CN101420758B - Method for resisting simulated main customer attack in cognitive radio - Google Patents

Abstract

A method for resisting primary user impersonation attacks in cognitive radio belongs to the technical field of radio communication, particularly relates to sensing and processing of the signals in the cognitive radio system and identification of a signal source. The method comprises the following steps: reducing noise of a composite signal which is received during a spectrum sensing process, estimating number of the signal sources in the composite signal, and reconstructing various signals by a Capon beamformer; separating the signals of different radios, focusing the signals of different radios to separate out a transient signal and a modulation signal, analyzing the transient signal by wavelet transformation; extracting characteristic parameters and realizing identification of the signal source. The method for resisting the primary user impersonation attacks helps realize effective identification of the signal source and effectively resist the primary user impersonation attacks; and the method has a wide application scope, does not change a working mode of the existing primary user network, and satisfies operation requirements of FCC for a cognitive network.

Description

The method of antagonism simulated main customer attack in a kind of cognitive radio

Technical field

A kind of method of resisting simulated main customer attack belongs to wireless communication technology field in the cognitive radio, the particularly perception of signal and processing in the cognitive radio system, and to the identification of signal source identity.

Background technology

Cognitive radio networks is as a kind of intelligent radio communication system, and it can perception, the study surrounding environment, and the messaging parameter of adjusting oneself adapts to environmental change on every side.Cognitive user is dynamically utilized idle mandate frequency range with " chance mode " on the basis that does not influence the authorized user proper communication, can well solve present frequency spectrum resource problem of shortage.

Cognitive user has the ability of transmitting on each frequency of most of frequency bands, can use many communication channels in the consolidated network, and this makes it more complicated more than the physical layer of conventional wireless network, also is faced with more attack.Wherein mainly be simulated main customer and forge the frequency spectrum perception data and attack at what dynamic spectrum inserted.Simulated main customer attack can have a strong impact on the frequency spectrum perception process in the cognition network, and can obviously reduce legal understanding user's available channel resources.For fear of disturbing or make cognitive user miss the access chance because of the frequency spectrum perception mistake produces main user, thereby cognitive user must have the ability antagonism simulated main customer attack of the main user identity of identification.One of major technique challenge that is frequency spectrum perception is exactly how to distinguish main subscriber signal and cognitive user signal accurately.

The method that detects main user's transmission at present mainly contains non-cooperative detection, cooperative detection and Interference Detection.Wherein non-cooperative detection (also claiming emission source to detect) comprises: detect (waveform-based sensing) based on energy measuring (Energy Detection) or based on signal characteristic.Wherein using filter match detection (Matched Filter Detection) or cyclic stationary process to detect (Cyclostationary Feature Detection) technology based on the signal characteristic detection technique realizes signal essence Feature Recognition.Energy measuring belongs to blind Detecting, and detection signal be need not any priori.But it can only detect the signal appearance, can not distinguish the type of signal, and promptly it can not distinguish modulated signal noise and interference.When perception is easy to be subjected to selfish or during the invasion of the cognitive user of malice, can not discerns main user identity.Present a kind of improved plan is to introduce " quiet period ", and its prerequisite is that all cognitive user all are cooperations, if exist this scheme of malice cognitive user to collapse fully.Matched filtering detection and cyclic stationary process detect the substantive characteristics that can detect signal, the identification source identity, but need have priori and at different main user's matched filtering detections independent matched filter will be arranged the cognitive user signal.In addition feature detection is combined with analysis of spectrum, and in conjunction with improving the performance and the efficient of detection based on the pattern recognition of neural net.

At the influence of multipath fading in the wireless transmission and shadow effect, especially overcome " concealed terminal " problem that shadow fading causes, introduce the cooperative detection technology, its structure can be centralized, also can be distributed.Introduce the cooperation diversity technique cooperative detection can produce perceptual performance more accurately, but can have a negative impact, and study under the main customer location known case of its hypothesis for resource-constrained network.Propose " local oscillator leakage " for main customer location condition of unknown and detect, utilize the local oscillator leakage power that detects to come positioning main user, the while can utilize the cooperation between CR to improve the detection probability of weak signal.Consideration is by the interference of cognitive user to main custom system, and FCC proposes the interferometry model---and the interference temperature detection model mainly is to consider the interference of restriction to main subscriber signal.

For the cognitive radio networks that uses the TV frequency range,, therefore can utilize sender's position and signal power level to distinguish main subscriber signal and cognitive user signal because main user is the TV television tower.Can determine sender's identity by the power level of uniting the positional information of using the sender and received signal.How accurately to obtain sender's positional information at the problem difficult point at present two kinds of technology are arranged: distance is than verification and range difference verification (Distance Ratio Test (DRT) and Distance Difference Test (DDT)).The former determines that according to the intensity of the signal source that location confirmation person receives sender position, its scope of application are large scale decline models, and will priori be arranged to main customer location; The dependent phase difference that the same signal of The latter produces through different paths (distance) in-position affirmant is determined sender's position.At main when being the littler portable terminal of transmitted power, a kind of scheme of effective antagonism simulated main customer attack is to utilize " Radio Environment Map " notion (REM) at present, REM is a database that comprises main location of user equipment and action message, can utilize these information to discern main user, its concrete enforcement schemes also need further be studied.

Summary of the invention

The purpose of this invention is to provide in a kind of cognitive radio and can discern user identity, effectively resist the method for simulated main customer attack simultaneously variety of network types is general---utilize the method for signal " transient state fingerprint characteristic " identification user identity.So-called transient signal is meant the non-stationary signal that is full of the radio station fine feature of the non-modulated that the radio station powered on moment produces.The present invention is by utilizing the transient signal in wavelet transformation analysis radio station, and the binding pattern recognition technology is judged the identity of signal sender then.Utilize the present invention can in various forms of cognitive radio networks, realize differentiation, thereby immediately vacate the frequency range effectively attack of antagonism simulated main customer simultaneously for the main user who occurs to the signal sender identity.

Technical solution of the present invention:

A kind of method of resisting simulated main customer attack in the cognitive radio networks, form by following step:

Step 1: the mixed signal that the frequency spectrum perception process receives is carried out noise reduction process.

Consider that a cognitive radio system is made up of M main user and N cognitive user, each cognitive user can both independently be carried out frequency spectrum perception, and the mixed signal that perceives in a certain cognitive user of synchronization comprises main subscriber signal and the tentative q of being of non-main subscriber signal number (because the uncertain q that main user occurs is a random number).Only consider the white noise in the signal, utilize the soft-threshold method of wavelet analysis that mixed signal is carried out noise reduction, can be good at keeping the former characteristic spikes point that makes signal, and computational speed is fast.Because the waveform of small echo has important value to the signal noise silencing effect, therefore to select suitable small echo according to the original mixed signal waveform, thereby reach good de-noising effect choosing of small echo.

Step 2: the number to information source in the mixed signal is estimated.

Correct estimation information source number is the basis of subsequent analysis and processing, the signal that contains what radio station in the mixed signal is estimated the accuracy of its estimation directly influences subsequent analysis and processing.Each signal in the mixed signal that the consideration cognitive user receives is incoherent, and noise is white Gaussian noise (AWGN), noise and signal independence and noise power the unknown.Estimate the information source number in this employing based on information-theoretical method (MDL criterion and AIC criterion).If antenna has p array element, receive signal s from q information source _i(t) (i=1,2..., q).The ergodic Gaussian random process of steadily answering of signal.N snap X (t) is:

$x\left(t\right)=\underset{i=1}{\overset{q}{\mathrm{\Σ}}}A\left({\mathrm{\Φ}}_i\right)s_i\left(t\right)+n\left(t\right)---\left(1\right)$

S wherein _i(t) be t i signal constantly, p * 1 complex vector A (Φ _i) be the incident angle of i signal to each array element.P * 1 complex vector n (t) represents t additive white noise constantly, is the ergodic Gaussian random process of answering stably, and average is zero, and covariance is σ ²I (σ ²The unknown, I is a unit matrix).Signal and noise are separate.The number q in requirement estimated signal source from the N sample.

Consider X={x (1) ..., x (N) } and be independent identically distributed situation, wherein each value all is the multiple Gaussian random vector of zero-mean.Each covariance matrix: R ^(k)=Ψ ^(k)+ σ ²I

Ψ wherein ^(k)Represent order be k ∈ 0,1 ..., the positive semidefinite matrix of p-1}, σ ²Represent unknown noise power.Ψ=ASA ^H, A ^HThe clothes conjugation of expression A, the covariance matrix of S representation signal has

A is the matrix of p * q: Suppose that A is a full rank, S is nonsingular, and then the order of Ψ is q, and p-q the minimal eigenvalue that Ψ is described is zero.The list of feature values of R is shown λ ₁〉=λ ₂〉=... 〉=λ _p, its minimum p-q characteristic value equals σ ²AIC criterion is:

$\mathrm{AIC}\left(k\right)=-2\log {\left[\frac{\underset{i=k+1}{\overset{p}{\mathrm{\Π}}}{l}_i^{1/(p-k)}}{\underset{i=k+1}{\overset{p}{\mathrm{\Σ}}}{l}_i/(p-k)}\right]}^{N(p-k)}+2k(2p-k)---\left(2\right)$

The MDL criterion is:

$\mathrm{MDL}\left(k\right)=-\log {\left[\frac{\underset{i=k+1}{\overset{p}{\mathrm{\Π}}}{l}_i^{1/(p-k)}}{\underset{i=k+1}{\overset{p}{\mathrm{\Σ}}}{l}_i/(p-k)}\right]}^{N(p-k)}+\frac{1}{2}k(2p-k)\log N---\left(3\right)$

The information source number equals to make the k value of AIC or MDL criterion minimum, k ∈ 0,1 ..., p-1}.

Step 3: adopt each signal of Capon Beam-former reconstruct, with the Signal Separation in different radio station.

Wave beam forms and is divided into two classes: classical wave beam is formed on and carries out the information source recovery under the condition of knowing expectation information source direction of arrival in advance; Blind wave beam forms the recovery that then requires to carry out information source under the situation of the unknown expectation information source direction of arrival.When adopting the Capon Beam-former to recover source signal, the direction of supposing a certain signal source is a desired orientation, and the direction of other signal sources is an interference radiating way.Calculate the optimum right vector of this direction, and finally recover the source signal of this direction.

Restoring signal output: $y\left(n\right)=w^{H}x\left(n\right)=\underset{m=1}{\overset{M}{\mathrm{\Σ}}}{w}_m^{*}{x}_m\left(n\right)---\left(4\right)$

W=[w wherein ₁..., w _M] ^TBe weight vector, x (n) is each array element received signal vector.Adopt different weight vector following formulas that the different directions electric wave is had different responses, thereby form the spatial beams of different directions.

Step 4: the signal in each radio station is focused on separation transient signal and modulation signal.

Thereby the singular point that utilizes wavelet transformation to find out between transient signal and the modulation signal separates both.The Lipschitz index can be used for describing singularity of signal, metric signal on one point regularity or in the consistent regularity of a time period.Generally, Signal Singularity in two kinds of situation: signal is carved at a time and is undergone mutation, and makes signal waveform discontinuous, this Lipschitz index α=0; Signal waveform is continuous, and amplitude is sudden change not, but the single order differential of signal is discontinuous at certain point, this Lipsehitz index 0＜α＜1.Function certain point continuously, can be little, be 1 then at the Lipschitz of this some index α.Lipschitz index 0＜α＜1 of transient signal; Modulation signal is a continuously smooth, its Lipschitz index α 〉=1.Therefore can separate transient signal and modulation signal by the Lipschitz index that calculates singular point.

If function f in certain neighborhood that v is ordered be consistent Lipschitz α's and α＞m, then f in this neighborhood must be m time continuously differentiable.Lipschitz index α is big more, and function is smooth more.Utilize small echo then can provide function in an interval even the regularity of a bit.At this, employed small echo must have the vanishing moment of corresponding exponent number.

The Taylor expansion form of function: f (t)=p _v(t)+ε _v(t) (| ε _v(t) |≤K|t-v| ^α).p _vBe the m order polynomial,

Carry out wavelet transformation, have the small echo Ψ (t) and the multinomial quadrature that is less than or equal to the n-1 rank of n rank square, then:

Wf(u，s)＝Wp _v(u，s)+Wε _v(u，s)＝Wε _v(u，s) (5)

ε _v(t) be the amount of weighing Lipschitz.Therefore, can pass through | and Wf (u, s) | approach Lipschitz α.

Log ₂| Wf (u, s) |≤log ₂A+ (α+1/2) log ₂S wherein A is greater than zero constant.(6)

The Lipschitz regularity of point v equals log ₂| Wf (u, s) | as log ₂The function of s deducts 1/2 along the greatest gradient of the very big curve that converges on v, thereby obtains the Lipschitz index of signal.

Step 5: utilize wavelet transformation that transient signal is analyzed, extract characteristic parameter.

The transient signal that separation is obtained carries out wavelet analysis, obtains wavelet coefficient, therefrom chooses a certain proportion of wavelet coefficient as characteristic vector, draws eigenvectors matrix.Continuous wavelet transform (CWT) is:

$X_{a,b}=\frac{1}{\sqrt{\left|a\right|}}\∫x\left(t\right){\mathrm{\Ψ}}^{*}\left(\frac{t-b}{a}\right)\mathrm{dt}---\left(7\right)$

Wherein Ψ (t) is female small echo.A, b are called scale factor and shift factor.

Wavelet transform (DWT) is:

$X_{m,n}=\underset{k}{\mathrm{\Σ}}x\left(k\right){\mathrm{\Ψ}}_{m,n}\left(k\right)---\left(8\right)$

Wherein

Usually select a ₀=2, b ₀=1.

Step 6: realize identification to the signal source identity.

The structure of cognitive radio networks can be divided into centralized and distributed, hypothesis cognitive user end equipment (CPE) can obtain main user's signal characteristic from management of base station (BS) in the centralized network framework, be that cognitive user end equipment has priori to main subscriber signal, the identity that adopts this moment template matching method that the step 5 extraction signal characteristic parameter and the template that prestores are mated to come identification source.In the distributed network framework, it between the cognitive user stand-alone terminal of self-organizing, do not provide the base station of main subscriber signal feature to them, be that the cognitive user terminal does not have priori to main subscriber signal, need this moment to adopt and cut apart the blind identification that clustering algorithm is realized transient signal, after extracting characteristic parameter, will carry out the class number and estimate, determine that Initialization Center carries out Classification and Identification then.

Core innovation of the present invention is to introduce " transient signal fingerprint " notion, on this basis wavelet transformation and two kinds of technology of pattern recognition is combined, thereby reaches the purpose of resisting simulated main customer attack.Concretism is: cognitive user is by putting radio environment around the perception, utilize wavelet transformation to carry out noise reduction, separation, extraction transient signal characteristic parameter to the received signal, according to different network models signal source is discerned then, the first kind (base station is arranged) adopts template matching method to discern main user, and second class (no base station) adopts comprehensive cluster to carry out blind identification.

Beneficial effect of the present invention:

1, the signal source identification of adopting the inventive method to realize can be discerned the identity of Any user, overcome based on the accurate identification source identity and be subjected to restriction of energy measuring main user's priori, introduce " transient signal fingerprint " notion, can avoid interference immediately for the main user that may occur vacates frequency range by the identification transient signal to main user's proper communication.And the method all can be suitable for for centralized and distributed cognition radio network, also all is suitable for for non-cooperation formula and cooperative frequency spectrum perception, and just the precision of discerning when cognitive user is cooperated the formula frequency spectrum perception can be higher.

2, adopt the inventive method not change the mode of operation of existing transmitter, noting be used in increases extra content in transmitting, increase burden for main user, meets the requirement of FCC to cognitive user in the cognitive radio.The method does not have both provisioning requests to main user's priori, and cognitive user only need know that the feature of main user's transient signal gets final product.For main user's transmission mode, coded system does not require, and does not influence main user's communications signal, and the scope of application is wider.

3, adopt the realization signal source identification of the inventive method not only can avoid but also can effectively resist simulated main customer attack the interference of main user's proper communication, thereby and the method can make the precision of identification higher in conjunction with present existing technology.Can effective recognition go out main subscriber signal, effectively reduce the threat of simulation main customer attack.

Description of drawings

Fig. 1 is the whole realization flow figure of the inventive method.

Provided among the figure from receiving mixed signal to signal being carried out preliminary treatment at last to the differentiation that realizes the signal source identity.

Fig. 2 is centralized cognition network flow diagram of portions.

Provided the centralized network structure main user has been had identification process under the situation of priori from obtaining part after the transient signal among the figure.

Fig. 3 is a distributed cognition network flow diagram of portions.

Provided the distributed network structure main user has not been had identification process under the situation of priori from obtaining part after the transient signal among the figure.

Embodiment

The realization of antagonism simulated main customer attack of the present invention, as shown in Figure 1, consider that a cognitive radio system is familiar with the user by M main user and N and forms, each cognitive user can both independently be carried out frequency spectrum perception, consider the perception radio environment on every side that the cognitive user terminal can both be continuous, can receive the transient signal that each radio station powered on moment produces.It is the array antenna of p that cognitive user adopts array number, and the mixed signal that at a time a certain cognitive user perceives comprises main subscriber signal and the tentative q of being of non-main subscriber signal number (because the uncertain q that main user occurs is a random number).The signal source personal identification method that the present invention realizes divides following step:

Step 1: the mixed signal that the frequency spectrum perception process receives is carried out noise reduction process.

Only consider the white noise in the signal, utilize the soft-threshold method of wavelet analysis that mixed signal is carried out noise reduction, can be good at keeping the former characteristic spikes point that makes signal, and computational speed is fast.Because the waveform of small echo has important value to the signal noise silencing effect, therefore to select small echo according to the original mixed signal waveform, thereby reach good de-noising effect choosing of small echo.

Step 2: the number to information source in the mixed signal is estimated.

Correct estimation information source number is the basis of subsequent analysis and processing, the signal that contains what radio station in the mixed signal is estimated the accuracy of its estimation directly influences subsequent analysis and processing.Each signal in the mixed signal that the consideration cognitive user receives is incoherent, and noise is white Gaussian noise (AWGN), noise and signal independence and noise power the unknown.Estimate the information source number in this employing based on information-theoretical method (MDL criterion and AIC criterion).If antenna has p array element, receive signal s from q information source _i(t) (i=1,2..., q).The ergodic Gaussian random process of steadily answering of signal.N snap X (t) is:

$x\left(t\right)=\underset{i=1}{\overset{q}{\mathrm{\Σ}}}A\left({\mathrm{\Φ}}_i\right)s_i\left(t\right)+n\left(t\right)---\left(1\right)$

S wherein _i(t) be t i signal constantly, p * 1 complex vector A (Φ _i) be the incident angle of i signal to each array element.P * 1 complex vector n (t) represents t additive white noise constantly, is the ergodic Gaussian random process of answering stably, and average is zero, and covariance is σ ²I (σ ²The unknown, I is a unit matrix).Signal and noise are separate.The number q in requirement estimated signal source from the N sample.

Consider X={x (1) ..., x (N) } and be independent identically distributed situation, wherein each value all is the multiple Gaussian random vector of zero-mean.Each covariance matrix: R ^(k)=Ψ ^(k)+ σ ²I

Ψ wherein ^(k)Represent order be k ∈ 0,1 ..., the positive semidefinite matrix of p-1}, σ ²Represent unknown noise power.Ψ=ASA ^H, A ^HThe clothes conjugation of expression A, the covariance matrix of S representation signal has

A is the matrix of p * q:

Suppose that A is a full rank, S is nonsingular, and then the order of Ψ is q, and p-q the minimal eigenvalue that Ψ is described is zero.The list of feature values of R is shown λ ₁〉=λ ₂〉=... 〉=λ _p, its minimum p-q characteristic value equals σ ²AIC criterion is:

$\mathrm{AIC}\left(k\right)=-2\log {\left[\frac{\underset{i=k+1}{\overset{p}{\mathrm{\Π}}}{l}_i^{1/(p-k)}}{\underset{i=k+1}{\overset{p}{\mathrm{\Σ}}}{l}_i/(p-k)}\right]}^{N(p-k)}+2k(2p-k)$

The MDL criterion is:

$\mathrm{MDL}\left(k\right)=-\log {\left[\frac{\underset{i=k+1}{\overset{p}{\mathrm{\Π}}}{l}_i^{1/(p-k)}}{\underset{i=k+1}{\overset{p}{\mathrm{\Σ}}}{l}_i/(p-k)}\right]}^{N(p-k)}+\frac{1}{2}k(2p-k)\log N$

The information source number equals to make the k value of AIC or MDL criterion minimum, k ∈ 0,1 ..., p-1}.

Step 3: adopt each signal of Capon Beam-former reconstruct, with the Signal Separation in different radio station.

Wave beam forms and is divided into two classes: classical wave beam is formed on and carries out the information source recovery under the condition of knowing expectation information source direction of arrival in advance; Blind wave beam forms the recovery that then requires to carry out information source under the situation of the unknown expectation information source direction of arrival.When adopting the Capon Beam-former to recover source signal, the direction of supposing a certain signal source is a desired orientation, and the direction of other signal sources is an interference radiating way.Calculate the optimum right vector of this direction, and finally recover the source signal of this direction.

Restoring signal output: $y\left(n\right)=w^{H}x\left(n\right)=\underset{m=1}{\overset{M}{\mathrm{\Σ}}}{w}_m^{*}{x}_m\left(n\right)$

W=[w wherein ₁..., w _M] ^TBe weight vector, x (n) is each array element received signal vector.Adopt different weight vector following formulas that the different directions electric wave is had different responses, thereby form the spatial beams of different directions.

Step 4: the signal in each radio station is focused on separation transient signal and modulation signal.

The singular point that utilizes wavelet transformation to find out between transient signal and the modulation signal separates both.Lipschitz index 0＜α＜1 according to transient signal; Modulation signal is a continuously smooth, its Lipschitz index α 〉=1; Separate transient signal and modulation signal by the Lipschitz index that calculates singular point.

The small echo that utilization has a m rank vanishing moment is given in the regularity of ordering at v of m continuously differentiable function f in v point one neighborhood.Wherein the Taylor expansion form of function is: f (t)=p _v(t)+ε _v(t) (| ε _v(t) |≤K|t-v| ^α).p _vBe the m order polynomial,

Carry out wavelet transformation, have the small echo Ψ (t) and the multinomial quadrature that is less than or equal to the n-1 rank of n rank square, then:

Wf(u，s)＝Wp _v(u，s)+Wε _v(u，s)＝Wε _v(u，s)

Can pass through | and Wf (u, s) | approach Lipschitz α.

Log ₂| Wf (u, s) |≤log ₂A+ (α+1/2) log ₂S wherein A is greater than zero constant.

The Lipschitz regularity of point v equals log ₂| Wf (u, s) | as log ₂The function of s deducts 1/2 along the greatest gradient of the very big curve that converges on v, thereby obtains the approximation of the Lipschitz index of signal | Wf (u, s) |.

Step 5: utilize wavelet transformation that transient signal is analyzed, extract characteristic parameter.

The transient signal that separation is obtained carries out wavelet analysis, obtains wavelet coefficient, therefrom chooses a certain proportion of wavelet coefficient (as maximum and standard deviation etc.) as characteristic vector, draws eigenvectors matrix.Continuous wavelet transform (CWT) is:

$X_{a,b}=\frac{1}{\sqrt{\left|a\right|}}\∫x\left(t\right){\mathrm{\Ψ}}^{*}\left(\frac{t-b}{a}\right)\mathrm{dt}$

Wherein Ψ (t) is female small echo.A, b are called scale factor and shift factor.Wavelet transform (DWT) is:

$X_{m,n}=\underset{k}{\mathrm{\Σ}}x\left(k\right){\mathrm{\Ψ}}_{m,n}\left(k\right)$

Wherein

Usually select a ₀=2, b ₀=1.

Step 6: realize signal source is carried out identification.

The structure of cognitive radio networks can be divided into centralized and distributed, as shown in Figure 2, hypothesis cognitive user end equipment (CPE) can obtain main user's signal characteristic from management of base station (BS) in the centralized network framework, be that cognitive user end equipment has priori to main subscriber signal, the identity that adopts this moment template matching method that the step 5 extraction signal characteristic parameter and the template that prestores are mated to come identification source.As shown in Figure 3, in the distributed network framework, it between the cognitive user stand-alone terminal of self-organizing, do not provide the base station of main subscriber signal feature to them, be that the cognitive user terminal does not have priori to main subscriber signal, need this moment to adopt to cut apart the blind identification that clustering algorithm is realized transient signal, after extracting characteristic parameter, will carry out the class number and estimate, determine that Initialization Center carries out Classification and Identification then.

By above step, just can realize in the cognitive radio system identification to the signal source identity, promptly realize the antagonism simulated main customer attack.

Claims (1)

1. the method for antagonism simulated main customer attack in the cognitive radio, suppose that a cognitive radio system is made up of M main user and N cognitive user, the mixed signal number that the frequency spectrum perception process receives is random number q, it is characterized in that, is realized by following step:

Step 1: the mixed signal that the frequency spectrum perception process receives is carried out noise reduction process

Only consider the white noise in the signal, utilize the soft-threshold method of wavelet analysis that mixed signal is carried out noise reduction,, reach de-noising effect choosing of small echo according to the selected small echo of original mixed signal waveform;

Step 2: the number to information source in the mixed signal is estimated

If antenna has p array element, receive signal s from q information source _i(t) (i=1,2..., q), and the ergodic Gaussian random process of steadily answering of signal, N snap x (t) is:

$x\left(t\right)=\underset{i=1}{\overset{q}{\mathrm{\Σ}}}A\left({\mathrm{\Φ}}_i\right)s_i\left(t\right)+n\left(t\right)$

S wherein _i(t) be t i signal constantly, p * 1 complex vector A (Φ _i) be the incident angle of i signal to each array element; P * 1 complex vector n (t) represents t additive white noise constantly, is the ergodic Gaussian random process of answering stably, and average is zero, and covariance is σ ²I (σ ²The unknown, I is a unit matrix), signal and noise are separate, require the number q in estimated signal source from the N sample, consider X={x (1) ..., x (N) } be independent identically distributed situation, wherein each value all is the multiple Gaussian random vector of zero-mean, each covariance matrix:

R ^(k)＝Ψ ^(k)+σ ²I

Ψ wherein ^(k)Represent order be k ∈ 0,1 ..., the positive semidefinite matrix of p-1}, σ ²Represent unknown noise power; Ψ=ASA ^H, A ^HThe clothes conjugation of expression A, the covariance matrix of S representation signal has

A is the matrix of p * q:

, supposing that A is a full rank, S is nonsingular, and then the order of Ψ is q, and p-q the minimal eigenvalue that Ψ is described is zero, and the list of feature values of R is shown λ ₁〉=λ ₂〉=... 〉=λ _p, its minimum p-q characteristic value equals σ ², AIC criterion is:

$\mathrm{AIC}\left(k\right)=-2\log {\left[\frac{\underset{i=k+1}{\overset{p}{\mathrm{\Π}}}{l}_i^{1/(p-k)}}{\underset{i=k+1}{\overset{p}{\mathrm{\Σ}}}{l}_i/(p-k)}\right]}^{N(p-k)}+2k(2p-k)$

The MDL criterion is:

$\mathrm{MDL}\left(k\right)=-\log {\left[\frac{\underset{i=k+1}{\overset{p}{\mathrm{\Π}}}{l}_i^{1/(p-k)}}{\underset{i=k+1}{\overset{p}{\mathrm{\Σ}}}{l}_i/(p-k)}\right]}^{N(p-k)}+\frac{1}{2}k(2p-k)\log N$

The information source number equals to make the k value of AIC or MDL criterion minimum, k ∈ 0,1 ..., p-1};

Step 3: adopt each signal of Capon Beam-former reconstruct, with the Signal Separation in different radio station

$y\left(n\right)=w^{H}x\left(n\right)=\underset{m=1}{\overset{M}{\mathrm{\Σ}}}{w}_m^{*}{x}_m\left(n\right)$

W=[w wherein ₁..., w _M] ^TBe weight vector, x (n) is each array element received signal vector, adopts different weight vector following formulas that the different directions electric wave is had different responses, forms the spatial beams of different directions;

Step 4: the signal in each radio station is focused on separation transient signal and modulation signal

Thereby the singular point that utilizes wavelet transformation to find out between transient signal and the modulation signal separates both, because the Lipschitz index can be used for describing singularity of signal, in order to metric signal on one point regularity or in consistent regularity---Lipschitz index 0＜α＜1 of transient signal of a time period, modulation signal is a continuously smooth, its Lipschitz index α 〉=1; So separate transient signal and modulation signal by the Lipschitz index that calculates singular point;

The Taylor expansion form of function: f (t)=p _v(t)+ε _v(t) (| ε _v(t) |≤K|t-v| ^α), p _vBe the m order polynomial, Carry out wavelet transformation, have the small echo Ψ (t) and the multinomial quadrature that is less than or equal to the n-1 rank of n rank square, then:

Wf(u，s)＝Wp _v(u，s)+Wε _v(u，s)＝Wε _v(u，s)

By | Wf (u, s) | approach Lipschitz α, log ₂| Wf (u, s) |≤log ₂A+ (α+1/2) log ₂S, wherein A is the constant greater than zero, the Lipschitz regularity log of some v ₂| Wf (u, s) | as log ₂The function of s deducts 1/2 along the greatest gradient of the very big curve that converges on v, obtains the Lipschitz index of signal;

Step 5: utilize wavelet transformation that transient signal is analyzed, extract characteristic parameter

The transient signal that separation is obtained carries out wavelet analysis, obtains wavelet coefficient, therefrom chooses a certain proportion of wavelet coefficient as characteristic vector, draws eigenvectors matrix, and continuous wavelet transform CWT is:

$X_{a,b}=\frac{1}{\sqrt{\left|a\right|}}\∫x\left(t\right){\mathrm{\Ψ}}^{*}\left(\frac{t-b}{a}\right)\mathrm{dt}$

Wherein Ψ (t) is female small echo, and a, b are called scale factor and shift factor,

Wavelet transform DWT is:

$X_{m,n}=\underset{k}{\mathrm{\Σ}}x\left(k\right){\mathrm{\Ψ}}_{m,n}\left(k\right)$

Ψ wherein _{M, n}(t)=a ₀ ^-1/2Ψ _{M, n}(a ₀ ^-mT-nb ₀), select a usually ₀=2, b ₀=1;

Step 6: realize signal source is carried out identification

The structure of cognitive radio networks is divided into centralized and distributed, cognitive user end equipment CPE can obtain main user's signal characteristic from management of base station BS in the centralized network framework, the identity that adopts this moment template matching method that the step 5 extraction signal characteristic parameter and the template that prestores are mated to come identification source, in the distributed network framework, it between the cognitive user stand-alone terminal of self-organizing, do not provide the base station of main subscriber signal feature to them, adopt this moment and cut apart the blind identification that clustering algorithm is realized transient signal, after extracting characteristic parameter, carry out the class number and estimate, determine that Initialization Center carries out Classification and Identification then.

Images