KR101164761B1 - Accuracy enhancement method of distributed spectrum sensing in security required cognitive radio system - Google Patents

Abstract

PURPOSE: A method for improving the sensing accuracy of distributed spectrums in a cognitive radio system is provided to rapidly eliminate influence caused by an attack by excluding a sensing node having high attack capability from a fusion process in a following sensing step. CONSTITUTION: A plurality of sensing values is inputted according to the sequence of reputation values of the plurality of sensing values. A final result value is calculated by applying a weighted value to the plurality of sensing values(S100). The reputation values of the plurality of sensing values are controlled according to the final result value. A trend value is calculated for the reputation values of the plurality of sensing values(S200). The abnormality of the plurality of sensing values is determined based on the trend value. A corresponding sensing node showing abnormal symptom is excluded from fusion(S300), IDs of the plurality of sensing nodes are arranged in a descending order based on the reputation values of the plurality of sensing values(S400).

Description

ACCURACY ENHANCEMENT METHOD OF DISTRIBUTED SPECTRUM SENSING IN SECURITY REQUIRED COGNITIVE RADIO SYSTEM}

The present invention relates to a method for enhancing the accuracy of distributed spectrum sensing in a cognitive radio system that requires security. More particularly, the present invention relates to fusion of a sensing value obtained by sensing a distributed spectrum in a cognitive radio system. In this case, RDSSS (Robust Distributed Spectrum Sensing) receives reputation values in descending order and performs fusion on the influential sensing nodes first to quickly synthesize them. Improved reputation-based fusion for secure distributed spectrum sensing in cognitive wireless networks with high security and ability to respond to external attacks by lowering the number of attacking nodes from the next fusion by analyzing the trend value of the reputation value. It is about a method.

Cognitive radio (CR) technology has been steadily researched and standardized to meet the increasing demand for radio resources and to solve the problem of inefficient use of frequency bands. The CR technology refers to a technology that detects an empty channel and performs data communication in a corresponding channel by recognizing the usage status of a radio frequency band of a main user (a licensed user who uses a fixed frequency allocated from the FCC). The core of such CR technology is channel sensing which accurately recognizes whether the main user uses the channel. However, due to various factors (eg, fading, shadowing, etc.) in the wireless communication environment, the sensing result of one node may be less accurate.

Therefore, recently, in order to increase the sensing accuracy, distributed spectrum sensing (DSS) technology, in which several nodes sense at the same time and combine the results into one, has been actively studied.

However, the biggest vulnerability in distributed spectrum sensing is the possibility of attack by nodes participating in the sensing. This can result in forged sensing results that can lead to errors in the final composite result. In the worst case, a forged sensing result of one node could break the CR technology's goal of protecting the main user's channel usage and maximizing channel efficiency. In order to make up for these weaknesses, safe distributed spectrum technologies have recently been studied.

RDSS, a representative safe spread spectrum mechanism, collects sensing values from surrounding sensing nodes and makes a final decision. It uses a weighted sequential probability ratio test (WSPRT) hypothesis test. Unlike general SPRT, the reputation value can be adjusted according to the accuracy of sensing value and used as weight to reduce the influence of attack by the modulated sensing value. However, the fusion mechanism may require a large number of WSPRT iterations according to the order of sensing values input, and may eventually lead to abnormal results with a few forged sensing values. In addition, there is a problem that a considerable time is required to reduce the impact from the attack node.

The present invention has been made to solve the above problems, but WSPRT can be performed according to the reputation value as in RDSS, but WSPRT can be performed from the result of the sensing value of the sensing node having a high reputation value to increase the accuracy and speed of sensing. . In addition, by calculating the change in reputation value as a trend value, the attacking potential of the sensing node can be calculated, and thus, the effect of the attack can be quickly removed by excluding the sensing node with the high possibility of attack from the fusion process during the next sensing. The aim of the present invention is to provide an enhanced reputation-based fusion method for secure distributed spectrum sensing in cognitive wireless networks with improved performance and performance.

According to the present invention, a method for enhancing accuracy of distributed spectrum sensing in a cognitive radio system requiring security includes fusion of a plurality of sensing values collected from a plurality of sensing nodes by distributed spectrum sensing in a cognitive radio network. In the method for enhancing the accuracy of distributed spectrum sensing in a cognitive radio system requiring security for calculating a result value, the plurality of sensing values are input in accordance with the reputation value order of the plurality of sensing values, and weighted to the plurality of sensing values. Calculating a final result by fusion by applying (S100); Adjusting a flat plate value of the plurality of sensing values according to the final result value and calculating a trend value for the adjusted flat plate value of the plurality of sensing values (S200); Determining whether the plurality of sensing values are abnormal based on the trend value, and excluding the corresponding sensing node indicating an abnormal indication from the fusion (S300); And sorting IDs of the plurality of sensing nodes in descending order based on the reputation values of the plurality of sensing values (S400).

In addition, according to an embodiment of the present invention, the step of calculating a trend value for the plate value of the plurality of sensing values (S200),

는 시간 t _n 의 트랜드값, t _n 은 퓨전 시간, W는 트랜드 계산을 위한 윈도우 사이즈, r _tn 는 시간 t _n 의 평판값)(here,

Is the trend value of time t _n , t _n is the fusion time, W is the window size for trend calculation, r _tn is the plate value of time t _n )

It is characterized by calculating the trend value by.

)이 소정의 임계치(α)와 비교하여 하향하거나 계속적으로 변화하는 트랜드값인 경우에 상기 이상 징후로 판단하는 것을 특징으로 한다.
In addition, according to an embodiment of the present invention, the step (S300) of excluding the corresponding sensing node from the fusion may include the trend value of the sensing node (

) Is determined as the abnormal symptom when the trend value is a trend value that is downwardly or continuously changing compared to a predetermined threshold value α.

In the security-recognized wireless system according to the present invention, a method for enhancing the accuracy of distributed spectrum sensing is performed by performing WSPRT according to a reputation value, as in RDSS, but performing WSPRT from the result of a sensing node having the highest reputation value. Provides the effect of speeding up. In addition, by calculating a trend value indicating the rate of change of reputation value, the attack potential of the sensing node can be calculated to exclude the high-sensing sensing node from the next fusion process. to provide.

1 is a diagram illustrating a cognitive radio network model according to an embodiment of the present invention;
2 is a flow diagram of an improved flatbed-based fusion method for secure distributed spectrum sensing in a cognitive radio network according to an embodiment of the present invention;
3 is a diagram illustrating a simulation configuration of a cognitive radio network and an attack setup according to an embodiment of the present invention;
4 is a view showing a trend value of each node according to D (3000m) in the absence of an attack;
5 is a view showing a trend value of each node according to D (5000m) in the absence of an attack;
6 is a diagram showing a trend value of each node according to D (3000m) in case of an attack;
7 is a view showing the attack response performance of the RDSS and eRDSS in the absence of an attack,
8 is a diagram showing the number of iterations of WSPRT in the absence of an attack;
9 is a view showing the attack response performance of the RDSS and eRDSS when there is an attack,
10 is a view showing the attack response performance according to the sensing time when there is an attack,
11 is a diagram showing the number of repetitions of the WSPRT when there is an attack.

DETAILED DESCRIPTION Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a diagram illustrating a cognitive radio network model according to the present invention.

Referring to FIG. 1, a network model in an RDSS may include a plurality of secondary users (SUs), a base station (BS), and a primary user transmitter (PUtx). A plurality of secondary user nodes (SU) can form an ad hoc cognitive radio networks around the base station (BS), periodically sensing the presence or absence of the transmission signal of the main user transmitter (PUtx), such as a TV transmission tower do. As described above, auxiliary user nodes that perform sensing may configure a sensing node set. Preferably, a separate sensing node may be arranged. The sensing result of the sensing node may be delivered to a fusion node (for example, a base station) that synthesizes the result, thereby producing a final sensing result, and performing CR network communication through channel scheduling based on the sensing result. This network model is identical to the network model in IEEE 802.22, which is a standard of representative CR technology.

Referring to FIG. 1, an attack model by a compromised SU may be an "Always-False attack". This is an attack method of transmitting a value opposite to the signal sensing value of the main user transmitter PUtx, that is, "0" when a signal is detected and "1" when a signal is not detected. These attacks are powerful attacks that increase the miss detection ratio (MDR) and decrease the correct sensing ratio (CSR) in distributed spectrum sensing. The non-detection rate is a ratio making the case of H ₁ (signal present) of the total sensing finally determined as "0" (empty), H _0, to a sensing accuracy is "1" (which filled) if H ₁ (signal member ) Is the rate at which the final decision is made to "0".

In addition, injecting attacks by arbitrary sensing values from unlawful nodes can be handled by basic authentication mechanisms (eg, MAC (message authentication code) verification or RF fingerprint). Replaying attacks that capture and retransmit can be detected by inserting timestamps or sequence numbers, and the physical layer jamming attack is a basic jamming-resistant technique called FSSS (frequency). It is possible to respond by hopping spread spectrum (DSP) or direct sequence spread spectrum (DSSS) technique.

IEEE 802.22 wireless regional areal networks (WRAN) have been proposed to provide wireless data services in rural areas with relatively poor communication environment by utilizing TV broadcasting bands that are not used locally. Referring to FIG. 1, an IEEE 802.22 network manages a user's channel and provides a service using a centralized media access control (MAC) protocol through one base station (BS). The IEEE 802.22 network performs sensing for each available channel through a quiet period in which all nodes stop transmitting. The base station (BS) collects the sensing results of the sensing nodes to derive the final sensing results, and based on this schedule for the channel use. As the sensing method, an energy detection technique and a characteristic detection technique may be used.

In addition, as active researches on distributed spectrum sensing in terms of accuracy have raised the issue of safety in the mechanism. That is, when a sensing value of a participating sensing node is forged (spectrum sensing data falsification attack), a final abnormal result may be derived.

In order to cope with such an attack, the suspicious level of each sensing node is calculated through prior knowledge of the signal patterns (H ₁ , H ₂ ) of the main user (PU). A method of OR-fusion of the remaining values has been proposed. However, the PU signal pattern assumed in this method is difficult to obtain by measuring experimentally accurately. In addition, based on the log-normal shadowing model, the ADSP method using the received signal model of the sensing value was established. In this ADSP method, a cluster is formed between sensing nodes to compare the similarity of sensing values in the cluster, and the low similarity sensing value can be excluded during fusion to reduce the impact of the attack. However, this method will not be able to cope if more than one-third of the nodes in the cluster are attacked.

RDSS is proposed as another method to counter an attack by forged sensing value. RDSS does not require prior knowledge of main user (PU) signal patterns, similar to ADSP. Instead, we establish a PU signal propagation model based on shadowing, have a distance value between the PU and the sensing node, and use this model to estimate the strength of the PU signal reception of each sensing node. can do. Based on this, a conditional probability ratio of the received sensing values (0 or 1) is obtained and multiplied by these values to perform a weighted SPRT that leads to a final decision when the threshold is reached. In this case, by calculating the reputation value according to the sensing value accuracy of each sensing node and using it as a weight for the conditional probability ratio, the influence from the forged sensing value can be reduced.

However, in the conventional RDSS, the number of executions can be increased according to the order of the (falsified) sensing values input to the first WSPRT, and the number of nodes is reduced by allowing an attacker to input the forged sensing values initially during the second WSPRT iteration. Attack can easily lead to the final abnormal result, and if the node with the third highest reputation is attacked, considerable time may be required to reduce the impact of the attack.

2 is a flowchart illustrating a method for enhancing accuracy of distributed spectrum sensing in a cognitive radio system requiring security according to the present invention. Referring to FIG. 2, in the method for enhancing the accuracy of distributed spectrum sensing according to the present invention, the final result is calculated by fusion by applying a weight to a plurality of sensing values (S100), for a flat plate value of a plurality of sensing values. Computing a trend value (S200), excluding the corresponding sensing node indicating an abnormal indication from the fusion (S300) and sorting the IDs of the plurality of sensing nodes in descending order (S400).

Table 1 below shows an eRDSS procedure showing a method for enhancing the accuracy of distributed spectrum sensing in a cognitive radio system requiring security.

Referring to Table 1, the step (S100) of calculating a final result by fusion by applying a weight to a plurality of sensing values will be described in more detail as follows.

이면 H₁ 결정, S_n≤

이면 H₀ 결정), 그렇지 않은 경우 다른 센싱값을 다시 적용하여 계산할 수 있다.In the present invention, the SPRT statistic basically used for fusion is based on the sensing value (u _i = 1 or 0) from each sensing node, as shown in Equation 1 below. ₁ , multiplied by the conditional probability ratio of the member H ₀ ), and when the value reaches the threshold, the final result is obtained (S _n ≥

H ₁ crystal, S _n ≤

If H ₀ ), otherwise it can be calculated by applying another sensing value again.

와

는 허용된 오탐지율(tolerated false alarm probability) P₀₁과 허용된 미탐지율(tolerated miss detection probability) P₁₀에 의해

으로 계산된다. 또한, 조건부 확률값은 다음의 수학식 2와 같이 주사용자(PU)와 센싱 노드 사이의 거리(d)에 따른 로그-노멀 쉐도우잉 거리손실모델로서 계산된다. 이때, X_σ ~ N(0, σ²)는 확률 변수, d₀는 참조거리,l은 경로손실지수(path loss exponent)이다.Where two thresholds

Wow

Is given by the tolerated false alarm probability P ₀₁ and by the tolerated miss detection probability P ₁₀ .

. In addition, the conditional probability value is calculated as a log-normal shadowing distance loss model according to the distance d between the main user PU and the sensing node as shown in Equation 2 below. In this case, X _σ to N (0, σ ² ) are random variables, d ₀ is a reference distance, and l is a path loss exponent.

Based on this loss model, the received energy for H ₁ is calculated as P _r = P _t - PL (d) , and for H ₀ , P _r = n ₀ The following equations 3 to 6 are derived.

는 노이즈 확률 변수이다. Γ is the detection threshold, P _t is the transmission energy of the main user (PU),

Is a noise random variable.

In order to correspond to the forged sensing value, the weight f ( r _i ) by the reputation value r _i may be applied according to the accuracy of the sensing value of each sensing node (see step 5 of Table 1). In this case, f (r _i ) is f (r _i ) = 0 when r _i = -g according to the system parameter g. If r _i > -g, f ( r _i ) = (r _i + g ) / (max ( r _i ) + g ). After deriving the final result u, each plate value is adjusted according to whether it is equal to the sensing value (see step 9 in Table 1).

Referring to Table 1, the WSPRT weighted by the reputation value is performed, and after obtaining the final sensing result, the reputation value of each node is adjusted (see steps 3 to 9 of Table 1).

When fusion is started for a plurality of sensing values by spread spectrum sensing, the sensing values of the sensing nodes having a high reputation value may be input according to the reputation value of each node's sensing value. The fusion method of the plurality of input sensing values may be basically performed by OR, AND, and majority fusion methods.

The final decision method may be a method of determining that there is a final signal used by the primary user when the total number of sensing nodes is N, and the number of nodes determined as H ₁ is greater than or equal to k. However, this does not indicate good performance in terms of both performance, detection rate and sensing accuracy.

Therefore, in the preferred embodiment according to the present invention, the final signal used by the primary user is weighted by using the final result obtained by adding the weighted sensing values received from all sensing nodes and adding the plurality of weighted sensing values. Soft decision can be used.

In operation S200, a trend value for a flat plate value of a plurality of sensing values may be used to model a change pattern of a flat plate value, which is dynamically changing data, as shown in Equation 7 below. .

Where r _t is the plate value at time t , t _n ( n = 0, 1, ... n ) is the time series, and time t is With a relationship of t _n <t < t _{n + l} , a ( t _n ) is a model parameter representing the rate of change of the plate value at t _n .

According to the present invention, in order to quickly detect an attacked node, a change pattern of a reputation value r , that is, a trend value, may be calculated to determine a change pattern. Trend data is calculated The algorithm used in financial time series forecasting can be used to estimate the rate of change. When the data value changes with time, the trend value of the data at time t may be defined as the slope of the expected value of the instance value as shown in Equation 8 below.

은 식 (9)와 같이 각 값들의 차이값의 평균으로 구할 수 있다.Periodic observations can be used in a moving window of size W to estimate a (t _n ) . That is, an estimate for estimating a ( t _n ) with past W data values

Can be obtained as the average of the difference between the values as shown in equation (9).

In step S300 of excluding a corresponding sensing node indicating an abnormal symptom from the fusion, when the trend value of the plate value shows a downward or continuous change, the sensing value of the corresponding sensing node is excluded in the next sensing. (Steps 11-12 of Table 1).

이 다음의 수학식 10과 같이 특정 임계치와의 비교를 통해 공격의 가능성, 즉 하향 또는 계속적인 변화(fluctuation)로 감지되는 경우, 해당 센싱 노드의 값을 다음의 WSPRT 퓨전 시 배제할 수 있다. 트랜드값의 임계치 α는 시스템 파라미터로서 0 혹은 작은 음의 수이며, 적정 임계치는 네트워크 노드 설치 후 실측치에 기반하여 설정할 수 있다.In more detail, the trend value estimated through Equation 9

As shown in Equation 10 below, when it is detected as a possibility of attack, that is, a downward or continuous fluctuation through comparison with a specific threshold value, the value of the sensing node may be excluded at the next WSPRT fusion. The threshold value α of the trend value is 0 or a small negative number as a system parameter, and an appropriate threshold value can be set based on actual values after network node installation.

In the next sensing after the step S300, sorting IDs of the plurality of sensing nodes in descending order based on the reputation values of the plurality of sensing values so that the sensing values of the nodes having the highest reputation values may be input first (S400). (See Table 1 step 10). The ID sorting of the plurality of sensing nodes performs a descending sorting order from the large reputation value to the smallest value. Also, in order to detect the attacked node quickly, the pattern of change of reputation value r is examined to determine whether or not the fusion is performed (steps 11 to 12).

As discussed above, the improved fusion mechanism according to the present invention (hereinafter referred to as eRDSS) may further comprise two characteristic steps as compared to the conventional fusion mechanism. The first is to perform the fusion by inputting the sensing value of the node with high reputation value for fast WSPRT execution (step 10 in Table 1). Second, when the trend of the change in reputation value is downward or continually changing, the sensing value of the corresponding sensing node is excluded at the next sensing (steps 11 to 12 of Table 1).

Since these two additional steps are performed in the fusion node after the sensing is completed, there is no change in time required for sensing, so that the problem of increasing the fusion time does not occur even when applied to the conventional fusion method.

Figure 3 shows a simulated schematic diagram of a cognitive radio network and attack setup according to the present invention.

Referring to FIG. 3, the CR network has a 2000m × 2000m area in which a base station (BS) is located at the center, and the BS is separated from PUtx by D (1000 to 5000m), and has 500 randomly located secondary users ( Secondary User: SU) node may be configured to serve.

Each SU node may have a communication radius of 250m. Each SU node has a maximum speed of 10m / s, mobility according to a random waypoint mobility model, and may have a maximum downtime of 120s.

)는 -106dBm으로 설정할 수 있다. The SU node may perform sensing using an energy sensing method every 30 s, and the BS performs sensing and fusion functions. Minimum power (γ) for signal detection, -94 dBm, average noise (

) Can be set to -106dBm.

In FIG. 3, 10 ⁻⁵ and 10 ⁻⁶ may be used as P ₀₁ and P ₁₀ values for calculating the WSPRT threshold value, and 5 may be used as a parameter value g for weight calculation. The duty cycle of PUtx is 0.2, and a total of 2 hours of simulation can be performed. The path loss model for obtaining PL (d) can use the HATA model proposed by the IEEE 802.22 working group for a specific environment. The PUtx channel can be set to 617MHz UHF channel, and the PUtx and receiver antenna heights can be set to 100m and 1m, respectively. have.

The attack scenario is based on the attack model described above (Always-False attack), but at the time of the attack, a randomly selected 30% of the total SU nodes (Compromised SU; CSU) can continuously execute the Always-False attack.

4 and 5 illustrate trend values of respective sensing nodes according to D (3000m or 5000m) in the absence of an attack.

4 and 5, the value represents the 100th value of the sensing time. In the case of FIG. 4, most values have a value of 0.6 or more and 1 or less, while in FIG. 5, the trend value is between 0.4 and 0.8 as the received signal strength decreases due to a longer distance (D = 5000m). We can see that there are also very few nodes with a value of -0.2.

6 is a diagram illustrating a trend value of each sensing node according to D (3000m) in case of an attack. It can be seen that many nodes have negative values compared to the case where there is no attack (see FIGS. 4 and 5), and in most cases, they correspond to the attacked node.

It can be performed by setting to a value of -0.2 as the appropriate trend threshold value α of this simulation. In the case of a real network, an appropriate trend threshold may be obtained by measuring trend values and general cases after initial network node installation.

7 is a diagram showing the performance of the RDSS and eRDSS in the absence of an attack. Referring to FIG. 7, in a general case without an attack, the difference in attack response performance between the RDSS and the eRDSS may not be large. However, as the distance increases, the performance of the eRDSS fused from a higher reputation value may be better. That is, the maximum CSR increases by 1.3% and the MDR decreases to 0 compared to the RDSS). As RDSS decreases the energy detection accuracy of each sensing node on average over long distances, CSR tends to decrease slightly and MDR increases slightly.

8 shows the number of repetitions of the WSPRT in the absence of an attack. Referring to FIG. 8, the eRDSS reduces the number of WSPRT iterations performed to obtain the final sensing result by up to 12.4% compared to the RDSS. This is because, in the case of eRDSS, the plate values are input in a high order to reach the threshold of the WSPRT faster.

9 is a diagram showing the attack response performance of the RDSS and eRDSS when there is an attack. Referring to FIG. 9, a performance comparison between RDSS and eRDSS, in which a randomly selected 30% of the nodes attacked continuously carry out an attack that delivers the opposite result to the original sensing result, can be seen. It can be seen that the performance of eRDSS is excellent in both MDR and CSR (i.e., eRDSS has a 55.4% increase in CSR and 89.5% decrease in MDR compared to RDSS). This is because there is a high probability that the trend value of the node performing the attack has a negative value, so it is more likely to be excluded during fusion.

10 is a diagram illustrating attack response performance according to a sensing time when an attack is present.

Referring to FIG. 10, the RDSS can confirm that the agility corresponding to the attack is considerably inferior. In other words, the reputation value of the attacking node decreases by 1, and its influence is reduced after several sensing times. As a result, the performance of MDR and CSR is poor for a considerable sensing time. However, since eRDSS is input from the sensing value with high reputation value, even though the reputation value of attack node is gradually lowered, it is possible to reduce the possibility of entering reputation value of attack node. Excluding the node can lead to high performance.

11 is a diagram showing the number of repetitions of the WSPRT when there is an attack. Referring to FIG. 11, it can be seen that the number of repetitions of the WSPRT is also similar to that of no attack, but in the case of RDSS, the number of repetitions cannot be reached quickly by entering the reputation value of the attack node. ERDSS is reduced by up to 52.5% compared to RDSS).

In addition, preferred embodiments of the present invention are disclosed for the purpose of illustration, those skilled in the art will be able to make various modifications, changes, additions, etc. within the spirit and scope of the present invention, such modifications, changes, etc. fall within the scope of the claims Should be seen.

Claims (3)

A method for enhancing the accuracy of distributed spectrum sensing in a cognitive radio system that requires security of fusion of a plurality of sensing values collected from a plurality of sensing nodes by distributed spectrum sensing to produce a final result,
Calculating a final result by inputting the plurality of sensing values and fusion by applying weights to the plurality of sensing values according to the order of the plate values of the plurality of sensing values (S100);
Adjusting a flat plate value of the plurality of sensing values according to the final result value and calculating a trend value for the adjusted flat plate value of the plurality of sensing values (S200);
Determining whether the plurality of sensing values are abnormal based on the trend value, and excluding the corresponding sensing node indicating an abnormal indication from the fusion (S300); And
And arranging the IDs of the plurality of sensing nodes in descending order based on the reputation values of the plurality of sensing values (S400).
The method of claim 1,
Computing a trend value for the plate value of the plurality of sensing values (S200),

(here,

Is the trend value of time t _n , t _n is the fusion time, W is the window size for trend calculation, r _tn is the plate value of time t _n )
The method for enhancing the accuracy of distributed spectrum sensing in a security-aware cognitive wireless system, characterized in that for calculating the trend value by.
The method of claim 1,
Excluding the corresponding sensing node from the fusion (S300),
The trend value of the sensing node (

) Is determined as the abnormal indication when the trend value is downward or continuously changing compared to a predetermined threshold α, and the accuracy of distributed spectrum sensing in a cognitive radio system in need of security.

Images