A method, device and medium for intelligently identifying communication interference signals

Abstract

The embodiment of the present invention discloses an intelligent recognition method, device and medium for communication interference signals. The method comprises: dividing the acquired spectrum data of the communication interference signal into a training set and a verification set; constructing a recognition model of the communication interference signal by using a DCNN capable of estimating a confidence score and calculating a recognition probability; training the recognition model by using the training set; recognizing the verification set by the trained recognition model, and setting a determination threshold for determining an unknown interference signal according to the distribution of the data confidence score of the verification set in the recognition result; recognizing the communication interference signal to be identified by the trained recognition model, and obtaining the confidence score and probability distribution of the communication interference signal to be identified; if the confidence score corresponding to the communication interference signal to be identified is less than the determination threshold, determining that the communication interference signal to be identified is unknown interference; otherwise, determining that the communication interference signal to be identified is the communication interference signal with the highest probability.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of wireless communication technology, and in particular to a method, device and medium for intelligently identifying communication interference signals. Background Art

[0002] The electromagnetic environment in which wireless communication systems are currently located is becoming increasingly complex and is more susceptible to a variety of interference signals and noise. For civilian communications, the rapid growth in the number of mobile communications has aggravated the problem of spectrum congestion. For military communications, with the emphasis on research on communication countermeasure technology, more and more types of interference have emerged, making the electromagnetic environment of the battlefield increasingly complex; therefore, in order to ensure reliable communication, current wireless communication systems must have sufficient anti-interference capabilities.

[0003] For a wireless communication system with anti-interference capability, it is necessary to accurately identify the interference pattern emitted by the enemy, and then select the corresponding anti-interference decision to suppress or eliminate the enemy interference, and finally ensure the normal operation of the communication system. Such a wireless communication system usually includes: interference perception, interference signal identification, and anti-interference decision; among which, interference signal identification is the premise and basis of anti-interference capability. During the communication process, if the wireless communication system can accurately identify the type of interference signal, it can take the corresponding anti-interference decision to suppress, eliminate or avoid the interference signal to the greatest extent, thereby minimizing the damage of the interference signal to the reliable communication of the wireless communication system.

[0004] At present, conventional interference identification technologies are interference identification technologies based on characteristic parameters and interference identification methods based on artificial neural networks. However, wireless communication systems usually face the problem of receiving a large number of unlabeled signal samples and interference signals of unknown types during communication. If only known communication interference signal label samples are available for learning in machine learning, the recognition function of the trained model will be greatly limited. When encountering a signal type that is not in the training set, it will also be mistakenly identified as a known interference type in the training set, which will have an immeasurable impact on the next step of interference suppression. Therefore, the model that needs to be trained can not only accurately identify the existing signal types in the training set, but also solve the problem of encountering unknown types of signals that are not in the training set. Summary of the invention

[0005] In view of this, the embodiments of the present invention hope to provide a method, device and medium for intelligently identifying communication interference signals; it is capable of identifying unknown interference type signals in actual open interference scenarios, and improving the accuracy and real-time performance of interference signal identification when characteristic parameters are invalid.

[0006] The technical solution of the embodiment of the present invention is achieved as follows:

[0007] In a first aspect, an embodiment of the present invention provides a method for intelligently identifying a communication interference signal, the method comprising:

[0008] Divide the acquired spectrum data of the communication interference signal into a training set and a validation set;

[0009] A recognition model for communication interference signals is constructed using a deep convolutional neural network (DCNN) that can estimate confidence scores and calculate recognition probabilities.

[0010] Using the training set to train the recognition model;

[0011] The trained recognition model is used to recognize the verification set, and a determination threshold for determining an unknown interference signal is set according to the distribution of data confidence scores of the verification set in the recognition result;

[0012] Using the trained recognition model to identify the communication interference signal to be identified, and obtaining the confidence score and probability distribution of the communication interference signal to be identified;

[0013] If the confidence score corresponding to the communication interference signal to be identified is less than the determination threshold, the communication interference signal to be identified is determined to be unknown interference; otherwise, the communication interference signal to be identified is determined to be the communication interference signal with the highest probability.

[0014] In a second aspect, an embodiment of the present invention provides an intelligent recognition device for communication interference signals, the intelligent recognition device comprising: a division part, a construction part, a training part, a setting part and a recognition part:

[0015] Wherein, the division part is configured to divide the acquired spectrum data of the communication interference signal into a training set and a verification set;

[0016] The construction part is configured to construct a recognition model of the communication interference signal using a deep convolutional neural network DCNN capable of estimating confidence scores and calculating recognition probabilities;

[0017] The training part is configured to train the recognition model using the training set;

[0018] The setting part is configured to use the trained recognition model to recognize the verification set, and set a determination threshold for determining an unknown interference signal according to the distribution of data confidence scores of the verification set in the recognition result;

[0019] The identification part is configured to identify the communication interference signal to be identified by using the trained identification model to obtain the confidence score and probability distribution of the communication interference signal to be identified; and, if the confidence score corresponding to the communication interference signal to be identified is less than the judgment threshold, the communication interference signal to be identified is determined to be unknown interference; otherwise, the communication interference signal to be identified is determined to be the communication interference signal with the highest probability.

[0020] In a third aspect, an embodiment of the present invention provides a computer storage medium, wherein the computer storage medium stores a program for intelligently identifying communication interference signals, and when the program for intelligently identifying communication interference signals is executed by at least one processor, the steps of the method for intelligently identifying communication interference signals described in the first aspect are implemented.

[0021] The embodiments of the present invention provide a method, device and medium for intelligently identifying communication interference signals; during the training process, a confidence score is added to the DCNN model to judge whether the signal to be identified is a communication interference signal of a known type in the training set, so that communication interference signals of unknown types different from the known signal types can be identified, avoiding erroneous identification of the type of communication interference signals of unknown types as a certain known communication interference signal type in the training set, and then using erroneous means to suppress, eliminate or avoid the communication interference signals of unknown types; thereby improving the accuracy of interference signal identification and the real-time performance of identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A schematic flow chart of a method for intelligently identifying communication interference signals provided by an embodiment of the present invention;

[0023] Figure 2 A schematic diagram of a DCNN structure provided by an embodiment of the present invention;

[0024] Figure 3 A schematic diagram of the composition of an intelligent identification device for communication interference signals provided by an embodiment of the present invention;

[0025] Figure 4 A schematic diagram of a specific hardware structure of a computing device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0026] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0027] At present, in the process of identifying communication interference signals based on neural network technology and machine learning algorithms, the trained neural network model can only identify the types of communication interference signals known during the training process. This will cause the unknown type of communication interference signal to be mistakenly identified as a certain known type of communication interference signal in the training set when encountering a signal type that is not in the training set during the training process, and then the wrong means will be used to suppress, eliminate or avoid the unknown type of communication interference signal. As a result, the communication interference signal cannot be identified in an open interference scenario, reducing the accuracy of identification.

[0028] In order to avoid the above situation and improve the recognition accuracy of communication interference signals, the embodiment of the present invention is expected to add an unknown type of communication interference signal recognition strategy to the neural network model, thereby improving the recognition accuracy and real-time performance of interference signals. Figure 1 , which shows an intelligent identification method for communication interference signals provided by an embodiment of the present invention, the method may include:

[0029] S101: Divide the acquired spectrum data of the communication interference signal into a training set and a verification set;

[0030] S102: constructing a recognition model for communication interference signals using a deep convolutional neural network (DCNN) capable of estimating confidence scores and calculating recognition probabilities;

[0031] S103: training the recognition model using the training set;

[0032] S104: using the trained recognition model to recognize the verification set, and setting a determination threshold for determining an unknown interference signal according to the distribution of data confidence scores of the verification set in the recognition result;

[0033] S105: using the trained recognition model to recognize the communication interference signal to be recognized, and obtaining a confidence score and a probability distribution of the communication interference signal to be recognized;

[0034] S106: If the confidence score corresponding to the communication interference signal to be identified is less than the determination threshold, determining that the communication interference signal to be identified is unknown interference;

[0035] Otherwise, go to S107: determine the communication interference signal to be identified as the communication interference signal with the highest probability.

[0036] pass Figure 1In the technical solution shown, during the training process, a confidence score is added to the DCNN model to judge whether the signal to be identified is a communication interference signal of a known type in the training set, so that an unknown type of communication interference signal different from the known signal type can be identified, avoiding erroneous identification of the unknown type of communication interference signal as a certain known type of communication interference signal in the training set, and then using erroneous means to suppress, eliminate or avoid the unknown type of communication interference signal; the interference signal identification accuracy and recognition real-time performance are improved.

[0037] for Figure 1 In some possible implementations of the technical solution shown, the acquired spectrum data of the communication interference signal is divided into a training set and a verification set, including:

[0038] Collecting data of the I channel and Q channel of the acquired communication interference signal;

[0039] Performing discrete Fourier transform on the data of the I channel and the Q channel to obtain spectrum data of the acquired communication interference signal;

[0040] 80% of the data in each type of communication interference signal data set in the acquired spectrum data of the communication interference signal is randomly selected as the training set, and the remaining 20% ​​of the data is used as the verification set.

[0041] For the above implementation, specifically, first, the I and Q channel data x(n) of the collected communication interference signal are discretely Fourier transformed according to the following formula to obtain the spectrum data X(k) of x(n):

[0042]

[0043] Where N represents the number of sampling points.

[0044] Then, a spectrum graph can be drawn based on the spectrum data X(k).

[0045] The above process can be a preprocessing process for the collected communication interference signal. After the above preprocessing process, the embodiment of the present invention preferably divides the preprocessed communication interference signal into a training set and a verification set. The specific process is: randomly select 80% of the data from each type of communication interference signal data set obtained by simulation as a training set, and use the remaining 20% ​​of the data as a verification set.

[0046] for Figure 1 In some possible implementations of the technical solution shown, the use of a DCNN capable of estimating confidence scores and calculating recognition probabilities to construct a recognition model for communication interference signals includes:

[0047] Based on the recognition requirements, the nodes of the input and output layers, the number of convolutional layers, the number of convolutional kernels, the activation function, the number of pooling layers, and the number of fully connected layers of the DCNN are determined and initialized;

[0048] The DCNN is divided into a feature extraction part, a feature integration part and a result processing part; wherein the feature extraction part includes a convolution layer and a pooling layer; the feature integration part includes two fully connected layers; the result processing part includes two fully connected layers and the two fully connected layers respectively process the results output by the feature integration part to obtain the matching probability of the interference signal to be identified on each known interference signal and the confidence score of the interference signal to be identified.

[0049] For the DCNN described in the above implementation, in some examples, the feature extraction part has a total of 13 layers, the number of nodes in the input layer is 150528, the number of nodes in the output layer is 25088, the number of convolutional layers is 8, and the number of pooling layers is 5; the number of neurons in the two fully connected layers of the feature integration part are 25088 and 512 respectively; the number of neurons in the two fully connected layers of the result processing part is 128.

[0050] For the DCNN described in the above implementation method, in some examples, the result processing part includes a confidence score estimation unit and a recognition probability calculation unit; wherein each unit corresponds to a fully connected layer; the recognition probability calculation unit is used to obtain the probability of identifying the interference signal to be identified as each known interference signal after passing the result output by the feature integration part through the corresponding fully connected layer and Softmax layer; the confidence score estimation unit is used to obtain the reliability of the result of the recognition probability of the interference signal to be identified after passing the result output by the feature integration part through the corresponding fully connected layer, and then process the reliability through the sigmoid activation function to obtain a number between 0 and 1 as the confidence score of the interference signal to be identified.

[0051] For the above implementation and its examples, specifically, the embodiment of the present invention preferably determines the nodes of the input and output layers of the DCNN, the number of convolutional layers, the number of convolutional kernels, the activation function, the number of pooling layers, and the number of fully connected layers according to actual needs; then, the parameters of the nodes of each layer are initialized. As shown in the above implementation, in the embodiment of the present invention, the DCNN preferably includes a feature extraction part, a feature integration part, and a result processing part; wherein, as shown in Table 1, the specific structure of the neural network model of the feature extraction part includes 13 layers, the number of input layer nodes is 150528, the number of output layer nodes is 25088, and the number of convolutional layers is 8, and the number of pooling layers is 5;

[0052] Table 1

[0053] Layer Type Specific structure Input layer size Covn+relu kernel_size=(3,3), stride=(1,1), padding=(1,1) 224×224×3 MaxPool kernel_size=(2,2),stride=(2,2) 224×224×64 Covn+relu kernel_size=(3,3), stride=(1,1), padding=(1,1) 112×112×64 MaxPool kernel_size=2,stride=2 112×112×128 Covn+relu kernel_size=(3,3), stride=(1,1), padding=(1,1) 56×56×128 Covn+relu kernel_size=(3,3), stride=(1,1), padding=(1,1) 56×56×256 MaxPool kernel_size=2,stride=2 56×56×256 Covn+relu kernel_size=(3,3), stride=(1,1), padding=(1,1) 28×28×256 Covn+relu kernel_size=(3,3), stride=(1,1), padding=(1,1) 28×28×512 MaxPool kernel_size=2,stride=2 28×28×512 Covn+relu kernel_size=(3,3), stride=(1,1), padding=(1,1) 14×14×512 Covn+relu kernel_size=(3,3), stride=(1,1), padding=(1,1) 14×14×512 MaxPool kernel_size=2,stride=2 14×14×512

[0054] In Table 1, Covn represents the convolution layer, MaxPool represents the pooling layer, relu represents the activation function, kernel_size represents the convolution kernel size, stride represents the step size, and padding represents padding.

[0055] As shown in Table 2, the feature integration part has two layers, including two fully connected layers, and the number of neurons in these two fully connected layers is 25088 and 512 respectively;

[0056] Table 2

[0057] Layer structure Specific structure Input size Linear+relu Dropout: 0.5, Neurons: 512 25088 Linear+relu Dropout: 0.5, Neurons: 128 512

[0058] In Table 2, Linear represents the fully connected layer, Dropout represents the proportion of randomly dropped neurons, and Neurons represents the number of neurons.

[0059] As shown in Table 3, the result processing part includes two parts: a confidence score estimation unit and a recognition probability calculation unit, and the result processing part includes two fully connected layers with 128 neurons each, which correspond to the confidence score estimation unit and the recognition probability calculation unit, respectively.

[0060] Table 3

[0061] Layer structure Specific structure Input size Linear+softmax Neurons:6 128 Linear+sigmoid Neurons:1 128

[0062] In Table 3, the recognition probability calculation unit is used to obtain the probability of identifying the interference signal to be identified as each known interference signal after passing the result output by the feature integration part through the corresponding fully connected layer and Softmax layer; the confidence score estimation unit is used to obtain the reliability of the result of the recognition probability of the interference signal to be identified after passing the result output by the feature integration part through the corresponding fully connected layer, and then process the reliability through the sigmoid activation function to obtain a number between 0 and 1 as the confidence score of the interference signal to be identified.

[0063] For the above implementations and examples, based on the structures described in Tables 1 to 3, the recognition model structure constructed by DCNN used in the embodiment of the present invention is as follows: Figure 2As shown, it can be understood that the DCNN adopted in the embodiment of the present invention has a convolutional layer that is a two-dimensional convolutional neural network. The convolution kernel of the two-dimensional convolutional neural network can automatically extract the deep features of the communication interference signal, which solves the problem of low real-time performance of the technical recognition process of the conventional scheme using feature parameters for recognition, and also solves the problem that the recognition accuracy of the communication interference signal of the conventional scheme method drops rapidly when the feature is invalid; in addition, the embodiment of the present invention adds a confidence score estimation unit to the DCNN, which can not only realize the recognition of the communication interference signal in the closed set scenario by the prior art, but also detect whether the signal to be recognized is an unknown communication interference signal, which overcomes the problem that the conventional scheme cannot be applied in the actual open set scenario, so that the present invention has the advantages of detecting unknown communication interference and accurately identifying the known communication interference type.

[0064] In combination with the foregoing implementation manner and examples thereof, in some examples, the using the training set to train the recognition model includes:

[0065] Batch processing is adopted for the training data, and the amount of data inputted into the recognition model from the training set is set each time;

[0066] Set the data sampling method to random sampling;

[0067] The task loss E for recognition probability calculation is defined based on the following formula P and the confidence estimate loss E of the confidence score estimate C :

[0068]

[0069] Among them, p′ i =c×p i +(1-c)y i represents the corrected communication interference signal recognition probability, p i represents the original predicted communication interference signal recognition probability obtained by the recognition probability calculation unit in the recognition model, y i represents the target probability distribution, M represents the number of known communication interference signal types, and C is the confidence score output by the confidence score estimation unit in the recognition model;

[0070] The task loss E calculated according to the recognition probability P and the confidence estimate loss E of the confidence score estimate C The loss function E of the recognition model is determined according to the following formula:

[0071] E=E P +λE C

[0072] The budget hyperparameter β is introduced to represent the confidence loss allowed by the recognition model; the hyperparameter λ is introduced to balance the task loss and the confidence estimation loss, and λ changes dynamically in the model. C >β, increase λ, when E C When <β, reduce λ;

[0073] Data is extracted from the randomly sampled data set according to the set parameters and input into the DCNN, and training is performed based on the set SDG optimization algorithm according to the set number of iterations to obtain a recognition model for the communication interference signal that has been trained.

[0074] For the above example, it should be noted that after completing the training of the recognition model, the trained recognition model can be used to identify the verification set described in the above implementation method, and then the distribution of confidence scores of known types of communication interference signals on the verification set is obtained, and the judgment threshold T is set according to the actual sensitivity requirements for unknown signals. Next, the communication interference signal to be identified in the real environment is identified using the trained recognition model, so as to obtain the confidence score of the recognition model for the identified communication interference signal and the probability distribution of the communication interference signal being identified as a known communication interference signal label set. If the confidence score obtained above is less than the threshold set above, it means that the type of the communication interference signal to be identified does not belong to a known type, and it can be determined as unknown interference; otherwise, the communication interference signal to be identified is determined to be the communication interference signal with the highest probability on the communication interference label signal set.

[0075] Based on the above technical solution, the embodiment of the present invention further illustrates the technical performance and effect of the above technical solution through specific simulation experiments. The specific simulation conditions and parameters are as follows:

[0076] The GPU model GeForce GTX2060 is used as the simulation environment, and the deep learning framework PyTorch is used to train and test the convolutional neural network. The simulation software Matlab is used to generate simulated communication interference signals. The simulated communication interference signals include six types of communication interference signals, namely single-tone interference, multi-tone interference, narrowband noise interference, broadband noise interference, linear frequency sweep interference, and broadband comb interference, as six known interferences. At the same time, multi-tone interference and narrowband noise interference are compounded as unknown communication interference signal 1, and single-tone interference and broadband comb interference are compounded as unknown communication interference signal 2. In the simulation process, the sampling rate is 10MHz, the number of sampling points is 2048, the channel noise is additive Gaussian white noise, and the signal-to-interference ratio range is -10dB to 15dB. For the six known interference signals, 500 samples are generated at each dB from -10dB to 15dB, of which 400 samples are used as training sets and 100 samples are used as validation sets. Six known interference signals and unknown interference signals 1 and 2 are simulated with 100 samples at each dB in the range of -5dB to 15dB as the test set. Finally, it can be seen that the total number of samples in the training set is 62400, the total number of samples in the validation set is 15600, and the number of samples in the test set is 16800.

[0077] The communication interference signal recognition model built based on the technical solution of the embodiment of the present invention (hereinafter referred to as the technical solution of the embodiment of the present invention) and the conventional communication interference recognition solution based on the convolutional neural network (hereinafter referred to as the prior art solution) are trained and tested using the above training set, validation set and test set, and the test results are compared, and the interference signal recognition accuracy rate α is used to evaluate the performance of the technical solution of the embodiment of the present invention and the prior art solution. Among them, the interference signal recognition accuracy rate is specifically shown in the following formula:

[0078] Where T is the number of interference signals correctly identified, and A is the total number of interference signals identified;

[0079] The interference signal recognition accuracy shown in the above formula is used to calculate the recognition accuracy of each interference signal type in the test set by the prior art solution and the technical solution of the embodiment of the present invention, and the confusion matrix tables shown in Table 4 and Table 5 are obtained respectively. Among them, Table 4 is the test result confusion matrix obtained by the prior art solution for each interference signal recognition in the test set in the open set scenario. Table 5 is the test result confusion matrix obtained by the technical solution of the embodiment of the present invention for each interference signal recognition in the test set in the open set scenario.

[0080] Table 4

[0081]

[0082] Table 5

[0083]

[0084] It can be seen from Table 4 that the prior art solution, based on the limitations of its principles, identifies both unknown interference signal 1 and unknown interference signal 2 as known interferences in the training set, and is unable to detect unknown signals, and is obviously unable to be used in an open set scenario. From Table 5, it can be seen that the technical solution of the embodiment of the present invention has an 80% and 75% probability of identifying unknown interference 1 and unknown interference 2 as unknown interference types, respectively, and also achieves a high recognition accuracy rate for the six known interferences. The above experimental simulation results show that the technical solution of the embodiment of the present invention can not only accurately identify known interference signal types, but also detect unknown interference signals in open set scenarios, solving the problem that the prior art cannot be applied to signal recognition in open set scenarios.

[0085] Based on the same inventive concept as the above technical solution, see Figure 3 , which shows an intelligent recognition device 30 for communication interference signals provided by an embodiment of the present invention, the intelligent recognition device 30 includes: a division part 301, a construction part 302, a training part 303, a setting part 304 and a recognition part 305:

[0086] The division part 301 is configured to divide the acquired spectrum data of the communication interference signal into a training set and a validation set;

[0087] The construction part 302 is configured to construct a recognition model of the communication interference signal using a deep convolutional neural network DCNN capable of estimating confidence scores and calculating recognition probabilities;

[0088] The training part 303 is configured to train the recognition model using the training set;

[0089] The setting part 304 is configured to use the trained recognition model to recognize the verification set, and set a determination threshold for determining an unknown interference signal according to the distribution of data confidence scores of the verification set in the recognition result;

[0090] The identification part 305 is configured to identify the communication interference signal to be identified by using the trained identification model to obtain the confidence score and probability distribution of the communication interference signal to be identified; and, if the confidence score corresponding to the communication interference signal to be identified is less than the determination threshold, the communication interference signal to be identified is determined to be unknown interference; otherwise, the communication interference signal to be identified is determined to be the communication interference signal with the highest probability.

[0091] In some examples, the partitioning portion 301 is configured as follows:

[0092] Collecting data of the I channel and Q channel of the acquired communication interference signal;

[0093] Performing discrete Fourier transform on the data of the I channel and the Q channel to obtain spectrum data of the acquired communication interference signal;

[0094] 80% of the data in each type of communication interference signal data set in the acquired spectrum data of the communication interference signal is randomly selected as the training set, and the remaining 20% ​​of the data is used as the verification set.

[0095] In some examples, the building portion 302 is configured to:

[0096] Based on the recognition requirements, the nodes of the input and output layers, the number of convolutional layers, the number of convolutional kernels, the activation function, the number of pooling layers, and the number of fully connected layers of the DCNN are determined and initialized;

[0097] The DCNN is divided into a feature extraction part, a feature integration part and a result processing part; wherein the feature extraction part includes a convolution layer and a pooling layer; the feature integration part includes two fully connected layers; the result processing part includes two fully connected layers and the two fully connected layers respectively process the results output by the feature integration part to obtain the matching probability of the interference signal to be identified on each known interference signal and the confidence score of the interference signal to be identified.

[0098] In the above example, the feature extraction part has a total of 13 layers, the number of nodes in the input layer is 150528, the number of nodes in the output layer is 25088, the number of convolutional layers is 8, and the number of pooling layers is 5; the number of neurons in the two fully connected layers of the feature integration part are 25088 and 512 respectively; the number of neurons in the two fully connected layers of the result processing part is 128.

[0099] In the above example, the result processing part includes a confidence score estimation unit and an identification probability calculation unit; wherein each unit corresponds to a fully connected layer; the identification probability calculation unit is used to obtain the probability of identifying the interference signal to be identified as each known interference signal after passing the result output by the feature integration part through the corresponding fully connected layer and Softmax layer; the confidence score estimation unit is used to obtain the reliability of the result of the identification probability of the interference signal to be identified after passing the result output by the feature integration part through the corresponding fully connected layer, and then process the reliability through the sigmoid activation function to obtain a number between 0 and 1 as the confidence score of the interference signal to be identified.

[0100] In some examples, the training portion 303 is configured to:

[0101] Batch processing is adopted for the training data, and the amount of data input from the training set to the recognition model is set each time;

[0102] Set the data sampling method to random sampling;

[0103] The task loss E for recognition probability calculation is defined based on the following formula P and the confidence estimate loss E of the confidence score estimate C :

[0104]

[0105] Among them, p′ i =c×p i +(1-c)y i represents the corrected communication interference signal recognition probability, p i represents the original predicted communication interference signal recognition probability obtained by the recognition probability calculation unit in the recognition model, y i represents the target probability distribution, M represents the number of known communication interference signal types, and C is the confidence score output by the confidence score estimation unit in the recognition model;

[0106] The task loss E calculated according to the recognition probability P and the confidence estimate loss E of the confidence score estimate C The loss function E of the recognition model is determined according to the following formula:

[0107] E=E P +λE C

[0108] The budget hyperparameter β is introduced to represent the confidence loss allowed by the recognition model; the hyperparameter λ is introduced to balance the task loss and the confidence estimation loss, and λ changes dynamically in the model. C >β, increase λ, when E C When <β, reduce λ;

[0109] Data is extracted from the randomly sampled data set according to the set parameters and input into the DCNN, and training is performed based on the set SDG optimization algorithm according to the set number of iterations to obtain a recognition model for the communication interference signal that has been trained.

[0110] It can be understood that in this embodiment, "part" can be part of a circuit, part of a processor, part of a program or software, etc., and of course it can also be a unit, a module, or a non-modular one.

[0111] In addition, each component in this embodiment may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above integrated unit may be implemented in the form of hardware or in the form of software function modules.

[0112] If the integrated unit is implemented in the form of a software function module and is not sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this embodiment is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) or a processor to perform all or part of the steps of the method described in this embodiment. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0113] Therefore, this embodiment provides a computer storage medium, which stores a program for intelligently identifying communication interference signals. When the program for intelligently identifying communication interference signals is executed by at least one processor, the steps of the method for intelligently identifying communication interference signals in the above technical solution are implemented.

[0114] According to the above-mentioned intelligent identification device 30 for communication interference signals and computer storage medium, see Figure 4 , which shows the specific hardware structure of a computing device 40 of an intelligent identification device 30 capable of implementing the above-mentioned communication interference signal provided by an embodiment of the present invention. The computing device 40 can be a wireless device, a mobile or cellular phone (including a so-called smart phone), a personal digital assistant (PDA), a video game console (including a video display, a mobile video game device, a mobile video conferencing unit), a laptop computer, a desktop computer, a TV set-top box, a tablet computing device, an e-book reader, a fixed or mobile media player, etc. The computing device 40 includes: a communication interface 401, a memory 402 and a processor 403; the various components are coupled together through a bus system 404. It can be understood that the bus system 404 is used to realize the connection and communication between these components. In addition to the data bus, the bus system 404 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, in Figure 4 In the figure, various buses are labeled as bus system 404. Among them,

[0115] The communication interface 401 is used to receive and send signals during the process of sending and receiving information with other external network elements;

[0116] The memory 402 is used to store a computer program that can be run on the processor 403;

[0117] The processor 403 is used to execute the steps of the intelligent identification method of communication interference signals in the aforementioned technical solution when running the computer program, which will not be repeated here.

[0118] It can be understood that the memory 402 in the embodiment of the present invention can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory can be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct RAM bus random access memory (DRRAM). The memory 402 of the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0119] The processor 403 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the processor 403. The above processor 403 may be a general processor, a digital signal processor (Digital Signal Processor, DSP), an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field programmable gate array (Field Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present invention can be implemented or executed. The general processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the method disclosed in the embodiment of the present invention can be directly embodied as a hardware decoding processor to be executed, or the hardware and software modules in the decoding processor are combined to be executed. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 402, and the processor 403 reads the information in the memory 402 and completes the steps of the above method in combination with its hardware.

[0120] It is understood that the embodiments described herein may be implemented in hardware, software, firmware, middleware, microcode, or a combination thereof. For hardware implementation, the processing unit may be implemented in one or more application specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPD), programmable logic devices (PLD), field programmable gate arrays (FPGA), general purpose processors, controllers, microcontrollers, microprocessors, other electronic units for performing the functions described in the present application, or a combination thereof.

[0121] For software implementation, the techniques described herein can be implemented by modules (e.g., procedures, functions, etc.) that perform the functions described herein. The software code can be stored in a memory and executed by a processor. The memory can be implemented in the processor or outside the processor.

[0122] Specifically, the processor 403 is further configured to, when running the computer program,

[0123] It can be understood that the exemplary technical solutions of the intelligent identification device 30 and the computing device 40 for communication interference signals belong to the same concept as the technical solution of the intelligent identification method for communication interference signals, and therefore, the details not described in detail in the technical solutions of the intelligent identification device 30 and the computing device 40 for communication interference signals can be referred to the description of the technical solution of the intelligent identification method for communication interference signals. The embodiment of the present invention will not be elaborated on this.

[0124] It should be noted that the technical solutions described in the embodiments of the present invention can be combined arbitrarily without conflict.

[0125] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for intelligently identifying communication interference signals, characterized in that: The method comprises: Divide the acquired spectrum data of the communication interference signal into a training set and a validation set; A recognition model for communication interference signals is constructed using a deep convolutional neural network (DCNN) that can estimate confidence scores and calculate recognition probabilities. Using the training set to train the recognition model; The trained recognition model is used to recognize the verification set, and a determination threshold for determining an unknown interference signal is set according to the distribution of data confidence scores of the verification set in the recognition result; Using the trained recognition model to identify the communication interference signal to be identified, and obtaining the confidence score and probability distribution of the communication interference signal to be identified; If the confidence score corresponding to the communication interference signal to be identified is less than the determination threshold, the communication interference signal to be identified is determined to be unknown interference; otherwise, the communication interference signal to be identified is determined to be the communication interference signal with the highest probability; The method of using a DCNN capable of estimating confidence scores and calculating recognition probabilities to construct a recognition model for communication interference signals includes: Based on the recognition requirements, determine the nodes of the input and output layers, the number of convolutional layers, the number of convolutional kernels, the activation function, the number of pooling layers, and the number of fully connected layers of the DCNN and initialize them; The DCNN is divided into a feature extraction part, a feature integration part and a result processing part; wherein the feature extraction part includes a convolution layer and a pooling layer; the feature integration part includes two fully connected layers; the result processing part includes two fully connected layers and the two fully connected layers respectively process the results output by the feature integration part to obtain the matching probability of the interference signal to be identified on each known interference signal and the confidence score of the interference signal to be identified; The step of training the recognition model using the training set includes: Batch processing is adopted for the training data, and the amount of data inputted into the recognition model from the training set is set each time; Set the data sampling method to random sampling; The task loss for recognition probability calculation is defined based on the following formula and confidence estimate loss for confidence score estimates : in, represents the corrected communication interference signal identification probability, represents the original predicted communication interference signal recognition probability obtained by the recognition probability calculation unit in the recognition model, represents the target probability distribution, Indicates the number of known communication interference signal types, C A confidence score output by a confidence score estimation unit in the recognition model; The task loss calculated based on the recognition probability and confidence estimate loss for confidence score estimates The loss function of the recognition model is determined as follows: : Among them, the budget hyperparameter is introduced Represents the confidence loss allowed by the recognition model; introduces hyperparameters is used to balance the task loss and the confidence estimation loss, and In the model, the dynamic changes Increase ,when When reduced ; Data is extracted from the randomly sampled data set according to the set parameters and input into the DCNN, and training is performed based on the set SDG optimization algorithm according to the set number of iterations to obtain a recognition model for the communication interference signal that has been trained.

2. The method according to claim 1, characterized in that The step of dividing the acquired spectrum data of the communication interference signal into a training set and a verification set includes: Collecting data of the I channel and Q channel of the acquired communication interference signal; Performing discrete Fourier transform on the data of the I channel and the Q channel to obtain spectrum data of the acquired communication interference signal; Randomly extract 80% of the data from each type of communication interference signal data set in the acquired spectrum data of the communication interference signal as the training set, and use the remaining 20% ​​of the data as the verification set.

3. The method according to claim 1, characterized in that The feature extraction part has a total of 13 layers, the number of nodes in the input layer is 150528, the number of nodes in the output layer is 25088, the number of convolutional layers is 8, and the number of pooling layers is 5; the number of neurons in the two fully connected layers of the feature integration part are 25088 and 512 respectively; the number of neurons in the two fully connected layers of the result processing part is 128.

4. The method according to claim 1, characterized in that: The result processing part includes a confidence score estimation unit and an identification probability calculation unit; wherein each unit corresponds to a fully connected layer; the identification probability calculation unit is used to obtain the probability that the interference signal to be identified is identified as each known interference signal after passing the result output by the feature integration part through the corresponding fully connected layer and Softmax layer; the confidence score estimation unit is used to obtain the reliability of the result of the identification probability of the interference signal to be identified after passing the result output by the feature integration part through the corresponding fully connected layer, and then the reliability is processed by the sigmoid activation function to obtain a number between 0 and 1 as the confidence score of the interference signal to be identified.

5. An intelligent identification device for communication interference signals, characterized in that: The intelligent recognition device comprises: a division part, a construction part, a training part, a setting part and a recognition part: Wherein, the division part is configured to divide the acquired spectrum data of the communication interference signal into a training set and a verification set; The construction part is configured to construct a recognition model of the communication interference signal using a deep convolutional neural network DCNN capable of estimating confidence scores and calculating recognition probabilities; The training part is configured to train the recognition model using the training set; The setting part is configured to use the trained recognition model to recognize the verification set, and set a determination threshold for determining an unknown interference signal according to the distribution of data confidence scores of the verification set in the recognition result; The identification part is configured to identify the communication interference signal to be identified by the trained identification model, and obtain the confidence score and probability distribution of the communication interference signal to be identified; and, if the confidence score corresponding to the communication interference signal to be identified is less than the determination threshold, the communication interference signal to be identified is determined to be unknown interference; otherwise, the communication interference signal to be identified is determined to be the communication interference signal with the highest probability; Wherein, the construction part is configured as follows: Based on the recognition requirements, determine the nodes of the input and output layers, the number of convolutional layers, the number of convolutional kernels, the activation function, the number of pooling layers, and the number of fully connected layers of the DCNN and initialize them; The DCNN is divided into a feature extraction part, a feature integration part and a result processing part; wherein the feature extraction part includes a convolution layer and a pooling layer; the feature integration part includes two fully connected layers; the result processing part includes two fully connected layers and the two fully connected layers respectively process the results output by the feature integration part to obtain the matching probability of the interference signal to be identified on each known interference signal and the confidence score of the interference signal to be identified; The training part is configured as follows: Batch processing is adopted for the training data, and the amount of data inputted into the recognition model from the training set is set each time; Set the data sampling method to random sampling; The task loss for recognition probability calculation is defined based on the following formula and confidence estimate loss for confidence score estimates : in, represents the corrected communication interference signal identification probability, represents the original predicted communication interference signal recognition probability obtained by the recognition probability calculation unit in the recognition model, represents the target probability distribution, represents the number of known communication interference signal types, and C is the confidence score output by the confidence score estimation unit in the recognition model; The task loss calculated based on the recognition probability and confidence estimate loss for confidence score estimates The loss function of the recognition model is determined as follows: : Among them, the budget hyperparameter is introduced Represents the confidence loss allowed by the recognition model; introduces hyperparameters is used to balance the task loss and the confidence estimation loss, and In the model, the dynamic changes Increase ,when When reduced ; Data is extracted from the randomly sampled data set according to the set parameters and input into the DCNN, and training is performed based on the set SDG optimization algorithm according to the set number of iterations to obtain a recognition model for the communication interference signal that has been trained.

6. A computer storage medium, characterized in that: The computer storage medium stores a program for intelligently identifying communication interference signals, and when the program for intelligently identifying communication interference signals is executed by at least one processor, the steps of the method for intelligently identifying communication interference signals according to any one of claims 1 to 4 are implemented.

Images