CN110138591B - Method for acquiring dynamic spectrum access frequency point and dynamic spectrum access method - Google Patents

Method for acquiring dynamic spectrum access frequency point and dynamic spectrum access method Download PDF

Info

Publication number
CN110138591B
CN110138591B CN201910284056.7A CN201910284056A CN110138591B CN 110138591 B CN110138591 B CN 110138591B CN 201910284056 A CN201910284056 A CN 201910284056A CN 110138591 B CN110138591 B CN 110138591B
Authority
CN
China
Prior art keywords
graph
sample label
spectrum
sample
adjusted
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
Application number
CN201910284056.7A
Other languages
Chinese (zh)
Other versions
CN110138591A (en
Inventor
李大朋
江民民
邱昕
柴旭荣
徐波
孙志浩
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.)
Institute of Microelectronics of CAS
Original Assignee
Institute of Microelectronics of CAS
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 Institute of Microelectronics of CAS filed Critical Institute of Microelectronics of CAS
Priority to CN201910284056.7A priority Critical patent/CN110138591B/en
Publication of CN110138591A publication Critical patent/CN110138591A/en
Application granted granted Critical
Publication of CN110138591B publication Critical patent/CN110138591B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/08Configuration management of networks or network elements
    • H04L41/0893Assignment of logical groups to network elements
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/14Network analysis or design
    • H04L41/145Network analysis or design involving simulating, designing, planning or modelling of a network

Abstract

The invention discloses a method for acquiring dynamic spectrum access frequency points and a dynamic spectrum access method, which belong to the technical field of communication and comprise the following steps: superposing a primary user sample annotation graph and an initial secondary user sample annotation graph in a target environment to synthesize a comprehensive sample annotation graph; inputting the comprehensive sample label graph as a condition to a trained condition confrontation generation network model to generate a first simulation spectrum waterfall graph set of the target environment; inputting the first simulation spectrum waterfall graph set and the comprehensive sample label graph into a decision network to obtain a frequency band decision result; and adjusting the comprehensive sample label graph according to the frequency band decision result, circularly executing until the adjusted comprehensive sample label graph meets a preset condition, outputting the adjusted comprehensive sample label graph, and performing dynamic spectrum access. The training efficiency of the decision network is greatly improved, and the practicability of dynamic spectrum access decision network is enhanced.

Description

Method for acquiring dynamic spectrum access frequency point and dynamic spectrum access method
Technical Field
The invention relates to the technical field of communication, in particular to a method for acquiring a dynamic spectrum access frequency point and a dynamic spectrum access method.
Background
At present, the fierce artificial intelligence technology has deepened into the aspect of the information field, and the cognitive radio field is no exception. The problems to be solved by cognitive radio are as follows: the radio device automatically adjusts to the optimal working state through the sensed surrounding environment information, belongs to the optimal decision problem in the automatic control field, and can exactly solve the problem by the characteristics of intellectualization and self-adaption of an artificial intelligent decision network.
However, in recent years, some scholars and research institutions have proposed methods for performing intelligent spectrum dynamic access by using artificial intelligence decision networks, but the prior art has a problem of low efficiency in acquiring dynamic spectrum access frequency points.
Disclosure of Invention
The embodiment of the application solves the problem of low efficiency of acquiring the dynamic spectrum access frequency point in the related technology by providing a dynamic spectrum access decision network training method and a dynamic spectrum access method.
In a first aspect, the present application provides the following technical solutions through an embodiment of the present application:
a dynamic spectrum access decision network training method comprises the following steps:
superposing a primary user sample annotation graph and an initial secondary user sample annotation graph in a target environment to synthesize a comprehensive sample annotation graph;
inputting the comprehensive sample label graph as a condition to a trained condition confrontation generation network model to generate a first simulation spectrum waterfall graph set of the target environment;
inputting the first simulation spectrum waterfall graph set and the comprehensive sample label graph into a decision network to obtain a frequency band decision result;
adjusting the secondary user sample label graph according to the frequency band decision result, and obtaining an adjusted comprehensive sample label graph according to the adjusted secondary user sample label graph;
judging whether the adjusted comprehensive sample annotation graph meets a preset condition or not;
if not, circularly executing the steps until the adjusted comprehensive sample annotation graph meets the preset condition; and outputting the adjusted comprehensive sample label graph when the adjusted comprehensive sample label graph meets the preset condition.
Optionally, the training method for generating the network model by conditional countermeasure specifically includes:
randomly obtaining a diversified spectrum waterfall graph and a sample label graph corresponding to the spectrum waterfall graph;
and training to obtain a conditional countermeasure generation network model by taking the frequency spectrum waterfall chart and a sample labeled graph corresponding to the frequency spectrum waterfall chart as training samples, wherein the sample labeled graph is a condition.
Optionally, the method for randomly acquiring a diversified spectrum waterfall graph includes:
manually intervening a transmitter of the primary user to generate diversified spectrum information;
and collecting the frequency spectrum information in real time and forming a frequency spectrum waterfall diagram.
Optionally, the adjusted integrated sample annotation graph includes: and whether the frequency bands of the secondary user and the primary user collide or not.
Optionally, the preset conditions include: and within the preset continuous cycle times, the frequency bands of the secondary user and the primary user in the adjusted comprehensive sample label graph do not collide.
Optionally, the sample label graph corresponding to the spectrum waterfall graph includes at least one of a current working center frequency point, a bandwidth, and spectrum background information.
Optionally, the sample label graph of the primary user and the sample label graph of the secondary user in the target environment are prior information artificially compiled according to the target environment.
In a second aspect, the present application provides the following technical solutions according to an embodiment of the present application:
a system for acquiring dynamic spectrum access frequency points comprises:
the synthesis module is used for superposing a master user sample annotation graph and an initial secondary user sample annotation graph in the target environment to synthesize a comprehensive sample annotation graph;
the simulation module is used for inputting the comprehensive sample annotation graph as a condition to a trained condition countermeasure generation network model to generate a first simulation spectrum waterfall diagram set of the target environment;
the training module is used for inputting the first simulation spectrum waterfall pattern set and the comprehensive sample label graph into a decision network to obtain a frequency band decision result;
the decision control module is used for adjusting the secondary user sample label graph according to the frequency band decision result and obtaining an adjusted comprehensive sample label graph according to the adjusted secondary user sample label graph;
the judging module is used for judging whether the adjusted comprehensive sample annotation graph meets the preset condition or not; if not, circularly executing the steps until the adjusted comprehensive sample annotation graph meets the preset condition;
and the output module is used for outputting the adjusted comprehensive sample label graph when the adjusted comprehensive sample label graph meets the preset condition.
In a third aspect, the present application provides the following technical solutions through an embodiment of the present application:
a dynamic spectrum access method, comprising:
the dynamic spectrum access frequency points obtained according to any one of claims 1-8 are used for dynamic spectrum access.
In a fourth aspect, the present application provides the following technical solutions according to an embodiment of the present application:
a dynamic spectrum access system, comprising:
an access module, configured to perform dynamic spectrum access according to the dynamic spectrum access frequency point obtained by the method according to any one of claims 1 to 8.
One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:
the invention utilizes a sample annotation graph of a primary user and a sample annotation graph of a secondary user in a target environment to superpose and synthesize a comprehensive sample annotation graph as a condition, inputs a trained condition countermeasure generation network model, can generate a first simulation spectrum waterfall graph set containing a large amount of spectrum waterfall graphs required by the target environment in an off-line manner, because the condition countermeasure generation network model is trained, namely off-line, for the condition countermeasure generation network model, as long as a priori input is given, corresponding output can be generated, in order to simulate the spectrum information of the target environment, a small amount of sample annotation graphs of the primary user and sample annotation graphs of the secondary users in the target environment are input as priori information, a large amount of vivid simulation spectrum waterfall graphs required by the target environment can be obtained to be used as the input of decision network training, thereby replacing the on-line training method that needs to wait for obtaining the spectrum waterfall graphs from a primary user transmitter, moreover, as the trained offline conditions are utilized to resist and generate the network model, waiting is not needed, and compared with the prior art, the speed of acquiring the target environment frequency spectrum information is greatly improved; then, inputting the first simulation spectrum waterfall graph set and the comprehensive sample label graph into a decision network to obtain a frequency band decision result; adjusting the secondary user sample label graph according to the frequency band decision result, and obtaining an adjusted comprehensive sample label graph according to the adjusted secondary user sample label graph; judging whether the adjusted comprehensive sample annotation graph meets a preset condition or not; if not, circularly executing the steps until the adjusted comprehensive sample annotation graph meets the preset condition; and when the adjusted comprehensive sample label graph meets the preset condition, outputting the adjusted comprehensive sample label graph, wherein the output adjusted comprehensive sample label graph comprises dynamic spectrum access frequency point information. The input of each circulation of the process of obtaining the decision network model by the circulation training is generated by the off-line trained conditional countermeasure generation network model, and the off-line trained conditional countermeasure generation network model can quickly generate a sufficient amount of vivid frequency spectrum waterfall patterns of a target environment in an off-line mode without waiting, so that the input obtained by controlling the training process of the decision network after the circulation feedback decision of each time is not dependent on obtaining a training sample from an on-line main user transmitter, therefore, the playing speed limit of the transmitting mode of the main user transmitter is avoided, the decision network training efficiency is greatly improved, and the problems that the efficiency of obtaining dynamic frequency spectrum access frequency points is low and the practical applicability is not strong due to the fact that the time is consumed for obtaining the training sample in the prior art are solved.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on the drawings without creative efforts.
Fig. 1 is a flowchart of a method for acquiring a dynamic spectrum access frequency point according to an embodiment of the present invention;
fig. 2 is a flowchart of a specific implementation of a method for acquiring a dynamic spectrum access frequency point according to an embodiment of the present invention;
FIG. 3 is a flow chart of a dynamic spectrum access decision network training method of the prior art;
fig. 4 is a flowchart of generating a network model by training an acquisition conditional countermeasure in a method for acquiring a dynamic spectrum access frequency point according to an embodiment of the present invention;
fig. 5 is a system diagram for acquiring a dynamic spectrum access frequency point according to an embodiment of the present invention.
Detailed Description
The embodiment of the application provides a dynamic spectrum access decision network training method, and solves the technical problem that the efficiency of acquiring dynamic spectrum access frequency points is low in the prior art.
In order to solve the technical problems, the general idea of the embodiment of the application is as follows:
superposing a primary user sample annotation graph and an initial secondary user sample annotation graph in a target environment to synthesize a comprehensive sample annotation graph;
inputting the comprehensive sample label graph as a condition to a trained condition confrontation generation network model to generate a first simulation spectrum waterfall graph set of the target environment;
inputting the first simulation spectrum waterfall graph set and the comprehensive sample label graph into a decision network to obtain a frequency band decision result;
adjusting the secondary user sample label graph according to the frequency band decision result, and obtaining an adjusted comprehensive sample label graph according to the adjusted secondary user sample label graph;
judging whether the adjusted comprehensive sample annotation graph meets a preset condition or not;
if not, circularly executing the steps until the adjusted comprehensive sample annotation graph meets the preset condition; and outputting the adjusted comprehensive sample label graph when the adjusted comprehensive sample label graph meets the preset condition.
In order to better understand the technical solution, the technical solution will be described in detail with reference to the drawings and the specific embodiments.
First, it is stated that the term "and/or" appearing herein is merely one type of associative relationship that describes an associated object, meaning that three types of relationships may exist, e.g., a and/or B may mean: a exists alone, A and B exist simultaneously, and B exists alone. In addition, the character "/" herein generally indicates that the former and latter related objects are in an "or" relationship.
Example one
The present embodiment provides a dynamic spectrum access decision network training method, and fig. 1 is a flowchart of the present embodiment. The method specifically comprises the following steps:
s101, superposing a primary user sample annotation graph and an initial secondary user sample annotation graph in a target environment to synthesize a comprehensive sample annotation graph;
s102, inputting the comprehensive sample label graph serving as a condition into a trained condition confrontation generation network model, and generating a first simulation spectrum waterfall graph set of the target environment;
s103, inputting the first simulation spectrum waterfall graph set and the comprehensive sample label graph into a decision network to obtain a frequency band decision result;
s104, adjusting the secondary user sample label graph according to the frequency band decision result, and obtaining an adjusted comprehensive sample label graph according to the adjusted secondary user sample label graph:
s105, judging whether the adjusted comprehensive sample annotation graph meets a preset condition; if not, circularly executing the steps until the adjusted comprehensive sample annotation graph meets the preset condition; and outputting the adjusted comprehensive sample label graph when the adjusted comprehensive sample label graph meets the preset condition.
To facilitate understanding of the invention, a brief description is made of a conventional dynamic spectrum access decision network training method, for example, fig. 3 is a flowchart of the conventional dynamic spectrum access decision network training method, and in the diagram, a transmission mode is used to control a transmission behavior of a transmitter, for example: and (3) performing actions such as frequency sweeping, random emission and the like, wherein parameters control the working state of the transmitter, such as: center frequency point, bandwidth, gain, etc.; the PU transmitter, also called primary user transmitter, operates according to the above-mentioned transmission mode and parameters, and simultaneously records the spectrum related information for each transmission time, for example: current working center frequency point, bandwidth, etc. The recorded information is used as a basis for judging the quality of a decision generated by the decision network, and can be regarded as a reference Label (Label) in a training sample set; the receiver is adjusted to the working frequency range of PU (master user) and SU (sub user) in advance, spectrum information is acquired on line in real time and stored as a spectrum waterfall graph, when a decision network trains on line, a decision is generated at intervals, the decision controls the working central frequency point of the SU transmitter, and the recorded transmitted spectrum marking information is combined to judge whether the current decision is good or bad, for example: and if the current SU transmitter working frequency band and the PU transmitter transmitting frequency band are overlapped and collided, the current decision is not good.
The prior art method is characterized in that: the sample acquisition rate of the on-line training is completely limited by the play rate of the PU transmitter transmission mode, for example: one sweep period of the swept mode is one mode sample. Because a large number of independent pattern samples are needed for neural network training, if a sufficient number of pattern training samples are to be acquired in an online manner, a large amount of time is required.
As can be seen from the above, the current intelligent spectrum dynamic access method is far from practical use because the following fatal problems exist in the training of the decision network:
low training efficiency and low speed: dynamic spectrum access belongs to a dynamic optimal decision problem, and different from a classification and regression problem, each step of training of the dynamic spectrum access requires environmental feedback. Thus, each step of such network training is completely dependent on the environmental feedback time after each decision. Taking the repetition period of the electromagnetic environment spectrum activity mode as 10s (usually much more than 10s), if 5000 independent mode samples are needed for model training (actually, the number of the needed samples is much more than 5000), the sample acquisition will take about 13.9 hours. Assuming sufficient computing power and a model training time of 0, the model generation time is 13.9+0 hours. Such an hour-scale model generation time is not acceptable in practical applications.
The following describes in detail specific implementation steps of the method provided by the embodiment of the present application with reference to fig. 1 to 3:
firstly, executing a step S101, and superposing a sample annotation graph of a primary user and a sample annotation graph of a secondary user in a target environment to synthesize a comprehensive sample annotation graph;
optionally, the sample label graph of the primary user and the sample label graph of the secondary user in the target environment are prior information artificially compiled according to the target environment.
Specifically, in this embodiment, as shown in fig. 2, the sample label graph of the primary user and the sample label graph of the secondary user in the target environment are manually pre-programmed according to the target environment through the PU mode generator and the SU mode generator, and may be manually and rapidly manufactured, or manufactured by collecting a small amount of current environment frequency spectrum, or a combination of the two; and superposing the sample label graph of the primary user and the sample label graph of the secondary user to form a sample label graph, namely a comprehensive sample label graph.
It can be understood that the PU mode generator may simulate the PU transmitter to generate the spectrum information corresponding to the target environment, and the SU mode generator simulates the spectrum information generated by the secondary user transmitter pre-accessing the frequency point. Since there is no spectrum basis for the secondary user initially, a sample annotation map for one secondary user can be initialized randomly.
Next, step S102 is executed, and the comprehensive sample label graph is used as a condition, and is input to a trained conditional countermeasure generation network model to generate a first simulated spectrum waterfall graph set of the target environment;
specifically, the comprehensive sample label graph obtained in step S101 is input into a trained conditional countermeasure generation network model to generate a realistic target environment simulation spectrum waterfall graph. Theoretically, as long as an input meeting the requirement is given, the mode can quickly generate infinite target environment frequency spectrum waterfall diagrams, as the conditional countermeasure generation network model is trained, namely offline, for the conditional countermeasure generation network model, corresponding output can be generated as long as a priori input is given, here, in order to simulate the frequency spectrum information of the target environment, a small number of sample labeled diagrams of a primary user and a secondary user in the target environment are taken as the priori information input, a large number of vivid simulated frequency spectrum waterfall diagrams needed by the target environment can be obtained and taken as the input of decision network training, so that the situation that the online training method needs to wait for obtaining the frequency spectrum waterfall diagrams from a primary user transmitter is replaced, and as the trained offline conditional countermeasure generation network model is utilized, waiting is not needed, compared with the prior art, the speed for obtaining the frequency spectrum information of the target environment is greatly improved, hardly influenced by the play rate of the transmitter transmission mode of the current environment (the manual drawing of the mode condition chart of the current environment takes a certain time, but is far shorter than the time required by online training).
More specifically, the trained conditional countermeasure generation network model is obtained by training the conditional countermeasure generation network with the sample label graph as the condition, and can quickly generate a large number of models simulating the spectrum waterfall graph when the prior information is input, similar to the spectrum environment simulation generator.
Next, step S103 is executed, the first simulated spectrum waterfall plot set and the comprehensive sample label graph are input to a decision network, and a frequency band decision result is obtained;
specifically, in this embodiment, the decision network is a decision neural network known in the art, and performs network training according to the simulated spectrum waterfall pattern and the current decision result, and generates a frequency band decision at intervals.
More specifically, the frequency band decision result may be a parameter control decision generated after the training and used for controlling the secondary user transmitter to transmit a sample label graph, or may be directly a frequency point information, as long as the input secondary user sample label graph can be updated.
Next, executing step S104, adjusting the secondary user sample label graph according to the frequency band decision result, and obtaining an adjusted comprehensive sample label graph according to the adjusted secondary user sample label graph;
specifically, the step is a process of feedback control input of decision network training, wherein a sample label graph of a secondary user is updated according to a decision, so that the frequency band in the sample label graph of the secondary user is adjusted according to the decision, and then the corresponding sample label graph of the primary user is synthesized to obtain an adjusted comprehensive sample label graph, namely, the adjustment of the decision network input is completed, so as to prepare for the next cycle input training.
More specifically, the adjusted integrated sample labeling diagram includes: and whether the frequency bands of the secondary user and the primary user collide or not. The comprehensive sample label graph formed by superposing the sample label graph of the primary user and the sample label graph of the secondary user comprises PU working frequency point information and PU working frequency point information which are simulated by the PU mode generator and the SU mode generator, and the quality of a current frequency point decision result can be judged according to whether the secondary user and the transmitting frequency band of the primary user are collided or not. For example, in the figure, the frequency points overlap by more than 50%, which is recorded as collision, and the specific collision definition can be set by itself as required, which is not limited herein. Accordingly, in this embodiment, the current decision result includes two results, i.e., a collision result and a non-collision result. And the result is contained in the adjusted comprehensive sample label graph and is used as the input of the decision network together.
Finally, step S105 is executed to determine whether the adjusted integrated sample label graph meets a preset condition; if not, circularly executing the steps until the adjusted comprehensive sample annotation graph meets the preset condition; and outputting the adjusted comprehensive sample label graph when the adjusted comprehensive sample label graph meets the preset condition.
As an optional specific implementation, the preset condition includes: and within the preset continuous cycle times, the frequency bands of the secondary user and the primary user in the adjusted comprehensive sample label graph do not collide.
Specifically, theoretically, when the adjusted comprehensive sample label graph displays that the frequency bands of the secondary user and the primary user do not collide for the first time in a certain circulation process, the available frequency points can be considered to be obtained, but in order to obtain reliable frequency point information more accurately, the frequency bands of the secondary user and the primary user in the adjusted comprehensive sample label graph do not collide for the continuous circulation times, the adjusted comprehensive sample label graph is considered to be a more reliable training end condition, and the obtained access frequency points are more reliable.
The technical scheme in the embodiment of the application at least has the following technical effects or advantages:
the method of the embodiment utilizes a sample annotation graph of a primary user and a sample annotation graph of a secondary user in a target environment to superpose and synthesize a comprehensive sample annotation graph as a condition, inputs a trained condition countermeasure generation network model, can generate a first simulation spectrum waterfall graph set containing a large amount of spectrum waterfall graphs required by the target environment in an off-line manner, and can generate corresponding output as long as a priori input is given to the condition countermeasure generation network model because the condition countermeasure generation network model is trained, namely the off-line manner, and in order to simulate the spectrum information of the target environment, a small amount of sample annotation graphs of the primary user and the sample annotation graphs of the secondary users in the target environment are used as the priori information input, so that a large amount of realistic simulation spectrum waterfall graphs required by the target environment can be obtained to be used as the input of decision network training, thereby replacing the on-line training method that needs to wait for obtaining the spectrum waterfall graphs from a primary user transmitter, moreover, as the trained offline conditions are utilized to resist and generate the network model, waiting is not needed, and compared with the prior art, the speed of acquiring the target environment frequency spectrum information is greatly improved; then, inputting the first simulation spectrum waterfall graph set and the comprehensive sample label graph into a decision network to obtain a frequency band decision result; adjusting the secondary user sample label graph according to the frequency band decision result, and obtaining an adjusted comprehensive sample label graph according to the adjusted secondary user sample label graph; judging whether the adjusted comprehensive sample annotation graph meets a preset condition or not; if not, circularly executing the steps until the adjusted comprehensive sample annotation graph meets the preset condition; and when the adjusted comprehensive sample label graph meets the preset condition, outputting the adjusted comprehensive sample label graph, wherein the output adjusted comprehensive sample label graph comprises dynamic spectrum access frequency point information. The input of each circulation of the process of obtaining the decision network model by the circulation training is generated by the off-line trained conditional countermeasure generation network model, and the off-line trained conditional countermeasure generation network model can quickly generate a sufficient amount of vivid frequency spectrum waterfall patterns of a target environment in an off-line mode without waiting, so that the input obtained by controlling the training process of the decision network after the circulation feedback decision of each time is not dependent on obtaining a training sample from an on-line main user transmitter, therefore, the playing speed limit of the transmitting mode of the main user transmitter is avoided, the decision network training efficiency is greatly improved, and the problems that the efficiency of obtaining dynamic frequency spectrum access frequency points is low and the practical applicability is not strong due to the fact that the time is consumed for obtaining the training sample in the prior art are solved.
Meanwhile, due to the wide and diverse spectrum space, if the collected spectrum pattern samples are not comprehensive enough or a new spectrum pattern appears in practical application, the performance of the decision network is reduced or even fails. It is clear that it is not possible to collect all band spectra and modes in advance, because the pattern of the spectra and their combination pattern is infinite, and in fact, there is a high chance of encountering new spectral modes. Even if spectrum samples are collected on line on site and a transfer learning technology is adopted, the training efficiency is still low. Therefore, in this embodiment, due to the existence of the environment spectrum simulator, the spectrum environment is described according to only a small number of independent samples, a sample label graph of the target environment is made, and a vivid and practically close spectrum waterfall graph is generated, so that the application range of the model is wider, and the model is faster and easier to update.
Example two
In an embodiment, as shown in fig. 4, the training method for generating a network model by conditional countermeasure specifically includes:
s201, randomly acquiring a diversified spectrum waterfall graph and a sample label graph corresponding to the spectrum waterfall graph;
s202, training to obtain a conditional countermeasure and generating a network model by taking the frequency spectrum waterfall graph and a sample label graph corresponding to the frequency spectrum waterfall graph as training samples, wherein the sample label graph is the condition.
The following describes in detail, with reference to fig. 4, specific implementation steps of the method provided in the embodiment of the present application:
firstly, executing step S201, randomly obtaining a diversified spectrum waterfall graph and a sample annotation graph corresponding to the spectrum waterfall graph;
specifically, the spectrum waterfall graph is a graph containing spectrum information generated by a spectrum environment, and,
the sample label graph corresponding to the frequency spectrum waterfall graph comprises at least one of a current working center frequency point, a bandwidth and frequency spectrum background information.
More specifically, when the PU transmitter operates according to the transmission mode and parameters, the PU transmitter simultaneously records the spectrum related information at each transmission time, for example: the current working center frequency point, bandwidth and the like also comprise some frequency spectrum background information in the activity at the moment. The recorded information is used as a Label of a frequency spectrum waterfall diagram in the working environment of the PU transmitter, and can be regarded as a reference Label (Label) in a training sample set, namely a frequency spectrum Label diagram.
In order to randomly obtain a diversified spectrum waterfall graph and a sample label graph corresponding to the spectrum waterfall graph, the following method can be adopted:
manually intervening with the primary user transmitter to generate diversified spectrum information;
and collecting the frequency spectrum information in real time and forming a frequency spectrum waterfall diagram.
Specifically, the transmission behavior of the transmitter is controlled by using a transmission mode, such as: the actions of frequency sweep, random emission and the like (the frequency spectrum waterfall graph of the upper graph is generated in a random emission mode), and the working state of the transmitter is controlled through parameters, such as: the PU transmitter works according to a transmitting mode and parameters, and is adjusted to be within the PU working frequency range by using the receiver, so that diversified frequency spectrum information can be acquired, and a corresponding frequency spectrum waterfall graph is generated. Theoretically, the more diversified training samples, the better the trained network model is, but in this embodiment, in actual operation, a commonly-used spectrum waterfall chart or a spectrum waterfall chart which is diversified as much as possible is collected to meet the requirement.
Next, step S202 is executed, and a network model is generated by training an acquisition condition pair with the frequency spectrum waterfall graph and a sample labeled graph corresponding to the frequency spectrum waterfall graph as training samples, wherein the sample labeled graph is a condition;
it should be noted that the conditional countermeasure generating network is used here to set an inputtable condition, that is, a sample labeled graph, so that the environmental spectrum simulator can purposefully obtain a desired spectrum waterfall graph according to a pre-programmed prior condition, and the environmental spectrum simulator is a trained conditional countermeasure generating network model.
The technical scheme in the embodiment of the application at least has the following technical effects or advantages:
the method comprises the steps of firstly, randomly obtaining a diversified frequency spectrum waterfall graph and a sample label graph corresponding to the frequency spectrum waterfall graph, taking the frequency spectrum waterfall graph and the sample label graph corresponding to the frequency spectrum waterfall graph as training samples, and performing neural network training to generate an environment frequency spectrum simulator; the neural network is a conditional countermeasure generation network, wherein a sample label graph is a condition; thus, a spectrum waterfall graph capable of quickly generating a large amount of simulated target environments in an off-line manner is obtained by utilizing a small amount of on-line spectrum information training; in this way, the sample annotation graph of the primary user and the sample annotation graph of the secondary user in the target environment are superposed to synthesize a comprehensive sample annotation graph; compared with the prior art, the method has the advantages that the environment spectrum simulator is utilized, only the condition of the preset target environment is input, the frequency spectrum waterfall graph with the sufficient target environment vivid can be generated in an off-line mode rapidly, training samples are obtained from an online main user transmitter without waiting, and therefore the limitation of the play rate of the transmitting mode of the main user transmitter is avoided.
Based on the same inventive concept, another embodiment of the present application provides a system for acquiring a dynamic spectrum access frequency point according to the embodiment of the present application.
EXAMPLE III
In this embodiment, as shown in fig. 5, the method includes:
the synthesis module is used for superposing a master user sample annotation graph and an initial secondary user sample annotation graph in the target environment to synthesize a comprehensive sample annotation graph;
the simulation module is used for inputting the comprehensive sample annotation graph as a condition to a trained condition countermeasure generation network model to generate a first simulation spectrum waterfall diagram set of the target environment;
the training module is used for inputting the first simulation spectrum waterfall pattern set and the comprehensive sample label graph into a decision network to obtain a frequency band decision result;
the decision control module is used for adjusting the secondary user sample label graph according to the frequency band decision result and obtaining an adjusted comprehensive sample label graph according to the adjusted secondary user sample label graph;
the judging module is used for judging whether the adjusted comprehensive sample annotation graph meets the preset condition or not; if not, circularly executing the steps until the adjusted comprehensive sample annotation graph meets the preset condition;
and the output module is used for outputting the adjusted comprehensive sample label graph when the adjusted comprehensive sample label graph meets the preset condition.
Since the system completely corresponds to the method of the first embodiment, the specific implementation described in the first embodiment is also applicable to each module of the system of the present embodiment.
Based on the same inventive concept, another embodiment of the present application provides a method for implementing the dynamic spectrum access described in the embodiment of the present application.
Example four
In this embodiment, the dynamic spectrum access method includes: the dynamic spectrum access frequency points obtained according to any one of claims 1-8 are used for dynamic spectrum access.
Specifically, in a target environment to be accessed, the adjusted comprehensive sample label graph is obtained by using the method of the first embodiment, the adjusted comprehensive sample label graph comprises frequency points accessible by a secondary user, network access can be quickly completed by accessing the network according to the frequency points, online network training is not relied on, access efficiency is improved, and practicability of a dynamic spectrum access decision network is greatly enhanced.
EXAMPLE five
In this embodiment, the system includes: an access module, configured to perform dynamic spectrum access according to the dynamic spectrum access frequency point obtained by the method according to any one of claims 1 to 8.
Since the system completely corresponds to the method of the first embodiment, the specific implementation described in the first embodiment is also applicable to each module of the system of the present embodiment.
The above-mentioned embodiments are intended to illustrate the objects, technical solutions and advantages of the present invention in further detail, and it should be understood that the above-mentioned embodiments are merely exemplary embodiments of the present invention, and are not intended to limit the scope of the present invention, and any modifications, equivalent substitutions, improvements and the like made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims (10)

1. A method for obtaining a dynamic spectrum access frequency point is characterized by comprising the following steps:
superposing a primary user sample annotation graph and an initial secondary user sample annotation graph in a target environment to synthesize a comprehensive sample annotation graph;
inputting the comprehensive sample label graph as a condition to a trained condition confrontation generation network model to generate a first simulation spectrum waterfall graph set of the target environment;
inputting the first simulation spectrum waterfall graph set and the comprehensive sample label graph into a decision network to obtain a frequency band decision result;
adjusting the secondary user sample label graph according to the frequency band decision result, and obtaining an adjusted comprehensive sample label graph according to the adjusted secondary user sample label graph;
judging whether the adjusted comprehensive sample annotation graph meets a preset condition or not;
if not, circularly executing the steps until the adjusted comprehensive sample annotation graph meets the preset condition; and outputting the adjusted comprehensive sample label graph when the adjusted comprehensive sample label graph meets the preset condition.
2. The method according to claim 1, wherein the training method for conditional countermeasure generation of the network model specifically comprises:
randomly obtaining a diversified spectrum waterfall graph and a sample label graph corresponding to the spectrum waterfall graph;
and training to obtain a conditional countermeasure generation network model by taking the frequency spectrum waterfall chart and a sample labeled graph corresponding to the frequency spectrum waterfall chart as training samples, wherein the sample labeled graph is a condition.
3. The method of claim 2, wherein the randomly obtaining a diversified spectral waterfall graph comprises:
manually intervening a transmitter of the primary user to generate diversified spectrum information;
and collecting the frequency spectrum information in real time and forming a frequency spectrum waterfall diagram.
4. The method of claim 1, wherein the adjusted composite sample label graph comprises: and whether the frequency bands of the secondary user and the primary user collide or not.
5. The method according to claim 4, wherein the preset conditions include: and within the preset continuous cycle times, the frequency bands of the secondary user and the primary user in the adjusted comprehensive sample label graph do not collide.
6. The method according to claim 2, wherein the sample label graph corresponding to the spectrum waterfall graph includes at least one of a current operating center frequency point, a bandwidth and spectrum background information.
7. The method according to claim 1, wherein the sample label graph of the primary user and the sample label graph of the secondary user in the target environment are a priori information artificially compiled according to the target environment.
8. A system for obtaining dynamic spectrum access frequency points is characterized by comprising:
the synthesis module is used for superposing a master user sample annotation graph and an initial secondary user sample annotation graph in the target environment to synthesize a comprehensive sample annotation graph;
the simulation module is used for inputting the comprehensive sample annotation graph as a condition to a trained condition countermeasure generation network model to generate a first simulation spectrum waterfall diagram set of the target environment;
the training module is used for inputting the first simulation spectrum waterfall pattern set and the comprehensive sample label graph into a decision network to obtain a frequency band decision result;
the decision control module is used for adjusting the secondary user sample label graph according to the frequency band decision result and obtaining an adjusted comprehensive sample label graph according to the adjusted secondary user sample label graph;
the judging module is used for judging whether the adjusted comprehensive sample annotation graph meets the preset condition or not; if not, circularly executing the steps until the adjusted comprehensive sample annotation graph meets the preset condition;
and the output module is used for outputting the adjusted comprehensive sample label graph when the adjusted comprehensive sample label graph meets the preset condition.
9. A dynamic spectrum access method, comprising:
the dynamic spectrum access frequency points obtained according to any one of claims 1-7 are used for dynamic spectrum access.
10. A dynamic spectrum access system, comprising:
an access module, configured to perform dynamic spectrum access according to the dynamic spectrum access frequency point obtained by the method according to any one of claims 1 to 7.
CN201910284056.7A 2019-04-09 2019-04-09 Method for acquiring dynamic spectrum access frequency point and dynamic spectrum access method Active CN110138591B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910284056.7A CN110138591B (en) 2019-04-09 2019-04-09 Method for acquiring dynamic spectrum access frequency point and dynamic spectrum access method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910284056.7A CN110138591B (en) 2019-04-09 2019-04-09 Method for acquiring dynamic spectrum access frequency point and dynamic spectrum access method

Publications (2)

Publication Number Publication Date
CN110138591A CN110138591A (en) 2019-08-16
CN110138591B true CN110138591B (en) 2022-04-26

Family

ID=67569597

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910284056.7A Active CN110138591B (en) 2019-04-09 2019-04-09 Method for acquiring dynamic spectrum access frequency point and dynamic spectrum access method

Country Status (1)

Country Link
CN (1) CN110138591B (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102455859A (en) * 2010-11-03 2012-05-16 北京普源精电科技有限公司 Method and device for displaying spectrum data
CN105634624A (en) * 2014-10-27 2016-06-01 中国科学院沈阳自动化研究所 Cognitive radio network-oriented frequency-domain collaborative spectrum sensing method
CN109586820A (en) * 2018-12-28 2019-04-05 中国人民解放军陆军工程大学 The anti-interference model of dynamic spectrum and intensified learning Anti-interference algorithm in fading environment

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104105099B (en) * 2013-04-07 2018-12-14 电信科学技术研究院 A kind of frequency spectrum dynamic allocation method and equipment

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102455859A (en) * 2010-11-03 2012-05-16 北京普源精电科技有限公司 Method and device for displaying spectrum data
CN105634624A (en) * 2014-10-27 2016-06-01 中国科学院沈阳自动化研究所 Cognitive radio network-oriented frequency-domain collaborative spectrum sensing method
CN109586820A (en) * 2018-12-28 2019-04-05 中国人民解放军陆军工程大学 The anti-interference model of dynamic spectrum and intensified learning Anti-interference algorithm in fading environment

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Adversarial Deep Learning for Cognitive Radio Security: Jamming Attack and Defense Strategies;Yi Shi 等;《2018 IEEE International Conference on Communications Workshops (ICC Workshops)》;20180705;全文 *
基于寡头定价的动态频谱接入控制;李彤 等;《电讯技术》;20190128;第59卷(第1期);第7-12页 *

Also Published As

Publication number Publication date
CN110138591A (en) 2019-08-16

Similar Documents

Publication Publication Date Title
CN112615379B (en) Power grid multi-section power control method based on distributed multi-agent reinforcement learning
CN109682380A (en) A kind of communication unmanned plane method for optimizing route and equipment
Wilhelmi et al. Usage of network simulators in machine-learning-assisted 5G/6G networks
CN105204009A (en) Broadband multifunctional radar signal simulation system and method
Zheng et al. Synthetic dynamic PMU data generation: A generative adversarial network approach
CN111178496A (en) Method for exchanging knowledge among agents under multi-agent reinforcement learning cooperative task scene
CN107481598A (en) A kind of emulation training platform of intelligent substation
CN107563011A (en) A kind of Design of Simulation System method of Unmanned Aerial Vehicle Data Link reliability
Mahmood et al. Modeling, simulation and forecasting of wind power plants using agent-based approach
CN110138591B (en) Method for acquiring dynamic spectrum access frequency point and dynamic spectrum access method
CN114139377A (en) Power system modeling method and system, readable storage medium and server
Kaliappan et al. Flexible power consumption management using Q learning techniques in a smart home
CN107404405B (en) Intelligent substation IEC61850 consistency test method and system
CN113382060B (en) Unmanned aerial vehicle track optimization method and system in Internet of things data collection
CN107959598A (en) A kind of communication network reliability test profile construction method based on business
TWI748794B (en) Beam selection method based on neural network and management server
CN111182556B (en) Wireless network planning design method based on intelligent agent
CN112632710A (en) Intelligent integrated simulation method for ship electric propulsion system
CN116506863A (en) Decision optimization method, decision optimization device, electronic equipment and readable storage medium
CN106325069A (en) Method for designing optimal linear control strategy for wireless network control system
Zhao et al. Design of jiutian intelligent network simulation platform
Gonzalez et al. Vehicle model generation and optimization for embedded simulation
Möllerstedt et al. Model Based Residual Policy Learning with Applications to Antenna Control
Robertson et al. A pseudo-real time distribution network simulator for analysis of coordinated ANM control strategies
Pruyt Smart transition management to smarten energy systems in a deeply uncertain world

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