CN114866158B - Channel modeling and simulating method for underwater laser digital communication system - Google Patents
Channel modeling and simulating method for underwater laser digital communication system Download PDFInfo
- Publication number
- CN114866158B CN114866158B CN202210791128.9A CN202210791128A CN114866158B CN 114866158 B CN114866158 B CN 114866158B CN 202210791128 A CN202210791128 A CN 202210791128A CN 114866158 B CN114866158 B CN 114866158B
- Authority
- CN
- China
- Prior art keywords
- signal
- model
- sample
- training
- communication system
- 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
Images
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
- H04B10/80—Optical aspects relating to the use of optical transmission for specific applications, not provided for in groups H04B10/03 - H04B10/70, e.g. optical power feeding or optical transmission through water
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
- H04B10/07—Arrangements for monitoring or testing transmission systems; Arrangements for fault measurement of transmission systems
- H04B10/073—Arrangements for monitoring or testing transmission systems; Arrangements for fault measurement of transmission systems using an out-of-service signal
-
- 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
- H04B10/11—Arrangements specific to free-space transmission, i.e. transmission through air or vacuum
- H04B10/112—Line-of-sight transmission over an extended range
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B17/00—Monitoring; Testing
- H04B17/30—Monitoring; Testing of propagation channels
- H04B17/391—Modelling the propagation channel
- H04B17/3912—Simulation models, e.g. distribution of spectral power density or received signal strength indicator [RSSI] for a given geographic region
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A20/00—Water conservation; Efficient water supply; Efficient water use
- Y02A20/152—Water filtration
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Signal Processing (AREA)
- Computer Networks & Wireless Communication (AREA)
- Electromagnetism (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Geometry (AREA)
- Computer Hardware Design (AREA)
- Medical Informatics (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Optical Communication System (AREA)
Abstract
The invention provides a channel modeling and simulating method for an underwater laser digital communication system, which comprises the following steps: firstly, an underwater laser digital communication experimental system is set up, actual measurement receiving signals under various water quality conditions are obtained, a data set is set up, and a training set and a testing set are divided; then, establishing a condition to generate a confrontation network model CGAN, wherein the model comprises a generator model G and a discriminator model D, and simultaneously introducing a gradient punishment algorithm to improve the training process; then, setting network parameters and training parameters, adopting a small batch random gradient descent algorithm, importing a training set alternative training generator model and a discriminator model, obtaining a final channel model and realizing channel simulation. The reliability of the model established by the method is better; the accuracy of channel simulation is better, and the obtained simulation signals are more in line with random distribution in a real application environment, so that the method is more suitable for guiding the design and practical application of the underwater laser digital communication system.
Description
Technical Field
The invention belongs to the technical field of underwater digital information transmission, and particularly relates to a channel modeling and simulating method for an underwater laser digital communication system.
Background
Due to the complexity of the seawater channel, underwater visible light communication currently faces a series of difficulties. The ocean is a complex system composed of various living beings and non-living beings, and different ocean environment parameters, solutes and suspended particles have complex absorption and scattering properties for light with different wavelengths. When an optical signal is transmitted in seawater, photons randomly collide with water molecules or other particles in the seawater to deviate from the original transmission direction, so that a beam divergence phenomenon is formed, the power of the optical signal at a receiving end is attenuated, and the beam divergence phenomenon is more serious as the transmission distance is longer. Meanwhile, the attenuation characteristics of channels with different marine environment parameters (different water depths and different water qualities) on optical signals are different, and the complex absorption and scattering characteristics of the same channel on optical signals with different wavelengths are also greatly different. The complexity and uncertainty of the underwater visible light digital signal transmission channel therefore presents significant difficulties to channel modeling and simulation. Meanwhile, a real underwater visible light digital channel transmission model also needs to consider the influence of each part of a communication system on transmission signals in all aspects, and the pulse response of a sensing device and a corresponding subsequent processing circuit also has certain interference on received information, which is also a very complex and difficult-to-quantify process. It is difficult to have a method to fully simulate the complex characteristics of an underwater optical channel.
At present, the two common underwater visible light digital channel modeling methods at home and abroad can be roughly summarized into two methods, namely a theoretical model calculation method and a Monte Carlo numerical simulation method. The theoretical model research is to statistically describe the characteristics of the underwater optical channel from a mathematical angle, and a function expression of the whole transmission diffusion process of the light beam is fitted by calculating parameters such as transmission time, spatial position, deflection angle and the like of photons in different water bodies. The method is mainly used for qualitative analysis research, has large quantitative analysis error and cannot be used as an instructive tool in system design. The Monte Carlo simulation method of the underwater optical channel, also called as a numerical simulation method, simulates the propagation process of a light beam in water by simulating the motion trail of a large number of photons, and finally obtains a series of parameters of the channel based on statistical characteristics. According to the method, the water body medium is subdivided into small enough areas through grid division to realize medium uniformity and carry out calculation, and the calculation process is often not converged in the inhomogeneous water body medium with strict boundary conditions, so that an effective calculation result cannot be obtained.
In addition, the research of the conventional numerical channel modeling method focuses mainly on the physical attenuation process of different types of water channels on optical signals, and generally does not consider the response of the photoelectric devices at the transmitting end and the receiving end, such as the dark current noise of a Photomultiplier Tube (PMT) which is a commonly used receiving end device, the frequency response of an amplifier, the nonlinear output of a laser, and the absolute error in the D/a conversion and a/D conversion processes. Therefore, when modeling and simulating are carried out by the traditional theoretical analysis method and the Monte Carlo method, the physical loss process of the water environment on the optical signals is more important to research, and the influence of system errors caused by devices such as photoelectric devices is ignored. In recent years, with the rapid development of deep learning technology, a new idea is provided for solving the problem of modeling of an underwater visible light digital channel and improving the accuracy and reliability of simulation.
Disclosure of Invention
Aiming at the problems, the invention provides a channel modeling and simulating method for an underwater laser digital communication system, which comprises the following steps:
in the above formula θGRepresenting the generator model parameters, θDRepresenting discriminator model parameters, y representing transmitted signal samples as condition information, x representing true received signal samples,g (z | y) represents the simulated signal sample, z = pz(z) noise samples satisfying a Gaussian distribution, pdataAnd pgRespectively representing the true sample distribution and the sample distribution learned by the generator model, lambda represents a gradient penalty coefficient,is shown in pdataAnd pgUniformly sampling the two distributed samples to obtain new sample distribution; the goal of G is to minimize the probability that D identifies G (z | y) as false, the goal of D is to minimize the probability that G (z | y) is identified as true, while maximizing the probability that the true sample x is identified as true, G and D are trained alternately, thereby forming a mechanism for countertraining;
In one possible design, the generator model G adopts a network structure of five-layer cascaded one-dimensional convolution layers, the dimensions of each layer of output space are 256, 128, 64, 32, 1, the lengths of one-dimensional convolution windows are 5, 3, 2, the first three layers of activation functions all adopt Tanh, the fourth layer of activation functions adopts ReLU, and the fifth layer directly linearly outputs results; the generator model G is used for simulating a sample generator of a real sample, receives a noise signal and condition information as input, and generates a simulation sample after processing;
the discriminator model D adopts a six-layer network structure, wherein the first four layers are one-dimensional convolution layers, the second two layers are full-connection layers, the dimensions of output spaces of each convolution layer are respectively 256, 128, 64 and 16, the lengths of one-dimensional convolution windows are respectively 5, 3 and 3, and the number of neurons in the fifth layer is 100; the first three layers of activation functions all adopt ReLU, the fourth layer is linearly output to the fifth layer, the fifth layer of activation functions adopts ReLU, and the sixth layer directly and linearly outputs results; the discriminator model D is used for receiving the samples and the condition information as input, judging whether the current samples are from real samples or false samples generated by the generator under the guidance of the condition information, and judging whether the current samples correspond to correct condition information; the condition information is a sending signal, and the noise signal is gaussian white noise.
In one possible design, the specific training process of the model is:
s31, initializing model parameters thetaDAnd thetaG;
S32: in a training batch of D, a sending signal sample y and a noise sample z-p are obtained firstlyz(z) calculating a simulated signal sample therefrom←G(z|y);
S33: obtaining a true signal sample x-pdata(x) And a random number delta-U0, 1]From which a true and false joint distribution sample is computed;
S34: inputting the three samples and the transmission signal into D respectively to obtain corresponding output values, and then updating and recording the loss functionA value of (d);
s35: repeating the steps S32-S34 for 500 times, and updating theta by using an Adam optimizerD;
S36: one training batch at GIn the second step, step S32 is repeated first to obtain a simulation signal sampleInputting the sample into D to obtain an output value of D, and then updating and recording the value of the loss function (1);
s37: repeating the step S36 for 500 times, and updating theta by utilizing Adam optimizerG;
S38: repeating the steps S35 and S37 until the training set is traversed once;
s39: and (6) repeatedly executing the S38 step for about 1000 times to reach a convergence state, and finishing the training.
In a possible design, the underwater laser digital information communication system in the step 1 comprises a sending end and a receiving end; the transmitting end consists of an AWG and a laser, and the output end of the AWG is connected to the analog signal input end of the laser through a BNC connecting line; the receiving end comprises a PMT, an amplifying module, an oscilloscope, a PC, a high-voltage power supply and a direct-current stabilized power supply, wherein the output end of the PMT is connected to the input end of the amplifying module through a BNC-to-SMA connecting wire, the output end of the amplifying module is connected to a BNC input interface of the oscilloscope through an SMA-to-BNC connecting wire, and the oscilloscope is communicated with the PC through a USB port; the PMT is powered by a high-voltage power supply, and the amplification module is powered by a direct-current stabilized power supply.
In a possible design, the specific process of establishing the data set in step 1 is as follows:
s11, generating a pseudo-random binary sequence by a PC, converting the pseudo-random binary sequence into a non-return-to-zero code-on-off keying modulation signal as a sending signal and inputting the signal into an arbitrary waveform generator, and outputting an analog signal to directly drive a laser to carry out intensity modulation after the AWG finishes D/A conversion of the signal;
s12, converting the electric signal into an optical signal by a laser, transmitting the optical signal to a receiving end after the optical signal passes through a water body environment, converting the optical signal into the electric signal by the receiving end, amplifying the electric signal by an amplifier module, and inputting the electric signal into an oscilloscope for sampling to obtain a corresponding discrete signal;
s13, performing off-line processing on the discrete signal in the S12 in a PC to obtain a synchronized receiving signal, collecting the synchronized signal as the receiving signal, and performing off-line demodulation processing on the synchronized signal and calculating an error rate for use in subsequent verification of the modeling effect of the model;
s14: a data set is made in conjunction with the received signal in S13 and the corresponding transmitted signal in S11.
In one possible design, the setting of the parameters in step 3 includes setting the batch size to 500, setting the initial learning rate to 0.0001, and setting the gradient penalty factor to 5, where θ is set every time θ is optimizedGOptimizing thetaDIs set to 6.
In a possible design, step 4 is further included, namely, a process of testing the trained network model is specifically:
and (2) inputting the test set data obtained in the step (1) into the generator model trained in the step (3), and taking the absolute amplitude spectrum error and the bit error rate error between the model simulation signal and the real received signal as the modeling effect of a standard test model, wherein the calculation method is shown in a formula (3) and a formula (4).
In the above formula, X andrespectively representing the real received signal and the corresponding model simulation signal.
Compared with the prior art, the invention provides the channel modeling and simulation method for the underwater laser digital communication system, which designs and constructs an end-to-end channel model of the underwater laser digital communication system by fitting actual measurement data through a generated countermeasure network, truly and reliably restores the characteristic information of the underwater optical channel according to the actual measurement data, overcomes the limitation that the traditional modeling method only can model the physical environment of a water body, and effectively improves the accuracy of channel simulation. The method considers the integral influence of the response of the communication equipment on the system during modeling, so that the model has better reliability. The model is more suitable for the practical underwater application scene, the specific design and parameter selection of the underwater optical communication system can be more effectively guided by the invention, and the research progress of the underwater optical communication direction is accelerated.
Drawings
Fig. 1 is a diagram of an experimental device of an underwater laser communication system.
Fig. 2 is a schematic diagram of a conditional generation countermeasure network structure.
Fig. 3 is a schematic diagram of a generator network architecture.
Fig. 4 is a schematic diagram of a network structure of the discriminator.
FIG. 5 is an overall flow chart of the modeling method of the present invention.
Fig. 6 is a schematic diagram of a network prediction process.
Fig. 7 is a comparison graph showing simulation results of the model from a spectrum perspective.
Fig. 8 is a graph showing a comparison of simulation results of the model from the viewpoint of the bit error rate.
Detailed Description
The invention is further illustrated by the following specific examples.
The invention realizes an underwater laser digital communication system channel modeling and simulation method based on a generation countermeasure network, which mainly comprises the following four steps.
1. And (4) building an underwater laser digital communication system and manufacturing a data set.
And (3) setting up an underwater laser digital communication system, acquiring actually-measured receiving signals under various water quality conditions, establishing a data set by combining corresponding transmitting signals, and randomly dividing a training set and a testing set. The specific experimental device is shown in fig. 1, and the overall process is as follows:
s1, a pseudo-random binary sequence is generated by a PC (personal computer), converted into a non-return-to-zero code-on-off keying (NRZ-OOK) modulation signal as a sending signal and input into an Arbitrary Waveform Generator (AWG), and the AWG outputs an analog signal after completing the D/A conversion of the signal and directly drives a laser to carry out intensity modulation.
And S2, converting the electric signal into an optical signal by the laser, transmitting the optical signal to a receiving end after passing through the water body environment, converting the optical signal into the electric signal by the receiving end, amplifying the electric signal by the amplifier module, and inputting the electric signal into an oscilloscope for sampling to obtain a corresponding discrete signal.
And S3, performing off-line processing on the discrete signal in the S2 in a PC to obtain a synchronized receiving signal, collecting the synchronized signal as the receiving signal, and performing off-line demodulation processing on the synchronized signal and calculating an error rate for use in subsequent verification of the modeling effect of the model.
S4: a data set is made combining the received signal in S3 and the corresponding transmitted signal in S1.
The underwater laser communication system comprises a sending end and a receiving end. The transmitting end consists of AWG and a laser, and the output end of the AWG is connected to the analog signal input end of the laser through a BNC connecting line; the receiving end consists of a PMT, an amplifying module, an oscilloscope, a PC, a high-voltage power supply and a direct-current stabilized power supply, wherein the output end of the PMT is connected to the input end of the amplifying module through a BNC-to-SMA connecting wire, the output end of the amplifying module is connected to a BNC input interface of the oscilloscope through an SMA-to-BNC connecting wire, and the oscilloscope is communicated with the PC through a USB port; in addition, the PMT is powered by a high-voltage power supply, and the amplifying module is powered by a direct-current stabilized power supply.
2. Building CGAN model
The condition-building generation countermeasure network model (CGAN) comprises a generator model (G) and a discriminator model (D), and a schematic structural diagram of the CGAN used in the invention is shown in fig. 2.
The generation of the confrontation network is a generative network model for learning data distribution by means of confrontation. The original GAN comprises two networks, a Generator (Generator) and a Discriminator (Discriminator), and what is called a countermeasure is the mutual countermeasure between the Generator network and the Discriminator network. The generator can be regarded as a sample generator for simulating real samples, which takes a noise signal as an input and generates simulated samples after being processed by a generator network. The discriminator accepts as input a sample whose function is to determine whether the sample is from a true sample or a false sample generated by the generator, a larger value of the discriminator output indicating that the sample is more likely to belong to the set of true samples, and vice versa. The generator is targeted to increase the probability that the discriminator will discriminate the generated glitch as true, and the discriminator is targeted to increase the probability that the true signal will be discriminated as true and the glitch as false. The two can be alternately trained and continuously confronted to make progress together. With the increase of the confrontation times, the learning capacity of the generator and the identification capacity of the identifier are stronger and stronger, finally the identifier cannot distinguish true and false samples, the Nash equilibrium state is achieved, the network capacity cannot be further improved, the training is finished, and the generator is the required modeling network.
The difference between the CGAN and the original GAN is that condition information is added, and constraint is added to the original GAN, so that a generator can generate data under the guidance of the condition information, and the problem that the sample type generated by the original GAN is uncontrollable is solved. The discriminator in this case determines not only whether the input sample is true or false, but also whether the sample corresponds to correct condition information.
In the invention, a sending signal is used as condition information and is input into G along with Gaussian white noise, and simultaneously, the sending signal is input into D along with a real signal sample or a simulation signal sample generated by G. The objective functions of G and D in the conventional CGAN are expressed as formula (1) and formula (2), respectively.
In the above two formulae thetaGRepresenting the generator model parameters, θDRepresenting discriminator model parameters, y representing transmitted signal samples as condition information, x representing true received signal samples,g (z | y) represents the simulated signal sample, z = pz(z) noise samples satisfying a Gaussian distribution, pdataAnd pgRepresenting the true sample distribution and the sample distribution learned by the generator model, respectively. The goal of G is to minimize the probability that D identifies G (z | y) as false, while the goal of D is to minimize the probability that G (z | y) is identified as true, while maximizing the probability that the true sample x is identified as true, thereby forming a mechanism to combat training.
In order to solve the problem of instability of traditional GAN training, the invention introduces a Wasserstein GAN Gradient Penalty (WGAN-GP) algorithm to improve the training process, and the improved objective function is shown as formula (3) and formula (4).
In the above equation λ represents the gradient penalty coefficient,is shown in pdataAnd pgAnd uniformly sampling the two distributed samples to obtain a new sample distribution.
3. Setting of model parameters and training strategies
The third step mainly comprises the following steps: setting network parameters and training parameters, determining a training strategy, and importing a training set to alternately train a generator model and a discriminator model.
Specific network structures of G and D in the present invention are shown in fig. 3 and 4, respectively, and a one-dimensional convolutional Layer is adopted to replace a Multi-Layer Perceptron (MLP) in a conventional GAN to improve the capability of model feature extraction.
The generator G adopts a network structure of five-layer cascaded one-dimensional convolution layers, the dimensions of output spaces of all layers are respectively 256, 128, 64, 32 and 1, the lengths of one-dimensional convolution windows are respectively 5, 3 and 2, the first three layers of activation functions all adopt Tanh, the fourth layer of activation functions adopts ReLU, and the fifth layer directly outputs results in a linear mode.
The discriminator D adopts a six-layer network structure, wherein the first four layers are one-dimensional convolution layers, the second two layers are full connection layers, the dimensions of output spaces of each convolution layer are respectively 256, 128, 64 and 16, the lengths of one-dimensional convolution windows are respectively 5, 3 and 3, and the number of neurons in the fifth layer is 100. The first three layers of activation functions all adopt ReLU, the fourth layer is linearly output to the fifth layer, the fifth layer of activation functions adopts ReLU, and the sixth layer directly and linearly outputs results.
The analytical expressions of the Tanh activation function and the ReLU activation function are respectively as follows:
in the invention, a small-batch random gradient descent algorithm is adopted in the network model training process, the batch size is set to be 500, the initial learning rate is set to be 0.0001, and the gradient penalty coefficient in the formula (4) is set to be 5. FIG. 5 illustrates the training process of the model in detail. In each iteration, the training of G and D is alternated, and when one model is trained, the parameters of the other model are kept unchanged. G and D are both trained by using an Adam optimizer, D is trained firstly, and theta is optimized for six timesDOptimizing by thetaG. In a training batch of D, firstly obtaining a real signal sample and a generated signal sample, then calculating a true and false joint distribution sample, inputting the three samples and the transmitting signal into D together, respectively obtaining corresponding output values, then updating and recordingRecording the value of the loss function (4), and updating theta by using an Adam optimizer after the calculation of 500 batches is finishedD. In a training batch of G, firstly, acquiring a noise signal sample and a sending signal sample, inputting the noise signal sample and the sending signal sample into G together to obtain a simulation signal sample output by G, then inputting the sample into D to obtain an output value of D, then updating and recording the value of a loss function (3), and after the calculation of 500 batches is finished, updating theta by using an Adam optimizerG. The model can reach the convergence state when the iteration number (epoch) of the training set reaches about 1000 times.
The specific training process of the model is as follows:
s1, initializing model parameters thetaDAnd thetaG;
S2: in a training batch of D, firstly obtaining a sending signal sample y and a noise sample z-pz(z) calculating a simulated signal sample therefrom←G(z|y);
S3: obtaining a true signal sample x-pdata(x) And a random number delta-U0, 1]From which a true and false joint distribution sample is computed;
S4: inputting the three samples and the transmission signal into D respectively to obtain corresponding output values, and then updating and recording the loss functionA value of (d);
s5: repeating the steps S2-S4 for 500 times, and updating theta by using an Adam optimizerD;
S6: in a training batch of G, the step S2 is firstly repeated to obtain a simulation signal sampleInputting the sample into D to obtain the output value of D, then updating and recordingThe value of the loss function (1);
s7: repeating the step S6 for 500 times, and updating theta by using an Adam optimizerG;
S8: repeating the steps S5 and S7 until the training set is traversed once;
s9: and (5) repeatedly executing the step S8 for 1000 times or so to reach a convergence state, and finishing the training.
4. Testing of network models
Inputting the test set data into a generator model trained in the third step, and taking the absolute amplitude spectrum error (absolute amplitude spectrum mismatch) and the bit error rate error (BER mismatch) between the model simulation signal and the real received signal as the modeling effect of a standard test model, wherein the calculation method is shown in formula (5) and formula (6).
In the above formula, X andrespectively representing the real received signal and the corresponding model simulated signal.
The whole complete channel modeling method of the underwater laser digital communication system based on the generation countermeasure network is shown in figure 5.
Fig. 6 shows a network model prediction process, where a received signal approximately complying with the real channel distribution can be obtained by inputting white gaussian noise and a transmitted signal into the trained generator model, and at this time, the generator model realizes the function of an end-to-end underwater optical communication system channel simulator.
Taking a clear tap water channel as an example, a transmission signal with a symbol length of 500 is input into a model each time, fig. 7 shows a simulation result of the model from a spectrum perspective, fig. 7 (a) shows a true received signal waveform and a spectrum thereof (only the first 50 symbols of the signal are cut out in a signal waveform diagram for demonstration, the same applies below), fig. 7 (b) shows a simulated signal waveform and a spectrum thereof, fig. 7 (c) shows an absolute amplitude spectrum error between two signals, and an average error is 0.34dB. Taking two artificial turbid water channels as an example, continuously inputting a transmission signal with the code element length of 20000 into a model, calculating the error rate, measuring for multiple times and averaging, wherein the simulation result is shown in fig. 8, and the maximum value of the error rate is 0.5dB. In conclusion, the method of the invention has good modeling effect in both spectrum and bit error rate aspects.
The above description is only a preferred embodiment of the present application and is not intended to limit the present application, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, improvement and the like made within the spirit and principle of the present application shall be included in the protection scope of the present application.
Although the present invention has been described with reference to the specific embodiments, it should be understood by those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the invention.
Claims (6)
1. A channel modeling and simulation method for an underwater laser digital communication system is characterized by comprising the following steps:
step 1, constructing an underwater laser digital communication system, acquiring actually-measured receiving signals under various water quality conditions, establishing a data set by combining corresponding transmitting signals, and randomly dividing a training set and a testing set;
step 2, building conditions to generate a confrontation network model CGAN, wherein the CGAN model comprises a generator model G and a discriminator model D, the generator model G adopts a network structure of five-layer cascaded one-dimensional convolution layers, the dimension of each layer of output space is 256, 128, 64, 32 and 1, the length of a one-dimensional convolution window is 5, 3 and 2, the first three layers of activation functions all adopt Tanh, and the fourth layer of activation functions all adopt TanhThe layer activation function adopts ReLU, and the fifth layer directly and linearly outputs the result; the generator model G is a sample generator used for simulating a real sample, receives a noise signal and condition information as input, and generates a simulation sample after internal processing; the discriminator model D adopts a six-layer network structure, wherein the first four layers are one-dimensional convolution layers, the second two layers are full-connection layers, the dimensions of output spaces of each convolution layer are respectively 256, 128, 64 and 16, the lengths of one-dimensional convolution windows are respectively 5, 3 and 3, and the number of neurons in the fifth layer is 100; the first three layers of activation functions all adopt ReLU, the fourth layer is linearly output to the fifth layer, the fifth layer of activation functions adopts ReLU, and the sixth layer directly and linearly outputs results; the discriminator model D is used for receiving the samples and the condition information as input, judging whether the current samples are from real samples or false samples generated by the generator under the guidance of the condition information, and judging whether the current samples correspond to correct condition information; the condition information is a sending signal, and the noise signal is gaussian white noise; meanwhile, wasserstein GAN gradient penalty algorithm is introduced to improve the training process, and the improved objective functionAndas shown in equation (1) and equation (2):
in the above formula θGRepresenting the generator model parameters, θDRepresenting discriminator model parameters, y representing transmitted signal samples as condition information, x representing true received signal samples,g (z | y) denotes the simulated signal sample, z ~ pz(z) noise samples satisfying a Gaussian distribution, pdataAnd pgRespectively representing the true sample distribution and the sample distribution learned by the generator model, lambda represents a gradient penalty coefficient,is shown in pdataAnd pgUniformly sampling the two distributed samples to obtain new sample distribution; the goal of G is to minimize the probability that D identifies G (z | y) as false, the goal of D is to minimize the probability that G (z | y) is identified as true, while maximizing the probability that the true sample x is identified as true, G and D are trained alternately, thereby forming a mechanism for countertraining;
step 3, setting network parameters and training parameters of the CGAN model set up in the step 2, adopting a small batch random gradient descent algorithm, and importing the training set alternative training generator model G and the discriminator model D obtained in the step 1; in each iteration process, training of G and D is performed alternately, training of D is performed at first, when one model is trained, parameters of the other model are kept unchanged, and both G and D are trained by an Adam optimizer to update network parameters.
2. The method for modeling and simulating the channel of the underwater laser digital communication system according to claim 1, wherein the specific training process of the model is as follows:
s31: initializing model parameters θDAnd thetaG;
S32: in a training batch of D, a sending signal sample y and a noise sample z-p are obtained firstlyz(z) calculating a simulated signal sample therefrom←G(z|y);
S33: obtaining a true signal sample x-pdata(x) And a random number delta-U0, 1]From which a true and false joint distribution sample is computed;
S34: inputting the three samples and the sending signal into D to obtain corresponding output values, and updating and recording the loss functionA value of (d);
s35: repeating the steps S32-S34 for 500 times, and updating theta by using an Adam optimizerD;
S36: in a training batch of G, step S32 is repeated first to obtain a simulation signal sampleInputting the sample into D to obtain an output value of D, and then updating and recording the value of the loss function (1);
s37: repeating the step S36 500 times, and updating theta by using an Adam optimizerG;
S38: repeating the steps S35 and S37 until the training set is traversed once;
s39: and (6) repeatedly executing the S38 step 1000 times to obtain a convergence state, and finishing training.
3. The method for modeling and simulating the channel of the underwater laser digital communication system according to claim 1, wherein: the underwater laser digital communication system in the step 1 comprises a sending end and a receiving end; the transmitting end consists of an AWG and a laser, and the output end of the AWG is connected to the analog signal input end of the laser through a BNC connecting line; the receiving end consists of a PMT, an amplifying module, an oscilloscope, a PC, a high-voltage power supply and a direct-current stabilized power supply, wherein the output end of the PMT is connected to the input end of the amplifying module through a BNC-to-SMA connecting wire, the output end of the amplifying module is connected to a BNC input interface of the oscilloscope through an SMA-to-BNC connecting wire, and the oscilloscope is communicated with the PC through a USB port; the PMT is powered by a high-voltage power supply, and the amplification module is powered by a direct-current stabilized power supply.
4. The method for modeling and simulating the channel of the underwater laser digital communication system according to claim 1, wherein the specific process of establishing the data set in the step 1 is as follows:
s11, generating a pseudo-random binary sequence by a PC (personal computer), converting the pseudo-random binary sequence into a non-return-to-zero code-on-off keying modulation signal as a transmission signal and inputting the transmission signal into an arbitrary waveform generator, and outputting an analog signal to directly drive a laser to perform intensity modulation after the AWG finishes D/A (digital to analog) conversion of the signal;
s12, converting the electric signal into an optical signal by a laser, transmitting the optical signal to a receiving end after the optical signal passes through a water body environment, converting the optical signal into the electric signal by the receiving end, amplifying the electric signal by an amplifier module, and inputting the electric signal into an oscilloscope for sampling to obtain a corresponding discrete signal;
s13, performing off-line processing on the discrete signal in the S12 in a PC to obtain a synchronized receiving signal, collecting the synchronized signal as the receiving signal, and performing off-line demodulation processing on the synchronized signal and calculating an error rate for use in subsequent verification of the modeling effect of the model;
s14: a data set is made in conjunction with the received signal in S13 and the corresponding transmitted signal in S11.
5. The method for modeling and simulating the channel of the underwater laser digital communication system according to claim 1, wherein: the setting of the parameters in the step 3 comprises setting the batch size to be 500, setting the initial learning rate to be 0.0001 and setting the gradient penalty coefficient to be 5, wherein theta is set every time theta is optimizedGOptimizing thetaDIs set to 6.
6. The method for modeling and simulating the channel of the underwater laser digital communication system according to claim 1, further comprising a step 4 of testing the trained network model, specifically:
inputting the test set data obtained in the step 1 into the generator model trained in the step 3, and checking the modeling effect of the model by taking the absolute amplitude spectrum error and the bit error rate error between the model simulation signal and the real received signal as the standard, wherein the calculation method is shown in a formula (3) and a formula (4):
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210791128.9A CN114866158B (en) | 2022-07-07 | 2022-07-07 | Channel modeling and simulating method for underwater laser digital communication system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210791128.9A CN114866158B (en) | 2022-07-07 | 2022-07-07 | Channel modeling and simulating method for underwater laser digital communication system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114866158A CN114866158A (en) | 2022-08-05 |
CN114866158B true CN114866158B (en) | 2022-11-01 |
Family
ID=82626023
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210791128.9A Active CN114866158B (en) | 2022-07-07 | 2022-07-07 | Channel modeling and simulating method for underwater laser digital communication system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114866158B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116346217B (en) * | 2023-05-25 | 2023-08-08 | 北京理工大学 | Deep learning-based optical communication system channel construction method |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113541835A (en) * | 2021-06-17 | 2021-10-22 | 江苏科技大学 | Time-varying underwater acoustic channel simulation method for generating countermeasure network based on conditions |
Family Cites Families (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20200257985A1 (en) * | 2019-02-08 | 2020-08-13 | DeepSig Inc. | Adversarially generated communications |
CN110097185B (en) * | 2019-03-29 | 2021-03-23 | 北京大学 | Optimization model method based on generation of countermeasure network and application |
DE102019210270A1 (en) * | 2019-05-23 | 2020-11-26 | Robert Bosch Gmbh | Method for training a Generative Adversarial Network (GAN), Generative Adversarial Network, computer program, machine-readable storage medium and device |
CN110289927B (en) * | 2019-07-01 | 2021-06-15 | 上海大学 | Channel simulation realization method for generating countermeasure network based on condition |
US11989829B2 (en) * | 2019-09-12 | 2024-05-21 | Schlumberger Technology Corporation | Generating geological facies models with fidelity to the diversity and statistics of training images using improved generative adversarial networks |
CN113364528B (en) * | 2021-06-07 | 2022-05-03 | 中山大学 | Underwater wireless optical communication receiver for generating countermeasure network based on depth condition |
CN114362859B (en) * | 2021-12-28 | 2024-03-29 | 杭州电子科技大学 | Adaptive channel modeling method and system for enhanced condition generation countermeasure network |
CN114676637A (en) * | 2022-03-31 | 2022-06-28 | 北京邮电大学 | Fiber channel modeling method and system for generating countermeasure network based on conditions |
-
2022
- 2022-07-07 CN CN202210791128.9A patent/CN114866158B/en active Active
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113541835A (en) * | 2021-06-17 | 2021-10-22 | 江苏科技大学 | Time-varying underwater acoustic channel simulation method for generating countermeasure network based on conditions |
Also Published As
Publication number | Publication date |
---|---|
CN114866158A (en) | 2022-08-05 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110875790A (en) | Wireless channel modeling implementation method based on generation countermeasure network | |
CN110926355B (en) | Brillouin frequency shift extraction method and device based on convolutional neural network | |
Zhang et al. | Impulse response modeling for general underwater wireless optical MIMO links | |
CN114866158B (en) | Channel modeling and simulating method for underwater laser digital communication system | |
CN112949387B (en) | Intelligent anti-interference target detection method based on transfer learning | |
CN112468230B (en) | Wireless ultraviolet light scattering channel estimation method based on deep learning | |
CN113381828B (en) | Sparse code multiple access random channel modeling method based on condition generation countermeasure network | |
CN114362859B (en) | Adaptive channel modeling method and system for enhanced condition generation countermeasure network | |
CN111143984A (en) | Magnetotelluric two-dimensional inversion method based on genetic algorithm optimization neural network | |
Qadar et al. | Monte Carlo–based channel estimation and performance evaluation for UWOC links under geometric losses | |
CN109728824A (en) | A kind of LDPC code iterative decoding method based on deep learning | |
Dalgleish et al. | Physical layer simulator for undersea free-space laser communications | |
CN117056680A (en) | Data noise reduction and signal detection method, device and system and storage medium | |
Pachpande et al. | Visible light communication using deep learning techniques | |
Mahmoud et al. | Statistical studies using goodness-of-fit techniques with dynamic underwater visible light communication channel modeling | |
CN111859241B (en) | Unsupervised sound source orientation method based on sound transfer function learning | |
CN104734793B (en) | The energy detection method of wireless cooperation frequency spectrum perception based on p powers | |
CN117391463A (en) | River pollution tracing method | |
CN113640712A (en) | Prediction method for vertical component of vertical induced magnetic field of ship | |
CN115598714B (en) | Time-space coupling neural network-based ground penetrating radar electromagnetic wave impedance inversion method | |
CN103795436B (en) | Based on the robust multi-user test method of Quantum Hopfield Neural Network and quantum fish-swarm algorithm | |
CN115913424A (en) | Spectrum map accurate construction method for generating confrontation network based on conditions | |
CN116886174B (en) | Anti-interference efficiency evaluation method for satellite communication system | |
Yang et al. | Deep Reinforcement Learning For Secure Communication | |
CN115047408B (en) | Underwater multi-sound-source positioning method based on single-layer large convolution kernel neural network |
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 |