A Cognitive Radio Malicious User Identification Method Based on Graph Neural Network

Abstract

The present invention relates to the field of radio communication technologies. Specifically, it is a method for identifying malicious users in cognitive radio based on graph neural networks. The graph neural network is introduced into the identification of malicious users in a centralized multi-user cognitive radio. First, the degree of abnormality of each cognitive user in the cognitive radio network is obtained through the graph neural network. Then, the fusion center combines the global decision and the degree of conflict between each cognitive user and its neighbor users to establish an abnormality model. The malicious user identification method of the present invention can solve the problem of misjudgment caused by malicious users launching Byzantine attacks when judging the presence of primary users in cognitive radio, and can avoid malicious users preempting the frequency bands of primary users, enabling the frequency spectrum to be used by normal users.

Description

Technical Field

[0001] The present invention relates to the field of radio communication technologies, and specifically, to a method for identifying malicious users in cognitive radio based on graph neural networks. Background Art

[0002] In recent years, due to the sharp increase in radio devices and the static management of the spectrum, there is a shortage of available radio spectrum. One of the solutions to this problem is to use cognitive radio technology. Cognitive radio networks can use idle resources without affecting primary users, authorize frequency bands, improve the utilization rate of spectrum resources, and meet the needs of more wireless users. Essentially, cognitive radio networks are wireless communication networks with cognitive characteristics, aiming to solve the problem of scarce wireless spectrum resources. Cognitive radio networks can observe the surrounding wireless network environment, use environmental perception to obtain relevant spectrum usage information, process and learn the obtained information for decision-making, dynamically access available spectra, and finally reconstruct the network to adapt to the dynamic cognitive radio network environment and maximize the use of the spectrum. Users in cognitive radio networks are called cognitive users. Cognitive users borrow channels that originally belong to primary users. Therefore, once a primary user uses the channel occupied by a cognitive user, the cognitive user must immediately vacate the channel and simultaneously search for other available channels.

[0003] Cognitive radio encompasses multiple aspects of technologies. Among them, spectrum sensing is a key technology for cognitive radio users to wait for opportunities to use primary user spectra and transmit data. At the same time, due to reasons such as equipment failures, channel shadow fading, and noise, individual cognitive users often misjudge the occupancy of channels by primary users. However, collaborative spectrum sensing among cognitive users can solve the above problems and effectively improve the efficiency and reliability of spectrum observation.

[0004] Although collaborative sensing improves the sensing accuracy, it exposes the system to the danger of possible malicious users: not performing any sensing operations and randomly reporting sensing results to save time and energy; or reporting that the channel is busy to the fusion center, causing the fusion center to make incorrect judgments to achieve the purpose of occupying the spectrum, or launching a denial-of-service attack on the network. Summary of the Invention

[0005] The present invention provides a method for identifying malicious users in cognitive radio based on graph neural networks, which is used to identify and remove malicious users in cognitive radio networks. The present invention introduces the anomaly degree obtained by reconstructing the cognitive radio network using graph neural networks, fully considers the graph structure model of the cognitive radio network, and effectively detects malicious users in the cognitive radio network.

[0006] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0007] A method for identifying malicious users in cognitive radio based on graph neural network, comprising the following steps:

[0008] Step S1. In the cognitive radio network, each cognitive user uses energy detection within a certain period of time, makes a local decision, and uploads the local decision to the fusion center FC: Let E L represent the local detection result, and E G represent the detection result from the fusion center FC, where E L = +1 indicates that the local judgment is that the primary user exists, and E L = -1 indicates that the local judgment is that the primary user does not exist, and E G = +1 indicates that the fusion center FC judges that the primary user exists, and E G = -1 indicates that the fusion center FC judges that the primary user does not exist;

[0009] Step S2. The fusion center FC stores the local decision results of each cognitive user within a certain period of time in a first-in, first-out manner. At the same time, the fusion center FC obtains the geographical locations of the cognitive users through the geographical location database, and connects the nearby cognitive users to form a graph;

[0010] Step S3. The fusion center takes the local decision results of each user stored in Step S2 and the geographical location information of the cognitive users as the input information of the graph neural network. The graph neural network autoencoder reconstructs the input and takes the reconstruction error as the output. The goal of the graph neural network is to minimize the reconstruction error, and stop when the number of iterations exceeds a certain limit;

[0011] Step S4. Perform min-max normalization on the reconstruction error output by the graph neural network in Step S3 as the anomaly degree of each cognitive user in the current cognitive radio network. In the present invention, an anomaly degree is used to preliminarily judge whether a cognitive user is a malicious user, as shown in the following formula:

[0012]

[0013] where rec is the reconstruction error output by the graph neural network, is the preliminary anomaly degree;

[0014] Step S5. Set up a cognitive user list, set the value to 1 in the corresponding list for those with an anomaly degree exceeding the threshold, indicating that the user is a malicious user, and set it to 0 for those whose anomaly degree does not exceed the threshold, indicating that the user is a normal honest user, as shown in the following formula:

[0015]

[0016] where S is the established cognitive user list and λ is the threshold;

[0017] Step S6: Calculate the similarity between the local decisions made by the cognitive user within a certain time period and the global decisions made by the fusion center, using the following formula:

[0018]

[0019] where R is the number of times the local decision is the same as the global decision within a certain time period, and T is the number of times the cognitive user detects the presence of the primary user within a certain time;

[0020] Step S7: Normalize the similarity using the normalization formula:

[0021]

[0022] Step S8: Calculate the conflict degree between the reconstruction error and the global similarity degree d, using the following formula:

[0023] c = rec×(1 - d)+(1 - rec)×d

[0024] where c refers to the conflict degree;

[0025] Step S9: For the cognitive user SU with a relatively serious conflict degree i , analyze the proportion of malicious users among its neighbor users. For those with a proportion of malicious users exceeding half, remove them from the malicious user set, and for those with a proportion less than half, maintain the original judgment.

[0026] Advantages of the present invention: On the one hand, by introducing a graph neural network to detect malicious users in cognitive users, the graph neural network can better represent the correlation information among cognitive users, improve the detection performance of malicious users in the cognitive radio network detection network, and improve the accuracy of the fusion center FC when fusing local detection results; on the other hand, fully considering the environment where the cognitive user is located and combining the situation of neighbor users, it makes a beneficial supplement to the output of the graph neural network, effectively improving the detection performance of the cognitive radio system. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a schematic diagram of the cognitive radio system architecture provided by the present invention.

[0028] Figure 2 It is a schematic diagram of the neural network structure in the present invention.

[0029] Figure 3 It is a flowchart of the algorithm in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0030] To deepen the understanding of the present invention, the present invention will be further described in detail below in conjunction with the drawings and embodiments. The embodiments are only used to explain the present invention and do not limit the protection scope of the present invention.

[0031] Example: As Figure 1 shown, the cognitive radio network on which the present invention is based includes a primary user (PU), a fusion center (FC), and several cognitive users (SU). The cognitive users SU are divided into honest users (HU) and malicious users (MU). Generally, cognitive radio can be summarized as the following points: 1) The primary user PU has the highest priority and is the true owner of the spectrum temporarily occupied by cognitive users. Therefore, cognitive radio must first ensure the normal operation of the primary user PU, and the primary user is not restricted by cognitive users under any circumstances; 2) Cognitive users usually continuously sense the spectrum situation of the primary user according to a certain algorithm and only attempt to access when the primary user does not occupy the spectrum; 3) In centralized spectrum sensing, the fusion center is responsible for fusing the local decisions of each cognitive user in the cognitive network to determine whether the primary user occupies the spectrum; 4) The cognitive radio system is only effective for cognitive users, and the primary user does not need to consider the cognitive radio system; 5) There is a two-way link between cognitive users and the fusion center, and two-way communication can be carried out; 6) In the invention, the fusion center integrates the information of each cognitive user through a graph neural network, as Figure 2 shown.

[0032] See Figure 3 shown, the method for identifying malicious users of cognitive radio based on graph neural network in the present invention has the following operation steps:

[0033] Step S1: The cognitive user SU i performs spectrum sensing by using energy detection, accumulates the energy of the primary user signal within a certain time, and compares the accumulated energy with a predetermined threshold to make a local decision: if the accumulated energy exceeds the threshold, it is determined that the primary user exists, and it is represented by E L = +1; otherwise, it is determined that the primary user does not exist, and it is represented by E L = -1. Since the fusion center also fuses the detection information of other cognitive users at the same time, similar to the local detection result representation, E G = +1 indicates that the central node determines that the primary user exists, and E G = -1 indicates that the central node determines that the primary user does not exist;

[0034] Step S2: The several local decisions obtained by the cognitive user in Step S1 form a sequence P i = [p i (1), p i (2),..., p i (j),...]. A new sequence P i ′ = [p i (k - M), p i (k - M + 1),..., p i(k - 1), where k represents the current sample period. The function of this sliding window is to store the local decision numbers of cognitive users in the most recent M sample periods in a FIFO manner and continuously update as the sample period changes. Meanwhile, the fusion center constructs a graph based on the distance factors of each user in the cognitive network. If the distance between cognitive users is relatively close, the two are connected;

[0035] Step S3, refer to Figure 3 As shown, the cognitive user takes the sequence P i ′ obtained in Step S2 and the cognitive network graph as the input vectors of the graph neural network, takes the reconstruction error of the graph neural network in the current sample period as the actual output, and takes the original data in the current sample period as the expected output. When the graph neural network iterates a certain number of times, the graph neural network stops working.

[0036] Step S4, the fusion center FC normalizes the reconstruction error input into the graph neural network and obtained in Step S3,

[0037]

[0038] where rec is the reconstruction error output by the graph neural network, is the preliminary degree of abnormality;

[0039] Step S5, establish a cognitive user list. The fusion center FC sets the value of the cognitive user whose abnormality degree exceeds the threshold to 1 in the corresponding list, indicating that the cognitive user is a malicious user, and sets the value of the cognitive user whose abnormality degree does not exceed the threshold to 0, indicating that the cognitive user is a normal and honest user.

[0040]

[0041] where S is the established cognitive user list and λ is the threshold;

[0042] Step S6, after the detection period ends, the fusion center compares the local detection result E L of the cognitive user and the global decision result E G . The comparison results are represented by R and T as follows: For the j - th detection period, if E L = E G , record R incremented by 1; if E L ≠ E G , record R incremented by 0. From these data, the cognitive user can obtain the similarity degree between itself and the global decision.

[0043]

[0044] where Indicates the preliminary result of the similarity degree, where R is the number of times that the local decision is the same as the global decision within a certain period of time, and T is the number of times that the cognitive user detects the presence of the primary user within a certain period of time;

[0045] Step S7: Normalize the similarity degree through the normalization formula.

[0046]

[0047] Step S8: Calculate the conflict degree between the reconstruction error and the global similarity degree d

[0048] c = rec×(1 - d)+(1 - rec)×d

[0049] Step S9: After the above steps are completed, the fusion center FC determines whether it is necessary to correct the result obtained by the graph neural network according to the following conditions: (1) If the conflict degree c value of a cognitive user is too large, then compare whether there are too many malicious users among the neighbor users of this cognitive user. If so, regard this cognitive user as a normal user. If not, no correction is made, and it returns to step S2 in the next cycle.

[0050] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and the descriptions in the specification only illustrate the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for identifying malicious users in cognitive radio based on graph neural network, characterized in that, It includes the following steps: Step S1: In a cognitive radio network, each cognitive user uses energy detection within a certain period of time, makes a local decision, and uploads the local decision result to the fusion center FC; Step S2: The fusion center FC stores the local decision results of each cognitive user within a certain period of time in a first-in, first-out manner. At the same time, the fusion center FC obtains the geographical locations of the cognitive users through a geographical location database and connects the neighboring cognitive users to form a cognitive network graph; Step S3: The fusion center takes the local decision results of each user stored in Step S2 and the cognitive network graph of the cognitive users as the input information of the graph neural network. The graph neural network autoencoder reconstructs the input and takes the reconstruction error as the output. The goal of the graph neural network is to minimize the reconstruction error and stop when the number of iterations exceeds a certain limit; Step S4: Perform max-min normalization on the reconstruction error output by the graph neural network in Step S3 as the anomaly degree of each cognitive user in the current cognitive radio network; Step S5: Set up a cognitive user list. For those with an anomaly degree exceeding the threshold, set the value to 1 in the corresponding list, indicating that the user is a malicious user. For those with an anomaly degree not exceeding the threshold, set the value to 0, indicating that the user is a normal and honest user; Step S6: Calculate the similarity between the local decision made by a cognitive user within a certain period of time and the global decision made by the fusion center; Step S7: Normalize the similarity through a normalization formula to obtain the global similarity degree; Step S8: Calculate the conflict degree between the reconstruction error and the global similarity degree; Step S9: For cognitive users with a serious conflict degree, analyze the proportion of malicious users among their neighbor users. For those with a proportion of malicious users exceeding half, remove them from the malicious user set, and for those with a proportion less than half, maintain the original judgment.

2. The method for identifying malicious users of cognitive radio based on graph neural network according to claim 1, characterized in that, In step S1, use E L to represent the local detection result, and E G to represent the detection result from the fusion center FC, where E L = +1 indicates that the local judgment is that the primary user exists, and E L = -1 indicates that the local judgment is that the primary user does not exist, and E G = +1 indicates that the fusion center FC judges that the primary user exists, and E G = -1 indicates that the fusion center FC judges that the primary user does not exist.

3. The cognitive radio malicious user identification method based on a graph neural network according to claim 2, characterized in that, In Step S4, initially judge whether a cognitive user is a malicious user based on the anomaly degree. The following formula: Among them, rec is the reconstruction error output by the graph neural network, which is the preliminary degree of abnormality.

4. The method for identifying malicious users of cognitive radio based on graph neural network according to claim 3, characterized in that, In Step S5, set up a cognitive user list. For those with an anomaly degree exceeding the threshold, set the value to 1 in the corresponding list, indicating that the user is a malicious user. For those with an anomaly degree not exceeding the threshold, set the value to 0, indicating that the user is a normal and honest user. The following formula: Among them, S is the established cognitive user list, and λ is the threshold.

5. The cognitive radio malicious user identification method based on a graph neural network according to claim 4, wherein In Step S6, calculate the similarity between the local decision made by a cognitive user within a certain period of time and the global decision made by the fusion center. The following formula: Among them, R is the number of times when the local decision and the global decision are the same within a certain period of time, and T is the number of times when the cognitive user detects whether the primary user exists within a certain period of time.

6. The method for identifying malicious users of cognitive radio based on graph neural network according to claim 5, wherein In Step S7, normalize the similarity through a normalization formula:

7. The method for identifying malicious users of cognitive radio based on graph neural network according to claim 6, wherein In Step S8, calculate the conflict degree between the reconstruction error and the global similarity degree d. The following formula: c = rec×(1 - d)+(1 - rec)×d Among them, c refers to the conflict degree.

Images