ACGAN signal modulation format identification method and system based on complex neural network
Through the ACGAN framework based on complex neural networks, signal modulation format recognition is performed in the optical fiber communication system, the problems of high signal processing complexity and phase information loss in the prior art are solved, and efficient and accurate modulation format recognition and signal recovery are achieved.
Patent Information
- Application Number
- CN202411062479.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-05
- Publication Date
- 2024-11-15
AI Technical Summary
In the existing optical fiber communication system, the signal modulation format recognition technology is limited to the real domain, resulting in high signal processing complexity and the inability to effectively identify the modulation format under complex channel conditions, and lacks the ability to retain and process phase information.
The ACGAN signal modulation format recognition method based on complex neural networks is used to preprocess the signal through IQ orthogonal, dispersion compensation, clock recovery, polarization demultiplexing and frequency bias estimation, and the ACGAN framework of complex neural networks is trained and Identification, operate directly in the complex domain to extract multidimensional features, and achieve accurate identification of modulation format.
It improves the accuracy and complex signal processing capabilities of signal recognition, retains phase information, reduces the complexity of signal processing, enhances the robustness and adaptability of the system, and can accurately identify modulation formats in various environments.
Smart Images

Figure CN118972216A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to, but is not limited to, the field of optical fiber communication technology, and particularly relates to an ACGAN signal modulation format recognition method and system based on complex neural networks. Background Technology
[0002] High-capacity, low-latency, and high-reliability communication technologies are currently a crucial development direction. To meet the ever-increasing demands for capacity and bandwidth, optical fiber communication will play an even more vital role in the future. Signal modulation format recognition is an essential requirement in the offline DSP modules of optical fiber communication systems. Artificial intelligence and machine learning, with their unique advantages, can process and identify various modulation signals more accurately and quickly compared to traditional signal processing methods. However, current signal modulation format recognition technology in optical fiber communication systems is limited to the real number domain, i.e., using only I-channel or Q-channel signals, increasing the complexity of signal processing. Developing a new method and system for signal modulation format recognition in coherent optical communication systems is crucial for modulation format recognition in future dynamic network environments and complex channel conditions, injecting new vitality into the continued development of optical communication technology. Summary of the Invention
[0003] To address the problems existing in the prior art, this invention provides an ACGAN signal modulation format recognition method based on complex neural networks.
[0004] This invention is implemented as follows: an ACGAN signal modulation format recognition method based on complex neural networks, the method comprising:
[0005] S1: The original modulated signal output from the coherent optical receiver is processed offline by the DSP for decision output, and then undergoes IQ quadrature, dispersion compensation, clock recovery, polarization demultiplexing and frequency offset estimation.
[0006] S2: Perform signal preprocessing on the data after frequency offset estimation; then use 80% of the data as the training set and 20% of the data as the test set, and label each set of data as originating from which modulation format data (label);
[0007] S3: ACGAN modulation format recognition based on complex neural network;
[0008] S4: Output modulation format;
[0009] S5: Carrier phase recovery, decision output.
[0010] Furthermore, the IQ orthogonality adopts the Gram-Schmidt Orthogonal Normalization (GSOP) algorithm; the dispersion compensation utilizes electrical domain compensation and employs an FIR filter for compensation; the clock recovery adopts a digital squared filter clock recovery algorithm; the polarization demultiplexing adopts a constant mode algorithm to compensate for polarization mode dispersion and is used for depolarization multiplexing; and the frequency offset estimation adopts a VV (Viterbi-Viterbi) algorithm.
[0011] Furthermore, in S2, signal preprocessing is performed on the data after frequency offset estimation, specifically including: normalizing signals of different modulation formats, rearranging the data and labeling it.
[0012] Furthermore, S3 specifically includes:
[0013] The fully connected neural network in the ACGAN framework is replaced with a complex neural network, and the framework is divided into two networks: a generation network and a recognition network.
[0014] The training process of the ACGAN architecture is as follows: Training data (label) and noise (z) are input into the generator network (net_G), generating fake data (i.e., data affected by noise). Then, the training data (label) and the fake data are simultaneously input into the recognition network (net_D). The recognition network ultimately outputs the signal format recognition result (c1, c2...cn represent different signal format types; correct recognition is real=1, incorrect recognition is fake=0). The core idea is that the fake data generated by the generator network is continuously compared and trained with the correct data in the recognition network to achieve correct recognition. The part actually used for recognition is the recognition network. The evaluation criterion is the loss function of the recognition network.
[0015] loss_D = E1[logPD] real (x)]+E2[logPD fake (x)]+E3[logPD real (c,x)]+E4[logPD fake (c,x)]
[0016] If the loss function does not converge, it is fed back to the generator network for further training. If the loss function converges, the network is usable. The network is then tested using test set data to determine whether it outputs the correct recognition format.
[0017] Another objective of this invention is to provide a complex neural network-based ACGAN signal modulation format recognition system based on the aforementioned complex neural network-based ACGAN signal modulation format recognition method, characterized in that the system specifically includes:
[0018] The offline processing module processes the original modulated signal output from the coherent optical receiver through DSP offline processing for decision output, including IQ orthogonality, dispersion compensation, clock recovery, polarization demultiplexing, and frequency offset estimation.
[0019] The preprocessing module, connected to the offline processing module, performs signal preprocessing on the data after frequency offset estimation; then 80% of the data is used as the training set, and 20% of the data is used as the test set, with each set of data labeled according to the modulation format from which it originates.
[0020] The modulation format recognition module is connected to the preprocessing module and is based on the complex neural network ACGAN modulation format recognition.
[0021] The output module connects to the modulation format recognition module and outputs the modulation format.
[0022] The recovery module, connected to the output module, recovers the carrier phase and outputs a decision.
[0023] Another object of the present invention is to provide a computer device, the computer device including a memory and a processor, the memory storing a computer program, the computer program being executed by the processor causing the processor to perform the steps of the ACGAN signal modulation format recognition method based on complex neural networks.
[0024] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the ACGAN signal modulation format recognition method based on a complex neural network.
[0025] Another objective of this invention is to provide an information data processing terminal, which is used to implement the ACGAN signal modulation format recognition system based on complex neural networks.
[0026] Based on the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solution to be protected by this invention are as follows:
[0027] First, current signal modulation format identification methods operate in the real domain, meaning they extract I-channel or Q-channel data from the signal separately for analysis and computation, increasing signal processing complexity. However, complex neural networks (ANNs) have both real and complex domain input data, eliminating the need to extract I-channel or Q-channel data. The signal can be directly used as input to complete ACGAN training and identification based on a complex neural network. Furthermore, the network can extract phase features, resulting in richer feature extraction and significantly improved recognition capabilities. The core idea of this invention is to utilize the frequency offset estimation data while preserving the original data features, and then directly train the signal data using the ACGAN framework based on a complex neural network to extract multi-dimensional features. This achieves the goal of accurately identifying modulation formats and restoring the signal with optimal phase compensation.
[0028] Second, the technical solution of this invention solves a technical problem that people have long desired to solve but have never been able to achieve:
[0029] a. Preservation of phase information
[0030] Complex neural networks in ACGANs can naturally preserve and process the phase information of signals, which is crucial for modulation format recognition. The loss of phase information can severely affect the accuracy of recognition, and complex neural networks naturally handle this information through their architecture.
[0031] b. Complex signal processing
[0032] Complex signals typically contain more information that is difficult to extract and utilize effectively in traditional real-number neural networks. ACGAN, a complex neural network, operates and learns directly in the complex domain through its complex weights and activation functions. This allows the network to learn and process the inherent patterns of complex signals more naturally and effectively, enabling it to better capture and process this information and thus improve the accuracy of modulation format recognition.
[0033] c. More accurate recognition capabilities and better learning abilities
[0034] The generator in a complex neural network like ACGAN can create additional training data, helping the network better understand various modulation types, which is especially useful when the number of samples is limited. The discriminator learns to distinguish between real and generated signals, a process that enhances the network's ability to recognize complex signal patterns. The addition of auxiliary training promotes the completion of training and avoids the disadvantage of GANs struggling to converge and getting stuck in local optima. As training progresses, the generator and discriminator continuously optimize, the quality of the generated signals continuously improves, thereby increasing the overall recognition accuracy.
[0035] (4) The technical solution of the present invention overcomes technical bias:
[0036] Traditional neural networks typically only process real-valued signals, making the processing of complex signals less natural and efficient. This necessitates a transformation from the complex to the real domain, inevitably increasing signal processing complexity and neglecting phase information. Complex neural networks, on the other hand, can directly process complex signals, avoiding the information loss and additional computational complexity associated with converting complex signals to real-valued signals. While retaining the advantages of traditional neural networks in signal processing, they also enhance the extraction and processing of phase features. ACGAN, based on complex neural networks, not only inherits the processing advantages of complex neural networks but also fully leverages GAN's data learning and feature enhancement capabilities, further improving recognition accuracy. Furthermore, through auxiliary condition training, it avoids the disadvantages of GANs, such as difficulty converging and susceptibility to local optima. This approach overcomes technical biases, making processing more efficient and rational.
[0037] Third, the technical solution of this invention solves several key problems in the prior art and brings about significant technological progress. The following is a detailed explanation of these technical problems and technological progress:
[0038] ###Technical problems solved:
[0039] 1) Signal Distortion and Dispersion Issues: During signal transmission, various physical effects (such as dispersion, polarization mode dispersion, etc.) can cause signal distortion. This invention effectively reduces these distortions and improves signal quality by employing dispersion compensation and polarization demultiplexing techniques.
[0040] 2) Clock Synchronization Issue: Clock synchronization is crucial in communication systems. If the receiver's clock is out of sync with the transmitter's, decoding errors will occur. This invention achieves precise clock synchronization by employing a digital squared filter clock recovery algorithm.
[0041] 3) Frequency offset problem: Due to equipment instability or environmental factors, signal frequency offset may occur during transmission. This invention utilizes the VV algorithm for frequency offset estimation, effectively correcting this offset.
[0042] 4) Signal recognition accuracy issues: In complex communication environments, accurately identifying signals with different modulation formats is a challenge. Traditional signal recognition methods cannot cope with multiple modulation formats and noise interference.
[0043] Significant technological advancements:
[0044] 1) Improve signal quality: By adopting advanced signal processing technologies (such as GSOP algorithm, dispersion compensation, polarization demultiplexing, etc.), this invention significantly improves the quality of received signals, laying a solid foundation for subsequent signal identification and decoding.
[0045] 2) Enhanced system robustness: The application of clock recovery and frequency offset estimation techniques makes the communication system more robust and able to maintain high performance under various environmental conditions.
[0046] 3) Introduction of deep learning models: This invention innovatively applies deep learning models (especially the ACGAN architecture) to signal recognition, which not only improves the accuracy of recognition, but also enhances the system's adaptability to unknown or complex modulation formats.
[0047] 4) Application of complex neural networks: Traditional deep learning models are limited when processing complex signals. This invention introduces complex neural networks to process complex signals more effectively, further improving the accuracy of signal recognition.
[0048] 5) Adaptive learning capability: Through the training and optimization of deep learning models, the technical solution of this invention possesses adaptive learning capability. This means that the system can automatically adjust its parameters and strategies according to changes in the actual communication environment to maintain optimal performance.
[0049] In summary, the technical solution of the present invention not only solves several key problems in the prior art, but also achieves significant technological progress by introducing advanced signal processing and deep learning technologies, providing strong support for the efficient, accurate and robust operation of communication systems. Attached Figure Description
[0050] Figure 1 This is a flowchart of the ACGAN signal modulation format recognition method based on complex neural networks provided in this embodiment of the invention;
[0051] Figure 2 This is a compensation diagram of the FIR filter provided in the embodiment of the present invention;
[0052] Figure 3 This is a diagram of the clock recovery algorithm using digital square filtering provided in an embodiment of the present invention;
[0053] Figure 4 This is a diagram of the constant modulus algorithm provided in an embodiment of the present invention;
[0054] Figure 5 This is a diagram of the VV (Viterbi-Viterbi) algorithm provided in an embodiment of the present invention;
[0055] Figure 6 This is a complex neural network diagram provided in an embodiment of the present invention;
[0056] Figure 7 This is a data preprocessing flowchart provided in an embodiment of the present invention;
[0057] Figure 8This is a schematic diagram of normalized data rearrangement provided in an embodiment of the present invention;
[0058] Figure 9 This is a flowchart of ACGAN based on complex neural networks provided in an embodiment of the present invention;
[0059] Figure 10 This refers to the recognition rate of ACGAN based on complex neural networks for signals with different modulation formats provided in this embodiment of the invention.
[0060] Figure 11 This is the recognition of different modulation format signals by ACGAN based on complex neural networks provided in this embodiment of the invention. Detailed Implementation
[0061] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.
[0062] This invention describes an advanced signal processing workflow that covers multiple steps from signal preprocessing to deep learning model application. The following is a detailed explanation of this technical solution:
[0063] 1) Signal preprocessing and compensation:
[0064] Carrier phase recovery and decision output are critical steps in communication systems to ensure accurate demodulation of signals.
[0065] IQ orthogonality employs the Gram-Schmidt Orthogonal Normalization (GSOP) algorithm, a mathematical method used to ensure the orthogonality of signals and improve the accuracy of signal processing.
[0066] Dispersion compensation is achieved through electrical compensation and FIR filters, which helps reduce signal distortion during transmission.
[0067] The clock recovery uses a digital squared filter clock recovery algorithm to ensure that the receiver can accurately synchronize the clock of the transmitter and prevent decoding errors caused by clock deviation.
[0068] The polarization demultiplexing uses a constant mode algorithm, which not only compensates for polarization mode dispersion but also effectively demultiplexes polarization, improving signal transmission efficiency.
[0069] Frequency offset estimation uses the Viterbi-Viterbi (VV) algorithm, which is an effective frequency offset estimation method that can ensure that the receiver accurately decodes the transmitted signal.
[0070] 2) Data preprocessing:
[0071] After frequency offset estimation, signals with different modulation formats are normalized, which helps to unify the scale of different signals and facilitates subsequent processing.
[0072] The data is rearranged and labeled in order to prepare a training dataset for deep learning.
[0073] 3) Applications of deep learning models:
[0074] This invention innovatively replaces the fully connected neural network in the ACGAN (Auxiliary Classifier Generative Adversarial Network) framework with a complex neural network to better handle complex signals.
[0075] The framework consists of two parts: a generation network and a recognition network. The generation network is responsible for generating fake data that is affected by noise, while the recognition network is responsible for distinguishing real data from the fake data generated by the generation network.
[0076] Through continuous comparative training, the recognition network can learn to correctly identify different signal debugging format types.
[0077] The loss function of the recognition network is a key metric for evaluating model performance. If the loss function does not converge, it is fed back to the generator network for further training; if the loss function converges, it indicates that the network has been trained successfully and its recognition performance can be verified using test set data.
[0078] The technical solution of this invention overcomes the limitations of traditional signal processing methods in complex communication environments. By introducing a deep learning model, it significantly improves the accuracy and efficiency of signal recognition. This approach, combining traditional signal processing and deep learning, represents an innovative trend in the field of communication technology and is expected to play a vital role in future communication systems.
[0079] like Figure 1 As shown, this embodiment of the invention provides a method for identifying ACGAN signal modulation formats based on complex neural networks. The method includes:
[0080] S1: The original modulated signal output from the coherent optical receiver is processed offline by the DSP for decision output, and then undergoes IQ quadrature, dispersion compensation, clock recovery, polarization demultiplexing and frequency offset estimation.
[0081] S2: Perform signal preprocessing on the data after frequency offset estimation; then use 80% of the data as the training set and 20% of the data as the test set, and label each set of data as originating from which modulation format data (label);
[0082] S3: ACGAN modulation format recognition based on complex neural network;
[0083] S4: Output modulation format;
[0084] S5: Carrier phase recovery, decision output.
[0085] This invention provides a method for recognizing ACGAN (Generative Adversarial Network) signal modulation formats based on complex neural networks. Its working principle can be divided into the following steps:
[0086] 1. Raw signal processing
[0087] First, the coherent optical receiver receives the original modulated signal, which is then processed offline by digital signal processing (DSP) for decision output. The processing includes IQ quadrature (decomposing the signal into in-phase and quadrature components), dispersion compensation (correcting dispersion effects generated during fiber transmission), clock recovery (restoring the clock information of the signal), polarization demultiplexing (separating polarization modes), and frequency offset estimation (correcting frequency shift).
[0088] 2. Data preprocessing and set partitioning
[0089] After frequency offset estimation, the signal data is preprocessed to prepare for machine learning training. Data preprocessing steps include normalization and denoising. The processed data is divided into a training set (80%) and a test set (20%), and each set is labeled with its source modulation format (label) to ensure that the model can learn and validate the characteristics of different modulation formats during training and testing.
[0090] 3. ACGAN recognition based on complex neural networks
[0091] Modulation format recognition is performed using an ACGAN model based on complex neural networks. Complex neural networks can handle complex signal data, and ACGAN combines the generator and discriminator architecture of generative adversarial networks (GANs) to improve the model's recognition accuracy through adversarial training. The generator produces modulation format samples, while the discriminator evaluates the authenticity of the samples, thereby optimizing the model's recognition performance.
[0092] 4. Modulation format output and carrier phase recovery
[0093] Finally, the modulation format is determined based on the output of the ACGAN model, and carrier phase recovery is performed. The carrier phase recovery step corrects errors caused by phase changes during signal transmission, ensuring the accuracy of the decision output. This process ultimately generates accurate modulation format identification results, thereby achieving correct classification and processing of the modulated signal.
[0094] The IQ orthogonality is achieved using the Gram-Schmidt Orthogonal Normalization (GSOP) algorithm, as detailed below:
[0095]
[0096]
[0097]
[0098] Where the correlation coefficient ρ=E{r1(t)·r Q (t)}, P Q =E{Q′ 2 (t)}.
[0099] like Figures 2-5 As shown, the dispersion compensation utilizes electrical domain compensation and employs an FIR filter for compensation; the clock recovery uses a digital squared filter clock recovery algorithm; the polarization demultiplexing uses a constant mode algorithm to compensate for polarization mode dispersion and is used for depolarization multiplexing; the frequency offset estimation uses a VV (Viterbi-Viterbi) algorithm.
[0100] like Figure 7 , 8 As shown, further, in S2, signal preprocessing is performed on the data after frequency offset estimation, specifically including: normalizing signals of different modulation formats, rearranging the data and labeling it.
[0101] like Figure 6 As shown, further, S3 specifically includes:
[0102] The fully connected neural network in the ACGAN framework is replaced with a complex neural network, and the framework is divided into two networks: a generation network and a recognition network.
[0103] like Figure 9-11 As shown, the training process of the ACGAN architecture is as follows: Training data (label) and noise (z) are input into the generator network (net_G), generating fake data (i.e., data affected by noise). Then, the training data (label) and the fake data are simultaneously input into the recognition network (net_D). The recognition network ultimately outputs the signal format recognition result (c1, c2...cn represent different signal format types; correct recognition is real=1, incorrect recognition is fake=0). The core idea is that the fake data generated by the generator network and the correct data are continuously compared and trained in the recognition network to achieve correct recognition. The part actually used for recognition is the recognition network. The evaluation criterion is the loss function of the recognition network.
[0104] loss_D = E1[logPD] real (x)]+E2[logPD fake (x)]+E3[logPD real (c,x)]+E4[logPD fake (c,x)]
[0105] If the loss function does not converge, it is fed back to the generator network for further training. If the loss function converges, the network is usable. The network is then tested using test set data to determine whether it outputs the correct recognition format.
[0106] This invention provides a complex neural network-based ACGAN signal modulation format recognition system based on the aforementioned complex neural network-based ACGAN signal modulation format recognition method. The system specifically includes:
[0107] The offline processing module processes the original modulated signal output from the coherent optical receiver through DSP offline processing for decision output, including IQ orthogonality, dispersion compensation, clock recovery, polarization demultiplexing, and frequency offset estimation.
[0108] The preprocessing module, connected to the offline processing module, performs signal preprocessing on the data after frequency offset estimation; then 80% of the data is used as the training set, and 20% of the data is used as the test set, with each set of data labeled according to the modulation format from which it originates.
[0109] The modulation format recognition module is connected to the preprocessing module and is based on the ACGAN modulation format recognition of complex neural networks.
[0110] The output module connects to the modulation format recognition module and outputs the modulation format.
[0111] The recovery module, connected to the output module, recovers the carrier phase and outputs a decision.
[0112] This invention provides a computer device, which includes a memory and a processor. The memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the ACGAN signal modulation format recognition method based on complex neural networks.
[0113] This invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the ACGAN signal modulation format recognition method based on a complex neural network.
[0114] This invention provides an information data processing terminal, which is used to implement the ACGAN signal modulation format recognition system based on complex neural networks.
[0115] This embodiment provides a signal modulation format identification method for coherent optical communication systems, including: normalized modulation signal data preprocessing; and ACGAN modulation format identification based on complex neural networks.
[0116] In this embodiment, preferably, the preprocessing of the modulation signal data based on power normalization includes: after performing DSP processing on the communication complex signal, extracting each group of modulation signals to form a data sequence, and normalizing the data sequence; reconstructing and shuffling each group of normalized complex signal data sequences with 1024 signals per group; and adding a set of label sequences to the expanded matrix sequence for classification.
[0117] In some applications, conventional machine learning methods for identifying modulation formats train constellation diagrams, directly extracting features from the diagrams. However, the training method described in this embodiment involves randomly selecting a portion of the signal and using the complex signal as input to the neural network. This batch-based training approach is equivalent to local feature extraction, using local features to progressively train the network.
[0118] Traditional training methods for constellation maps require preprocessing, such as density labeling or clustering of constellation points, and distinguishing different regions by color or brightness to facilitate feature extraction. This embodiment, however, uses local complex signals as input, allowing direct extraction of full features from the complex signal. Furthermore, when processing the signal set, constellation maps contain blank areas (regions without any points), which are considered invalid. This embodiment extracts full signal features, eliminating invalid feature extraction.
[0119] Figure 10 To assess the recognition rate of signals with different modulation formats based on complex neural networks, it can be found that signal format recognition can be achieved when the optical signal-to-noise ratio is greater than 20dB. Figure 11 The display further illustrates the recognition performance of signals with different modulation formats. The horizontal axis represents the modulation format of the signal output by the network, and the vertical axis represents the actual modulation format of the signal. For example, the first column on the horizontal axis shows QPSK signals with a 100% success rate and zero incorrect recognitions.
[0120] It should be noted that embodiments of the present invention can be implemented in hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by a suitable instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the above-described devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuitry such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of the above-described hardware circuitry and software, such as firmware.
[0121] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. A method for identifying modulation format of ACGAN signal based on complex neural network, characterized in that: The method includes: S1: The original modulated signal output by the coherent optical receiver is processed offline by DSP for decision output, and then goes through IQ orthogonality, dispersion compensation, clock recovery, polarization demultiplexing and frequency offset estimation; S2: Perform signal preprocessing on the data after frequency offset estimation; then 80% of the data is used as a training set, and 20% of the data is used as a test set, and each group of data is marked with which modulation format data (label); S3: ACGAN modulation format recognition based on complex neural network; S4: output modulation format; S5: Carrier phase recovery and decision output.
2. According to claim 1, the ACGAN signal modulation format recognition method based on complex neural network is characterized in that: The IQ orthogonality adopts the Gram-Schmidt Orthogonal Normalization (GSOP) algorithm; the dispersion compensation utilizes electrical domain compensation and adopts an FIR filter for compensation; the clock recovery adopts a digital square filter clock recovery algorithm; the polarization demultiplexing adopts a constant modulus algorithm to compensate for polarization mode dispersion and is used for polarization demultiplexing; the frequency offset estimation adopts a VV (Viterbi-Viterbi) algorithm.
3. According to claim 1, the ACGAN signal modulation format recognition method based on complex neural network is characterized in that: In S2, signal preprocessing is performed on the data after the frequency offset estimation, specifically including: normalizing the signals of different modulation formats, rearranging the data and labeling them.
4. The ACGAN signal modulation format recognition method based on complex neural network according to claim 1, characterized in that: The S3 specifically includes: The fully connected neural network in the ACGAN framework is replaced with a complex neural network. The framework is divided into two networks, the generation network and the recognition network. The training process of ACGAN architecture is as follows: input the training set data (label) and noise (z) into the generation network (net_G), and generate fake data (i.e. data affected by noise), then input the training set data (label) and fake data into the recognition network (net_D) at the same time, and the recognition network finally outputs the signal format recognition result (c1, c2..cn represent different signal debugging format types, correct recognition is real=1, wrong recognition is fake=0); the core idea is to continuously compare and train the fake data generated by the generation network with the correct data in the recognition network to achieve correct recognition, and the part that is actually used for recognition is the recognition network. The evaluation criteria are the Loss function of the recognition network: loss_D=E1[logPD real (x)]+E2[logPD fake (x)]+E3[logPD real (c,x)]+E4[logPD fake (c,x)] If the loss does not converge, it is fed back to the generating network to continue training. If the loss converges, the network is usable and tested through the test set data to determine whether the recognition network outputs the correct recognition format.
5. A complex neural network-based ACGAN signal modulation format recognition system based on the complex neural network-based ACGAN signal modulation format recognition method according to claims 1-4, characterized in that: The system specifically includes: Offline processing module: the original modulated signal output by the coherent optical receiving end is processed offline by DSP for decision output, and then undergoes IQ orthogonality, dispersion compensation, clock recovery, polarization demultiplexing and frequency offset estimation; The preprocessing module is connected to the offline processing module to perform signal preprocessing on the data after frequency offset estimation; then 80% of the data is used as a training set, and 20% of the data is used as a test set, and each group of data is marked with which modulation format data (label); Modulation format recognition module, connected to the preprocessing module, based on ACGAN modulation format recognition of complex neural network; An output module, connected to the modulation format identification module, outputs the modulation format; The recovery module is connected to the output module, recovers the carrier phase, and determines the output.
6. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the ACGAN signal modulation format identification method based on a complex neural network as described in any one of claims 1 to 4.
7. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the ACGAN signal modulation format recognition method based on a complex neural network as described in any one of claims 1 to 4.
8. An information data processing terminal, characterized in that: The information data processing terminal is used to implement the ACGAN signal modulation format recognition system based on complex neural network as described in claim 5.
Citation Information
Cited By
Asynchronous modulation identification method and device based on plural neural networks
CN120110856A
A method and apparatus for identifying asynchronous modulation based on complex neural networks
CN120110856B
