Optical fiber transmission damage compensation method for fast random convolution kernel transformation
Through the combination of fast random convolution kernel transformation and linear classifier, the signal timing characteristics in the OAM-MDM fiber optic communication system are extracted, and the bit error rate problem caused by nonlinear damage to optoelectronic devices is solved, and low-complexity and efficient signals are achieved Compensation improves the quality of fiber optic communication.
Patent Information
- Application Number
- CN202410960276.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-17
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-07-17
AI Technical Summary
In OAM-MDM fiber optic communication system, the bit error rate performance caused by nonlinear damage to optoelectronic devices is poor, the traditional digital signal processing algorithm is poor, and the deep convolutional neural network training is slow and the calculation complexity is high, making it difficult to effectively compensate. Nonlinear damage.
The signal timing characteristics are extracted by fast random convolution kernel transformation, and signal classification is performed through a random convolution kernel with diverse weights, expansion coefficients and biases and a linear classifier to achieve nonlinear damage compensation and signal equalization, and improve bit error rate performance.
Low complexity and high accuracy fiber transmission damage compensation is achieved, and the communication quality of large-capacity and high-speed fibers is significantly improved, and the computing overhead and training complexity of deep neural networks is avoided.
Smart Images

Figure CN118972018A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a transmission damage equalization method of an orbital angular momentum mode multiplexing (OAM-MDM) direct modulation and direct detection (IM / DD) optical fiber transmission system based on fast random convolution kernel transformation, and in particular to an optical fiber transmission damage compensation method based on fast random convolution kernel transformation, belonging to the technical field of optical fiber communication. Background Art
[0002] With the continuous development of modern optical fiber communication systems, the system transmission capacity requirements are constantly increasing. Traditional multiplexing modulation technologies such as amplitude, phase, and frequency based on single-mode fiber (SMF) have gradually failed to meet the requirements. In this context, space division multiplexing technology using multi-core fiber (MCF) or multimode fiber (MMF) has attracted the attention of many researchers and has been widely and deeply studied. Since the OAM beam was proposed in 1992, related research has developed rapidly. OAM has infinite-dimensional characteristics in Hilbert space, so it has broad application prospects. For a long time after it was proposed, the research on OAM beams in communication has been limited to free space transmission systems. This is because the structure of traditional optical fibers limits the propagation of OAM beams, making OAM beams considered unsuitable for optical fiber communications. It was not until the optical fiber with a vortex structure was proposed that research on OAM beam transmission in optical fibers gradually emerged. After continuous research and updates, optical fibers with a vortex structure have developed into the current ring core fiber (RCF) and graded refractive index ring core fiber (GIRCF) that can support more OAM modes. As the capacity of OAM-MDM optical fiber transmission systems increases, the corresponding offline digital signal processing becomes more critical. Therefore, in order to meet the bit error rate (BER) performance requirements of the system, it is necessary to study the offline processing part of the digital signal processing of the OAM-MDM optical fiber communication system and appropriately apply algorithms with better performance.
[0003] In OAM-MDM optical fiber communication systems, the capacity and transmission distance of the system are limited by transmission damage caused by many factors, among which the most important factor is nonlinear damage, especially nonlinear damage caused by optoelectronic devices in the system, such as spatial light modulators (SLMs) and photodiodes. Spatial light modulators contain many nonlinear materials, such as liquid crystals, nonlinear polymers, and photorefractive materials, which can cause serious nonlinear damage. Generally, digital signal processing is the preferred method to alleviate nonlinear damage. Digital signal processing algorithms, such as nonlinear equalization (VNE) based on Volterra series, equalize distorted signals by fitting the nonlinear mathematical model of the system. However, in OAM-MDM systems, the nonlinear damage of the entire system is highly complex due to the nonlinear mutual coupling of multiple optoelectronic devices, and the mode coupling in the system also makes the nonlinear damage have a strong random characteristic. Therefore, the damage compensation effect of VNE in the optical fiber communication system is poor. In addition, in order to solve the problem that traditional digital signal processing algorithms are difficult to extract signal timing features, deep convolutional neural networks have been proposed for transmission damage compensation in optical fiber communication systems. However, its deep network design leads to slow training convergence, high computational complexity, and consumption of a large amount of computing resources. In summary, it is necessary to study a fiber transmission damage compensation method based on fast random convolution kernel transformation. Summary of the invention
[0004] Aiming at the problem that in the OAM mode multiplexing optical transmission system, the nonlinear damage caused by the graded refractive index ring core fiber (GIRCF), spatial light modulator (SLM), Mach Zehnder modulator (MZM) and photoelectric balanced detector (PD) causes serious signal damage and poor bit error rate performance, the main purpose of the present invention is to provide a fiber transmission damage compensation method of fast random convolution kernel transformation, which extracts rich timing features of the received signal through random convolution kernel transformation with diverse weights, expansion coefficients and biases, and inputs them into a linear classifier to classify the signal with high precision, thereby realizing nonlinear damage compensation for the signal and effective signal equalization, improving the bit error rate performance of the system, and improving the quality of large-capacity and high-speed optical fiber communication.
[0005] The objective of the present invention is achieved through the following technical solutions:
[0006] The present invention discloses a method for compensating optical fiber transmission damage by fast random convolution kernel transformation, comprising the following steps:
[0007] Step 1: Signal transmission and acquisition of orbital angular momentum mode multiplexing (OAM-MDM) direct modulation and direct detection (IM / DD) optical fiber transmission system;
[0008] The orbital angular momentum mode is modulated for the G-channel signal multiplexing. For the i-th channel transmission signal, an arbitrary waveform generator (AWG) is used to generate a modulated electrical signal. After the modulated electrical signal passes through a Mach-Zehnder modulator (MZM), it is loaded onto a Gaussian beam. The Gaussian beam is irradiated onto a spatial light modulator and modulated into a topological charge satisfying |l|=k i vortex beam.
[0009] The above-mentioned G-path vortex beam, after mode multiplexing, is sent into the graded refractive index ring core fiber for transmission, and then splits the G path, passes through the spiral phase plate (VPP) corresponding to the G block, and is converted back to a Gaussian beam, and then passes through a photodetector (PD) to be converted into an electrical signal, which is captured and collected by an oscilloscope.
[0010] Step 2: Perform preliminary digital signal processing on the signal collected in step 1 and define the data set;
[0011] The signal collected in step 1 is processed by low-pass filtering, resampling and matched filtering digital signal processing and is recorded as {S i},i=1,2,...,G, where S i =[x i,1 ,x i,2 ,...,x i,N ], N is the length of the signal sequence. The sliding window length is set to l win . The received signal of each channel is used as the feature vector and the symbol sequence is used as the label. win Each group is packaged into a feature vector sequence and a label sequence. The feature vector sequence of the i-th path is:
[0012]
[0013] Where X i (j) represents X i The jth row of Nl, j = 1, 2, ..., Nl input +1.
[0014] The symbol label sequence of the i-th path is: M is the modulation order of the signal.
[0015] The feature vector sequence and the label sequence are taken as the data set (X i ,Y i ), select the first 40% as the training data set (X i (k),Y i (k)), k = 1, 2, ..., 0.4N, and the last 60% are selected as the application data set (X i (p),Y i (p)),p=1,2,...,0.6N.
[0016] Step 3: Use a classification model based on fast random convolution kernel transformation to classify and identify the signal;
[0017] For all convolution kernels, set their characteristic parameters: length and weight.
[0018] First, the length of all convolution kernels is l kenel Fixed to 9;
[0019] In order to reduce the amount of calculation while maintaining accuracy, the weight values of the convolution kernel are limited to {1, -2}, and the 6 weights of each convolution kernel are limited to 1 and the 3 weights are limited to -2. Under this restriction, there are 84 possible weight values of the convolution kernel. K convolution kernels are selected from them. K is an integer less than or equal to 84 and greater than 1. The e-th convolution kernel is represented as
[0020]
[0021] For a convolution kernel, set the bias and dilation coefficient of the random convolution transformation to be performed.
[0022] The dilation coefficient d of each convolution kernel e Used to expand the convolution kernel on a time scale. For each convolution kernel, four expansion coefficients are selected. The expansion coefficient is randomly selected from the expansion coefficient set D. The expansion coefficient set D is defined as:
[0023] D={floor(2 0 ),floor(2 1 ),...,floor(2 max )},
[0024] in l win is the length of the input feature vector.
[0025] The bias b of each convolution kernel e Determined by the output of the convolution kernel. Extract the mean from the quantiles of the convolution output of multiple randomly selected training samples. For the randomly selected samples, calculate the convolution output W e *X i The quartile of (k) is randomly selected from the three quartiles as the bias. The quartiles are the values at the 25%, 50%, and 75% positions after the convolution output data is sorted from large to small.
[0026] For a given feature vector X of a training data set i (k), and combine it with all convolution kernels W e , and its corresponding expansion coefficient d e and bias b e Perform random convolution kernel convolution transformation, expressed as:
[0027]
[0028] After convolution operation with random convolution kernel, the proportion of positive values (PPV) is calculated as the feature.
[0029]
[0030] Where I(·) is the indicative function. i (k) PPV and symbolic label sequence Y of the training data set calculated after performing 4 expansion coefficients with K convolution kernels and the corresponding bias output of each convolution kernel i (k) Input to the ridge regression linear classifier CLF to train the linear classifier.
[0031] Step 4: Use the random convolution kernel and linear classifier in step 3 to perform transmission damage compensation on the received optical fiber communication signal;
[0032] The feature vector sequence in the application data set is convolved with the random convolution kernel in step 3 to calculate the PPV feature, which is then input into the linear classifier CLF trained in step 3 to obtain the corresponding symbol sequence after transmission damage compensation. Symbol sequence after transmission impairment compensation and apply the symbolic labels Y in the dataset i (p) Perform bit mapping, compare the mapped bit sequences one by one and calculate the bit error rate. Low-complexity and high-accuracy damage compensation is achieved through random convolution kernel transformation transmission damage compensation processing, which effectively alleviates the impact of transmission damage on the orbital angular momentum mode division multiplexing signal during optical fiber transmission and improves the communication quality of the optical communication system.
[0033] Beneficial Effects
[0034] 1. The present invention discloses a method for optical fiber transmission damage using fast random convolution kernel transformation, which uses a variety of random convolution kernels to perform convolution transformation on signals to extract timing features. It can compensate for transmission damage to signals more specifically targeting the signal damage characteristics of the transmission system. Compared with the traditional damage compensation algorithm, the bit error rate after processing is lower, achieving effective equalization and damage compensation, and improving the quality of large-capacity and high-speed optical fiber communications.
[0035] 2. The present invention discloses a method for optical fiber transmission damage using a fast random convolution kernel transformation, which can convert multiplication in the convolution operation process into addition through the convolution kernel weight setting, thereby achieving efficient and fast convolution transformation and avoiding the huge computational overhead when training deep convolutional networks.
[0036] 3. The present invention discloses a method for optical fiber transmission damage based on fast random convolution kernel transformation, which can extract the timing characteristics of signals at multiple scales by setting different expansion coefficients, thereby extracting the long-term timing dependence of signals, which is helpful for damage compensation and signal recognition of received signals.
[0037] 4. The present invention discloses a method for optical fiber transmission damage using a fast random convolution kernel transformation. After the fast random convolution kernel transformation, only a simple linear classifier is needed to compensate for the transmission damage of the damaged signal, thus avoiding the structural design of a multi-layer neural network and a large amount of computational overhead. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] Exemplary embodiments of the present invention will be described in detail below with reference to the accompanying drawings so that those skilled in the art will understand the above and other features and advantages of the present invention more clearly. In the accompanying drawings:
[0039] Figure 1 It is a flow chart of a method for compensating optical fiber transmission damage using fast random convolution kernel transformation disclosed in the present invention;
[0040] Figure 2 A schematic diagram of a fast convolution kernel transformation of a method for compensating optical fiber transmission damage using a fast random convolution kernel transformation disclosed in the present invention;
[0041] Figure 3 The OAM-MDM IM / DD optical fiber transmission system experimental device disclosed in this embodiment;
[0042] Figure 4 This is a comparison chart of the bit error rates of 32GBaudPAM8 signals without equalization, after fast random convolution kernel transformation, and after VNE compensation under different received optical powers in the embodiment;
[0043] Figure (a) shows the comparison of the bit error rates of 32GBaud PAM8 signal without equalization, after fast random convolution kernel transformation compensation and after VNE compensation when |l|=3; Figure (b) shows the comparison of the bit error rates of 32GBaud PAM8 signal without equalization, after fast random convolution kernel transformation compensation and after VNE compensation when |l|=2. DETAILED DESCRIPTION
[0044] The present invention will be described in detail below with reference to the accompanying drawings and embodiments. The technical problems solved by the technical solution of the present invention and the beneficial effects are also described. It should be noted that the described embodiments are only intended to facilitate the understanding of the present invention and do not have any limiting effect on it.
[0045] In this embodiment, an OAM-MDM IM / DD optical fiber communication system experimental platform is built to transmit 32GBaud PAM8 (8th order pulse amplitude modulation) signals multiplexed by two OAM modes. The experimental device is as follows: Figure 3 As shown. In the signal transmission part, the digital signal processing program generates the modulated signal and sends it to a Mach-Zehnder modulator (MZM) by an arbitrary waveform generator (AWG). The MZM modulates the signal onto an optical carrier. In the OAM multiplexing part, the signal is divided into two paths by an optical coupler (OC). After passing through an erbium-doped fiber amplifier (EDFA) and a polarization controller (PC), each beam is injected into a spatial light modulator (SLM) through a collimator (Col) and linear polarization (LP). The two spatial light modulators modulate the beams in different OAM modes, namely, the topological charge number |l|=2 and |l|=3. Then, a polarization beam combiner (PBC) is used to combine the two OAM beams with |l|=2 and |l|=3. The combined beam is converted to circular polarization by a quarter-wave plate (QWP) and then transmitted through a 2.3-kilometer GIRCF. In the OAM demultiplexing section, the transmitted beams are first separated by a beam splitter (BS), then demultiplexed into Gaussian beams by a vortex phase plate (VPP) with opposite topological charges, and then coupled to a single-mode fiber (SMF). In the signal receiving section, the two beams are amplified by an EDFA and converted into electrical signals by a photon detector (PD). For PAM signals, the BER is calculated after low-pass filtering, resampling, clock recovery, damage compensation, and signal decoding.
[0046] Since some active optoelectronic devices are used in the optical fiber communication system, such as spatial light modulators, Mach-Zehnder modulators, etc., these active optoelectronic devices have their own nonlinearity, and they interact with each other in the system to produce nonlinear effects, which deteriorates the system bit error rate performance and reduces the communication quality. This embodiment discloses a method for compensating optical fiber transmission damage by fast random convolution kernel transformation, such as Figure 1 As shown, the following steps are included:
[0047] Step 1: Orbital angular momentum mode multiplexing direct modulation and direct detection optical fiber transmission and signal acquisition;
[0048] Two-way signal multiplexing is used for orbital angular momentum mode modulation. For the i-th transmission signal, a Keysight 8194A arbitrary waveform generator (AWG) is used to generate a modulated PAM8 signal, the sampling frequency is set to 96GSa / s, the peak-to-peak value of the output electrical signal is set to 150mV, the baud rate of the signal is 32GBaud, and the signal rate is 96Gbps. After the modulated electrical signal passes through the Mach-Zehnder modulator (MZM), it is loaded onto the Gaussian beam. The Gaussian beam is irradiated onto the spatial light modulator and modulated into a vortex beam with a topological charge satisfying |l|=2 and |l|=3. After mode multiplexing, the above two vortex beams are sent into a gradient refractive index ring core fiber for transmission, and then split into two paths. After passing through two corresponding spiral phase plates, they are converted back to Gaussian beams, and then converted into electrical signals after passing through a photodetector, which are captured and collected by an oscilloscope.
[0049] Step 2: Perform preliminary digital signal processing on the signal collected in step 1 and define the data set;
[0050] The signal collected in step 1 is processed by low-pass filtering, resampling, and matched filtering digital signal processing and is recorded as {S i},i=1,2,...,G, where S i =[x i,1 ,x i,2 ,...,x i,N ], N is the signal sequence length, set to 65536. The sliding window length is set to l win =40. The received signal of each channel is used as the feature vector, and the symbol sequence is used as the label. win Each group is packaged into a feature vector sequence and a label sequence. The feature vector sequence of the i-th path is:
[0051]
[0052] Where X i (j) represents X i The jth row of Nl, j = 1, 2, ..., Nl input +1.
[0053] The symbol label sequence of the i-th path is: M is the modulation order of the signal.
[0054] The feature vector sequence and the label sequence are taken as the data set (X i ,Y i ), select the first 40% or the first 26214 as the training data set (X i (k),Y i (k)), k = 1, 2, ..., 0.4N, select the last 60%, i.e. the last 39322, as the application data set (X i(p),Y i (p)),p=1,2,...,0.6N.
[0055] Step 3: Use a classification model based on fast random convolution kernel transformation to classify and identify the signal
[0056] For all convolution kernels, set their characteristic parameters: length and weight.
[0057] First, the length of all convolution kernels is l kenel Fixed to 9;
[0058] In order to reduce the amount of calculation while maintaining accuracy, the weight values of the convolution kernel are limited to {1, -2}, and the 6 weights of each convolution kernel are limited to 1 and the 3 weights are limited to -2. Under this restriction, 60 convolution kernels are selected from the convolution kernel. The e-th convolution kernel is represented as
[0059] For a convolution kernel, set the bias and dilation coefficient of the random convolution transformation to be performed.
[0060] The dilation coefficient d of each convolution kernel e Used to expand the convolution kernel on a time scale. For each convolution kernel, four expansion coefficients are selected. The expansion coefficient is randomly selected from the expansion coefficient set D. The expansion coefficient set D is defined as:
[0061] D={floor(2 0 ),floor(2 1 ),...,floor(2 max )},
[0062] in l win is the length of the input feature vector.
[0063] The bias b of each convolution kernel e Determined by the convolution kernel output. Extract the mean from the quantiles of the convolution output of multiple randomly selected training samples. For the randomly selected samples, calculate the convolution output W e *X i The quartile of (k) is randomly selected from the three quartiles as the bias. The quartiles are the values at the 25%, 50%, and 75% positions after the convolution output data is sorted from large to small.
[0064] For a given feature vector X of a training data set i (k), and combine it with all convolution kernels W e , and its corresponding expansion coefficient d e and bias b e Perform random convolution kernel convolution transformation, expressed as:
[0065]
[0066] After convolution operation with random convolution kernel, the proportion of positive values (PPV) is calculated as the feature.
[0067]
[0068] Where I(·) is the indicative function. i (k) PPV and symbolic label sequence Y of the training data set calculated after performing 4 expansion coefficients with K convolution kernels and the corresponding bias output of each convolution kernel i (k) Input to the ridge regression linear classifier CLF to train the linear classifier.
[0069] Step 4: Use the random convolution kernel and linear classifier in step 3 to perform damage compensation on the received optical fiber communication signal.
[0070] The feature vector sequence in the application data set is convolved with the random convolution kernel in step 3 to calculate the PPV feature, which is then input into the linear classifier CLF trained in step 3 to obtain the corresponding symbol sequence after damage compensation. Symbol sequence after damage compensation and apply the symbolic labels Y in the dataset i (p) Perform bit mapping, compare the mapped bit sequences one by one and calculate the bit error rate. Figure 4 The comparison of the bit error rate between the fast convolution kernel transform-based damage compensation method and VNE at different received optical powers (ROP) is shown when |l|==2 or 3 (the left figure is |l|=2, and the right figure is |l|=3). The training data length is 26214, and the test data length is 39322. The experimental results show that the fast convolution kernel transform-based damage compensation method has better results than the Volterra-based damage compensation method. For |l|=2, the fast convolution kernel transform-based damage compensation method has a 2.7dB receiver sensitivity gain compared with the Volterra-based damage compensation method; for |l|=3, the fast convolution kernel transform-based damage compensation method has a 2.6dB receiver sensitivity gain compared with the Volterra-based damage compensation method. It can be seen that the bit error rate performance is significantly improved.
[0071] The specific description above further illustrates the purpose, technical solutions and beneficial effects of the invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.
Claims
1. A method for compensating optical fiber transmission damage by fast random convolution kernel transformation, characterized in that: The following steps are included: Step 1: Signal transmission and acquisition of orbital angular momentum mode multiplexing direct modulation and direct detection optical fiber transmission system; Step 2: Perform preliminary digital signal processing on the signal collected in step 1 and define the data set; Step 3: Use a classification model based on fast random convolution kernel transformation to classify and identify the signal; For all convolution kernels, set their characteristic parameters: length, weight; First, the length of all convolution kernels is l kenel Fixed to 9; In order to reduce the amount of calculation while maintaining accuracy, the weight values of the convolution kernel are limited to {1, -2}, and the 6 weights of each convolution kernel are limited to 1 and the 3 weights are limited to -2. Under this restriction, there are 84 possible weight values of the convolution kernel. K convolution kernels are selected from them; K is an integer less than or equal to 84 and greater than 1. The e-th convolution kernel is expressed as For a convolution kernel, set the bias and dilation coefficient of the random convolution transformation to be performed; The dilation coefficient d of each convolution kernel e Used to expand the convolution kernel on a time scale; for each convolution kernel, four expansion coefficients are selected; the expansion coefficient is randomly selected from the expansion coefficient set D; the expansion coefficient set D is defined as: D={floor(2 0 ),floor(2 1 ),...,floor(2 max )}, in l win is the length of the input feature vector; The bias b of each convolution kernel e Determined by the output result of the convolution kernel; Extract the mean from the quantiles of the convolution output of multiple randomly selected training samples; for the randomly selected samples, calculate the convolution output W e *X i (k) quartile, one of the three quartiles is randomly selected as the bias; the quartiles are the values at the 25%, 50%, and 75% positions after the convolution output data is sorted from large to small; For a given feature vector X of a training data set i (k), and combine it with all convolution kernels W e , and its corresponding expansion coefficient d e and bias b e Perform random convolution kernel convolution transformation, expressed as: After the random convolution kernel convolution operation, the positive value ratio PPV is calculated as the feature; Where I(·) is the indicative function; the input feature vector X i (k) PPV and symbolic label sequence Y of the training data set calculated after performing 4 expansion coefficients with K convolution kernels and the corresponding bias output of each convolution kernel i (k) Input to the ridge regression linear classifier CLF to train the linear classifier; Step 4: Use the random convolution kernel and linear classifier in step 3 to compensate for the transmission damage of the received optical fiber communication signal.
2. The optical fiber transmission damage compensation method of fast random convolution kernel transformation according to claim 1, characterized in that: The implementation method of step one is: The G-channel signal multiplexing is subjected to orbital angular momentum mode modulation; for the i-th channel transmission signal, an arbitrary waveform generator is used to generate a modulated electrical signal; After the modulated electrical signal passes through the Mach-Zehnder modulator, it is loaded onto the Gaussian beam; the Gaussian beam is irradiated onto the spatial light modulator and modulated into a topological charge satisfying |l|=k i Vortex beam; The above-mentioned G-path vortex beam, after mode multiplexing, is sent into the graded refractive index ring core fiber for transmission, and then splits the G-path, passes through the spiral phase plate corresponding to the G block, and is converted back to a Gaussian beam, and then passes through a photodetector to be converted into an electrical signal, which is captured and collected by an oscilloscope.
3. The optical fiber transmission damage compensation method of fast random convolution kernel transformation according to claim 2, characterized in that: The implementation method of step 2 is: The signal collected in step 1 is processed by low-pass filtering, resampling and matched filtering digital signal processing and is recorded as {S i },i=1,2,...,G, where S i =[x i,1 ,x i,2 ,...,x i,N ], N is the signal sequence length; the sliding window length is set to l win ; Take the received signal of each channel as the feature vector and the symbol sequence as the label, l win Each group is packaged into a feature vector sequence and a label sequence; the feature vector sequence of the i-th path is: Where X i (j) represents X i The jth row of Nl, j = 1, 2, ..., Nl input +1; The symbol label sequence of the i-th path is: M is the modulation order of the signal; The feature vector sequence and the label sequence are taken as the data set (X i ,Y i ), select the first 40% as the training data set (X i (k),Y i (k)), k = 1, 2, ..., 0.4N, and the last 60% are selected as the application data set (X i (p),Y i (p)),p=1,2,...,0.6N.
4. The optical fiber transmission damage compensation method of fast random convolution kernel transformation according to claim 3, characterized in that: The implementation method of step 4 is: The feature vector sequence in the application data set is convolved with the random convolution kernel in step 3 to calculate the PPV feature, which is then input into the linear classifier CLF trained in step 3 to obtain the corresponding symbol sequence after transmission damage compensation. Symbol sequence after transmission impairment compensation and apply the symbolic labels Y in the dataset i (p) Perform bit mapping, compare the mapped bit sequences one by one and calculate the bit error rate; realize low-complexity and high-accuracy damage compensation through random convolution kernel transformation transmission damage compensation processing, effectively alleviate the influence of transmission damage on orbital angular momentum mode division multiplexing signal during optical fiber transmission, and improve the communication quality of optical communication system.
Citation Information
Patent Citations
Nonlinear compensation method for optical fiber communication based on residual neural network
CN115865199A
Method for constructing mode division multiplexing optical fibre communication system, and constructed optical fibre communication system
WO2018196056A1
Cited By
Satellite signal quality prediction model construction method and device and prediction method
CN117830828A
A satellite signal quality prediction model construction method, device and prediction method
CN117830828B
