CN113411122A - Solar blind ultraviolet light communication self-adaptive signal detection method based on deep learning - Google Patents

Solar blind ultraviolet light communication self-adaptive signal detection method based on deep learning Download PDF

Info

Publication number
CN113411122A
CN113411122A CN202110500783.XA CN202110500783A CN113411122A CN 113411122 A CN113411122 A CN 113411122A CN 202110500783 A CN202110500783 A CN 202110500783A CN 113411122 A CN113411122 A CN 113411122A
Authority
CN
China
Prior art keywords
signal
ultraviolet light
signal detection
deep learning
frequency
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.)
Granted
Application number
CN202110500783.XA
Other languages
Chinese (zh)
Other versions
CN113411122B (en
Inventor
赵太飞
吕鑫喆
张海军
王玮
苏芊芊
张爽
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xian University of Technology
Original Assignee
Xian University of Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Xian University of Technology filed Critical Xian University of Technology
Priority to CN202110500783.XA priority Critical patent/CN113411122B/en
Publication of CN113411122A publication Critical patent/CN113411122A/en
Application granted granted Critical
Publication of CN113411122B publication Critical patent/CN113411122B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • H04B10/07—Arrangements for monitoring or testing transmission systems; Arrangements for fault measurement of transmission systems
    • H04B10/075—Arrangements for monitoring or testing transmission systems; Arrangements for fault measurement of transmission systems using an in-service signal
    • H04B10/079—Arrangements for monitoring or testing transmission systems; Arrangements for fault measurement of transmission systems using an in-service signal using measurements of the data signal
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • 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
    • H04B10/11—Arrangements specific to free-space transmission, i.e. transmission through air or vacuum
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B10/00—Transmission systems employing electromagnetic waves other than radio-waves, e.g. infrared, visible or ultraviolet light, or employing corpuscular radiation, e.g. quantum communication
    • H04B10/25—Arrangements specific to fibre transmission
    • H04B10/2507—Arrangements specific to fibre transmission for the reduction or elimination of distortion or dispersion
    • H04B10/2537—Arrangements specific to fibre transmission for the reduction or elimination of distortion or dispersion due to scattering processes, e.g. Raman or Brillouin scattering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching
    • 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
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Abstract

A solar blind ultraviolet light communication self-adaptive signal detection method based on deep learning comprises the following steps: step 1, solar blind ultraviolet light communication modeling simulation; step 2, processing signal time-frequency images; step 3, extracting characteristics of the ultraviolet light transmission signal; step 4, designing a signal detection model based on a convolutional neural network; the invention utilizes the autonomous learning capability and the excellent calculation capability of deep learning to solve the problems of sensitive threshold selection, weak adaptability to complex environment, need of a large amount of prior signal information and the like in the traditional ultraviolet signal detection method. The deep learning and the wireless ultraviolet light signal detection are combined, and the analysis is carried out from the frequency domain characteristic angle of the signal, so that the specific target signal can be accurately detected and identified, and the receiving and transmitting accuracy and reliability of the communication system are improved.

Description

Solar blind ultraviolet light communication self-adaptive signal detection method based on deep learning
Technical Field
The invention belongs to the technical field of wireless ultraviolet light communication, and particularly relates to a solar blind ultraviolet light communication signal detection method based on deep learning.
Background
The wireless ultraviolet communication is a novel wireless communication technology based on atmospheric particle scattering, and solar blind ultraviolet light with the wave band of 200-280 nm is used as an information transmission carrier, so that the wireless ultraviolet communication has the potential advantages of all weather, short distance, safety, interference resistance, flexible operation and the like. Through continuous exploration and research for more than two hundred years, the ultraviolet light communication technology can be successfully applied to special application scenes such as non-direct vision and the like.
When solar blind ultraviolet light signals are transmitted in the atmosphere, the solar blind ultraviolet light signals are influenced by turbulence while being subjected to atmospheric absorption and scattering. Turbulence effects cause random fluctuations in the velocity of motion, temperature, and refractive index of the atmosphere, both in time and space, where the fluctuations in refractive index directly affect the propagation characteristics of the optical signal. The light beam flicker caused by turbulence can cause background noise in the light signal, thereby reducing the receiving signal-to-noise ratio and causing the error rate of the ultraviolet light communication to increase. The light beam jitter phenomenon caused by the turbulence effect can make the light wave deviate from the receiving aperture, and the signal intensity is reduced. Scattering and absorption of atmospheric particles can cause obvious multipath effect to occur in the process of optical signal propagation, pulse broadening is caused to signals, and inter-code interference phenomenon can occur in serious cases, so that the error rate is increased, and even information misjudgment can be caused. The research on an effective signal processing method has important significance for correctly receiving ultraviolet light signals, the distortion of a non-line-of-sight ultraviolet light communication channel can be compensated through a channel estimation and equalization technology, the problem of intersymbol interference in a system is solved, and the signals need to be judged and processed after the more correct estimated values of the signals are obtained. The transmission signal is disturbed by background and noise when propagating in turbulent atmosphere, thereby affecting the correct decision. Therefore, it is very critical to design a practical and effective method for detecting solar blind ultraviolet light signals, and accurate reception of signals is of great significance to the application of solar blind ultraviolet light communication systems.
The traditional signal detection method is sensitive to selection of a decision threshold, and cannot be efficiently adapted to a complex scene of a solar blind ultraviolet communication system under turbulent flow. In recent years, deep learning has been attracting attention from various researchers as one of the most popular techniques in machine learning. The neural network model in deep learning is introduced into the signal processing process of ultraviolet light communication, so that the communication system has the capabilities of autonomous learning, autonomous decision making and autonomous updating, and can adapt to more complex communication environments and complete more unknown challenges.
Based on the research background, the deep learning is introduced into the signal detection of the solar blind ultraviolet communication system, and the transmission signal is accurately detected and identified by learning the deep signal characteristics from the frequency domain characteristics of the signal.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention aims to provide a signal detection method of a solar blind ultraviolet communication system, which utilizes a neural network model and an optimization algorithm in deep learning to solve the problems and the defects of the traditional method, including the problems of sensitive threshold selection, weak complex environment adaptability and need of prior signal information; the method has the characteristics of flexible threshold setting, self-adaptive detection under a time-varying channel and high detection accuracy.
In order to achieve the purpose, the invention adopts the technical scheme that: a solar blind ultraviolet light communication self-adaptive signal detection method based on deep learning comprises the following steps:
step 1, solar blind ultraviolet light communication modeling simulation:
comprehensively considering scattering effect and turbulence effect in the atmosphere, constructing an ultraviolet light transmission model in turbulent atmosphere by deducing a turbulence model formula of non-direct-view ultraviolet light communication and utilizing simulation software, and analyzing the influence of different parameters on single scattering irradiance in a random turbulent medium;
step 2, signal time-frequency image processing:
carrying out time-frequency analysis on the ultraviolet light transmission signal by using short-time Fourier transform to obtain a signal time-frequency diagram, and taking image data of the signal time-frequency diagram as the input of a neural network to provide a basis for the subsequent signal detection process of the convolutional neural network;
and 3, extracting characteristics of the ultraviolet light transmission signals:
extracting characteristic parameters for detection and identification through signal time-frequency processing in the step 2, compressing a high-dimensional data space into a low-dimensional data space through mathematical transformation, and extracting the characteristics of the existence of a target signal for accurately finishing a classification matching task in the follow-up process;
step 4, designing a signal detection model based on the convolutional neural network:
and (3) building a convolutional neural network model, inputting the processed signal time-frequency image, and outputting a two-classification model of whether a signal exists, wherein the neural network is used for monitoring a classification task in learning and mainly comprises an offline training process and an online testing process.
In the step 1, a non-direct-view single scattering channel is selected for the scattering effect to represent the scattering effect in solar blind ultraviolet communication, namely, photons emitted by an emitting end receive only one scattering in a common scattering body and then are transmitted to a receiving end, the non-direct-view communication mode selects an NLOS (c) mode, the elevation angles of the receiving end and the transmitting end are all indeterminate values and are less than 90 degrees, and the visual angle overlapping area of the communication mode is limited and can support larger bandwidth;
atmospheric scattering in sunny days is mainly considered, atmospheric molecules mainly generate Rayleigh scattering due to the low concentration of aerosol in the air, and the Rayleigh scattering coefficient is represented by the following formula:
Figure BDA0003056151460000031
wherein n (lambda) is the refractive index of the atmosphere,
Figure BDA0003056151460000032
NAis the concentration of atmospheric particles, and d (x) is the effective average diameter of the particles.
For the turbulent effect in step 1, for the most cases of weak turbulent intensity, a lognormal (log-normal) distribution is adopted to represent the probability distribution of light intensity flicker, the logarithmic fluctuation amplitude of the probability distribution is gaussian distribution, and the probability density function is:
Figure BDA0003056151460000041
in the formula, E [ X ]]And
Figure BDA0003056151460000042
representing the expectation and variance, respectively, of the random variable X, since the intensity I and the log-amplitude fluctuations are correlated, then:
I＝I0 exp(2X-E[X]) (3)
wherein E [ X ]]Is the mean value of X, I0Is the mean value of the light intensity, defined
Figure BDA0003056151460000043
And σ2 ln(I)The light intensity variance and the logarithmic intensity variance, respectively, can be derived from equation (3):
Figure BDA0003056151460000044
Figure BDA0003056151460000045
Figure BDA0003056151460000046
the light intensity flicker satisfies a log-normal distribution:
Figure BDA0003056151460000047
step 2, the short-time fourier transform is a mathematical transform, which converts a time domain signal collected by a receiving end of the communication system into frequency domain data, and uses a window function to perform sliding calculation on a time domain to obtain information parameters of the signal in the frequency domain, where the signal is set to be s (t), the window function is g (t), and a calculation formula of the short-time fourier transform (STFT) is:
Figure BDA0003056151460000048
for simple calculation, the signal is generally discretized, i.e. equation (8) can be changed to:
Figure BDA0003056151460000049
the STFT is characterized in that the parameters are set as follows: the sampling frequency is 25600Hz, a Hamming window is adopted as a window function, the length of the window function is 160ms, and the window function is visually researched by drawing a time-frequency graph of a signal, so that the subsequent depth feature extraction is facilitated.
Step 3, further depth feature extraction is carried out on the signal time-frequency diagram generated in step 2, and the calculation amount of parameters can be reduced to a great extent; and correct input parameters and matching basis are provided for the classification detection task of the neural network through feature extraction.
In the step 4, a Convolutional Neural Network (CNN) is selected as a Convolutional Neural network learning model, the Convolutional Neural network learning model is a Neural network with a special structure, the working principle of the Convolutional Neural network learning model can be regarded as a mode for processing visual information by the human brain Neural network, a signal time-frequency image after preprocessing and feature extraction is input, and a binary classification model for representing whether a signal exists is output.
And 4, the convolutional neural network learning model is of a 7-layer network structure and comprises two convolutional layers, two pooling layers, two full-connection layers and one softmax layer.
The invention has the beneficial effects that:
1) the solar blind ultraviolet communication signal detection method provided by the invention can be suitable for a complex communication scene of an ultraviolet scattering channel under turbulence, and the signal detection classification model is trained by using a deep learning method, so that the problem of sensitive threshold selection in the signal detection judgment process can be solved, the threshold value does not need to be changed along with different channel environments when the detection performance reaches a high level, and the method has a certain practical value.
2) Aiming at the problem of processing ultraviolet communication signals in random turbulence media, the traditional method faces the problems that a large amount of signal data often has high calculation complexity and low detection accuracy, introduces a neural network with autonomous learning capability into signal detection of solar blind ultraviolet communication, can overcome the problems and improve communication performance, utilizes iterative training of data to enable a network model to adapt to differentiated characteristics existing in random turbulence medium scene data, and ensures correct judgment of final signals.
3) The deep learning is mostly carried on the GPU for calculation, and has excellent calculation capability. Therefore, compared with the traditional signal detection algorithm, the signal detection method based on deep learning has better detection real-time performance, which is very important for the performance of the communication system.
Aiming at the scattering effect and the turbulence effect of ultraviolet light waves in the atmospheric transmission process, an ultraviolet light transmission model in turbulent atmosphere is constructed by using the characteristics of a fluctuating field; then, carrying out short-time Fourier transform on the signal of the communication receiving end to obtain a time-frequency diagram of the signal; and secondly, performing further characteristic extraction on the signal time-frequency diagram, and performing a series of image processing operations to provide effective input and matching basis for the subsequent training of the neural network. Finally, a convolutional neural network model is constructed, and two classification tasks for representing whether signals exist are completed through network testing and training processes, so that signal detection of solar blind ultraviolet communication is realized; the invention utilizes the autonomous learning capability and excellent calculation capability of deep learning, and solves the problems of sensitive threshold selection, weak adaptability to complex environment, need of a large amount of prior signal information and the like in the traditional ultraviolet signal detection method. The deep learning and the wireless ultraviolet light signal detection are combined, and the analysis is carried out from the frequency domain characteristic angle of the signal, so that the specific target signal can be accurately detected and identified, and the receiving and transmitting accuracy and reliability of the communication system are improved.
Drawings
Fig. 1 is an overall scheme diagram of a solar blind ultraviolet light communication signal detection method based on deep learning.
Fig. 2 is a block diagram of a wireless ultraviolet light communication system under atmospheric turbulence.
Fig. 3 is a non-line-of-sight solar-blind uv single-scattering channel geometry model.
Fig. 4 is a schematic view of a spherical coordinate system.
Fig. 5 is a diagram of a CNN detection model.
Detailed Description
The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.
The invention discloses a solar blind ultraviolet communication signal detection method based on deep learning, the overall scheme of which is shown in figure 1 and is implemented according to the following steps:
step 1, firstly, modeling simulation is carried out on a non-line-of-sight ultraviolet light turbulence model, a scattering effect and a turbulence effect which are suffered by an ultraviolet light signal when the ultraviolet light signal is transmitted in the atmosphere are comprehensively considered, and a solar blind ultraviolet light communication system under the atmosphere turbulence is constructed, as shown in fig. 2;
according to the geometric model of the non-direct-view solar blind ultraviolet single scattering channel in turbulent atmosphere shown in FIG. 3, the propagation and scattering problems of ultraviolet light waves in a random turbulent medium are researched, a non-direct-view single scattering transmission model in the medium is deduced, and a beta in FIG. 3TAnd betaRElevation angle, theta, of the transmitting end and the receiving end, respectivelyTAnd thetaRRespectively beam divergence angle and reception field angle, thetaSIs the scattering angle, V is the effective scatteringA projectile body, C is the coverage of a random turbulent flow medium, a volume infinitesimal delta V is taken in C, R is the distance from a transmitting end Tx to delta V, and a unit vector in the direction is taken
Figure BDA0003056151460000071
r is the distance from delta V to the receiving end Rx, and the unit vector in the direction is taken
Figure BDA0003056151460000072
d is the distance between the transmitting end and the receiving end, and part of the ultraviolet light signal transmitted by Tx is scattered due to the randomness of the medium and received by Rx, specifically:
step 1.1, firstly, the influence of a turbulence effect is not considered, the average received optical power is calculated according to the single scattering characteristic under the model, the total energy of a receiving end is calculated by using a infinitesimal method to calculate the volume fraction, and according to a scattering theory, the energy received by the volume infinitesimal delta V is as follows:
Figure BDA0003056151460000073
in the formula, ETAnd ERRepresenting transmitted and received energy, respectively, P (mu) being a scattering phase function, kSAnd keRespectively representing the atmospheric scattering coefficient and the absorption coefficient, ArIs the area of the receiving hole, omegaTAnd zeta is the included angle between the vector from Rx to deltaV of the receiving end and the axis of the receiving field angle. A spherical coordinate system is established as shown in fig. 4. For the convenience of analysis, assuming that the scatterer V is very small, cos (ζ) is approximated as a constant, and infinitesimal transformation is performed in a spherical coordinate system, and the total energy of the single scattering receiving end is approximated as:
Figure BDA0003056151460000081
after the above analysis of the energy of the receiving end, taking ζ as ct/d, calculating the channel impulse response under the model, and then performing approximation and simplification, wherein the expression is as follows:
Figure BDA0003056151460000082
step 1.2, further considering turbulence effect, calculating total scattering power detected by Rx according to the characteristics of atmospheric turbulence motion, and assuming that a point source is placed in a random turbulence medium C and the transmitting power is PTThrough analyzing loss attenuation on two links of non-direct-view communication and an equivalent scattering cross section capable of representing scattering power, the total power of the received signals is finally obtained as follows:
Figure BDA0003056151460000083
in the formula tau1Representing the attenuation, τ, of the ultraviolet light as it travels along the Tx to δ V path2Representing the attenuation of the ultraviolet light along the deltav to Rx path,
Figure BDA0003056151460000084
an equivalent scattering cross-section per unit volume expressed as:
Figure BDA0003056151460000085
in the formula L0In order to be the outer dimension of the turbulence,
Figure BDA0003056151460000086
expressing the intensity of turbulence, k-2 pi/lambda is the wave number, d0Is the relative length representing the average size of turbulent eddies;
step 2, performing time-frequency analysis on the time domain signal acquired by the receiving end of the communication system by using a short-time Fourier transform (STFT), specifically:
step 2.1, a time-frequency localized window function is set, the window function is subjected to sliding processing on a time domain signal, fast Fourier transform is carried out in each time window to obtain a power spectrum in the time window, then frequency spectrum information of the signal changing along with time is obtained through dynamic summation, the signal is set to be s (t), the window function is g (t), and a calculation formula of a short time Fourier transform method (STFT) is as follows:
Figure BDA0003056151460000091
for computational simplicity, the signal is typically discretized, i.e.:
Figure BDA0003056151460000092
step 2.2, acquiring the received signal in the communication system in the step 1, and drawing a time-frequency graph of the transmission signal after short-time Fourier transform by using simulation software, wherein the abscissa of the time-frequency graph is time, the unit is ns, the ordinate of the time-frequency graph is frequency, and the unit is Hz;
step 3, aiming at the signal time-frequency diagram in the step 2, the feature extraction processing of time-frequency data is required to be carried out on the signal time-frequency diagram, so that correct input parameters and matching bases are provided for the subsequent neural network detection and classification tasks, and the method specifically comprises the following steps:
step 3.1, performing graying processing on a signal time-frequency diagram, wherein the time-frequency diagram obtained by the signal through STFT is in a jpg or png format, the signal time-frequency diagram belongs to a three-channel RGB format, graying of the image is performed by using single-channel gray information to replace color information, the calculated amount during training can be reduced, preparation is made for upper-layer operations such as image compression, image identification and the like, an average value method is adopted for graying of the image, namely, three components in the color image are averaged to obtain a gray value, and the method is as follows:
Figure BDA0003056151460000093
and 3.2, image compression processing is carried out, the general size of an original signal time-frequency graph is large, a large amount of memory is occupied under the condition that a data set is large, the picture needs to be compressed into a format with a uniform size, the two-dimensional pixel array is converted into a data set which is irrelevant statistically, redundant data on the image is removed, the calculated amount of parameters is further reduced, and the network training efficiency is improved.
And 3.3, performing two-dimensional image filtering processing, wherein image compression in the step 3.2 may cause some important data of signals to be lost, so that data discontinuity is caused, even noise is introduced, so that image noise needs to be suppressed under the condition of keeping image detail characteristics as much as possible, effective image characteristics are extracted for image identification, a Gaussian filtering method is adopted for performing two-dimensional image filtering operation, and each original pixel point and other pixel values in a neighborhood are weighted and averaged to obtain a new pixel point. The zero-mean discrete gaussian filter function applied in the two-dimensional image is:
Figure BDA0003056151460000101
wherein x and y are pixel point coordinates, muxAnd muyThe coordinate mean values of the pixel points in the x and y directions respectively,
Figure BDA0003056151460000102
and
Figure BDA0003056151460000103
respectively representing the coordinate variances of the pixel points in the x direction and the y direction;
step 3.4, normalization processing is carried out, pixel values of the gray-scale picture after graying are between 0 and 255, normalization processing is needed for facilitating subsequent data processing calculation, each pixel value is divided by 255 and is between 0 and 1, and quantization loss caused by parameter range inconsistency among weights of different layers is avoided;
and 4, designing a convolutional neural network model, and detecting the solar blind ultraviolet communication signal. The core of this part is training and testing stage of Neural network, the superparameter of the preliminary setting network structure, utilize the continuous iterative optimization of sufficient training data, and then make the network model adapt to the channel environment more, obtain the mapping relation of the actual ultraviolet ray channel through the learning ability of network, thus realize the signal detection classification task, accomplish accurate judgement, (learning model chooses Convolutional Neural Network (CNN) for use, it is a Neural network with special structure, its theory of operation can be seen as the mode that human brain Neural network processes visual information, the input is the signal time frequency image after the preprocessing and characteristic are extracted, the output is the binary classification model whether the representation signal exists) specifically:
step 4.1, the CNN model constructed by the present solution is a 7-layer structure including a convolutional layer, a pooling layer, a full-link layer, and a softmax layer, as shown in fig. 5. Inputting grayscale picture data set to 64 × 1 by a network, wherein the size of a convolution kernel in a convolution layer is 5 × 16, and selecting a ReLU function by an activation function; the pooling size of the pooling layer is 2 x 2, the stepping size is 2, and the pooling mode is max-pooling; selecting a sigmoid function by an activation function in a full connection layer; the softmax layer maps the output features to classification results, and the results are divided into two types, namely target signals and non-target signals; since the model is used to accomplish the binary task, the loss function selects the cross entropy function as follows:
Figure BDA0003056151460000111
wherein m is the total number of output nodes, hθ(x(i)) Is the ith output of the CNN model, y(i)Is the target value corresponding to the training set;
step 4.2, adopting a back propagation algorithm to train the CNN, and initializing the value of each hidden layer parameter in the first step; secondly, transmitting an input signal along a network, applying different calculation methods when the input signal passes through different structural layers, and obtaining an output value by a forward propagation algorithm; thirdly, calculating the error gradient of the output layer by using the loss function; fourthly, the error is propagated reversely along the network, and the parameter of each layer is updated according to different structural layers by different formulas; and fifthly, repeating the first step to the fourth step until the training times reach a preset value or the error reaches a threshold value.
And 4.3, performing online detection on the time-frequency diagram by using the CNN detection model trained in the step 4.2, outputting whether a target signal exists or not, and finishing two classification tasks of whether the signal exists or not.

Claims (8)

1. A solar blind ultraviolet light communication self-adaptive signal detection method based on deep learning is characterized by comprising the following steps:
step 1, solar blind ultraviolet light communication modeling simulation:
comprehensively considering scattering effect and turbulence effect in the atmosphere, constructing an ultraviolet light transmission model in turbulent atmosphere by deducing a turbulence model formula of non-direct-view ultraviolet light communication and utilizing simulation software, and analyzing the influence of different parameters on single scattering irradiance in a random turbulent medium;
step 2, signal time-frequency image processing:
carrying out time-frequency analysis on the ultraviolet light transmission signal by using short-time Fourier transform to obtain a signal time-frequency diagram, and taking image data of the signal time-frequency diagram as the input of a neural network to provide a basis for the subsequent signal detection process of the convolutional neural network;
and 3, extracting characteristics of the ultraviolet light transmission signals:
extracting characteristic parameters for detection and identification through signal time-frequency processing in the step 2, compressing a high-dimensional data space into a low-dimensional data space through mathematical transformation, and extracting the characteristics of the existence of a target signal for accurately finishing a classification matching task in the follow-up process;
step 4, designing a signal detection model based on the convolutional neural network:
and (3) building a convolutional neural network learning model, inputting the processed signal time-frequency image, and outputting a two-classification model of whether a signal exists, wherein the neural network is used for monitoring a classification task in learning and mainly comprises an offline training process and an online testing process.
2. The solar-blind ultraviolet communication adaptive signal detection method based on deep learning of claim 1, characterized in that, in step 1, a scattering effect in the solar-blind ultraviolet communication is characterized by selecting a non-direct-view single-scattering channel for the scattering effect, that is, photons emitted from an emitting end are transmitted to a receiving end after receiving only one scattering in a common scatterer, the non-direct-view communication mode selects an nlos (non line of sight) (c) mode, elevation angles of the transmitting and receiving ends are all indeterminate and less than 90 degrees, and a view angle overlapping region of the communication mode is limited and can support a larger bandwidth;
atmospheric scattering in sunny days is mainly considered, atmospheric molecules mainly generate Rayleigh scattering due to the low concentration of aerosol in the air, and the Rayleigh scattering coefficient is represented by the following formula:
Figure FDA0003056151450000021
wherein n (lambda) is the refractive index of the atmosphere,
Figure FDA0003056151450000022
NAis the concentration of atmospheric particles, and d (x) is the effective average diameter of the particles.
3. The solar blind ultraviolet light communication adaptive signal detection method based on deep learning according to claim 1, wherein the turbulent effect of step 1 adopts a lognormal distribution to represent probability distribution of light intensity flicker under the condition that most turbulent intensities are weak, the logarithmic fluctuation amplitude of the probability distribution is gaussian distribution, and the probability density function is as follows:
Figure FDA0003056151450000023
in the formula, E [ X ]]And
Figure FDA0003056151450000024
representing the expectation and variance, respectively, of the random variable X, since the intensity I and the log-amplitude fluctuations are correlated, then:
I＝I0exp(2X-E[X]) (3)
wherein E [ X ]]Is the mean value of X, I0Is the mean value of the light intensity, defined
Figure FDA0003056151450000025
And σ2 ln(I)The light intensity variance and the logarithmic intensity variance, respectively, can be derived from equation (3):
Figure FDA0003056151450000026
Figure FDA0003056151450000027
Figure FDA0003056151450000028
the light intensity flicker satisfies a log-normal distribution:
Figure FDA0003056151450000031
4. the deep learning-based solar-blind ultraviolet light communication adaptive signal detection method according to claim 1, wherein in step 2, the short-time fourier transform is a mathematical transform, which converts time-domain signals collected by a receiving end of a communication system into frequency-domain data, and performs sliding calculation on a time domain by using a window function to obtain information parameters of the signals in the frequency domain, where the signal is set to be s (t), and the window function is g (t), then the short-time fourier transform is calculated according to a formula:
Figure FDA0003056151450000032
for simple calculation, the signal is generally discretized, i.e. equation (8) can be changed to:
Figure FDA0003056151450000033
5. the deep learning-based solar-blind ultraviolet light communication adaptive signal detection method according to claim 4, wherein the short-time Fourier transform is parameterized as: the sampling frequency is 25600Hz, a Hamming window is adopted as a window function, the length of the window function is 160ms, and the window function is visually researched by drawing a time-frequency graph of a signal, so that the subsequent depth feature extraction is facilitated.
6. The deep learning-based solar-blind ultraviolet light communication adaptive signal detection method according to claim 1, characterized in that, in the step 3, further depth feature extraction is performed on the signal time-frequency diagram generated in the step 2; and correct input parameters and matching basis are provided for the classification detection task of the neural network through feature extraction.
7. The solar-blind ultraviolet communication adaptive signal detection method based on deep learning of claim 1, wherein in the step 4, the convolutional neural network learning model is a convolutional neural network, the input of which is a signal time-frequency image after preprocessing and feature extraction, and the output of which is a binary classification model for representing whether a signal exists.
8. The deep learning-based solar-blind ultraviolet light communication adaptive signal detection method according to claim 1, wherein in the step 4, the convolutional neural network learning model is a 7-layer network structure comprising two convolutional layers, two pooling layers, two full-link layers and a softmax layer.
CN202110500783.XA 2021-05-08 2021-05-08 Solar blind ultraviolet light communication self-adaptive signal detection method based on deep learning Active CN113411122B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110500783.XA CN113411122B (en) 2021-05-08 2021-05-08 Solar blind ultraviolet light communication self-adaptive signal detection method based on deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110500783.XA CN113411122B (en) 2021-05-08 2021-05-08 Solar blind ultraviolet light communication self-adaptive signal detection method based on deep learning

Publications (2)

Publication Number Publication Date
CN113411122A true CN113411122A (en) 2021-09-17
CN113411122B CN113411122B (en) 2022-09-06

Family

ID=77678283

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110500783.XA Active CN113411122B (en) 2021-05-08 2021-05-08 Solar blind ultraviolet light communication self-adaptive signal detection method based on deep learning

Country Status (1)

Country Link
CN (1) CN113411122B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114696933A (en) * 2022-04-01 2022-07-01 西安交通大学 AI receiver based on deep learning technology and use method

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108052956A (en) * 2017-11-07 2018-05-18 西安理工大学 Wireless light communication subcarrier modulation constellation recognition methods under a kind of atmospheric turbulance
CN108446631A (en) * 2018-03-20 2018-08-24 北京邮电大学 The smart frequency spectrum figure analysis method of deep learning based on convolutional neural networks
US20180262291A1 (en) * 2017-03-07 2018-09-13 Government Of The United States Of America, As Represented By The Secretary Of The Navy Method for free space optical communication utilizing patterned light and convolutional neural networks
CN109031654A (en) * 2018-09-11 2018-12-18 安徽农业大学 A kind of adaptive optics bearing calibration and system based on convolutional neural networks
CN111064512A (en) * 2019-12-06 2020-04-24 中山大学 Deep learning-based optical orthogonal frequency division multiplexing modulation method and system
CN111786724A (en) * 2020-06-30 2020-10-16 中山大学 Multi-wavelength LED underwater visible light communication modulation method based on deep learning
CN112468230A (en) * 2020-11-19 2021-03-09 西安理工大学 Wireless ultraviolet light scattering channel estimation method based on deep learning

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180262291A1 (en) * 2017-03-07 2018-09-13 Government Of The United States Of America, As Represented By The Secretary Of The Navy Method for free space optical communication utilizing patterned light and convolutional neural networks
CN108052956A (en) * 2017-11-07 2018-05-18 西安理工大学 Wireless light communication subcarrier modulation constellation recognition methods under a kind of atmospheric turbulance
CN108446631A (en) * 2018-03-20 2018-08-24 北京邮电大学 The smart frequency spectrum figure analysis method of deep learning based on convolutional neural networks
CN109031654A (en) * 2018-09-11 2018-12-18 安徽农业大学 A kind of adaptive optics bearing calibration and system based on convolutional neural networks
CN111064512A (en) * 2019-12-06 2020-04-24 中山大学 Deep learning-based optical orthogonal frequency division multiplexing modulation method and system
CN111786724A (en) * 2020-06-30 2020-10-16 中山大学 Multi-wavelength LED underwater visible light communication modulation method based on deep learning
CN112468230A (en) * 2020-11-19 2021-03-09 西安理工大学 Wireless ultraviolet light scattering channel estimation method based on deep learning

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
牟迪等: "基于条件GAN的无人机智能光通信", 《激光与红外》 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114696933A (en) * 2022-04-01 2022-07-01 西安交通大学 AI receiver based on deep learning technology and use method
CN114696933B (en) * 2022-04-01 2023-02-07 西安交通大学 AI receiver based on deep learning technology and use method

Also Published As

Publication number Publication date
CN113411122B (en) 2022-09-06

Similar Documents

Publication Publication Date Title
CN109993280B (en) Underwater sound source positioning method based on deep learning
CN112468230B (en) Wireless ultraviolet light scattering channel estimation method based on deep learning
WO2019232846A1 (en) Speech differentiation method and apparatus, and computer device and storage medium
Jiang et al. Multi-scale spectral feature extraction for underwater acoustic target recognition
CN111723701B (en) Underwater target identification method
Huynh-The et al. Accurate LPI radar waveform recognition with CWD-TFA for deep convolutional network
CN105760825A (en) Gesture identification system and method based on Chebyshev feed forward neural network
CN113376600B (en) Pedestrian radar echo denoising method based on RSDNet
CN113296087B (en) Frequency modulation continuous wave radar human body action recognition method based on data enhancement
CN114595732B (en) Radar radiation source sorting method based on depth clustering
CN113411122B (en) Solar blind ultraviolet light communication self-adaptive signal detection method based on deep learning
Li A new robust signal recognition approach based on holder cloud features under varying SNR environment
Zhao et al. Radar HRRP target recognition via semi-supervised multi-task deep network
Hu et al. Signal denoising based on wavelet threshold denoising and optimized variational mode decomposition
CN114117912A (en) Sea clutter modeling and inhibiting method under data model dual drive
CN115797772A (en) Classified early warning method and system for icing thickness of overhead transmission line
CN105656577B (en) Towards the cluster-dividing method and device of channel impulse response
CN114428234A (en) Radar high-resolution range profile noise reduction identification method based on GAN and self-attention
CN111859241B (en) Unsupervised sound source orientation method based on sound transfer function learning
Guo et al. Modulation Signal Classification Algorithm Based on Denoising Residual Convolutional Neural Network
CN115426055B (en) Noise-containing underwater acoustic signal blind source separation method based on decoupling convolutional neural network
CN112801065B (en) Space-time multi-feature information-based passive sonar target detection method and device
CN114580468A (en) Interference signal identification method based on time-frequency waterfall graph and convolutional neural network
Wei et al. A multi-resolution channel structure learning estimation method of geometry-based stochastic model with multi-scene
CN113723253B (en) Clutter pre-classification-based moving target detection method

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