Carrier phase recovery method and system based on convolutional recurrent neural network

Abstract

The invention relates to the technical field of optical fiber communication, and particularly discloses a carrier phase recovery method and system based on a convolutional recurrent neural network. The method comprises the following steps: firstly, performing Gaussian blur on a received signal by adopting a convolutional layer to reduce the influence of amplified spontaneous emission; then, a recurrent relationship between symbols is constructed by means of a recurrent neural network, so that interference of additive white Gaussian noise is effectively removed under the condition of low signal-to-noise ratio, and the performance of a blind phase search algorithm in a probabilistic shaping system is improved. According to the method, the convolutional recurrent neural network is combined with the blind phase search algorithm, effective compensation of phase noise in the probabilistic shaping system is achieved, the problem that the performance of a traditional blind phase search algorithm at the low signal-to-noise ratio in the probabilistic shaping system is limited is solved, the advantages of probabilistic shaping in optical communication are further explored, and the optical communication performance is improved. The mutual information performance of a transmission system is effectively improved, and the method has an important application prospect in the field of carrier phase recovery of optical communication.

Description

Technical Field

[0001] The present invention relates to the technical field of optical fiber communication, and in particular to a carrier phase recovery method and system based on a convolutional recurrent neural network. Background Art

[0002] In the field of optical communication, greater bandwidth, longer transmission distance, and higher receiving sensitivity are the current goals. The development of coherent optical communication technology and constellation shaping technology has greatly improved the capacity of optical fiber communication systems. Coherent optical communication systems can be used in the communication of various modulation format signals and meet the above goals at the same time. To meet the growing bandwidth requirements, a probability shaping system is often used to improve the spectral utilization rate. Probability shaping technology adjusts the probability distribution of signal constellation points, improves the achievable information rate of the system, and reduces the signal-to-noise ratio required for transmission. However, when the signal propagates in the probability shaping system, the characteristic of uniform distribution of the transmitted signal is broken, resulting in the degradation of the performance of most existing digital signal processing algorithms.

[0003] Blind phase search is an algorithm that equally divides the phase interval to obtain a series of test phases, and uses this to compensate for carrier phase noise. This algorithm does not require pilot overhead, has a significant effect on compensating for phase noise damage under high-order quadrature amplitude modulation, is widely used in coherent optical communication systems, and is often used as the limit reference value for most carrier phase recovery algorithms.

[0004] However, after using probability shaping to adjust the probability distribution of signal constellation points, the probability of high-amplitude symbols in the transmitted signal is reduced, resulting in serious mutual information performance damage to the blind phase search algorithm. On the other hand, in the case of a low signal-to-noise ratio, in order to mitigate the influence of additive white Gaussian noise, the blind phase search algorithm often needs to increase the length of the averaging filter to improve the system performance, which increases the complexity of the algorithm. The current problem is how to reduce the influence of additive white Gaussian noise, amplify the gain of high-amplitude symbols, improve the robustness of the algorithm, and enhance the fast tracking ability of phase noise without increasing the filter length.

[0005] The existing optical fiber communication technology has the following technical problems: (1) When the signal propagates in the probability shaping system, the characteristic of uniform distribution of the transmitted signal is broken, resulting in the degradation of the performance of most existing digital signal processing algorithms; (2) After using probability shaping to adjust the probability distribution of signal constellation points, the probability of high-amplitude symbols in the transmitted signal is reduced, resulting in serious mutual information performance damage to the blind phase search algorithm; (3) In the case of a low signal-to-noise ratio, in order to mitigate the influence of additive white Gaussian noise, the blind phase search algorithm improves the system performance by increasing the length of the averaging filter, increasing the complexity of the algorithm. Summary of the Invention

[0006] To solve the above technical problems, the present invention provides a carrier phase recovery method and system based on a convolutional recurrent neural network, which improves the prior art, reduces the influence of additive white Gaussian noise, amplifies the gain of high-amplitude symbols, and enhances the robustness of the algorithm and the ability to quickly track phase noise.

[0007] To achieve the above object, the technical solution adopted by the present invention is: First, a carrier phase recovery method based on a convolutional recurrent neural network is provided, including the following steps:

[0008] S1 Normalize and batch the input signal to obtain a batched signal;

[0009] S2 Select a test phase, rotate the batched signal according to the test phase , and make a decision between the rotated signal and the standard constellation points, and calculate the square of the Euclidean distance between the rotated signal and the decision signal ;

[0010] S3 Use the square of the Euclidean distance as the input of the convolutional recurrent neural network to construct a recursive relationship of the sum of the squares of the Euclidean distances;

[0011] S4 Take the test phase corresponding to the minimum value of the sum of the squares of the Euclidean distances as the carrier phase estimation value; perform carrier phase recovery on the batched signal according to the obtained carrier phase estimation value;

[0012] S5 Repeat steps S2 - S4 until the input signal is completely recovered.

[0013] Preferably, the input signal in S1 is:

[0014] ,

[0015] where is the input signal, is the ideal noise-free signal, is the additive white Gaussian noise in the optical fiber link, is the phase noise.

[0016] Preferably, the process of selecting the test phase is as follows:

[0017] Bisection the phase angle to obtain B test phases as follows:

[0018] ,

[0019] where B is the number of test angles, indicating that the phase angle The number of equal - interval divisions.

[0020] Preferably, the calculation process of the squared Euclidean distance in S2 is as follows:

[0021]

[0022] Wherein, is the symbol after decision.

[0023] Preferably, the construction process of the recursive relationship of the sum of squared distances in S3 is as follows:

[0024] Construct a convolution kernel M according to the amplitude of the received signal:

[0025]

[0026] Wherein, represents the amplitude of the n - th symbol, and L represents the length of the convolution kernel;

[0027] Perform a convolution operation on the squared Euclidean distance:

[0028] ,

[0029] Wherein, represents convolution calculation, M is the convolution kernel, and this convolution kernel is related to the amplitude characteristics of the constellation points;

[0030] Calculate the sum of squared Euclidean distances of the symbols before and after the current symbol, that is, the current squared Euclidean distance sum, and obtain:

[0031] ,

[0032] Wherein, is the number of filter taps;

[0033] Send the current squared Euclidean distance sum into a recurrent neural network to construct the recursive relationship between it and the squared Euclidean distance sum of the previous batch:

[0034] ,

[0035] Wherein, is the forgetting factor.

[0036] Preferably, the formation method of the input signal in S1 includes the following steps:

[0037] S1.1 Send the binary sequence into a constant - composition distribution matcher for probability shaping to generate a bit sequence that conforms to the Maxwell - Boltzmann distribution;

[0038] S1.2 Map the bit sequence to generate corresponding 64QAM symbols;

[0039] S1.3 Modulate the signal onto the optical carrier by an IQ modulator and then send it into a standard single-mode fiber link for transmission; at the receiving end, the local oscillator laser has the same parameters as the laser at the transmitting end, and the optical signal is filtered by an optical band-pass filter.

[0040] S1.4 The coherent receiver receives the optical signal, and the electrical signal obtained after photoelectric conversion is used as the input signal .

[0041] Preferably, the convolutional recurrent neural network described in S3 includes an input layer, a hidden layer, and an output layer, and the layers are connected in a fully connected manner; the hidden layer is divided into a CNN layer and an RNN layer. The CNN layer includes a convolutional layer and a fully connected layer, and the RNN layer is a long short-term memory network structure (LSTM), including layer units.

[0042] Preferably, the convolutional recurrent neural network uses the rectified linear unit as the activation function, uses the mean square error function as the loss function, and is trained using the backpropagation algorithm.

[0043] The present invention also provides a carrier phase recovery system based on a convolutional recurrent neural network, including:

[0044] Signal input module:

[0045] Used to normalize and batch the input signal to obtain a batched signal;

[0046] Euclidean distance square calculation module:

[0047] Used to select a test phase, rotate the batched signal according to the test phase and make a decision between the rotated signal and the standard constellation point, and calculate the square of the Euclidean distance between the rotated signal and the decision signal ;

[0048] Recursive relationship construction module:

[0049] Used to use the square of the Euclidean distance as the input of the convolutional recurrent neural network and construct a recursive relationship of the sum of the squares of the Euclidean distances;

[0050] Carrier phase recovery module:

[0051] Used to take the test phase corresponding to the minimum value of the sum of the squares of the Euclidean distances as the carrier phase estimation value; perform carrier phase recovery on the batched signal according to the obtained carrier phase estimation value.

[0052] The above technical solution has the following advantages or beneficial effects:

[0053] The present invention constructs a convolutional recurrent neural network. By introducing Gaussian blur through the convolutional layer, the impact of amplified spontaneous emission noise is reduced. Then, by constructing a recursive relationship through the recurrent neural network, the impact of additive white Gaussian noise under low signal-to-noise ratio is reduced, thereby solving the problem of the performance degradation of the carrier phase recovery algorithm caused by unequal signal transmission probabilities in the probability shaping coherent optical communication system. The accuracy of phase noise estimation is improved, the channel mutual information is enhanced, the algorithm complexity is reduced, and thus the performance of the probability shaping system is improved.

[0054] The above summary is only for the purpose of the specification and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments, and features described above, further aspects, embodiments, and features of the present invention will be readily apparent by reference to the drawings and the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In the drawings, unless otherwise specified, the same reference numerals throughout the several views represent the same or similar components or elements. These drawings are not necessarily drawn to scale. It should be understood that these drawings only depict some embodiments disclosed in accordance with the present invention and should not be regarded as limiting the scope of the present invention.

[0056] Figure 1 is a flowchart of a carrier phase recovery method based on a convolutional recurrent neural network provided by the present invention;

[0057] Figure 2 is a schematic structural diagram of the convolutional recurrent neural network model in the present invention;

[0058] Figure 3 is a flowchart of the steps of a preferred embodiment of a carrier phase recovery method based on a convolutional recurrent neural network provided by the present invention;

[0059] Figure 4 is a structural diagram of a carrier phase recovery system based on a convolutional recurrent neural network provided by the present invention; DETAILED DESCRIPTION OF THE EMBODIMENTS

[0060] In the following, only some exemplary embodiments are briefly described. As those skilled in the art can recognize, the described embodiments can be modified in various different ways without departing from the spirit or scope of the present invention. Therefore, the drawings and the description are considered to be exemplary in nature rather than restrictive.

[0061] The present invention first provides a carrier phase recovery method based on a convolutional recurrent neural network. The process is as Figure 1 shown and includes the following steps:

[0062] S1 Input signal Normalize and batch to obtain batched signals;

[0063] S2 Select test phases and rotate the batched signals according to the test phases and perform a decision between the signal after each rotation and the standard constellation points, and calculate the square of the Euclidean distance between the rotated signal and the decision signal ;

[0064] S3 Use the square of the Euclidean distance as the input of the convolutional recurrent neural network to construct a recursive relationship for the sum of the squares of the Euclidean distances;

[0065] S4 Take the test phase corresponding to the minimum value of the sum of the squares of the Euclidean distances as the carrier phase estimation value; perform carrier phase recovery on the batched signals according to the obtained carrier phase estimation value;

[0066] S5 Repeat steps S2 - S4 until the input signal is completely recovered.

[0067] Figure 3 is the flowchart of the steps of a preferred embodiment of a carrier phase recovery method based on a convolutional recurrent neural network provided by the present invention. As Figure 3 shown, according to an implementation manner of a carrier phase recovery method based on a convolutional recurrent neural network of the present invention, the specific implementation steps are as follows:

[0068] S1.1 Send a binary sequence with a sequence length of 65536 bit into a constant - composition distribution matcher for probability shaping to generate a bit sequence conforming to the Maxwell - Boltzmann distribution;

[0069] S1.2 Map the bit sequence to generate corresponding 64QAM symbols, set the wavelength of the transmitter laser to 1550 nm, and set the line width to 100 kHz;

[0070] S1.3 Modulate the signal onto an optical carrier by an IQ modulator and then send it into a standard single - mode fiber link with a transmission distance of 160 km for transmission; at the receiving end, the local oscillator laser uses the same parameters as the transmitter laser, and the optical signal is filtered by an optical band - pass filter with a bandwidth set to 60 GHz.

[0071] S1.4 The coherent receiver receives the optical signal, and the electrical signal obtained after photoelectric conversion is used as the input signal. The input signal is subjected to digital signal processing and every 32 symbols are divided into a batch. The digital signal processing includes resampling, normalization, orthogonalization, dispersion compensation, polarization mode dispersion compensation, and frequency offset estimation.

[0072] S2 Select 16 test phases and rotate the batched signals according to the test phases Rotate and make a decision on the signal after each rotation with the standard constellation points, and calculate the square of the Euclidean distance between the rotated signal and the decision signal of the Euclidean distance square;

[0073] The test phase is selected as follows:

[0074] For the phase angle make 16 equal divisions to obtain 16 test phases as follows:

[0075] .

[0076] The calculation process of the Euclidean distance square is as follows:

[0077] .

[0078] S3 uses the square of the Euclidean distance as the input of the convolutional recurrent neural network to construct a recursive relationship of the sum of the squares of the Euclidean distances, which is equivalent to using an infinitely long filter with an exponentially decaying coefficient to reduce the influence of additive white Gaussian noise.

[0079] The construction process of the recursive relationship of the sum of the squares of the Euclidean distances is as follows:

[0080] Construct the convolutional kernel M according to the amplitude of the received signal:

[0081]

[0082] where, represents the amplitude of the nth symbol, and L represents the length of the convolutional kernel;

[0083] First, the square of the Euclidean distance is convolved with the convolutional kernel in the convolutional layer to assign different weight coefficients to the squares of the Euclidean distances of different symbols:

[0084] ,

[0085] where, represents the convolutional calculation, M is the convolutional kernel, and this convolutional kernel is related to the amplitude characteristics of the constellation points;

[0086] Calculate the sum of the squares of the Euclidean distances of 16 symbols before and after the current symbol, that is, the current sum of the squares of the Euclidean distances, to obtain:

[0087] ;

[0088] Send the current sum of the squares of the Euclidean distances into the recurrent neural network to construct a recursive relationship between it and the sum of the squares of the Euclidean distances of the previous batch :

[0089] ,

[0090] Among them, is the forgetting factor.

[0091] The structure of the convolutional recurrent neural network model is as Figure 2 shown, including an input layer, a hidden layer and an output layer. The layers are connected in a fully connected manner. The hidden layer is divided into a CNN layer and an RNN layer. The CNN layer includes a convolutional layer and a fully connected layer. The RNN layer is a long short-term memory network structure (LSTM), including layer units. The rectified linear unit is used as the activation function, the mean square error function is used as the loss function, and the backpropagation algorithm is used for training to reduce the loss function until the optimal neural network is obtained. After training, the test data is used to verify the performance of the neural network model.

[0092] S4 takes the test phase corresponding to the minimum value of the sum of squared Euclidean distances as the carrier phase estimation value; the input signal is subjected to carrier phase recovery according to the obtained carrier phase estimation value;

[0093] S5 repeats steps S2 - S4 to batch implement the carrier phase recovery of the signal until all signals are recovered.

[0094] The present invention also provides a carrier phase recovery system based on a convolutional recurrent neural network, the structure of which is as Figure 4 shown, including:

[0095] Signal input module:

[0096] Used to normalize and batch the input signal to obtain a batched signal;

[0097] Euclidean distance square calculation module:

[0098] Used to select a test phase, rotate the batched signal according to the test phase , make a decision between the rotated signal and the standard constellation point, and calculate the square of the Euclidean distance between the rotated signal and the decision signal ;

[0099] Recursive relationship construction module:

[0100] Used to take the square of the Euclidean distance as the input of the convolutional recurrent neural network and construct a recursive relationship of the sum of squared Euclidean distances;

[0101] Carrier phase recovery module:

[0102] Used to take the test phase corresponding to the minimum value of the sum of squared Euclidean distances as the carrier phase estimation value; perform carrier phase recovery on the batched signal according to the obtained carrier phase estimation value.

[0103] The above technical solution has the following advantages or beneficial effects: In the present invention, a convolutional recurrent neural network is used to perform Gaussian blur on the received signal, reducing the influence of amplified spontaneous emission on phase noise compensation. At the same time, a recursive relationship between symbols is constructed through a recurrent neural network, solving the problem that the filter length required for removing additive white Gaussian noise by the traditional blind phase search algorithm is too long, and further improving the feasibility and robustness of the algorithm.

[0104] In the description of this specification, the description referring to terms such as "one embodiment", "one preferred embodiment", "example", "specific example", or "some examples" means that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. Moreover, the specific features, structures, materials, or characteristics described may be combined in a suitable manner in any one or more embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0105] Any process or method description shown in the flowchart or described in other ways herein can be understood as representing a module, segment, or part of code including one or more executable instructions for implementing a specific logical function or process, and the scope of the preferred embodiments of the present invention includes additional implementations, where the functions may be executed in a substantially simultaneous manner or in an opposite order according to the involved functions, rather than in the order shown or discussed, which should be understood by those skilled in the art to which the embodiments of the present invention belong.

[0106] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a definitional sequence list of executable instructions for implementing logical functions, and can be embodied specifically in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device), or used in conjunction with these instruction execution systems, apparatuses, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection portion having one or more wirings (electronic device), a portable computer diskette (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or other suitable medium on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpretation, or otherwise processing as appropriate, and then storing it in a computer memory.

[0107] It should be understood that various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application specific integrated circuits having appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0108] Those of ordinary skill in the art of this technology can understand that all or part of the steps carried by the methods of the above embodiments can be completed by instructing relevant hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiments. Among them, the device embodiments correspond to the method embodiments, so the description of the device embodiments is relatively brief, and the relevant descriptions can refer to the descriptions of the method embodiments.

[0109] In addition, each functional unit in various embodiments of the present invention may be integrated into a processing module, may exist separately physically for each unit, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium. The storage medium may be a read-only memory, a magnetic disk, an optical disc, or the like.

[0110] As described above, the above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of various changes or substitutions thereof, and these should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A carrier phase recovery method based on convolutional recurrent neural network, characterized in that: The steps include: S1 will input the signal Normalize and batch to obtain batch signals; S2 selects the test phase and performs the test on the batch signals according to the test phase. Rotate, judge the signal after each rotation with the standard constellation point, and calculate the difference between the rotated signal and the judgment signal The square of the Euclidean distance; S3 uses the square of Euclidean distance as the input of the convolutional recurrent neural network and constructs a recursive relationship of the sum of squared Euclidean distances; S4 takes the test phase corresponding to the minimum value of the sum of squares of the Euclidean distance as the carrier phase estimation value; performs carrier phase recovery on the batch signals according to the obtained carrier phase estimation value; S5 repeats steps S2-S4 until the input signal until the recovery is complete.

2. A carrier phase recovery method based on convolutional recurrent neural network as claimed in claim 1, characterized in that: The input signal of S1 is: ， in, is the input signal, For an ideal noise-free signal, is the additive white Gaussian noise in the optical fiber link, is the phase noise.

3. A carrier phase recovery method based on a convolutional recurrent neural network as claimed in claim 1, wherein the test phase The selection process is as follows: Phase Angle Divide into B equal parts and obtain B test phases as follows: ， Where B is the test angle, indicating that the phase angle The number of equal-interval divisions.

4. A carrier phase recovery method based on convolutional recurrent neural network as claimed in claim 1, characterized in that: The calculation process of the Euclidean distance squared described in S2 is: ， in, It is the symbol after the judgment.

5. A carrier phase recovery method based on convolutional recurrent neural network as claimed in claim 1, characterized in that: The recursive relationship construction process of the distance square sum described in S3 is as follows: Construct the convolution kernel M according to the amplitude of the received signal: in, represents the amplitude of the nth symbol, and L represents the convolution kernel length; The convolution operation is performed on the square of the Euclidean distance: ， in, represents convolution calculation, M is the convolution kernel, which is related to the amplitude characteristics of the constellation points; Calculate the current symbol before and after The sum of squared Euclidean distances of the symbols, that is, the sum of squared current Euclidean distances, is obtained: ， in, is the number of filter taps; The sum of the current Euclidean distance squares Send it to the recurrent neural network to construct the sum of squared Euclidean distances between it and the previous batch The recursive relationship between: ， in, For the forgetting factor.

6. A carrier phase recovery method based on convolutional recurrent neural network as claimed in claim 1, characterized in that: S1 Input signal The forming method comprises the following steps: S1.1 sends the binary sequence into the constant component distribution matcher for probability shaping to generate a bit sequence that conforms to the Maxwell-Boltzmann distribution; S1.2 maps the bit sequence to generate corresponding 64QAM symbols; S1.3 modulates the signal onto the optical carrier by the IQ modulator and then transmits it into the standard single-mode optical fiber link for transmission; at the receiving end, the local oscillator laser uses the same parameters as the transmitting end laser, and the optical signal is filtered by an optical bandpass filter; S1.4 The coherent receiver receives the optical signal and converts it into an electrical signal as the input signal. .

7. A carrier phase recovery method based on convolutional recurrent neural network as claimed in claim 1, characterized in that: The convolutional recurrent neural network described in S3 includes an input layer, a hidden layer and an output layer, and the layers are connected in a fully connected manner; the hidden layer is divided into a CNN layer and a RNN layer, the CNN layer includes a convolutional layer and a fully connected layer, and the RNN layer is a long short-term memory network structure (LSTM), including layer units.

8. A carrier phase recovery method based on convolutional recurrent neural network as claimed in claim 7, characterized in that: The convolutional recurrent neural network uses a rectified linear unit as an activation function, adopts a mean square error function as a loss function, and adopts a back propagation algorithm for training.

9. A carrier phase recovery system based on convolutional recurrent neural network, characterized in that: include: Signal input module: For input signal Normalize and batch; Euclidean distance square calculation module: The received signal is tested according to the phase Rotate, judge the signal after each rotation with the standard constellation point, and calculate the difference between the rotated signal and the judgment signal The square of the Euclidean distance; Recursive relation building blocks: The squared Euclidean distance is used as the input of the convolutional recurrent neural network to construct a recursive relationship of the sum of squared Euclidean distances; Carrier phase recovery module: The test phase corresponding to the minimum value of the sum of squared Euclidean distances is taken as a carrier phase estimation value; and the carrier phase of the input signal is recovered according to the obtained carrier phase estimation value.

Images