SSDF attack detection method for unmanned aerial vehicle spectrum sensing

Abstract

The invention belongs to the technical field of unmanned aerial vehicle communication, and particularly relates to an unmanned aerial vehicle spectrum sensing-oriented SSDF attack detection method, which comprises the following steps of: 1, establishing a cognitive unmanned aerial vehicle network based on centralized cooperative spectrum sensing; 2, each unmanned aerial vehicle performs spectrum sensing independently; 3, analyzing the influence of transmission errors on malicious user detection; and 4, identifying malicious users according to the credibility. According to the method, a cognitive unmanned aerial vehicle network model is constructed, the spectrum sensing probability of each unmanned aerial vehicle is independently calculated according to the coordinates of each unmanned aerial vehicle, the probability of transmission errors of SU and FC is also calculated, then, a credibility is allocated to each SU according to the probability of reporting an accurate sensing result by each SU, the SU with the credibility not exceeding a credibility threshold value is regarded as a malicious user, and the malicious user is regarded as a malicious user. And the unmanned aerial vehicles are prohibited from participating in cooperative spectrum sensing, so that the unmanned aerial vehicles have better spectrum sensing performance when the unmanned aerial vehicles are distributed dispersedly or are close to the PU, and the probability of mistaking legal users as malicious users is lower.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned aerial vehicle communication, and in particular relates to an SSDF attack detection method for unmanned aerial vehicle spectrum sensing. Background Art

[0002] Drones have broad application prospects. However, with the increase in development and application, the traditional static spectrum allocation mode cannot adapt to the fast-moving and dynamically changing communication needs of drones, and the problem of spectrum shortage in drones has become increasingly prominent. Cognitive radio technology is currently the most direct and effective means to solve the shortage of spectrum resources in the field of wireless communications. Applying it to drone networks can adaptively adjust internal communication parameters or communication mechanisms by sensing the occupancy of nearby spectrum to achieve automatic perception and call unoccupied spectrum, which can effectively alleviate the problem of spectrum resource shortage in drones. Drone spectrum sensing can effectively alleviate the problem of tight utilization of drone spectrum resources, which is of great significance for the promotion and utilization of drone technology. However, the openness of wireless channels brings many security threats to cognitive drone networks. For example, dynamic access to spectrum under the framework of collaborative perception provides opportunities for malicious users to participate in perception. At present, there is little research on the security of drone spectrum sensing. In addition, in the process of spectrum sensing, drone networks are often regarded as a unified entity, and the distance differences between drones are often ignored.

[0003] In a collaborative spectrum sensing network, malicious users can send erroneous local sensing results to the fusion center (FC) or other secondary users (SU) to interfere with the transmission of the primary user (PU) to occupy idle spectrum, greatly reducing the efficiency of collaborative sensing. This attack launched by malicious users is usually called spectrum sensing data falsification (SSDF) attack. Since SSDF attacks have multiple attack modes and complex and flexible attack strategies, it is difficult to effectively detect and defend against them. Although domestic and foreign scholars have conducted extensive research on spectrum sensing detection of SSDF attacks, most of them are based on conventional spectrum sensing, and few have studied how to apply SSDF attack detection methods to drone spectrum sensing. In addition, most existing works assume that the transmission between SU ​​and FC is error-free, which is difficult to achieve in practice.

[0004] Therefore, in-depth research on SSDF attack detection in cognitive UAV networks under the condition that transmission errors may occur has important theoretical and practical significance. Summary of the invention

[0005] The purpose of the present invention is to provide a SSDF attack detection method for UAV spectrum sensing, which has better spectrum sensing performance when the UAVs are dispersed or close to the PU, and has a lower probability of mistaking legitimate users for malicious users.

[0006] The technical solution adopted by the present invention is as follows:

[0007] A SSDF attack detection method for UAV spectrum sensing includes the following steps:

[0008] Step 1: Establish a cognitive drone network based on centralized collaborative spectrum sensing;

[0009] Step 2: Each drone performs spectrum sensing individually;

[0010] Step 3: Analyze the impact of transmission errors on malicious user detection;

[0011] Step 4: Identify malicious users based on reputation.

[0012] Furthermore, the cognitive drone network based on centralized cooperative spectrum sensing in step 1 includes malicious users, and the entire cognitive drone network includes N drones as SUs, assuming that the number of malicious users is K, and there is also a PU and an FC on the ground responsible for spectrum management of all SUs;

[0013] Malicious users will disguise themselves as legitimate users to launch SSDF attacks, and deliberately send forged spectrum sensing data to FC with a certain probability, thereby misleading FC to make wrong spectrum judgment results; symbol SU i To represent the i-th SU (i=1,2,...,N) in the network.

[0014] Furthermore, the spectrum sensing of the cognitive drone network is divided into three stages:

[0015] Phase 1: Local spectrum sensing: The legitimate users and malicious users participating in the collaboration perform independent local spectrum sensing on the authorized frequency band of the PU selected by the FC;

[0016] Phase 2, Report / SSDF attack: Legitimate users send their perception data or perception results to FC; malicious users launch SSDF attacks with a certain probability and send wrong perception data or perception results to FC;

[0017] Phase 3, decision reporting: FC uses the information transmitted by SU to obtain the FC's decision statistics according to appropriate data fusion rules, and judges the usage status of the frequency band based on the decision statistics. After making the final decision, the result is notified to SU.

[0018] Furthermore, the local spectrum sensing of each UAV in the step 2 adopts a sensing technology based on energy detection.

[0019] Furthermore, the step 2 includes the following steps:

[0020] Let H0 represent PU not using spectrum, H1 represent PU occupying spectrum; i The signal of receiving PU can be expressed as:

[0021]

[0022] Among them, x i (t) is SU i Receive the signal from PU; i (t) has a mean of 0 and a variance of The additive Gaussian white noise of ; s(t) is the signal of PU;

[0023] Energy detection statistic Y i It is expressed as:

[0024]

[0025] Where N0 is the single-sided noise power density spectrum, T w is the recording time window, Y i satisfy:

[0026]

[0027] in, and They represent central and noncentral chi-square distributions, respectively, with degrees of freedom of 2ω. The noncentral parameter of the latter is ω is the time-bandwidth product, From PU to SU i The signal-to-noise ratio of the received signal;

[0028] SU i The detection probability and false alarm probability can be expressed as:

[0029]

[0030] Where λ is the energy detection threshold, Γ(·) is the gamma function, and Γ(.,.) is the incomplete gamma function. The expression is:

[0031]

[0032] where ρ PU is the transmit power of PU; PU and SU i The path gain between is expressed as:

[0033]

[0034] Where κ and μ are path gain constant and path gain index respectively; d P d i PU and SU respectively i 3D position.

[0035] Furthermore, the step three includes the following steps:

[0036] Assume SU i The transmission with FC will not go wrong, and the legitimate user SU i The probability of reporting a correct perception result is:

[0037]

[0038] Assume SU i The transmission with FC may be erroneous. Given the BPSK modulation in Rayleigh fading environment, SU i The probability that the report to the FC is in error is:

[0039]

[0040] in, YesSU i The average signal-to-noise ratio of the bit report between FC is expressed as:

[0041]

[0042] Among them, ρ i For SU i The transmission power; h i It is FC and SU i The path gain between is expressed as:

[0043]

[0044] Among them, d F is the 3D position of FC;

[0045] based on and Legitimate user SU i In the case of possible errors in the transmission with FC, the probability of correctly reporting the perception result is:

[0046]

[0047] Here, the function F(x,y) is defined as F(x,y)=x(1-y)+(1-x)y.

[0048] set up and They are malicious users SU j The probability of reporting the correct perception result when the transmission is error-free and error-prone; similar to formula (12), we have:

[0049]

[0050] Suppose the probability that the result reported by a malicious user is opposite to the local perception is P mal , if the malicious user takes the independent attack approach, then:

[0051]

[0052] If the malicious user adopts the "L out of K" collaborative attack strategy, then:

[0053]

[0054] Furthermore, the step 4 includes the following steps:

[0055] Set SU i The credibility is r i , is the probability of reporting an accurate perception result:

[0056]

[0057] where τ i T w SU i The number of times the local report is consistent with the final decision of the FC;

[0058] Assume η i For SU i The credibility threshold, P i For SU i The probability that the local report of SU is consistent with the final decision of FC; if SU i Credibility i Greater than η i , then it is judged as a legitimate user, If SU i Credibility i Less than or equal to η i , then it is judged as a malicious user,

[0059] Assume z i (t) is the t time slot SU i The correctness of the local perception result, if SU i If the correct result is reported, then z i (t) = 1; otherwise, z i(t) = 0. Therefore P(z i (t) = 1) = P i , P(z i (t) = 0) = 1-P i ; Therefore z i (t) obeys Bernoulli distribution. According to formula (17), we have:

[0060]

[0061] Among them, t0 is the start time of the recording time window; due to z i (t) follows a Bernoulli distribution, if T w satisfy:

[0062] T w P i ≥10and T w (1-P i )≥10 (19)

[0063] Then r i Normal distribution Before starting spectrum sensing, it is necessary to find T that satisfies formula (19) w ,Since FC has no prior information about which SUs are malicious, it is reasonable to assume that all SUs are legitimate at the beginning, Therefore, T w Should meet:

[0064]

[0065] in, is the set of all SU. Obviously, such T w satisfy

[0066] For legitimate users SU i ,because Assume r i The probability density function is f H,i (x), the expression is:

[0067]

[0068] Legitimate user SU i Probability of being mistaken for a malicious user The expression is:

[0069]

[0070] Among them, erfc(x) is the complementary error function, expressed as:

[0071]

[0072] The technical effects achieved by the present invention are:

[0073] (1) The SSDF attack detection method for UAV spectrum perception of the present invention calculates the perception probability of each UAV based on its coordinates, detects the SSDF attack suffered by the cognitive UAV network when transmission errors are possible, and has better spectrum perception performance when the UAVs are dispersed or close to the PU. Compared with the method that does not consider transmission errors, the method proposed by the present invention has a lower probability of mistaking legitimate users for malicious users. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 is a schematic diagram of a cognitive drone network model used in the simulation of the present invention;

[0075] Figure 2 It is a schematic diagram of the projection of three different distributions of drone groups and PU and FC on the XOY plane in the simulation experiment of the present invention;

[0076] Figure 3 It is a schematic diagram of the influence of three different distributions of drone groups on unified probability and individual probability in the simulation experiment of the present invention;

[0077] Figure 4 It is a schematic diagram of the projection of the drone group, PU, ​​and FC on the XOY plane when the PU is far away from the drone group in the simulation experiment of the present invention;

[0078] Figure 5 is the different PU transmission power ρ in the simulation experiment of the present invention PU Below, a schematic diagram of the effect of PU away from the drone group on the unified probability and individual probability;

[0079] Figure 6 is the probability of different transmission errors in the simulation experiment of the present invention The comparison method obtained Schematic diagram of the impact of

[0080] Figure 7 The invention is obtained by the invention method in the simulation experiment Compared with the method obtained Schematic diagram of the comparison. DETAILED DESCRIPTION

[0081] In order to make the purpose and advantages of the present invention more clearly understood, the present invention is specifically described below in conjunction with embodiments. It should be understood that the following text is only used to describe one or several specific embodiments of the present invention, and does not strictly limit the scope of protection of the specific claims of the present invention.

[0082] The present invention provides a credibility-based UAV spectrum sensing method for detecting SSDF attacks, which calculates the perception probability of each UAV individually according to the coordinates, and detects SSDF attacks suffered by the cognitive UAV network in the case of possible transmission errors. Please refer to the following embodiments for details.

[0083] Embodiment 1:

[0084] like Figure 1-7 As shown, a SSDF attack detection method for UAV spectrum sensing includes the following steps:

[0085] Step 1: Establish a cognitive drone network based on centralized collaborative spectrum sensing;

[0086] The step 1 establishes a cognitive UAV network based on centralized cooperative spectrum sensing, including malicious users. The entire cognitive UAV network contains N UAVs as SUs. Assume that the number of malicious users is K. There is also a PU and a FC on the ground responsible for spectrum management of all SUs. Malicious users will disguise themselves as legitimate users to launch SSDF attacks, and deliberately send forged spectrum sensing data to FC with a certain probability, thereby misleading FC to make incorrect spectrum judgment results. For the convenience of representation, the symbol SU is used. i To represent the i-th SU (i=1,2,...,N) in the network.

[0087] The spectrum sensing of the cognitive drone network can be divided into three stages:

[0088] Phase 1: Local spectrum sensing: The legitimate users and malicious users participating in the collaboration perform independent local spectrum sensing on the authorized frequency band of the PU selected by the FC.

[0089] Phase 2, Report / SSDF attack: Legitimate users send their perception data or perception results to FC; malicious users launch SSDF attacks with a certain probability and send incorrect perception data or perception results to FC.

[0090] Phase 3, decision reporting: FC uses the information transmitted by SU to obtain the FC's decision statistics according to appropriate data fusion rules, and judges the usage status of the frequency band based on the decision statistics. After making the final decision, the result is notified to SU.

[0091] Step 2: Each drone performs spectrum sensing individually;

[0092] The second step calculates the spectrum sensing probability of each drone in the network according to their coordinates, and considering the small error of distance helps to improve the performance of spectrum sensing. The local spectrum sensing of each drone adopts the sensing technology based on energy detection.

[0093] The step 2 comprises the following steps:

[0094] Let H0 represent the spectrum not used by PU, and H1 represent the spectrum occupied by PU. i The signal of receiving PU can be expressed as:

[0095]

[0096] Among them, x i (t) is SU i Receive the signal from PU; i (t) has a mean of 0 and a variance of is the additive white Gaussian noise; s(t) is the signal of PU.

[0097] Energy detection statistic Y i It is expressed as:

[0098]

[0099] Where N0 is the single-sided noise power density spectrum, T w is the recording time window, Y i satisfy:

[0100]

[0101] in, and They represent central and noncentral chi-square distributions, respectively, with degrees of freedom of 2ω. The noncentral parameter of the latter is ω is the time-bandwidth product, From PU to SU i The signal-to-noise ratio of the received signal.

[0102] SU i The detection probability and false alarm probability can be expressed as:

[0103]

[0104] Where λ is the energy detection threshold, Γ(·) is the gamma function, and Γ(.,.) is the incomplete gamma function. The expression is:

[0105]

[0106] where ρ PU is the transmit power of PU; PU and SU i The path gain between is expressed as:

[0107]

[0108] Where κ and μ are path gain constant and path gain index respectively; d P d i PU and SU respectively i 3D position.

[0109] Step 3: Analyze the impact of transmission errors on malicious user detection;

[0110] Specifically, the step three includes the following steps:

[0111] Assume SU i The transmission with FC will not go wrong, so the legitimate user SU i The probability of reporting a correct perception result is:

[0112]

[0113] Assume SU i The transmission with FC may be erroneous. Given the BPSK modulation in Rayleigh fading environment, SU i The probability that the report to the FC is in error is:

[0114]

[0115] in, YesSU i The average signal-to-noise ratio of the bit report between FC is expressed as:

[0116]

[0117] Among them, ρ i For SU i The transmission power; h i It is FC and SU i The path gain between is expressed as:

[0118]

[0119] Among them, d F is the 3D position of FC.

[0120] based on and Legitimate user SU i In the case of possible errors in the transmission with FC, the probability of correctly reporting the perception result is:

[0121]

[0122] Here, the function F(x,y) is defined as F(x,y)=x(1-y)+(1-x)y.

[0123] set up and They are malicious users SU j The probability of reporting the correct perception result in the case where the transmission will not go wrong and the transmission may go wrong. Similar to formula (12), we have:

[0124]

[0125] Suppose the probability that the result reported by a malicious user is opposite to the local perception is P mal , if the malicious user takes the independent attack approach, then:

[0126]

[0127] If the malicious user adopts the "L out of K" collaborative attack strategy, then:

[0128]

[0129] Step 4: Identify malicious users based on reputation;

[0130] Step 4 relies on the concept of “reputation” to detect SSDF attacks by malicious users.

[0131] Specifically, the step 4 includes the following steps:

[0132] Set SU i The credibility is r i , is the probability of reporting an accurate perception result:

[0133]

[0134] where τ i T w SU i The number of times the local report is consistent with the final decision of FC.

[0135] Assume η i For SU i The credibility threshold, P i For SU i The probability that the local report of SU is consistent with the final decision of FC. i Credibility r i is greater than η i , then it is judged as a legitimate user, If SU i Credibility i Less than or equal to n i , then it is judged as a malicious user,

[0136] Assume z i(t) is the t time slot SU i The correctness of the local perception result, if SU i If the correct result is reported, then z i (t) = 1; otherwise, z i (t) = 0. Therefore P(z i (t) = 1) = P i , P(z i (t) = 0) = 1-P i Therefore, z i (t) obeys Bernoulli distribution. According to formula (17), we have:

[0137]

[0138] Among them, t0 is the start time of the recording time window. i (t) follows a Bernoulli distribution, if T w satisfy:

[0139] T w P i ≥10and T w (1-P i )≥10 (19)

[0140] Then r i Normal distribution Before starting spectrum sensing, it is necessary to find T that satisfies formula (19) w ,Since FC has no prior information about which SUs are malicious, it is reasonable to assume that all SUs are legitimate at the beginning,P i = Therefore, T w Should meet:

[0141]

[0142] in, is the set of all SU. Obviously, such T w satisfy

[0143] For legitimate users SU i ,because Assume r i The probability density function is f H,i (x), the expression is:

[0144]

[0145] Legitimate user SU i Probability of being mistaken for a malicious user The expression is:

[0146]

[0147] Among them, erfc(x) is the complementary error function, expressed as:

[0148]

[0149] Embodiment 2:

[0150] In order to compare with the SSDF attack detection method for drone spectrum sensing, this embodiment considers an SSDF attack detection method (hereinafter referred to as “comparison method”) in which the transmission of SU and FC is assumed to be error-free.

[0151] The contrast method divides the spectrum sensing process into a time window consisting of T sensing cycles. In a T sensing window, SU i The local decision can be made using u i =[u i [1],u i [2],…,u i [T]] represents. In the same time window, the final decision of FC is u0 = [μ0[1], u0[2], …, u0[T]]. Since FC knows that there may be attackers in the network, it assigns a reputation metric n to each user. i , represents the number of times the final decision differs from the local decision of a specific SU within the T time window. Therefore, n i The higher the value, the more likely SU i The less reliable the observations.

[0152] All legitimate users in the comparison method use the same false alarm probability and the detection probability All malicious users use the same false alarm probability and the detection probability The contrast method uses the “L out of-N” rule as the fusion rule. i The situation where the received observations are inconsistent with the final decision is represented as L (·) is an indicator function, and Defined as

[0153] P(s i [t]=1|H j )=P(u i [t]=0|H j )P(u0[t]=1|u i [t]=0,H j )+P(u i [t]=1|H j)P(u0[t]=0|u i [t]=1,H j ) (24)

[0154] where j ∈ {0, 1}. Usually, if the random variable V follows a binomial distribution with parameters ω and p, then the probability of obtaining exactly V successes in ω trials is

[0155]

[0156] In a given time window, there are K legitimate users and M malicious users participating in cooperative spectrum sensing. The FC uses the "L out of N" rule (where N = K + M is the total number of users participating in spectrum sensing in the network within the given time window) for data fusion.

[0157] For legitimate user SU i :

[0158]

[0159] where a = max(0, l - K + 1) and b = min(l, M)

[0160]

[0161] Therefore:

[0162]

[0163] If SU i is a legitimate user, then its credibility n i is defined as

[0164]

[0165] and

[0166] By isolating or disconnecting those users with credibility measures greater than a fixed threshold η (η < T) from the information fusion process, the SSDF attack can be countered. The probability that a legitimate user is isolated at the end of the time window T can be evaluated as:

[0167]

[0168] To better compare with the SSDF attack detection method for UAV-oriented spectrum sensing, assume that in the comparison method, each legitimate user calculates and each malicious user calculates and SUi The probability of an error in FC transmission is As shown in formula (9).

[0169] In the case that the transmission between SU ​​and FC does not go wrong, H j In this case, FC receives the legitimate user SU i Report i [t] The probability of the outcome is equal to P(u i [t]|H j )(j,u i [t]∈{0,1}).

[0170] based on and In the case where the transmission between SU ​​and FC may be erroneous, FC receives the legitimate user SU in case H1. i Report i The probability that [t] = 1 is:

[0171]

[0172] In case H0, FC receives the legitimate user SU i Report perception i The probability that [t] = 1 is:

[0173]

[0174] Similarly, and

[0175] For legitimate users SU i , assuming that its actual local perception result is u i [t]=1.

[0176] The transmission process has The probability of error is that FC receives the result u i [t] = 0, FC therefore derives the probability that u0[t] = 0 as:

[0177]

[0178] The transmission process has The probability of no error is that the result received by FC is u i [t] = 1, FC therefore derives the probability that u0[t] = 0 as:

[0179]

[0180] Therefore, in the case that the transmission between SU ​​and FC may be wrong, for the legitimate user SU i :

[0181]

[0182] Similarly,

[0183]

[0184] According to the above formula, the transmission error may occur.

[0185] This technical solution evaluates the accuracy improvement brought by independently calculating the spectrum sensing probability of each drone through simulation, and compares the SSDF attack detection method for drone spectrum sensing with the SSDF attack detection method that does not consider transmission errors (comparison method).

[0186] The simulation parameters are set as follows. The performance of the drones is the same and they are all at the same altitude. The drones are stationary, the 3D spatial coordinates of the FC are (0,0,0), and the PU can move from (2000,0,0) to (200,0,0) on the x-axis (unit: m). Noise variance Time-bandwidth product ω = 5, energy detection threshold λ = 14, PU transmission power ρ PU =0.03W, path gain constant κ = 1, path gain index μ = 3, SU i The transmission power ρ i =0.01W,SU i The credibility threshold η i =0.5. “Uniform probability” refers to the spectrum perception probability of the center of the drone group, and “independent probability” refers to the average spectrum perception probability of the drone group.

[0187] When comparing the impact of drone distribution on spectrum perception, assume that there are three drones in the system. The PU moves from (2000,0,0) to (200,0,0), and the x-axis coordinate interval decreases by 200. By changing the distribution of drones, the individual probability method is compared with the unified probability method through simulation to analyze the impact of drone distribution on spectrum perception.

[0188] There are three types of drone distribution used in the simulation:

[0189] Centralized distribution: Assume that the three-dimensional spatial coordinates of the three drones are A1(500,-150,5), A2(450,-150,5), and A3(350,-100,5), and the center coordinate is D1(433.333,-133.333,5).

[0190] Dispersed distribution: Assume that the three-dimensional spatial coordinates of the three drones are B1 (200, -300, 5), B2 (900, -300, 5), and B3 (0, 300, 5), and the center coordinate is D2 (366.667, -100, 5).

[0191] Conventional distribution: Assume that the three-dimensional spatial coordinates of the three drones are C1(200,-200,5), C2(600,-200,5), and C3(200,100,5), and the center coordinate is D3(333.333,-100,5).

[0192] The projections of the drone swarm, PU, ​​and FC on the XOY plane are shown in the figure below: Figure 2 Based on the above parameter simulation, we get Figure 3 The simulation results are shown.

[0193] When comparing the impact of the distance from the drone group to the PU on spectrum perception, assume that there are 3 drones in the system, and the PU moves from (2000, 0, 0) to (200, 0, 0), with the x-axis coordinate interval decreasing by 200. The three-dimensional spatial coordinates of the three drones are A(2200, -200, 5), B(2600, -200, 5), and C(2200, 100, 5), and the center coordinate is D(2333.333, -100, 5). When the PU moves toward the FC, the PU gradually moves away from the drone group. Change the PU's transmit power ρ PU , let ρ PU They are 0.01W, 0.02W, and 0.03W respectively. Through simulation, the individual probability method is compared with the unified probability method, and the impact of the distance from the drone swarm to the PU on spectrum perception is analyzed.

[0194] The projections of the drone swarm, PU, ​​and FC on the XOY plane are shown in the figure below: Figure 4 Based on the above parameter simulation, we get Figure 5 The simulation results are shown.

[0195] When comparing the effects of transmission errors between SU ​​and FC, in order to study the effect of transmission errors on the performance of the comparison method, the transmission error probability The simulation was performed under four conditions: 0.1, 0.2 and 0.3. P(H0)=P(H1)=0.5, T=4, η=2. The system used includes 100 users, and the proportion of malicious users is changed to explore The results are as follows Figure 6 shown.

[0196] When comparing SSDF attack detection methods for drone spectrum sensing, we use Figure 2 The two methods are compared in the same simulation environment. PU is fixed at (2000,0,0). The three-dimensional spatial coordinates of the three drones are C1(200,-200,5), C2(600,-200,5), and C3(200,100,5). C1 and C2 are malicious users, and C3 is a legitimate user. p(H0)=0.4,P(H1)=0.6,T=T w ,

[0197] The projections of the drone swarm, PU, ​​and FC on the XOY plane are shown in the figure below: Figure 2 For the same legitimate user, the simulation results are and Comparison of Figure 7 shown.

[0198] In summary, the present invention constructs a cognitive drone network model to meet the needs of future drone spectrum sensing data security. In order to improve the accuracy of spectrum sensing, the present invention calculates the spectrum sensing probability of each drone separately according to its coordinates; in order to be more in line with reality, the present invention also includes the probability of SU and FC transmission errors in the calculation. Subsequently, a reputation is assigned to each SU based on the probability of reporting accurate perception results. SUs whose reputation does not exceed the reputation threshold are regarded as malicious users and are prohibited from participating in collaborative spectrum sensing. Finally, the performance of the SSDF attack detection method that does not consider SU and FC transmission errors in transmission errors is studied and compared with the method proposed in the present invention. The simulation results show that the method proposed in the present invention has better spectrum sensing performance when the drones are dispersed or close to the PU; compared with the method that does not consider transmission errors, the method proposed in the present invention has a lower probability of mistaking legitimate users for malicious users.

[0199] The above is only a preferred embodiment of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications should also be considered as the protection scope of the present invention. The structures, devices and operating methods not specifically described and explained in the present invention shall be implemented according to the conventional means in the art unless otherwise specified and limited.

Claims

1. A SSDF attack detection method for UAV spectrum sensing, characterized by: The following steps are involved: Step 1: Establish a cognitive drone network based on centralized collaborative spectrum sensing; Step 2: Each drone performs spectrum sensing individually; Step 3: Analyze the impact of transmission errors on malicious user detection; Step 4: Identify malicious users based on reputation.

2. The SSDF attack detection method for UAV spectrum sensing according to claim 1 is characterized in that: The cognitive drone network based on centralized cooperative spectrum sensing in step 1 includes malicious users. The entire cognitive drone network includes N drones as SUs. It is assumed that the number of malicious users is K. There is also a PU and an FC on the ground responsible for spectrum management of all SUs. Malicious users will disguise themselves as legitimate users to launch SSDF attacks, and deliberately send forged spectrum sensing data to FC with a certain probability, thereby misleading FC to make wrong spectrum judgment results; symbol SU i To represent the i-th SU (i=1,2,...,N) in the network.

3. The SSDF attack detection method for UAV spectrum sensing according to claim 2 is characterized in that: The spectrum sensing of the cognitive drone network is divided into three stages: Phase 1: Local spectrum sensing: The legitimate users and malicious users participating in the collaboration perform independent local spectrum sensing on the authorized frequency band of the PU selected by the FC; Phase 2, Report / SSDF attack: Legitimate users send their perception data or perception results to FC; malicious users launch SSDF attacks with a certain probability and send wrong perception data or perception results to FC; Phase 3, decision reporting: FC uses the information transmitted by SU to obtain the FC's decision statistics according to appropriate data fusion rules, and judges the usage status of the frequency band based on the decision statistics. After making the final decision, the result is notified to SU.

4. The SSDF attack detection method for UAV spectrum sensing according to claim 1 is characterized in that: The second step calculates the spectrum sensing probability of each UAV in the network according to their coordinates.

5. The SSDF attack detection method for UAV spectrum sensing according to claim 1 is characterized in that: The local spectrum sensing of each UAV in the step 2 adopts a sensing technology based on energy detection.

6. The SSDF attack detection method for UAV spectrum sensing according to claim 1 is characterized in that: The step 2 comprises the following steps: Let H0 represent PU not using spectrum, H1 represent PU occupying spectrum; i The signal of receiving PU can be expressed as: Among them, x i (t) is SU i Receive PU signal; i (t) has a mean of 0 and a variance of The additive Gaussian white noise of ; s(t) is the signal of PU; Energy detection statistic Y i It is expressed as: Where N0 is the single-sided noise power density spectrum, T w is the recording time window, Y i satisfy: in, and They represent central and noncentral chi-square distributions, respectively, with degrees of freedom of 2ω. The noncentral parameter of the latter is ω is the time-bandwidth product, From PU to SU i The signal-to-noise ratio of the received signal; SU i The detection probability and false alarm probability can be expressed as: Where λ is the energy detection threshold, Γ(·) is the gamma function, and Γ(.,.) is the incomplete gamma function. The expression is: where ρ PU is the transmit power of PU; PU and SU i The path gain between is expressed as: Where κ and μ are path gain constant and path gain index respectively; d P ,d i PU and SU respectively i 3D position.

7. The SSDF attack detection method for UAV spectrum sensing according to claim 1 is characterized in that: The step three comprises the following steps: Assume SU i The transmission with FC will not go wrong, and the legitimate user SU i The probability of reporting a correct perception result is: Assume SU i The transmission with FC may be erroneous. Given the BPSK modulation in Rayleigh fading environment, SU i The probability that the report to the FC is in error is: in, It is SU i The average signal-to-noise ratio of the bit report between FC is expressed as: Among them, ρ i For SU i The transmission power; h i It is FC and SU i The path gain between is expressed as: Among them, d F is the 3D position of FC; based on and Legitimate user SU i In the case of possible errors in the transmission with FC, the probability of correctly reporting the perception result is: Here, the function F(x,y) is defined as F(x,y)=x(1-y)+(1-x)y. set up and They are malicious users SU j The probability of reporting the correct perception result when the transmission is error-free and error-prone; similar to formula (12), we have: Suppose the probability that the result reported by a malicious user is opposite to the local perception is P mal , if the malicious user takes the independent attack approach, then: If the malicious user adopts the "L out of K" collaborative attack strategy, then:

8. The SSDF attack detection method for UAV spectrum sensing according to claim 1 is characterized in that: The step 4 comprises the following steps: Set SU i The credibility is r i , is the probability of reporting an accurate perception result: where τ i T w SU i The number of times the local report is consistent with the final decision of the FC; Assume η i For SU i The credibility threshold, P i For SU i The probability that the local report of SU is consistent with the final decision of FC; if i Credibility i Greater than η i , then it is judged as a legitimate user, If SU i Credibility i Less than or equal to η i , then it is judged as a malicious user, Assume z i (t) is the t time slot SU i The correctness of the local perception result, if SU i If the correct result is reported, then z i (t) = 1; otherwise, z i (t) = 0. Therefore P(z i (t) = 1) = P i , P(z i (t) = 0) = 1-P i ; Therefore z i (t) obeys Bernoulli distribution. According to formula (17), we have: Among them, t0 is the start time of the recording time window; due to z i (t) follows a Bernoulli distribution, if T w satisfy: T w P i ≥10and T w (1-P i )≥10 (19) Then r i Normal distribution Before starting spectrum sensing, it is necessary to find T that satisfies formula (19) w ,Since FC has no prior information about which SUs are malicious, it is reasonable to assume that all SUs are legitimate at the beginning, Therefore, T w Should meet: in, is the set of all SUs. Obviously, such T w satisfy For legitimate users SU i ,because Assume r i The probability density function is f H,i (x), the expression is: Legitimate user SU i Probability of being mistaken for a malicious user The expression is: Among them, erfc(x) is the complementary error function, expressed as:

Images