CN115314118B - Optical fiber nonlinear compensation method and device - Google Patents
Optical fiber nonlinear compensation method and device Download PDFInfo
- Publication number
- CN115314118B CN115314118B CN202210899794.4A CN202210899794A CN115314118B CN 115314118 B CN115314118 B CN 115314118B CN 202210899794 A CN202210899794 A CN 202210899794A CN 115314118 B CN115314118 B CN 115314118B
- Authority
- CN
- China
- Prior art keywords
- nonlinear
- signal
- carrier phase
- recovery module
- value
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 239000013307 optical fiber Substances 0.000 title claims abstract description 53
- 238000000034 method Methods 0.000 title claims abstract description 33
- 238000013528 artificial neural network Methods 0.000 claims abstract description 78
- 238000011084 recovery Methods 0.000 claims abstract description 69
- 230000015654 memory Effects 0.000 claims abstract description 36
- 238000004891 communication Methods 0.000 claims abstract description 23
- 230000003287 optical effect Effects 0.000 claims abstract description 20
- 239000000835 fiber Substances 0.000 claims abstract description 17
- 238000012216 screening Methods 0.000 claims abstract description 13
- 230000001427 coherent effect Effects 0.000 claims abstract description 12
- 230000005540 biological transmission Effects 0.000 claims description 16
- 210000002569 neuron Anatomy 0.000 claims description 12
- 230000021615 conjugation Effects 0.000 claims description 10
- 238000010276 construction Methods 0.000 claims description 6
- 238000003491 array Methods 0.000 claims description 4
- 230000006870 function Effects 0.000 claims description 3
- 230000007787 long-term memory Effects 0.000 claims 1
- 230000006403 short-term memory Effects 0.000 claims 1
- 230000000694 effects Effects 0.000 abstract description 8
- 239000011159 matrix material Substances 0.000 description 20
- 238000004422 calculation algorithm Methods 0.000 description 11
- 238000004088 simulation Methods 0.000 description 10
- 210000004027 cell Anatomy 0.000 description 6
- 239000006185 dispersion Substances 0.000 description 5
- 238000012549 training Methods 0.000 description 5
- 230000009286 beneficial effect Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000012706 support-vector machine Methods 0.000 description 4
- 230000009022 nonlinear effect Effects 0.000 description 3
- 230000010287 polarization Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000012795 verification Methods 0.000 description 3
- 230000002457 bidirectional effect Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 230000006735 deficit Effects 0.000 description 2
- 230000005374 Kerr effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B10/00—Transmission systems employing electromagnetic waves other than radio-waves, e.g. infrared, visible or ultraviolet light, or employing corpuscular radiation, e.g. quantum communication
- H04B10/60—Receivers
- H04B10/61—Coherent receivers
- H04B10/616—Details of the electronic signal processing in coherent optical receivers
- H04B10/6163—Compensation of non-linear effects in the fiber optic link, e.g. self-phase modulation [SPM], cross-phase modulation [XPM], four wave mixing [FWM]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B10/00—Transmission systems employing electromagnetic waves other than radio-waves, e.g. infrared, visible or ultraviolet light, or employing corpuscular radiation, e.g. quantum communication
- H04B10/25—Arrangements specific to fibre transmission
- H04B10/2507—Arrangements specific to fibre transmission for the reduction or elimination of distortion or dispersion
- H04B10/2543—Arrangements specific to fibre transmission for the reduction or elimination of distortion or dispersion due to fibre non-linearities, e.g. Kerr effect
- H04B10/2557—Cross-phase modulation [XPM]
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B10/00—Transmission systems employing electromagnetic waves other than radio-waves, e.g. infrared, visible or ultraviolet light, or employing corpuscular radiation, e.g. quantum communication
- H04B10/25—Arrangements specific to fibre transmission
- H04B10/2507—Arrangements specific to fibre transmission for the reduction or elimination of distortion or dispersion
- H04B10/2543—Arrangements specific to fibre transmission for the reduction or elimination of distortion or dispersion due to fibre non-linearities, e.g. Kerr effect
- H04B10/2563—Four-wave mixing [FWM]
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B10/00—Transmission systems employing electromagnetic waves other than radio-waves, e.g. infrared, visible or ultraviolet light, or employing corpuscular radiation, e.g. quantum communication
- H04B10/60—Receivers
- H04B10/61—Coherent receivers
- H04B10/616—Details of the electronic signal processing in coherent optical receivers
Abstract
The invention discloses a fiber nonlinear compensation method and a device, which construct nonlinear disturbance items of cross phase modulation in a channel and four wave mixing in the channel according to output signals of a carrier phase recovery module in a receiving end of a coherent optical communication system; screening and reconstructing nonlinear disturbance items to obtain multidimensional input features; inputting the multidimensional input characteristics into a BiLSTM neural network to obtain an optical fiber nonlinear damage value; compensating the output signal of the carrier phase recovery module according to the nonlinear damage value of the optical fiber to obtain a nonlinear compensation signal; the invention combines the multidimensional input characteristic reconstructed by the nonlinear disturbance term with the time memory characteristic of the BiLSTM, thereby compensating the nonlinearity of the optical fiber and realizing remarkable compensation effect under different modulation formats.
Description
Technical Field
The invention relates to the technical field of optical fiber communication, in particular to an optical fiber nonlinear compensation method and device.
Background
The optical transmission network is an underlying infrastructure of the global communication network, and the capacity of the optical network based on the traditional single-core single-mode optical fiber is close to the shannon theoretical limit at present, so that the upgrading and capacity expansion of the communication system are urgent. The emerging technologies such as mode division multiplexing, space division multiplexing and the like are generated, so that the capacity expansion problem of a future communication system can be effectively solved, but the transmission capacity of the optical fiber communication system is limited and restrained by the common limitation and restriction of the linear damage and the nonlinear damage of the optical fiber. For the linear damage of the optical fiber, the effects of loss, dispersion, frequency offset, line width and the like on the optical fiber transmission can be relieved by utilizing the existing mature Digital Signal Processing (DSP) algorithm and devices. However, for nonlinear damage of the optical fiber, in the transmission process of a long distance with multiple spans, the nonlinear kerr effect of the optical fiber channel is significantly accumulated due to the improvement of the fiber entering power, so that serious nonlinear phase modulation is generated, and the signal quality is greatly reduced. Therefore, in high-speed long-distance optical fiber communication systems, the nonlinear disturbance of the optical fiber not only limits the maximum fiber-in power, but also limits the transmission distance and transmission capacity. Whereas nonlinear equalization techniques typically require very complex DSP algorithms, the compensation difficulty is increased compared to linear equalization. Thus, there is an urgent need for higher performance and lower complexity nonlinear equalization algorithms.
In recent years, machine learning algorithms have received widespread attention and have been applied to optical fiber communications to compensate for nonlinear impairments of optical fibers. Generally, support vector machine (Support Vector Machine, SVM) algorithms, clustering algorithms, and Neural Network (NN) algorithms are classified. Although the support vector machine algorithm and the clustering algorithm can obtain better nonlinear equalization effect, for the high-order modulation format and the large-scale data set, the support vector machine algorithm needs higher complexity, the classification speed of the clustering algorithm is reduced, and the classification result is greatly influenced by linear and nonlinear noise.
Disclosure of Invention
Aiming at the defects in the prior art, the optical fiber nonlinear compensation method and the device provided by the invention solve the problem that the existing method which has good effect and low complexity on compensating the nonlinear disturbance of the optical fiber is lacking.
In order to achieve the aim of the invention, the invention adopts the following technical scheme: a method of compensating for non-linearities of an optical fiber, comprising the steps of:
s1, constructing a nonlinear disturbance term of cross phase modulation in a channel and four-wave mixing in the channel according to an output signal of a carrier phase recovery module in a receiving end of a coherent optical communication system;
s2, screening and reconstructing nonlinear disturbance items to obtain multidimensional input features;
s3, inputting the multidimensional input characteristics into a BiLSTM neural network to obtain a fiber nonlinear damage value;
and S4, compensating the output signal of the carrier phase recovery module according to the nonlinear damage value of the optical fiber to obtain a nonlinear compensation signal.
Further, the expression of the nonlinear disturbance term in the step S1 is:
T t =A x/y,n+t A * x/y,m+n+t A x/y,m+t +A y/x,n+t A * y/x,m+n+t A x/y,m+t
wherein T is t Signal value pair at time t of output signal of carrier phase recovery moduleThe corresponding nonlinear disturbance term, when A x/y,n+t When a horizontally polarized signal representing a signal value at the n+t time of an output signal of a carrier phase recovery module is a y/x,n+t A vertically polarized signal representing the signal value at the n+t time of the output signal of the carrier phase recovery module, when A x/y,n+t When a vertically polarized signal representing a signal value at the n+t time of an output signal of a carrier phase recovery module is a y/x,n+t A horizontally polarized signal representing the signal value at the n+t time of the output signal of the carrier phase recovery module, when A * x/y,m+n+t When the conjugation of the horizontally polarized signal representing the signal value at the m+n+t-th time of the output signal of the carrier phase recovery module, then A * y/x,m+n+t Conjugation of vertically polarized signal representing signal value at m+n+t time of output signal of carrier phase recovery module, when A * x/y,m+n+t When the conjugation of the vertically polarized signal representing the signal value at the m+n+t-th time of the output signal of the carrier phase recovery module, then A * y/x,m+n+t Conjugation of horizontally polarized signal representing signal value at m+n+t time of output signal of carrier phase recovery module, A x/y,m+t The signal value of the signal value at the m+t time of the output signal of the carrier phase recovery module is horizontally polarized or vertically polarized, m represents the m time of the output signal of the carrier phase recovery module, and n represents the n time of the output signal of the carrier phase recovery module.
The beneficial effects of the above-mentioned further scheme are: the nonlinear disturbance term may include nonlinear impairment information that can characterize intra-channel cross-phase modulation and intra-channel four-wave mixing.
Further, the step S2 includes the following sub-steps:
s21, setting a disturbance threshold condition, and reserving a nonlinear disturbance term meeting the disturbance threshold condition:
s22, forming a two-dimensional real number array by real and imaginary parts of the two-dimensional complex number array of the reserved nonlinear disturbance term;
s23, dividing the two-dimensional real number array into three-dimensional real number arrays by taking the step length as M;
s24, taking the three-dimensional real number array as a multi-dimensional input feature, the method can be expressed as follows:
T t,M =[T t-k ,…,T t-1 ,T t ,T t+1 ,…,T t+k ]
wherein T is t A nonlinear disturbance term at the t-th moment; m is the time step, T t,M For the multidimensional input feature at time t, k represents the unit of the metering time step.
The beneficial effects of the above-mentioned further scheme are: the nonlinear disturbance term in the multidimensional input characteristic is reduced while the nonlinear effect correlation between the front and rear adjacent symbols is considered, so that the scheme complexity is reduced while the compensation performance is improved.
Further, the threshold condition in the step S21 is:
|m||n|≤C,|m|≤L,|n|≤L
wherein m represents the mth moment of the output signal of the carrier phase recovery module, n represents the nth moment of the output signal of the carrier phase recovery module, C is a disturbance threshold for balancing compensation performance, and L is a disturbance threshold for balancing complexity.
The beneficial effects of the above-mentioned further scheme are: according to the disturbance threshold condition, the nonlinear contribution in the disturbance item can be screened out, so that all nonlinear damage information is not needed to be contained in the input characteristics, and the complexity of the scheme is reduced.
Further, the BiLSTM neural network in step S3 includes: an input layer, a two-way long-short-term memory neural network layer, a flat layer and a full-connection layer;
the input end of the input layer is used as the input end of the BiLSTM neural network, and the output end of the input layer is connected with the input end of the two-way long-short-period memory neural network layer; the input end of the flat layer is connected with the output end of the two-way long-short-period memory neural network layer, and the output end of the flat layer is connected with the input end of the full-connection layer; the output end of the full connection layer is used as the output end of the BiLSTM neural network.
Further, the two-way long-short-term memory neural network layer comprises 10 neurons, the full-connection layer comprises 2 neurons, one neuron of the full-connection layer is used for outputting a real part of the optical fiber nonlinear damage value, and the other neuron of the full-connection layer is used for outputting an imaginary part of the optical fiber nonlinear damage value.
Further, the BiLSTM neural network further includes: a dropout layer; the value range of the dropoff value of the dropoff layer is [0.1,0.5], and the dropoff value setting method of the dropoff layer comprises the following steps: when the transmission power of the multi-dimensional input feature is reduced, the dropoff value is gradually increased, and when the transmission power of the multi-dimensional input feature is increased, the dropoff value is gradually reduced.
Further, the step S4 includes the following sub-steps:
s41, constructing a real part and an imaginary part of the nonlinear damage value of the optical fiber into a complex number array;
s42, subtracting the output signal of the carrier phase recovery module from the complex number array to obtain a nonlinear compensation signal.
Further, the specific operation process in the BiLSTM neural network is as follows:
c t =σ(W f [h t-1 ,T t,M ]+b f )*c t-1 +σ(W i [h t-1 ,T t,M ]+b i )*tanh(W c [h t-1 ,T t,M ]+b c )
h t =σ(W o [h t-1 ,T t,M ]+b o )*tanh(c t )
wherein c t T is the cell state at time T t,M For multidimensional input feature at time t, h t-1 Outputting the state of the two-way long-short-term memory neural network layer at the t-1 time, h t For the output of the state of the layer of the bidirectional long-short-term memory neural network at the t moment, W f Weight matrix for forgetting gate in two-way long-short-term memory neural network layer, W i Weight matrix W for input gate in two-way long-short-term memory neural network layer c Weight matrix for cell state in two-way long-short-term memory neural network layer,W o B is a weight matrix of output gates in the two-way long-short-term memory neural network layer f Bias matrix for forgetting gate in two-way long-short-term memory neural network layer, b i Bias matrix for input gate in two-way long-short term memory neural network layer, b c Bias matrix for cell state in two-way long-short-term memory neural network layer, b o A bias matrix for output gates in the two-way long-short-term memory neural network layer, j is a counting symbol, k represents an identifier for calculating the number, h j B, outputting the state of the layer of the two-way long-short-term memory neural network at the j-th moment j Bias matrix for full connection layer, W j Is a weight matrix of the full connection layer,and learning an output optical fiber nonlinear damage value for the BiLSTM neural network.
Further, during training, the loss function adopted by the fiber nonlinear equalization model is as follows:
wherein L is MSE In order to achieve a loss value, the value of the loss,the fiber nonlinear damage value which is learned and output by the BiLSTM neural network is B, the size of a sample batch is B, i is the ith sample, and +.>For the output signal of the carrier phase recovery module, +.>Is a signal at the transmitting end in a high-speed optical communication system.
Further, during training, the calculation formula of the label is as follows:
wherein H is t,label In the case of a label being a label,for the output signal of the carrier phase recovery module, +.>Is a signal at the transmitting end in a high-speed optical communication system.
An apparatus for a method of compensating for non-linearities of an optical fiber, comprising: the system comprises a nonlinear disturbance item construction unit, a screening and reconstruction unit, a neural network unit and a compensation unit;
the nonlinear disturbance term construction unit is used for constructing nonlinear disturbance terms of intra-channel cross phase modulation and intra-channel four-wave mixing according to output signals of the carrier phase recovery module in a receiving end of the coherent optical communication system; the screening and reconstructing unit is used for screening and reconstructing the nonlinear disturbance item to obtain multidimensional input characteristics; the neural network unit is used for inputting the multidimensional input characteristics into the BiLSTM neural network to obtain the nonlinear damage value of the optical fiber; the compensation unit is used for compensating the output signal of the carrier phase recovery module according to the nonlinear damage value of the optical fiber to obtain a nonlinear compensation signal.
In summary, the invention has the following beneficial effects: the invention combines the multidimensional input characteristic reconstructed by the nonlinear disturbance item with the time memory characteristic of the BiLSTM neural network, and learns the nonlinear damage value from the neural network more accurately, thereby compensating the cross phase modulation in the optical fiber channel and the four-wave mixing effect in the channel, and realizing remarkable compensation effect under different modulation formats; meanwhile, the BiLSTM neural network adopted by the invention has stronger nonlinear fitting capability and low complexity.
Drawings
FIG. 1 is a comprehensive flow chart of a signal processing process and a method of the present invention in a receiving end of a coherent optical communication system;
FIG. 2 is a flow chart of a method of compensating for fiber nonlinearity;
FIG. 3 is a schematic diagram of a BiLSTM neural network;
fig. 4 is a block diagram of a simulation of a single carrier and 5 channel WDM coherent optical transmission system in accordance with the present invention;
FIG. 5 is a graph of Q factor simulation of a 28GBaud PDM-16QAM signal under a single carrier in the present invention;
FIG. 6 is a graph of Q factor simulation of a 28GBaud PDM-16QAM signal in a WDM system according to the present invention;
FIG. 7 is a graph of Q factor simulation for an 85GBaud PDM-64QAM signal under single carrier in accordance with the present invention;
fig. 8 is a graph of Q factor simulation of an 85GBaud PDM-64QAM signal in accordance with the present invention in a WDM system.
Detailed Description
The following description of the embodiments of the present invention is provided to facilitate understanding of the present invention by those skilled in the art, but it should be understood that the present invention is not limited to the scope of the embodiments, and all the inventions which make use of the inventive concept are protected by the spirit and scope of the present invention as defined and defined in the appended claims to those skilled in the art.
As shown in fig. 1-2, a method for compensating for nonlinearity of an optical fiber includes the steps of:
s1, constructing a nonlinear disturbance term of cross phase modulation in a channel and four-wave mixing in the channel according to an output signal of a carrier phase recovery module in a receiving end of a coherent optical communication system;
after receiving signals at a receiving end of a coherent optical communication system, the receiving end sequentially carries out dispersion compensation, polarization demultiplexing, frequency offset estimation and carrier phase recovery processing on the received signals, and the processed output signals are used for constructing nonlinear disturbance items of cross phase modulation in a channel and four-wave mixing in the channel.
The expression of the nonlinear disturbance term in the step S1 is:
T t =A x/y,n+t A * x/y,m+n+t A x/y,m+t +A y/x,n+t A * y/x,m+n+t A x/y,m+t
wherein T is t A nonlinear disturbance term corresponding to a signal value at a t-th moment of an output signal of the carrier phase recovery module, when A x/y,n+t When a horizontally polarized signal representing a signal value at the n+t time of an output signal of a carrier phase recovery module is a y/x,n+t A vertically polarized signal representing the signal value at the n+t time of the output signal of the carrier phase recovery module, when A x/y,n+t When a vertically polarized signal representing a signal value at the n+t time of an output signal of a carrier phase recovery module is a y/x,n+t A horizontally polarized signal representing the signal value at the n+t time of the output signal of the carrier phase recovery module, when A * x/y,m+n+t When the conjugation of the horizontally polarized signal representing the signal value at the m+n+t-th time of the output signal of the carrier phase recovery module, then A * y/x,m+n+t Conjugation of vertically polarized signal representing signal value at m+n+t time of output signal of carrier phase recovery module, when A * x/y,m+n+t When the conjugation of the vertically polarized signal representing the signal value at the m+n+t-th time of the output signal of the carrier phase recovery module, then A * y/x,m+n+t Conjugation of horizontally polarized signal representing signal value at m+n+t time of output signal of carrier phase recovery module, A x/y,m+t The signal value of the signal value at the m+t time of the output signal of the carrier phase recovery module is horizontally polarized or vertically polarized, m represents the m time of the output signal of the carrier phase recovery module, and n represents the n time of the output signal of the carrier phase recovery module.
The output signal of the carrier phase recovery module can be in any square QAM modulation format.
S2, screening and reconstructing nonlinear disturbance items to obtain multidimensional input features;
the step S2 comprises the following sub-steps:
s21, setting a disturbance threshold condition, and reserving a nonlinear disturbance term meeting the disturbance threshold condition:
the threshold condition in the step S21 is:
|m||n|≤C,|m|≤L,|n|≤L
wherein m represents the mth moment of the output signal of the carrier phase recovery module, n represents the nth moment of the output signal of the carrier phase recovery module, C is a disturbance threshold for balancing compensation performance, and L is a disturbance threshold for balancing complexity.
In this embodiment, c=7, l=11. By the threshold condition, a nonlinear disturbance term with a large nonlinear influence is retained.
S22, forming a two-dimensional real number array by real and imaginary parts of the two-dimensional complex number array of the reserved nonlinear disturbance term;
s23, dividing the two-dimensional real number array into three-dimensional real number arrays by taking the step length as M;
in this embodiment, a three-dimensional real number array (B, M, N), B is the batch size of the multidimensional input feature of the input BiLSTM neural network, N is the real number of nonlinear disturbance terms, and the step size M is set to 11 in consideration of the range of nonlinear damage.
S24, taking the three-dimensional real number array as a multi-dimensional input feature, the method can be expressed as follows:
T t,M =[T t-k ,…,T t-1 ,T t ,T t+1 ,…,T t+k ]
s3, inputting the multidimensional input characteristics into a BiLSTM neural network to obtain a fiber nonlinear damage value;
in this embodiment, a data set composed of three-dimensional real number arrays is divided into 50% as a training set, 20% as a verification set, and 30% as a test set. The BiLSTM neural network used in step S3 is a trained, validated and tested BiLSTM neural network. When the BiLSTM neural network is trained, the label adopts the difference value between the output signal of the receiving end carrier phase recovery module and the signal of the transmitting end. When the verification accuracy of 10 samples of epochs is not improved when the BiLSTM neural network is trained, the early-stop method is used to enhance the generalization performance of the BiLSTM neural network.
As shown in fig. 3, the BiLSTM neural network in step S3 includes: an input layer, a two-way long-short-term memory neural network layer, a flat layer and a full-connection layer;
the input end of the input layer is used as the input end of the BiLSTM neural network, and the output end of the input layer is connected with the input end of the two-way long-short-period memory neural network layer; the input end of the flat layer is connected with the output end of the two-way long-short-period memory neural network layer, and the output end of the flat layer is connected with the input end of the full-connection layer; the output end of the full connection layer is used as the output end of the BiLSTM neural network.
The two-way long-short-term memory neural network layer comprises 10 neurons, the full-connection layer comprises 2 neurons, one neuron of the full-connection layer is used for outputting a real part of an optical fiber nonlinear damage value, and the other neuron of the full-connection layer is used for outputting an imaginary part of the optical fiber nonlinear damage value.
The specific operation process in the BiLSTM neural network is as follows:
c t =σ(W f [h t-1 ,T t,M ]+b f )*c t-1 +σ(W i [h t-1 ,T t,M ]+b i )*tanh(W c [h t-1 ,T t,M ]+b c )
h t =σ(W o [h t-1 ,T t,M ]+b o )*tanh(c t )
wherein c t T is the cell state at time T t,M For multidimensional input feature at time t, h t-1 Outputting the state of the two-way long-short-term memory neural network layer at the t-1 time, h t For the output of the state of the layer of the bidirectional long-short-term memory neural network at the t moment, W f Weight matrix for forgetting gate in two-way long-short-term memory neural network layer, W i Weight matrix W for input gate in two-way long-short-term memory neural network layer c Is a weight matrix of cell states in a two-way long-short-term memory neural network layer, W o B is a weight matrix of output gates in the two-way long-short-term memory neural network layer f Bias matrix for forgetting gate in two-way long-short-term memory neural network layer, b i Bias matrix for input gate in two-way long-short term memory neural network layer, b c Bias matrix for cell state in two-way long-short-term memory neural network layer, b o A bias matrix for output gates in the two-way long-short-term memory neural network layer, j is a counting symbol, k represents an identifier for calculating the number, h j B, outputting the state of the layer of the two-way long-short-term memory neural network at the j-th moment j Bias matrix for full connection layer, W j Is a weight matrix of the full connection layer,and learning an output optical fiber nonlinear damage value for the BiLSTM neural network.
During training, the loss function adopted by the fiber nonlinear equalization model is as follows:
wherein L is MSE In order to achieve a loss value, the value of the loss,the fiber nonlinear damage value which is learned and output by the BiLSTM neural network is B, the size of a sample batch is B, i is the ith sample, and +.>For the output signal of the carrier phase recovery module, +.>Is a signal at the transmitting end in a high-speed optical communication system.
During training, the calculation formula of the label is as follows:
wherein H is t,label In the case of a label being a label,for the output signal of the carrier phase recovery module, +.>Is a signal at the transmitting end in a high-speed optical communication system.
In this embodiment, the BiLSTM neural network may further include: a dropout layer; the value range of the dropoff value of the dropoff layer is [0.1,0.5], and the dropoff value setting method of the dropoff layer comprises the following steps: when the transmission power of the multi-dimensional input feature is reduced, the dropoff value is gradually increased, and when the transmission power of the multi-dimensional input feature is increased, the dropoff value is gradually reduced and is used for preventing the model from being over fitted.
And S4, compensating the output signal of the carrier phase recovery module according to the nonlinear damage value of the optical fiber to obtain a nonlinear compensation signal.
The step S4 includes the following sub-steps:
s41, constructing a real part and an imaginary part of the nonlinear damage value of the optical fiber into a complex number array;
s42, subtracting the output signal of the carrier phase recovery module from the complex number array to obtain a nonlinear compensation signal.
An apparatus for a method of compensating for non-linearities of an optical fiber, comprising: the system comprises a nonlinear disturbance item construction unit, a screening and reconstruction unit, a neural network unit and a compensation unit;
the nonlinear disturbance term construction unit is used for constructing nonlinear disturbance terms of intra-channel cross phase modulation and intra-channel four-wave mixing according to output signals of the carrier phase recovery module in a receiving end of the coherent optical communication system; the screening and reconstructing unit is used for screening and reconstructing the nonlinear disturbance item to obtain multidimensional input characteristics; the neural network unit is used for inputting the multidimensional input characteristics into the BiLSTM neural network to obtain the nonlinear damage value of the optical fiber; the compensation unit is used for compensating the output signal of the carrier phase recovery module according to the nonlinear damage value of the optical fiber to obtain a nonlinear compensation signal.
The compensation method and the system in the embodiment are suitable for an ultra-high speed optical communication system, and the highest information rate of a single channel can reach 1.020Tbps for a single carrier and WDM system.
In order to further verify the performance of the invention, a coherent optical communication simulation system shown in fig. 4 is built by using VPI11.1, MATLAB and Keras, and is simultaneously suitable for a single carrier and WDM system. Firstly, generating 28Gbaud PDM-16QAM and 85Gbaud PDM-64QAM signals at a transmitting end, setting the frequency offset of a laser to be 100MHz and the line width to be 100KHz, and switching a single carrier and a 5-channel WDM system by utilizing an optical switch. Transmitting the signal into an optical fiber loop, setting the length of each span of the optical fiber to be 80km, the loss coefficient to be 0.2dB/km, the dispersion coefficient to be 16 ps/(nm-km), the polarization mode dispersion to be 0.1ps/≡km, and the nonlinear coefficient to be 1.3W -1 /km. And adding EDFA in each span to compensate the loss in the link, and introducing ASE noise. The transmission distance of PDM-16QAM is 1600km, and the EDFA noise index is set to be 6dB; the transmission distance of PDM-64QAM is 400km, and the EDFA noise index is set to be 4dB. Photoelectric conversion is realized at the receiving end by utilizing a coherent receiver, and I in a transmission link is received x 、Q x 、I y 、Q y Four paths of electric signals. Finally, the received signal is processed by a series of DSP processes, such as dispersion compensation, polarization demultiplexing, frequency offset estimation and carrier phase recovery, and after carrier phase recovery, the method of the invention is used for realizing optical fiber nonlinear compensation. For a 5-channel WDM system, an intermediate 3 rd channel signal which is most affected by nonlinearity is selected for simulation, the center frequency is 193.4MHz, the channel interval when transmitting 28Gbaud PDM-16QAM signals is 50GHz, and the channel interval when transmitting 85Gbaud PDM-64QAM signals is 100GHz.
After compensation, the signal is decoded and BER calculated, and the Q factor is deduced by using the BER value to be used as an index for measuring the system performance. As shown in fig. 5-6, which are PDM-16QAM signal simulation diagrams, for a single carrier system, the optimal fiber-in power is increased from-1 dBm to 0dBm, and the signal-to-noise ratio can be increased by about 4.1dB under the optimal fiber-in power; for WDM systems, the optimum in-fiber power is also increased from-1 dBm to 0dBm, and at the optimum in-fiber power, the signal to noise ratio can be increased by about 2.9dB. As shown in fig. 7-8, which are PDM-64QAM signal simulation diagrams, for a single carrier system, the optimal fiber-in power is increased from 3dBm to 4dBm, and the signal-to-noise ratio can be increased by about 2.5dB under the optimal fiber-in power; for WDM systems, the optimum in-fiber power is raised to 2dBm to 3dBm, and at the optimum in-fiber power, the signal to noise ratio can be raised by about 2.2dB. Therefore, the invention has obvious compensation effect under the single carrier system, and the performance improvement under the WDM system is not higher than that of the single carrier system, but the compensation effect is obvious as well. And the higher the fiber-entering power is, namely the larger the nonlinear effect influence is, the more obvious the performance of the single carrier and the WDM system is improved.
In conclusion, through simulation verification, the method compensates the nonlinearity of the optical fiber by combining the multidimensional input characteristics reconstructed by the nonlinear disturbance term and the time memory of the BiLSTM, can accurately estimate the nonlinear damage value in the optical fiber, and has more obvious advantages under the condition that the nonlinear effect is more obvious.
Claims (7)
1. A method of compensating for non-linearities of an optical fiber, comprising the steps of:
s1, constructing a nonlinear disturbance term of cross phase modulation in a channel and four-wave mixing in the channel according to an output signal of a carrier phase recovery module in a receiving end of a coherent optical communication system;
s2, screening and reconstructing nonlinear disturbance items to obtain multidimensional input features;
s3, inputting the multidimensional input characteristics into a BiLSTM neural network to obtain a fiber nonlinear damage value;
s4, compensating the output signal of the carrier phase recovery module according to the nonlinear damage value of the optical fiber to obtain a nonlinear compensation signal;
the step S2 comprises the following sub-steps:
s21, setting a disturbance threshold condition, and reserving a nonlinear disturbance term meeting the disturbance threshold condition;
s22, forming a two-dimensional real number array by real and imaginary parts of the two-dimensional complex number array of the reserved nonlinear disturbance term;
s23, dividing the two-dimensional real number array into three-dimensional real number arrays by taking the step length as M;
s24, taking the three-dimensional real number array as a multi-dimensional input characteristic;
the threshold condition in the step S21 is:
|m||n|≤C,|m|≤L,|n|≤L
wherein m represents the mth moment of the output signal of the carrier phase recovery module, n represents the nth moment of the output signal of the carrier phase recovery module, C is a disturbance threshold for balancing compensation performance, and L is a disturbance threshold for balancing complexity;
the loss function adopted in the step S3 for calculating the nonlinear damage value of the optical fiber is as follows:
wherein L is MSE In order to achieve a loss value, the value of the loss,the fiber nonlinear damage value which is learned and output by the BiLSTM neural network is B, the size of a sample batch is B, i is the ith sample, and +.>For the output signal of the carrier phase recovery module, +.>Is a signal at the transmitting end in a high-speed optical communication system.
2. The method for compensating for optical fiber nonlinearity according to claim 1, wherein the expression of the nonlinear disturbance term in step S1 is:
wherein T is t A nonlinear disturbance term corresponding to a signal value at a t-th moment of an output signal of the carrier phase recovery module, when A x/y,n+t Representing the output of the carrier phase recovery moduleWhen the signal value at time n+t of the signal is horizontally polarized, then A y/x,n+t A vertically polarized signal representing the signal value at the n+t time of the output signal of the carrier phase recovery module, when A x/y,n+t When a vertically polarized signal representing a signal value at the n+t time of an output signal of a carrier phase recovery module is a y/x,n+t A horizontally polarized signal representing the signal value at the n+t time of the output signal of the carrier phase recovery module whenWhen the signal value representing the m+n+t-th time of the output signal of the carrier phase recovery module is conjugated, thenConjugation of a vertically polarized signal representing the signal value at the m+n+t time of the output signal of the carrier phase recovery module, when +.>Conjugation of a vertically polarized signal representing the signal value at the m+n+t time of the output signal of the carrier phase recovery module>Conjugation of horizontally polarized signal representing signal value at m+n+t time of output signal of carrier phase recovery module, A x/y,m+t The signal value of the signal value at the m+t time of the output signal of the carrier phase recovery module is horizontally polarized or vertically polarized, m represents the m time of the output signal of the carrier phase recovery module, and n represents the n time of the output signal of the carrier phase recovery module.
3. The method of claim 1, wherein the BiLSTM neural network in step S3 comprises: an input layer, a two-way long-short-term memory neural network layer, a flat layer and a full-connection layer;
the input end of the input layer is used as the input end of the BiLSTM neural network, and the output end of the input layer is connected with the input end of the two-way long-short-period memory neural network layer; the input end of the flat layer is connected with the output end of the two-way long-short-period memory neural network layer, and the output end of the flat layer is connected with the input end of the full-connection layer; the output end of the full connection layer is used as the output end of the BiLSTM neural network.
4. The method of claim 3, wherein the two-way long and short term memory neural network layer comprises 10 neurons, the full connection layer comprises 2 neurons, one neuron of the full connection layer is used for outputting a real part of the fiber nonlinear damage value, and the other neuron of the full connection layer is used for outputting an imaginary part of the fiber nonlinear damage value.
5. The method of claim 3, wherein the BiLSTM neural network further comprises: a dropout layer; the value range of the dropoff value of the dropoff layer is [0.1,0.5], and the dropoff value setting method of the dropoff layer comprises the following steps: when the transmission power of the multi-dimensional input feature is reduced, the dropoff value is gradually increased, and when the transmission power of the multi-dimensional input feature is increased, the dropoff value is gradually reduced.
6. The method of compensating for optical fiber nonlinearity according to claim 1, wherein said step S4 comprises the sub-steps of:
s41, constructing a real part and an imaginary part of the nonlinear damage value of the optical fiber into a complex number array;
s42, subtracting the output signal of the carrier phase recovery module from the complex number array to obtain a nonlinear compensation signal.
7. An apparatus of the optical fiber nonlinearity compensation method according to any one of claims 1 to 6, comprising: the system comprises a nonlinear disturbance item construction unit, a screening and reconstruction unit, a neural network unit and a compensation unit;
the nonlinear disturbance term construction unit is used for constructing nonlinear disturbance terms of intra-channel cross phase modulation and intra-channel four-wave mixing according to output signals of the carrier phase recovery module in a receiving end of the coherent optical communication system; the screening and reconstructing unit is used for screening and reconstructing the nonlinear disturbance item to obtain multidimensional input characteristics; the neural network unit is used for inputting the multidimensional input characteristics into the BiLSTM neural network to obtain the nonlinear damage value of the optical fiber; the compensation unit is used for compensating the output signal of the carrier phase recovery module according to the nonlinear damage value of the optical fiber to obtain a nonlinear compensation signal.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210899794.4A CN115314118B (en) | 2022-07-28 | 2022-07-28 | Optical fiber nonlinear compensation method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210899794.4A CN115314118B (en) | 2022-07-28 | 2022-07-28 | Optical fiber nonlinear compensation method and device |
Publications (2)
Publication Number | Publication Date |
---|---|
CN115314118A CN115314118A (en) | 2022-11-08 |
CN115314118B true CN115314118B (en) | 2024-02-20 |
Family
ID=83858106
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210899794.4A Active CN115314118B (en) | 2022-07-28 | 2022-07-28 | Optical fiber nonlinear compensation method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115314118B (en) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112291005A (en) * | 2020-08-20 | 2021-01-29 | 中国科学技术大学 | Bi-LSTM neural network-based receiving end signal detection method |
CN113285758A (en) * | 2021-05-18 | 2021-08-20 | 成都信息工程大学 | Optical fiber nonlinear equalization method based on IPCA-DNN algorithm |
CN113364527A (en) * | 2021-06-03 | 2021-09-07 | 聊城大学 | Nonlinear damage compensation method suitable for high-speed coherent polarization multiplexing system |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US11270200B2 (en) * | 2018-02-26 | 2022-03-08 | Nec Corporation | Single-step nonlinearity compensation using artificial intelligence for digital coherent transmission systems |
US10833770B2 (en) * | 2018-06-22 | 2020-11-10 | Nec Corporation | Optical fiber nonlinearity compensation using neural networks |
-
2022
- 2022-07-28 CN CN202210899794.4A patent/CN115314118B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112291005A (en) * | 2020-08-20 | 2021-01-29 | 中国科学技术大学 | Bi-LSTM neural network-based receiving end signal detection method |
CN113285758A (en) * | 2021-05-18 | 2021-08-20 | 成都信息工程大学 | Optical fiber nonlinear equalization method based on IPCA-DNN algorithm |
CN113364527A (en) * | 2021-06-03 | 2021-09-07 | 聊城大学 | Nonlinear damage compensation method suitable for high-speed coherent polarization multiplexing system |
Non-Patent Citations (2)
Title |
---|
Compensation of Fiber Nonlinearities in Digital Coherent Systems Leveraging Long Short-Term Memory Neural Networks;Stavros Deligiannidis 等;《Journal of Lightwave Technology》;第5991-5999页 * |
Low-Complexity Fiber Nonlinearity Impairments Compensation Enabled by Simple Recurrent Neural Network With Time Memory;Yan Zhao 等;《IEEE Access》;第160995-161004页 * |
Also Published As
Publication number | Publication date |
---|---|
CN115314118A (en) | 2022-11-08 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN113285758B (en) | Optical fiber nonlinear equalization method based on IPCA-DNN algorithm | |
Niu et al. | End-to-end deep learning for long-haul fiber transmission using differentiable surrogate channel | |
Li et al. | BER performance of FSO communication system with differential signaling over correlated atmospheric turbulence fading | |
Li et al. | End-to-end learning for optical fiber communication with data-driven channel model | |
CN113346957B (en) | Clustering nonlinear compensation method for OAM-QPSK transmission | |
CN113364527B (en) | Nonlinear damage compensation method suitable for high-speed coherent polarization multiplexing system | |
CN115314118B (en) | Optical fiber nonlinear compensation method and device | |
CN112613538B (en) | Nonlinear equalization method based on weighted principal component analysis | |
Guo et al. | Deep neural network based chromatic dispersion estimation with ultra-low sampling rate for optical fiber communication systems | |
Fang et al. | 6.4 Tb/s SSB WDM Transmission Over 320km SSMF With Linear Network-Assisted LSTM | |
CN115314119A (en) | Optical fiber nonlinear equalization method and system in high-speed optical communication system | |
Zhang et al. | DeepONet-Based Waveform-Level Simulation for a Wideband Nonlinear WDM System | |
Esteves et al. | Deep learning for BER prediction in optical connections impaired by inter-core crosstalk | |
CN114285715B (en) | Nonlinear equalization method based on bidirectional GRU-conditional random field | |
Cui et al. | Optical Fiber Channel Modeling Method Using Multi-BiLSTM for PM-QPSK Systems | |
Peng et al. | Complex long short-term memory neural networks for fiber nonlinearity equalization in long-haul transmission systems | |
Yang et al. | Fiber Nonlinear Compensation Using Bi-directional Recurrent Neural Network Model Based on Attention Mechanism | |
Yang et al. | Intelligent joint multi-parameter optical performance monitoring scheme based on HT images and MT-ResNet for elastic optical network | |
CN114553315B (en) | Optical fiber nonlinear equalization method and system based on CNN-biRNN | |
CN117318832A (en) | Nonlinear joint compensation method and device in and among WDM system channels | |
CN115208721B (en) | Volterra-like neural network equalizer construction method and system | |
CN112769497B (en) | Method for carrying out nonlinear compensation on high-capacity high-order QAM coherent light | |
Li et al. | MIMO-GRU for Fiber Nonlinearity Equalization in 880Gbit/s Long-distance Transmission System | |
CN116346230A (en) | Nonlinear damage compensation method and device suitable for high-order modulation wavelength division multiplexing system | |
Zhang et al. | Full-Spectrum INFT Algorithm for Dual-Polarization NFDM Transmission |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |