims to provide a low-complexity carrier phase recovery method. The method consists of two stages: the first stage uses an improved DBSCAN algorithm to cluster the signal and obtain cluster centers; the second stage applies the Blind Phase Search (BPS) algorithm to the cluster centers and uses the BPS results to recover the carrier phase, thus improving the signal processing accuracy at the optical communication receiver. This invention achieves carrier phase recovery with low complexity, realizing a low-complexity carrier phase recovery method.
[0005] The objective of this invention is achieved through the following technical solution:
[0006] This invention discloses a low-complexity optical communication carrier phase recovery method, comprising the following steps:
[0007] Step 1: Divide the input data into groups of Nc data points each. This batching facilitates the input into subsequent batch processing steps 2 through 4. The value of Nc is an empirical value, which should be adjusted appropriately based on different signal noise levels.
[0008] Step 2: Use the improved DBSCAN algorithm to perform a single clustering of the data after the batching in Step 1.
[0009] The DBSCAN algorithm has two parameters: the neighborhood search radius ε and the minimum number of points (min_points) that can stand alone as a class. If the number of clusters obtained is the same as the number of modulation points M of the input signal, the clustering result is considered correct; otherwise, ε or min_points needs to be adjusted, and the clustering process needs to be repeated. After the clustering result is correct, the average coordinates of data points in the same class are taken to calculate M cluster centers.
[0010] The specific implementation of the clustering method described in step two is as follows:
[0011] The DBSCAN algorithm primarily uses two parameters: the neighborhood search radius ε and the minimum number of points (min_points) that can stand alone as a class. The improved DBSCAN algorithm replaces the circular search neighborhood with a square neighborhood, where the extension length 'a' represents half the side length of the square neighborhood. The neighborhood boundaries are defined as the horizontal and vertical lines extending 'a' in the corresponding directions from the starting point. These boundaries form the square search neighborhood. By comparing the coordinates of the search point with those of the square neighborhood boundaries, it is determined whether the search point is within the square neighborhood of the center point. Using coordinate comparison instead of Euclidean distance calculation reduces the complexity of clustering.
[0012] The specific implementation method for step two is as follows:
[0013] Step 2.1: Select any point P as the starting point;
[0014] Step 2.2: Calculate point P(x) p ,y p The corresponding square neighborhood boundaries are the lines x = x max The line x = x min The straight line y = y max The straight line y = y min .
[0015] x max =x p +a
[0016] x min =xp -a
[0017] y max =y p +a
[0018] y min =y p -a
[0019] Step 2.3: Compare the x and y coordinates of all unclassified points except point P with the boundary values, and store the points within the square neighborhood of point P in the `seeds` array. The unclassified points include unprocessed points and points marked as noise. The coordinates of the points within the neighborhood (x, y, ...) are... i ,y i ) Must meet:
[0020] x min <x i <x max
[0021] y min <y i <y max
[0022] If the number of points in the neighborhood of point P is n p If the value is greater than or equal to min_points, then a class is formed, and point P and its neighboring points are labeled with the same label; otherwise, point P is temporarily marked as noise, and the seeds array is cleared.
[0023] Step 2.4: Access the next object in the seeds array, the neighborhood of point P. Perform the same search operation as in steps 2.2 to 2.3, label the points in its neighborhood with the same label as point P, and continue storing them in the seeds array. After accessing all unclassified points in the array with this point as the center point, delete this point from the seeds array.
[0024] Step 2.5: Repeat step 2.4 until the seeds array is empty, thus completing one clustering of a batch of data.
[0025] Step 2.6: After one clustering cycle, determine the number of categories. If the number of categories equals M, the clustering result is considered correct; otherwise, the clustering result is incorrect. When a clustering error occurs, perform the following operations: For the case of M-1 categories, determine whether to re-cluster the noisy points or the points with the most points by checking if the number of noise points after clustering exceeds a threshold; for the case of M+1 categories, average the cluster centers of the two categories with the fewest points and merge them; for other cases, adjust the min_points value according to the number of categories and re-cluster all points in this batch. Continue until the clustering result is correct, then average the coordinates of data points in the same category to calculate M cluster centers.
[0026] Step 3: Select a predetermined number of test phases and rotate the cluster centers obtained in Step 2 by the angle of the test phase. Compare the Euclidean distances between the rotated points and the decision constellation points. Since the phase noise levels of points in the same batch are similar, sum the distances between the M cluster centers with the same rotated test phase and the decision points to obtain the sum of distances between the rotated cluster points and the decision points. Compare the sum of distances under different rotated test phases, and take the test phase value corresponding to the point with the smallest sum as the carrier phase estimate. That is, use the BPS algorithm to perform carrier phase recovery on the cluster centers obtained in Step 2.
[0027] The specific implementation method for step three is as follows:
[0028] Step 3.1: Select B test phases as the preferred ones, where B is a power of 2. Test Phases The selection method is as follows:
[0029]
[0030] Step 3.2: Substitute the selected test phase into the following formula to rotate the cluster centers according to the test phase, and calculate the Euclidean distance between the rotated cluster centers and the decision signal:
[0031]
[0032] Where r k The input cluster center signal is denoted by `descision[·]`, which represents the constellation decision operation. k,b This represents the Euclidean distance between the signal at point k after incorporating the test phase b and the decision signal.
[0033] Step 3.3: Rotate the same angle as in step 3.2 The sum of the Euclidean distances obtained from the calculations of the cluster centers in the same batch is s. b :
[0034]
[0035] Take s b The rotation angle corresponding to the minimum value is the carrier phase estimate.
[0036] Step 4: The carrier phase estimates obtained in steps 2 and 3 are the carrier phase estimates for all points in this batch, thus realizing the carrier phase recovery of this batch of signals.
[0037] Step 5: Repeat steps 2 to 4 in batches to restore the carrier phase of the signal until all data carrier phases are restored, thus achieving low-complexity optical communication carrier phase restoration.
[0038] Beneficial effects:
[0039] 1. The present invention discloses a low-complexity optical communication carrier phase recovery method, which adopts clustering first and then performing blind phase search (BPS) processing on the cluster centers. Compared with the method of performing BPS on all data, it reduces the amount of data processed by BPS and improves the efficiency of carrier phase recovery.
[0040] 2. The present invention discloses a low-complexity optical communication carrier phase recovery method, which uses an improved DBSCAN algorithm for clustering. Compared with the original DBSCAN algorithm, it uses coordinate comparison instead of Euclidean distance calculation, which significantly reduces the number of multiplication operations in the clustering process and reduces the clustering complexity.
[0041] 3. The present invention discloses a low-complexity optical communication carrier phase recovery method, which constructs a processing method for erroneous clustering results when clustering data in batches, thereby reducing the number of times all data is re-clustered and improving the calculation speed.
[0042] 4. The present invention discloses a low-complexity optical communication carrier phase recovery method. When performing BPS carrier phase recovery on cluster centers, it determines that the phase noise of data in the same batch is the same. This provides a reference for the interval length of Euclidean distance summation when using clustering plus BPS, significantly reducing the number of multiplication operations in the clustering process and reducing the clustering complexity. Attached Figure Description
[0043] Figure 1 A schematic diagram of the improved DBSCAN algorithm parameters.
[0044] Figure 2 A flowchart of a low-complexity optical communication carrier phase recovery method.
[0045] Figure 3 Schematic diagram of the transmission system.
[0046] Figure 4 The proposed scheme is used to compare the constellation diagrams before and after carrier phase recovery.
[0047] Figure 5 A comparison chart of the complexity of carrier phase recovery for a single batch of data between the present invention and the original BPS algorithm. Detailed Implementation
[0048] To better illustrate the purpose and advantages of the present invention, the invention will be further described below in conjunction with the accompanying drawings and examples.
[0049] Example 1
[0050] For carrier phase recovery of 16QAM signals transmitted through standard single-mode fiber, such as Figure 2 As shown, it includes the following steps:
[0051] Will be by Figure 3 The data collected by the transmission system shown is input into the digital signal processing process. After steps such as low-pass filtering, I / Q imbalance compensation, dispersion compensation, clock recovery, polarization demultiplexing and polarization mode dispersion compensation, and frequency offset estimation, it is input into the proposed carrier phase recovery. This embodiment discloses a low-complexity optical communication carrier phase recovery method, the specific implementation steps of which are as follows:
[0052] Step 1: Scaling the 75,638 input data points to a standard size and dividing them into groups of 1,500 data points each. This batching of data makes it easier to input them into subsequent batch processing steps 2 to 4.
[0053] Step 2: Use the improved DBSCAN algorithm to perform a single clustering of the data after the batching in Step 1.
[0054] The DBSCAN algorithm primarily uses two parameters: the neighborhood search radius ε and the minimum number of points (min_points) that can stand alone as a class. For the improved DBSCAN algorithm, the circular search neighborhood is replaced with a square neighborhood, where the extension length 'a' represents half the side length of the square neighborhood. The neighborhood boundaries are defined as the horizontal and vertical lines extending 'a' in the corresponding directions from the starting point. A square search neighborhood is formed by these boundaries, as shown below. Figure 1 As shown, by comparing the coordinates of the search point with the boundary of the square neighborhood, it is determined whether the search point is within the square neighborhood of the center point. Using coordinate comparison instead of Euclidean distance calculation reduces the complexity of clustering.
[0055] The specific implementation method for step two is as follows:
[0056] Step 2.1: A batch of data can be represented as [(x1,y1),(x2,y2),...(x 2000 ,y 2000 Choose any point P(x) p ,y p (as the starting point);
[0057] Step 2.2: Take the initial value of the extension length 'a' as 0.3, and calculate the boundary of the square neighborhood corresponding to point P, which is the line x = x max The line x = x min The straight line y = y max The straight line y = y min .
[0058] x max =x p +a
[0059] x min =xp -a
[0060] y max =y p +a
[0061] y min =y p -a
[0062] Step 2.3: Set the min_points value to 25. Compare the x and y coordinates of all unclassified points except point P with the boundary values. Store the points within the square neighborhood of point P in the seeds array. The unclassified points include unprocessed points and points marked as noise. The coordinates of the points within the neighborhood (x, y, y) are... i ,y i ) Must meet:
[0063] x min <x i <x max
[0064] y min <y i <y max
[0065] If the number of points in the neighborhood of point P is n p If the number is greater than or equal to 20, then a class is formed, and point P and its neighboring points are labeled with the same label; otherwise, point P is temporarily marked as noise, and the seeds array is cleared.
[0066] Step 2.4: Access the next object in the seeds array, the neighborhood of point P. Perform the same search operation as in steps 2.2 to 2.3, label the points in its neighborhood with the same label as point P, and continue storing them in the seeds array. After accessing all unclassified points in the array with this point as the center point, delete this point from the seeds array.
[0067] Step 2.5: Repeat step 2.4 until the seeds array is empty, thus performing one clustering of a batch of data. The clustering process uses coordinate comparison instead of Euclidean distance calculation to determine whether a data point is within the neighborhood of the center point, without using multiplication operations, significantly reducing the clustering complexity.
[0068] Step 2.6: After one clustering cycle, check the number of clusters. If the number of clusters is 16, the clustering result is considered correct; otherwise, the clustering result is incorrect. When an error occurs, perform the following operations: For cases with 15 and 14 clusters, determine whether to re-cluster the noisy points or the points with the most points by checking if the number of noise points exceeds a threshold; for cases with 17 clusters, average the cluster centers of the two clusters with the fewest points and merge them; for other cases, adjust the `min_points` value according to the number of clusters and re-cluster all points in this batch. Continue until the clustering result is correct, then average the coordinates of data points within the same cluster to calculate 16 cluster centers.
[0069] The implementation method for step three is as follows:
[0070] Step 3.1: Select 16 test phases, requiring 16 multiplication operations to test the phases. The selection method is as follows:
[0071]
[0072] Step 3.2: Substitute the selected test phase into the following formula to rotate the cluster centers according to the test phase, and calculate the Euclidean distance between the rotated cluster centers and the decision signal:
[0073]
[0074] Where r k The input cluster center signal is denoted by `descision[·]`, which represents the constellation decision operation. k,b This represents the Euclidean distance between the signal at point k after incorporating the test phase b and the decision signal. A total of 16*16 points participate in this process. During the constellation point rotation, complex multiplication is converted into real number multiplication and addition, requiring a total of 16*16*4 multiplications. The Euclidean distance calculation does not involve square root extraction; each Euclidean distance calculation requires two multiplications, totaling 16*16*2 multiplications.
[0075] Step 3.3: Rotate the same angle as in step 3.2 The sum of the Euclidean distances calculated from the 16 cluster centers in the same batch yields s. b :
[0076]
[0077] Take s b The rotation angle corresponding to the minimum value is the carrier phase estimate. n represents the batch number.
[0078] Step 4: The carrier phase estimates obtained in steps 2 and 3 are the carrier phase estimates for all points in this batch, thus realizing the carrier phase recovery of this batch of signals.
[0079]
[0080] Where s in For all signals input in this batch, s out This is the output signal after phase noise recovery.
[0081] Step 5: Repeat steps 2 to 4 in batches to achieve carrier phase recovery of the signal until all data carrier phases are recovered. This achieves low-complexity optical communication carrier phase recovery. A comparison of the constellation diagrams before and after carrier phase recovery is shown below. Figure 4 As shown.
[0082] The above detailed description further illustrates the purpose, technical solution, and beneficial effects of the invention. It should be understood that the above description is only a specific embodiment of the present invention and does not limit the scope of protection of the present invention. Any modifications, substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A low-complexity optical communication carrier phase recovery method, characterized in that: Includes the following steps, Step 1: Divide the input data into groups of Nc data points each. This batching of data facilitates its input into subsequent batch processing steps 2 to 4. The value of Nc is an empirical value, which should be adjusted appropriately according to different signal noise levels. Step 2: Use the improved DBSCAN algorithm to perform a single clustering of the data after the batching in Step 1; The specific implementation of the clustering method described in step two is as follows: The DBSCAN algorithm mainly has two parameters: neighborhood search radius. And the minimum number of points (min_points) that can stand alone as a class; for the improved DBSCAN algorithm, the circular search neighborhood is replaced with a square neighborhood, and the extension length is... This represents half the side length of the neighborhood of the square; the lengths extending in the corresponding directions are the top, bottom, left, and right boundaries, centered on the starting point. The corresponding horizontal and vertical lines are defined as neighborhood boundaries; a square is formed by the neighborhood boundaries to search for the neighborhood; by comparing the coordinates of the search point with the square neighborhood boundary, it is determined whether the search point is within the square neighborhood of the center point; coordinate comparison is used instead of Euclidean distance calculation to reduce the complexity of clustering. Step 3: Select a predetermined number of test phases, rotate the cluster centers obtained in Step 2 by the angle of the test phase, and compare the Euclidean distance between the rotated points and the decision constellation points; sum the distances between the M cluster centers with the same rotated test phase and the decision points to obtain the sum of the distances between the rotated cluster points and the decision points; compare the magnitude of the sum of distances under different rotated test phases, and take the test phase value corresponding to the point with the smallest sum of distances as the carrier phase estimate, that is, use the BPS algorithm to perform carrier phase recovery on the cluster centers obtained in Step 2; Step 4: The carrier phase estimates obtained in steps 2 and 3 are the carrier phase estimates for all points in this batch, thus achieving carrier phase recovery for this batch of signals; Step 5: Repeat steps 2 to 4 in batches to restore the carrier phase of the signal until all data carrier phases are restored, thus achieving low-complexity optical communication carrier phase restoration.
2. The low-complexity optical communication carrier phase recovery method as described in claim 1, characterized in that: The specific implementation method for step two is as follows: Step 2.1: Select any point P as the starting point; Step 2.2: Calculate point P The corresponding square neighborhood boundaries are respectively straight lines. ,straight line ,straight line ,straight line ; Step 2.3: Compare the x and y coordinates of all unclassified points except point P with the boundary values, and store the points within the square neighborhood of point P in the `seeds` array; the unclassified points include unprocessed points and points marked as noise; the coordinates of the points within the neighborhood... Must meet: If the number of points in the neighborhood of point P If the value is greater than or equal to min_points, then a class is formed, and point P and its neighboring points are labeled with the same label; otherwise, point P is temporarily marked as noise, and the seeds array is cleared. Step 2.4: Access the next object in the seeds array of point P's neighborhood; perform the same search operation as in steps 2.2 to 2.3, label the points in its neighborhood with the same label as point P and continue to store them in the seeds array; after accessing all unclassified points in the array with this point as the center point, delete this point from the seeds array; Step 2.5: Repeat step 2.4 until the seeds array is empty, thus completing one clustering of a batch of data; Step 2.6: After one clustering operation, determine the number of categories. If the number of categories equals M, then the clustering result is considered correct. Otherwise, the clustering result is incorrect. When a clustering error occurs, perform the following operations: For the case where the number of categories is M-1, determine whether to re-cluster the noise points or re-cluster the points with the most points by judging whether the number of noise points after clustering exceeds the threshold value; For the case where the number of categories is M+1, take the average of the cluster centers corresponding to the two categories with the fewest points and merge them; For other cases, adjust the min_points value according to the number of categories, re-cluster all points in this batch until the clustering result is correct, and then calculate M cluster centers by taking the average of the coordinates of the data points of the same category.
3. The low-complexity optical communication carrier phase recovery method as described in claim 2, characterized in that: The specific implementation method for step three is as follows: Step 3.1: Select B test phases, where B is a power of 2; test phases The selection method is as follows: Step 3.2: Substitute the selected test phase into the following formula to rotate the cluster centers according to the test phase, and calculate the Euclidean distance between the rotated cluster centers and the decision signal: in The input cluster center signal, This refers to the operation of astrological judgment. This represents the Euclidean distance between the signal at the k-th point after incorporating the b-th test phase and the decision signal; Step 3.3: Rotate the same angle as in step 3.2 The Euclidean distances of the same batch of cluster centers are summed to obtain the result. : Pick The rotation angle corresponding to the minimum value is the carrier phase estimate.