CN102546059B - Non-supervision clustering-based distributed cooperative spectrum sensing method for cognitive self-organizing network - Google Patents

Abstract

A non-supervision clustering-based distributed cooperative spectrum sensing method for a cognitive self-organizing network relates to the field of cognitive radio in wireless communication technology. Aiming at solving the problems of difficult distributed cooperation of the cognitive self-organizing network and large overhead of whole-network cooperation, the method adopts the latest achievements of a non-supervision clustering theory and a co-recognition theory to achieve fully-distributed, steady and reliable distributed cooperative spectrum sensing under the condition of simplifying the overhead of network sensing; and users with potential optimal detection performance spontaneously gather only via information interaction between neighbors, further the users carry out cooperation spectrum sensing by utilizing an average co-recognition protocol, and a sensing result is broadcast to the whole network users. The method does not require local users to receive apriori information of noise-signal ratio, and does not need any central controllers, thereby greatly lowering sensing overhead and acquiring detection performances similar to optimal soft combination solution.

Description

Distributed cooperation frequency spectrum sensing method based on without supervision clustering in cognitive self-organizing network

Technical field

The present invention relates to the cognition wireless electrical domain in wireless communication technology, is specifically a kind of new method that realizes distributed cooperation frequency spectrum perception in cognitive self-organizing network without supervision clustering theory and the theoretical latest developments of knowing together of applying.

Background technology

At present, along with the rapid growth of radio communication service kind, the demand of radio spectrum resources is also to exponential increase, frequency spectrum resource " scarcity " problem in future wireless system is become increasingly conspicuous.Cognitive radio technology utilizes the idle frequency range of authorized user under the condition that guarantees authorized user service quality in the mode of " waiting for an opportunity to access ", greatly improve the service efficiency of frequency spectrum, be the effective ways that solve " frequency spectrum scarcity " problem, there is important practical significance and wide application prospect.Frequency spectrum perception technology is used to effectively detect the operating state of current authorized user, to find spectrum opportunities and to avoid the interference to authorized user or primary user's (primary user is called for short PU).Therefore, effectively frequency spectrum perception technology is prerequisite and the basis of the normal work of cognition wireless network.

Due to single cognitive user or secondary user's (secondary user, be called for short SU) frequency spectrum perception performance be very easily subject to the impact of the factors such as shadow effect in wireless channel, multipath fading, hidden terminal and exposed terminal and worsen, the method that people have proposed many SU cooperative spectrum sensing (cooperative spectrum sensing is called for short CSS) overcomes these problems.

From whether there is the angle of fusion center, current CSS method mainly comprises following two classes:

Center type CSS: in center type CSS, first each SU carries out local frequency spectrum perception, then sensing results is uploaded to fusion center, fusion center carries out the sensing results of each SU after data fusion, to make the judgement whether spectrum opportunities exists.Current, gradually ripe about the research of center type CSS, the advantage of this method is easily to realize obtaining and the optimization of the whole network perceptual performance of the whole network information; The deficiency of the method is too to rely on the infrastructure such as fusion center, easily loses efficacy because single point failure makes method, and extensibility and the robustness of network are poor.

Distributed C/S S: in Distributed C/S S, first each SU carries out local frequency spectrum perception, then only and between neighbours, carry out information interaction, fusion, through limited number of time iteration, final each SU independently makes the judgement whether spectrum opportunities exists to each SU.This Distributed C/S S method does not rely on the infrastructure such as fusion center, and robustness and the extensibility of network are better.Given this advantage, causes academia and industrial quarters broad interest gradually without center, adaptive cognitive self-organizing network in recent years, also starts to be gradually subject to research staff's close attention about the design of Distributed C/S S method.

Current Distributed C/S S method is only considered the scene that network size is less, and supposes that all users participate in cooperation.But in the time that the number of SU in considered cognitive self-organizing network is more, all SU participate in the huge perception expense that cooperation will bring; Meanwhile, considering the factors such as path loss, multipath fading and shadow effect, also can there is significant difference in the detecting reliability of the SU in different spatial.Therefore, how effectively to excavate and utilize these difference, under the brief condition of network aware expense, realize sane, frequency spectrum perception is one and has important theory significance and the problem of practical value reliably.

The powerful addressing the above problem without supervision clustering (Unsupervised clustering) theory.Its thought derives from the observation analysis to biocenose intelligence phenomenon, for example Flight of geese, bee colony gathering honey etc. at first.In recent years, be widely used (list of references: Pedro A F without supervision clustering theory in fields such as distributed control and decision-making, multiple agent cooperation and sensor network distribution type parameter Estimation, Alfonso C, Georgios B G, " Distributed clustering using wireless sensor networks; " IEEE J Sel Topics Signal Process, 2011,5 (4): 707-724).As the latest theories progress of pattern recognition and artificial intelligence field, core concept without supervision clustering theory is: " without tutor's self-study ", be that in network, each user obtains the observed quantity to environment first separately, based on without supervision clustering agreement, each user and neighbours carry out information interaction, through iteration repeatedly, the in the situation that of no center control telegon, the user that performance is close can spontaneously be brought together.

Common recognition (Consensus) theory is to realize the key technology that between user, distributed data merges.Its basic principle is: each user has at first different separately environments and measures, based on common recognition agreement, each user and neighbours carry out information interaction, through iteration repeatedly, the in the situation that of no center control telegon, between end user, form common recognition or consistency understanding (list of references: R.Olfati-Saber that environment is measured, J.Fax, and R.Murray, " Consensus and cooperation in networked multi-agent systems; " Proc IEEE, 2007,95 (1): 215-233).

Summary of the invention

The object of the invention is for distributed cooperation difficulty in cognitive self-organizing network, problem that the whole network cooperation expense is large, the latest developments that integrated application is theoretical without supervision clustering and common recognition is theoretical, realize sane, reliable distributed cooperation frequency spectrum perception under the brief condition of perception expense.

Technical scheme of the present invention is:

Distributed cooperation frequency spectrum sensing method based on without supervision clustering in a kind of cognitive self-organizing network, without control centre in the situation that, first determine the cognitive user S set BS with optimal perceived performance by unsupervised clustering, then detect based on the theoretical cooperation frequency spectrum of realizing between the multiple cognitive user SUs in SBS of common recognition, obtain the probability that corresponding frequency spectrum takies, finally utilize broadcast mechanism that testing result is informed to the multiple cognitive user SUs outside SBS.

The present invention specifically comprises the following steps:

Step 1. parameter initialization:

First each SU obtains self observed quantity to environment, then local class barycenter and the local Lagrange multiplier of random initializtion self to SBS class and non-SBS class, on this basis, the local class ownership of each SU initialization coefficient;

Described " class barycenter " refers to the weighted average that the environment of all cognitive user in class is measured;

Described " Lagrange multiplier " is the middle transition variable of clustering algorithm, there is no concrete physical significance;

Described " class ownership coefficient " is to characterize the possibility that it belongs to SBS class and non-SBS class;

Step 2. is determined SBS based on unsupervised clustering:

Each SU first with the mutual local class barycenter of a hop neighbor SUs, based on this interactive information, each SU upgrades local class barycenter, local Lagrange multiplier and local class ownership coefficient successively, this process iteration is carried out, until stopping criterion for iteration meets;

After algorithm iteration stops, in network, the local class barycenter of all SUs will be tending towards identical, reach " all SU class barycenter common recognitions ", but the local class ownership coefficient that each SU obtains is different.Based on this, each SU carries out the judgement of class ownership according to the local class ownership coefficient of self, realizes without supervision clustering;

Step 3. realizes SBS distribution within class formula cooperation frequency spectrum based on common recognition theory and detects:

First each SU in SBS class carries out local energy perception, and mutual local energy measured value then and between neighbours SUs, carries out data fusion based on common recognition agreement, and through iteration repeatedly, all SUs in final SBS class reach the common recognition to spectrum energy measured value;

Based on the Perspective of Energy measured value of common recognition, each SU carries out this locality judgement, and to obtain frequency spectrum state-detection result be the frequency spectrum free time or take;

Step 4. is broadcasted testing result:

Testing result is broadcast to the neighbours SU outside class by each SU of SBS class, realizes the whole network SU sensing results is reached common understanding.

Unsupervised clustering of the present invention comprises the following steps:

Step 1. parameter initialization:

1.1 based on historical perception information { E _i(m) | m=1 ..., M}, each SUi ∈ in network 1 ..., first N} obtains self observed quantity to spectrum environment:

$O_i=\frac{1}{M}\underset{m=1}{\overset{M}{\mathrm{\Σ}}}{E}_i\left(m\right)$

Wherein E _i(m) energy value detecting while being the m time perception, M is cumulative number of times; In the time that frequency spectrum is idle, E _i(m) only comprise noise energy; In the time that frequency spectrum takies, E _i(m) be the energy that not only comprises noise, also comprise the energy of PU signal.

Each SUi ∈ in 1.2 networks 1 ..., N} random initializtion

k ∈ 1,2} and k ∈ 1,2}, wherein

k ∈ 1,2} be respectively SUi ∈ 1 ..., the SBS class of N} this locality and the initial barycenter of non-SBS class, described barycenter refers to the weighted average of the observed quantity of all cognitive user in class,

k ∈ 1,2} be respectively SUi ∈ 1 ..., " Lagrange multiplier " described in the SBS class of N} this locality and the initial Lagrange multiplier of non-SBS class is the middle transition variable of clustering algorithm, there is no concrete physical significance;

Each cognitive user SUi ∈ in 1.3 networks 1 ..., the local class ownership of N} initialization coefficient

$a_i^0\left(k\right)=\frac{{\left|\right|O_i-c_i^0\left(k\right)\left|\right|}^{-\frac{2}{p-1}}}{{\mathrm{\Σ}}_{k=1}^K{\left|\right|O_i-c_i^0\left(k\right)\left|\right|}^{-\frac{2}{p-1}}},\∀i\∈\{1,...,N\},\∀k\∈\left\{\mathrm{1,2}\right\}$

Wherein, p > 1.Here

more approach 1, user SUi adds the possibility of class k larger; Otherwise,

more approach 0, user SUi adds the possibility of class k less;

Step 2. is based on determining SBS without supervision clustering:

Each SU first with the mutual local class barycenter of a hop neighbor SUs, based on this interactive information, each SU upgrades local class barycenter, local Lagrange multiplier and local class ownership coefficient successively, this process iteration is carried out, until stopping criterion for iteration meets; After iteration stops, each SU obtains local class ownership coefficient and local class barycenter (note: when after iteration convergence, in network, all SUs are identical to of a sort local class barycenter, reach " class barycenter common recognition ", but local class ownership coefficient is different); On this basis, each SU carries out the judgement of class ownership according to the local class ownership coefficient obtaining, and realizes without supervision clustering;

Concrete by carrying out following distributed iterative method realization: based on t=0,1,2 ... the local class barycenter that inferior iteration obtains each SUi, i ∈ 1 ..., N} carries out iteration the t+1 time:

2.1 each SUi ∈ 1 ..., N} is by local class barycenter

be broadcast to a hop neighbor user

here S _irefer to the set of a hop neighbor of SUi, d _ijrepresent the distance between cognitive user SUi and one hop neighbor user SUj, d _comrepresent between two SU can proper communication ultimate range;

2.2 each SUi ∈ 1 ..., N} upgrades local class barycenter, obtains:

$c_i^{t+1}\left(k\right)={\left(a_i^{t+1}\left(k\right)2\mathrm{\η}\right|S_i\left|\right)}^{-1}\{a_i^{t+1}\left(k\right)O_i-2{\mathrm{\λ}}_i^t\left(k\right)+\mathrm{\η}\underset{j\∈S_i}{\mathrm{\Σ}}[c_i^t\left(k\right)+c_J^t\left(k\right)\left]\right\},\∀i\∈\{1,...,N\},k\∈\left\{\mathrm{1,2}\right\}$

Wherein, η > 0; | S _i| the number of element in a hop neighbor S set i of expression SUi;

being a local Lagrange multiplier dynamically updating, is the middle transition variable of clustering algorithm, there is no concrete physical significance, and its update rule is according to following step 2.4.

2.3 each SUi ∈ 1 ..., N} upgrades local class ownership coefficient, obtains:

$a_i^{t+1}\left(k\right)=\frac{{\left|\right|O_i-c_i^{t+1}\left(k\right)\left|\right|}^{-\frac{2}{p-1}}}{{\mathrm{\Σ}}_{k=1}^K{\left|\right|O_i-c_i^{t+1}\left(k\right)\left|\right|}^{-\frac{2}{p-1}}},\∀i\∈\{1,...,N\},\∀k\∈\left\{\mathrm{1,2}\right\}$

Wherein, p > 1; In reality, often get p=2.Here

more approach 1, user SUi adds the possibility of class k larger; Otherwise,

more approach 0, user SUi adds the possibility of class k less;

2.4 each SUi ∈ 1 ..., N} upgrades local Lagrange multiplier, obtains:

${\mathrm{\λ}}_i^{t+1}\left(k\right)={\mathrm{\λ}}_i^t\left(k\right)+\frac{\mathrm{\η}}{2}\underset{j\∈S_i}{\mathrm{\Σ}}[c_j^{t+1}\left(k\right)-c_i^{t+1}\left(k\right)],\∀i\∈\{1,...,N\},k\∈\left\{\mathrm{1,2}\right\}$

Step 2.1-2.4 iteration is carried out, through iteration repeatedly, if condition

with meet, iteration stops simultaneously; Wherein, ε _a, ε _cand ε _λbe the positive number close to 0, in reality, often get ε _a=ε _c=ε _λ∈ [10 ^-6, 10 ^-3], value is less, restrains slowlyer, and convergence precision is higher; If end condition does not meet, rebound step 2.1, if condition meets, iteration stops; (note: what described end condition showed is key parameters

with

relative increment no longer there is significant change with the increase of iterations t.)

After 2.5 iteration stop, each SUi, i ∈ 1 ..., N} carries out the judgement of local class ownership according to following rule:

Each SUi, i ∈ 1 ..., definite SBS class or the non-SBS class of entering of N};

When

sUi enters the class of k=1, and will

be set to 1,

be set to 0;

Otherwise work as

sUi enters the class of k=2, and will be set to 0,

be set to 1;

Through the class ownership judgement of step 2.5, all SU that belong to SBS class spontaneously flock together, and form SBS user's collection:

Wherein,

represent barycenter larger class, i.e. SBS class,

represent that SUi belongs to SBS class k ^◇, S _sBSrepresent all SBS class k that belong to ^◇the set of SU;

Step 3. realizes SBS distribution within class formula cooperation frequency spectrum based on common recognition theory and detects:

In step 2, form on the basis of SBS class to the SUs self-organizing of optimal perceived performance, first each SU in SBS class carries out local energy perception, then mutual local energy measured value and between neighbours SUs, carry out data fusion based on common recognition agreement, through iteration repeatedly, all SUs in final SBS class reach the common recognition to spectrum energy measured value, based on this common recognition, each SU carries out this locality judgement, obtains frequency spectrum state-detection result and be frequency spectrum idle or take:

Concrete by carrying out following distributed iterative method realization: based on t=0,1,2 ... this locality common recognition variable that inferior iteration obtains

each SUi, i ∈ 1 ..., N} carries out iteration the t+1 time:

3.1 initialization: SBS class is optimal perceived performance cognitive user S set _sBSinterior each SUi ∈ S _sBScarry out local energy detection, obtain Perspective of Energy measured value E _iand its initial local common recognition variable is made as

3.2 each SUi ∈ S _sBSwith the one hop neighbor variable of knowing together alternately

be each SUi ∈ S _sBSby its common recognition variate-value be broadcast to a hop neighbor user

receive the common recognition variable from a hop neighbor cognitive user SU simultaneously

3.3 each SUi ∈ S _sBScarry out information fusion according to following common recognition agreement:

$x_i^{t+1}=x_i^t+s_x\underset{j\∈S_i}{\mathrm{\Σ}}(x_j^t-x_i^t)$

Wherein s _x> 0 is iteration step length, conventionally gets

If condition

meet, iteration stops, the whole network asymptotic reaching of on average knowing together, and consensus value is asymptotic is

wherein ε _xbe the positive number close to 0, in reality, often get ε _a=ε _c=ε _λ∈ [10 ^-6, 10 ^-3], value is less, restrains slowlyer, and convergence precision is higher; If condition does not meet, rebound step 3.2, if condition meets, iteration stops; (note: what described end condition showed is key parameters relative increment no longer there is significant change with the increase of iterations t.)

Once 3.4 iteration stop, each SU obtains final common recognition variate-value x ^*, carry out following local judgement,

Wherein λ is the decision threshold of frequency spectrum detection, and λ detects performance working point (P corresponding to one _fa, P _d), P _farefer to false alarm probability, i.e. the actual spectrum free time, court verdict is the probability that frequency spectrum takies; P _dbe detection probability, actual spectrum takies originally, the probability that court verdict also takies for frequency spectrum;

Step 4. is broadcasted testing result, completes following work:

Each SUi ∈ S in SBS class _sBStesting result d is broadcast to neighbours SU outside class (being the user who belongs to non-SBS class in the neighbours of SUi), thereby makes the user of non-SBS class upgrade the cognition for frequency spectrum free time/seizure condition, realize the whole network SU sensing results is reached common understanding.

In step 1 of the present invention, in the time that frequency spectrum is idle, E _i(m) only comprise noise energy; In the time that frequency spectrum takies, E _i(m) be the energy that not only comprises noise, also comprise the energy of PU signal.

P=2 of the present invention.

Beneficial effect of the present invention:

1, network operation is full distributed.Suggest plans, without any need for central coordinator (as base station, access point, bunch first-class), all information interactions only carry out between neighbours.Therefore, suggest plans and possess that robustness is strong, network scalability good and the advantage such as network overhead is little.

The complexity of 2, suggesting plans is very low.On the one hand, suggest plans in each SU do not need to carry out the estimation of himself received signal to noise ratio, do not need the prior information of PU position yet; On the other hand, suggest plans and do not need to spend extra time overhead and obtain the required observed quantity of cluster because this observed quantity utilization is historical detection information, this information can be learnt to obtain by the mode of off-line.

3, suggest plans, in obtaining compared with high detection reliability, greatly reduces network overhead.Emulation shows, the present invention suggests plans and can obtain the detection performance close with optimum soft information Merge Scenarios, but institute suggests plans and only need part SU to participate in cooperation, and information interaction amount reduces greatly, and while distributed iterative algorithm convergence rate is obviously accelerated.

Accompanying drawing explanation

Fig. 1 is cognitive radio system frame assumption diagram designed in the present invention.

Fig. 2 is method flow diagram of the present invention.

Fig. 3 is the result schematic diagram of instantiation artificial network model and cluster scheme in the present invention.

Fig. 4 is suggest plans in the present invention and the comparison schematic diagram of the receiver operating characteristic curves of traditional scheme.

Embodiment

Below in conjunction with drawings and Examples, the present invention is further illustrated.

As shown in Figure 1.A kind of cognitive radio system frame structure that the present invention is designed.This frame structure is made up of four essential parts: cluster period, perception period, radio slot and transfer of data period.The cluster period realizes distributed node and selects, and the SU with optimum detection performance spontaneously assembles formation SBS class; The perception period is realized the distributed cooperation frequency spectrum detection between SU in SBS class; In radio slot, sensing results is broadcast to the neighbours SU outside class by the SU in SBS class; In the transfer of data period, if sensing results is the frequency spectrum free time, carry out transfer of data, if frequency spectrum is taken by PU, mourn in silence and wait for the arrival of next frame.Make T _fthe total length that represents a basic frame, we define a basic frame and are made up of perception period, radio slot and transmission period.Notice that the cluster period is every N _f=T _c/ T _findividual basic frame activates once, wherein N _fthe network topology change frequency causing with SU mobility is relevant.

As shown in Figure 2.The flow chart of method of the present invention.

1. parameter initialization:

1.1 based on historical perception information { E _i(m) | m=1 ..., M}, each SUi ∈ in network 1 ..., first N} obtains self observed quantity to spectrum environment:

$O_i=\frac{1}{M}\underset{m=1}{\overset{M}{\mathrm{\Σ}}}{E}_i\left(m\right)$

Wherein E _i(m) energy value detecting while being the m time perception, M is cumulative number of times; M=100 in the following embodiments; In the time that frequency spectrum is idle, E _i(m) only comprise noise energy; In the time that frequency spectrum takies, E _i(m) be the energy that not only comprises noise, also comprise the energy of PU signal.

Each SUi ∈ in 1.2 networks 1 ..., N} random initializtion

k ∈ 1,2} and

k ∈ 1,2}, wherein

k ∈ 1,2} be respectively SUi ∈ 1 ..., the SBS class of N} this locality and the initial barycenter of non-SBS class, described barycenter refers to the weighted average of the observed quantity of all cognitive user in class, k ∈ 1,2} be respectively SUi ∈ 1 ..., " Lagrange multiplier " described in the SBS class of N} this locality and the initial Lagrange multiplier of non-SBS class is the middle transition variable of clustering algorithm, there is no concrete physical significance; In the following embodiments

Each cognitive user SUi ∈ in 1.3 networks 1 ..., the local class ownership of N} initialization coefficient

$a_i^0\left(k\right)=\frac{{\left|\right|O_i-c_i^0\left(k\right)\left|\right|}^{-\frac{2}{p-1}}}{{\mathrm{\Σ}}_{k=1}^K{\left|\right|O_i-c_i^0\left(k\right)\left|\right|}^{-\frac{2}{p-1}}},\∀i\∈\{1,...,N\},\∀k\∈\left\{\mathrm{1,2}\right\}$

Wherein, p > 1, often gets p=2 in reality.Here

more approach 1, user SUi adds the possibility of class k larger; Otherwise, more approach 0, user SUi adds the possibility of class k less;

2. based on determining SBS without supervision clustering:

Each SU first with the mutual local class barycenter of a hop neighbor SUs, based on this interactive information, each SU upgrades local class barycenter, local Lagrange multiplier and local class ownership coefficient successively, this process iteration is carried out, until stopping criterion for iteration meets; After iteration stops, each SU obtains local class ownership coefficient and local class barycenter (note: when after iteration convergence, in network, all SUs are identical to of a sort local class barycenter, reach " class barycenter common recognition ", but local class ownership coefficient is different); On this basis, each SU carries out the judgement of class ownership according to the local class ownership coefficient obtaining, and realizes without supervision clustering;

Concrete by carrying out following distributed iterative method realization: based on t=0,1,2 ... the local class barycenter that inferior iteration obtains

each SUi, i ∈ 1 ..., N} carries out iteration the t+1 time:

2.1 each SUi ∈ 1 ..., N} is by local class barycenter

be broadcast to a hop neighbor user

here S _irefer to the set of a hop neighbor of SUi, d _ijrepresent the distance between cognitive user SUi and one hop neighbor user SUj, d _comrepresent between two SU can proper communication ultimate range;

2.2 each SUi ∈ 1 ..., N} upgrades local class barycenter, obtains:

$c_i^{t+1}\left(k\right)={\left(a_i^{t+1}\left(k\right)2\mathrm{\η}\right|S_i\left|\right)}^{-1}\{a_i^{t+1}\left(k\right)O_i-2{\mathrm{\λ}}_i^t\left(k\right)+\mathrm{\η}\underset{j\∈S_i}{\mathrm{\Σ}}[c_i^t\left(k\right)+c_J^t\left(k\right)\left]\right\},\∀i\∈\{1,...,N\},k\∈\left\{\mathrm{1,2}\right\}$

Wherein, η > 0; | S _i| represent a hop neighbor S set of SUi _ithe number of middle element;

being a local Lagrange multiplier dynamically updating, is the middle transition variable of clustering algorithm, there is no concrete physical significance, and its update rule is according to following step 2.4.

2.3 each SUi ∈ 1 ..., N} upgrades local class ownership coefficient, obtains:

$a_i^{t+1}\left(k\right)=\frac{{\left|\right|O_i-c_i^{t+1}\left(k\right)\left|\right|}^{-\frac{2}{p-1}}}{{\mathrm{\Σ}}_{k=1}^K{\left|\right|O_i-c_i^{t+1}\left(k\right)\left|\right|}^{-\frac{2}{p-1}}},\∀i\∈\{1,...,N\},\∀k\∈\left\{\mathrm{1,2}\right\}$

Wherein, p > 1; In reality, often get p=2.Here

more approach 1, user SUi adds the possibility of class k larger; Otherwise,

more approach 0, user SUi adds the possibility of class k less;

2.4 each SUi ∈ 1 ..., N} upgrades local Lagrange multiplier, obtains:

${\mathrm{\λ}}_i^{t+1}\left(k\right)={\mathrm{\λ}}_i^t\left(k\right)+\frac{\mathrm{\η}}{2}\underset{j\∈S_i}{\mathrm{\Σ}}[c_j^{t+1}\left(k\right)-c_i^{t+1}\left(k\right)],\∀i\∈\{1,...,N\},k\∈\left\{\mathrm{1,2}\right\}$

Step 2.1-2.4 iteration is carried out, through iteration repeatedly, if condition

with

meet, iteration stops simultaneously; Wherein, ε _a, ε _cand ε _λbe the positive number close to 0, in reality, often get ε _a=ε _c=ε _λ∈ [10 ^-6, 10 ^-3], value is less, restrains slowlyer, and convergence precision is higher; ε in following embodiment _a=ε _c=ε _λ=10 ^-4; If end condition does not meet, rebound step 2.1; (note: what described end condition showed is key parameters

with

relative increment no longer there is significant change with the increase of iterations t.)

After 2.5 iteration stop, each SUi, i ∈ 1 ..., N} carries out the judgement of local class ownership according to following rule:

Each SUi, i ∈ 1 ..., definite SBS class or the non-SBS class of entering of N};

When

sUi enters the class of k=1, and will

be set to 1, be set to 0;

Otherwise work as

sUi enters the class of k=2, and will

be set to 0, be set to 1;

Through the class ownership judgement of step 2.5, all SU that belong to SBS class spontaneously flock together, and form SBS user's collection:

Wherein, represent barycenter larger class, i.e. SBS class,

represent that SUi belongs to SBS class k ^◇, S _sBSrepresent all SBS class k that belong to ^◇the set of SU;

3. realizing SBS distribution within class formula cooperation frequency spectrum based on common recognition theory detects:

In step 2, form on the basis of SBS class to the SUs self-organizing of optimal perceived performance, first each SU in SBS class carries out local energy perception, then mutual local energy measured value and between neighbours SUs, carry out data fusion based on common recognition agreement, through iteration repeatedly, all SUs in final SBS class reach the common recognition to spectrum energy measured value, based on this common recognition, each SU carries out this locality judgement, obtains frequency spectrum state-detection result and be frequency spectrum idle or take:

Concrete by carrying out following distributed iterative method realization: based on t=0,1,2 ... this locality common recognition variable that inferior iteration obtains

each SUi, i ∈ 1 ..., N} carries out iteration the t+1 time:

3.1 initialization: SBS class is optimal perceived performance cognitive user S set _sBSinterior each SUi ∈ S _sBScarry out local energy detection, obtain Perspective of Energy measured value E _iand its initial local common recognition variable is made as

3.2 each SUi ∈ S _sBSwith the one hop neighbor variable of knowing together alternately

be each SUi ∈ S _sBSby its common recognition variate-value

be broadcast to a hop neighbor user

receive the common recognition variable from a hop neighbor cognitive user SU simultaneously

3.3 each SUi ∈ S _sBScarry out information fusion according to following common recognition agreement:

$x_i^{t+1}=x_i^t+s_x\underset{j\∈S_i}{\mathrm{\Σ}}(x_j^t-x_i^t)$

Wherein s _x> 0 is iteration step length, conventionally gets

If condition

meet, iteration stops, the whole network asymptotic reaching of on average knowing together, and consensus value is asymptotic is

wherein ε _xbe the positive number close to 0, in reality, often get ε _a=ε _c=ε _λ∈ [10 ^-6, 10 ^-3], value is less, restrains slowlyer, and convergence precision is higher; ε in the following embodiments _x=10 ^-4; If condition does not meet, rebound step 3.2; (note: what described end condition showed is key parameters

relative increment no longer there is significant change with the increase of iterations t.)

Once 3.4 iteration stop, each SU obtains final common recognition variate-value x ^*, carry out following local judgement,

Wherein λ is the decision threshold of frequency spectrum detection, and λ detects performance working point (P corresponding to one _fa, P _d), P _farefer to false alarm probability, i.e. the actual spectrum free time, court verdict is the probability that frequency spectrum takies; P _dbe detection probability, actual spectrum takies originally, the probability that court verdict also takies for frequency spectrum;

4. broadcast testing result, completes following work:

Each SUi ∈ S in SBS class _sBStesting result d is broadcast to neighbours SU outside class (being the user who belongs to non-SBS class in the neighbours of SUi), thereby makes the user of non-SBS class upgrade the cognition for frequency spectrum free time/seizure condition, realize the whole network SU sensing results is reached common understanding.

Embodiment a: specific embodiment of the present invention is described as follows, system emulation adopts Matlab software, and setting parameter does not affect generality.Following embodiment is whether detect a certain channel of VHF/UHF frequency range idle is basic references object, mainly determines on this channel, whether there is PU signal by the mode of energy measuring.It is worth emphasizing that, this invention is suggested plans and is also suitable for the detection of signal in other frequency ranges.

N in the present embodiment _fbe taken as 100.Perceived bandwidth W is taken as 10MHz, and detecting period is 100 μ s.Noise power spectral density is N ₀=-174dBm, receiver noise figure is 11dB.The transmitting power of PU is made as 100mW.The path loss factor is 4, and shadow fading standard deviation is 5.5dB, and the average of multipath fading is 1.

As shown in Fig. 3 (a), in this embodiment, we consider the square area of a 10km × 10km, and 1 PU (representing with triangle in figure) is positioned at center, and its coordinate is (5000,5000).36 SUs (representing by empty circles in figure) are evenly distributed in square area, and its coordinate is respectively:

Table 1: all SU coordinates

SU numbering	Abscissa (m)	Ordinate (m)	SU numbering	Abscissa (m)	Ordinate (m)
						1	2924.1	1009.503	19	2730.718	6109.955
2	4426.882	750.4901	20	4187.477	5916.674
						3	5883.987	776.6619	21	6232.208	5484.891
4	7575.978	1152.288	22	7744.184	5874.95
						5	9108.442	461.2307	23	9248.488	5779.396
6	1146.945	2742.655	24	995.6626	7222.06
						7	2241.05	2232.496	25	2604.708	7501.983
8	3833.959	2498.827	26	3906.556	7210.746
						9	5830.178	2213.956	27	5750.613	7742.598
10	7576.491	2847.238	28	7933.851	7650.508
						11	9284.362	2893.584	29	9491.021	7472.757
12	618.506	3950.469	30	1149.068	9110.859
						13	2569.737	4421.017	31	2223.543	9420.612
14	4216.669	4323.365	32	4060.52	9430.696
						15	6224.181	3982.686	33	5827.415	9308.894
16	7364.333	3776.017	34	7217.564	9498.024
						17	9151.755	4694.266	35	8961.066	9191.028
18	1051.947	5897.68	36	724.6895	1252.52

According to classical wireless channel model (list of references: A.Goldsmith, Wireless Communications, Cambridge University Press, 2005.), consider path loss, shadow fading and multipath fading parameter, can obtain following average signal-to-noise ratio:

Table 2: all SU receive average signal-to-noise ratio

SU numbering	Average signal-to-noise ratio (dB)	SU numbering	Average signal-to-noise ratio (dB)	SU numbering	Average signal-to-noise ratio (dB)
						1	0.321903	13	22.15143	25	5.084683
2	6.54046	14	4.167672	26	19.49637
						3	3.433937	15	9.957042	27	-2.02053
4	0.043135	16	9.332119	28	-2.23853
						5	-9.59268	17	4.209668	29	-26.3339
6	-1.50782	18	4.860982	30	-10.4904

7	7.729074	19	19.50539	31	-3.7888
						8	4.584223	20	21.19825	32	1.16455
9	3.691429	21	17.19486	33	4.973656
						10	8.024405	22	18.884	34	-16.0538
11	-0.60406	23	-2.49563	35	2.711777
						12	14.32138	24	11.93699	36	-10.8365

Shown in (1), by M=100 historical energy measuring information { E of accumulation _i(m) | m=1 ..., M}, each SUi in network, i ∈ 1 ..., N} obtains classification observed quantity O _ibe respectively:

Table 3: the classification observed quantity of all SU

SU numbering	Classification observed quantity (dB)	SU numbering	Classification observed quantity (dB)	SU numbering	Classification observed quantity (dB)
						1	-89.9874	13	-89.6800	25	-89.9840
2	-89.9834	14	-89.9849	26	-89.8152
						3	-89.9855	15	-89.9682	27	-89.990
4	-89.9867	16	-89.9713	28	-89.9919
						5	-89.991	17	-89.9863	29	-89.9915
6	-89.9878	18	-89.9856	30	-89.9891
						7	-89.9386	19	-89.816	31	-89.9878
8	-89.9831	20	-89.7393	32	-89.9884
						9	-89.9842	21	-89.8895	33	-89.9828
10	-89.9769	22	-89.8411	34	-89.9883
						11	-89.9889	23	-89.9877	35	-89.986
12	-89.9184	24	-89.9298	36	-89.9905

Using table 3 data as input, Fig. 3 (b) has provided the result of the Distributed Cluster scheme of the present invention's put forward based on common recognition, in figure, 7 SUs (representing by solid circles in figure) self-organizing ground forms SBS class, and its coordinate, classification observed quantity and average signal-to-noise ratio are as shown in table 4.By contrast table 2, table 3 and table 4, we see: the classification observed quantity of the SUs in SBS class is greater than the SUs outside class, and corresponding average signal-to-noise ratio also has identical rule.Therefore, the classification observed quantity of SUs has reflected its average signal-to-noise ratio level well.

Table 4: the SU coordinate and the average signal-to-noise ratio that obtain after cluster that the present invention suggests plans

SU numbering	Abscissa (m)	Ordinate (m)	Classification observed quantity (dB)	Average signal-to-noise ratio (dB)
					12	618.5059621	3950.469241	-89.9184	14.32138339
13	2569.736697	4421.016728	-89.6800	22.15143138
					19	2730.718266	6109.954564	-89.816	19.50539124
20	4187.476877	5916.673989	-89.7393	21.19825018
					21	6232.208407	5484.890822	-89.8895	17.19485813
22	7744.184084	5874.949831	-89.8411	18.8839977
					26	3906.555554	7210.745656	-89.8152	19.49636712

As a comparison, the Distributed Cluster scheme that we have provided based on distance in Fig. 3 (c) (is that the nearest SU of distance P U is spontaneously brought together, notice that in this scheme, each SU need to possess stationkeeping ability) result, wherein 7 nearest SUs of distance P U form SBS class, and its coordinate and average signal-to-noise ratio are as follows:.

Table 5: SU coordinate and average signal-to-noise ratio based on apart from obtaining after cluster

SU numbering	Abscissa (m)	Ordinate (m)	Average signal-to-noise ratio (dB)
				13	2569.736697	4421.016728	22.15143138
14	4216.668528	4323.365253	4.167672237
				15	6224.180544	3982.685919	9.957042449
19	2730.718266	6109.954564	19.50539124
				20	4187.476877	5916.673989	21.19825018
21	6232.208407	5484.890822	17.19485813
				26	3906.555554	7210.745656	19.49636712

The difference of Fig. 3 (b) and Fig. 3 (c) comes from: the cluster scheme based on distance only considers that large scale path loss or distance are on detecting the impact of performance, and institute suggests plans and considered the impact that path loss, shadow fading and multipath Rayleigh decline.Meanwhile, contrast table 3 and table 4 can find out, except public SUs (13,19,20,21,26), the average signal-to-noise ratio of the SU that the cluster of suggesting plans obtains is higher than the average signal-to-noise ratio obtaining based on distance.

In Fig. 4, compare the detection performance of different schemes.Wherein, transverse axis represents false alarm probability (probability that mistaken verdict is " frequency spectrum takies " " frequency spectrum free time " in the situation that), and the longitudinal axis represents detection probability (probability that correct judgement is " frequency spectrum takies " " frequency spectrum takies " in the situation that).In figure, we can see, the in the situation that of given false alarm probability, the detection performance of suggesting plans is obviously better than traditional equal gain combining (Equal Gain Combination, EGC) scheme (list of references: S.P. Herath and N.Raj atheva, " Analysis of equal gain combining in energy detection for cognitive radio over Nakagami channels, " in Proc.IEEE GLOBECOM, Nov.2008.) the cluster scheme (list of references: Amy C.Malady and Claudio R.C.M.da Silva with based on distance, " Clustering methods for distributed spectrum sensing in cognitive radio systems, " in Proc.IEEE GLOBECOM, Nov.2008.), suggest plans and obtained and the soft merging of optimum linearity (Optimal Soft Combination simultaneously, OSC) scheme (list of references: J.Ma, G. Zhao, and G. Li, " Soft combination and detection for cooperative spectrum sensing in cognitive radio networks, " IEEE Transactions on Wireless Communications, vol.7, no.11, pp.4502-4507, Nov.2008.) close performance.Notice, than OSC scheme, the advantage of suggesting plans is: in network, each SU does not need to carry out the estimation of himself instantaneous received signal to noise ratio, and institute suggests plans simultaneously only needs part SU to participate in cooperation, and information interaction amount reduces greatly.

The part that the present invention does not relate to all prior art that maybe can adopt same as the prior art is realized.

Claims (4)

1. the distributed cooperation frequency spectrum sensing method based on without supervision clustering in a cognitive self-organizing network, it is characterized in that: without control centre in the situation that, first determine the cognitive user S set BS with optimal perceived performance by unsupervised clustering, then detect based on the theoretical cooperation frequency spectrum of realizing between the multiple cognitive user SUs in SBS of common recognition, obtain the probability that corresponding frequency spectrum takies, finally utilize broadcast mechanism that testing result is informed to the multiple cognitive user SUs outside SBS; It comprises the following steps:

Step 1. parameter initialization:

First each SU obtains self observed quantity to environment, then local class barycenter and the local Lagrange multiplier of random initializtion self to SBS class and non-SBS class, on this basis, the local class ownership of each SU initialization coefficient;

Described " class barycenter " refers to the weighted average that the environment of all cognitive user in class is measured;

Described " Lagrange multiplier " is the middle transition variable of clustering algorithm, there is no concrete physical significance;

Described " class ownership coefficient " is to characterize the possibility that it belongs to SBS class and non-SBS class;

Step 2. is determined SBS based on unsupervised clustering:

Each SU first with the mutual local class barycenter of a hop neighbor SUs, based on this interactive information, each SU upgrades local class barycenter, local Lagrange multiplier and local class ownership coefficient successively, this process iteration is carried out, until stopping criterion for iteration meets;

After algorithm iteration stops, in network, the local class barycenter of all SUs will be tending towards identical, reach " all SU class barycenter common recognitions ", but the local class ownership coefficient that each SU obtains is different, based on this, each SU carries out the judgement of class ownership according to the local class ownership coefficient of self, realizes without supervision clustering;

Step 3. realizes SBS distribution within class formula cooperation frequency spectrum based on common recognition theory and detects:

First each SU in SBS class carries out local energy perception, and mutual local energy measured value then and between neighbours SUs, carries out data fusion based on common recognition agreement, and through iteration repeatedly, all SUs in final SBS class reach the common recognition to spectrum energy measured value;

Based on the Perspective of Energy measured value of common recognition, each SU carries out this locality judgement, and to obtain frequency spectrum state-detection result be the frequency spectrum free time or take;

Step 4. is broadcasted testing result:

Testing result is broadcast to the neighbours SU outside class by each SU of SBS class, realizes the whole network SU sensing results is reached common understanding.

2. the distributed cooperation frequency spectrum sensing method based on without supervision clustering in cognitive self-organizing network according to claim 1, is characterized in that described unsupervised clustering comprises the following steps:

Step 1. parameter initialization:

1.1 based on historical perception information { E _i(m) | m=1 ..., M}, each SUi ∈ in network 1 ..., first N} obtains self observed quantity to spectrum environment:

$O_i=\frac{1}{M}\underset{m=1}{\overset{M}{\mathrm{\Σ}}}{E}_i\left(m\right)$

Wherein E _i(m) energy value detecting while being the m time perception, M is cumulative number of times, N represents the quantity of cognitive user in network;

Each SUi ∈ in 1.2 networks 1 ..., N} random initializtion

with

wherein k ∈ 1,2} be respectively SUi ∈ 1 ..., the SBS class of N} this locality and the initial barycenter of non-SBS class, described barycenter refers to the weighted average of the observed quantity of all cognitive user in class,

k ∈ 1,2} be respectively SUi ∈ 1 ..., the SBS class of N} this locality and the initial Lagrange multiplier of non-SBS class;

Each cognitive user SUi ∈ in 1.3 networks 1 ..., the local class ownership of N} initialization coefficient

$a_i^0\left(k\right)=\frac{\left|\right|O_i-c_i^0\left(k\right)|{|}^{-\frac{2}{p-1}}}{{\mathrm{\Σ}}_{k=1}^K\left|\right|O_i-c_i^0\left(k\right)|{|}^{-\frac{2}{p-1}}},\∀i\∈\{1,...,N\},\∀k\∈\left\{\mathrm{1,2}\right\}$

Wherein, p>1, p is the exponential factor calculating in local class ownership coefficient, here

Step 2. is based on determining SBS without supervision clustering:

Each SU first with the mutual local class barycenter of a hop neighbor SUs, based on this interactive information, each SU upgrades local class barycenter, local Lagrange multiplier and local class ownership coefficient successively, this process iteration is carried out, until stopping criterion for iteration meets; After iteration stops, each SU obtains local class ownership coefficient and local class barycenter; On this basis, each SU carries out the judgement of class ownership according to the local class ownership coefficient obtaining, and realizes without supervision clustering;

Concrete by carrying out following distributed iterative method realization: based on t=0,1,2 ... the local class barycenter that inferior iteration obtains

each SUi, i ∈ 1 ..., N} carries out iteration the t+1 time:

2.1 each SUi ∈ 1 ..., N} is by local class barycenter

be broadcast to a hop neighbor user

here S _irefer to the set of a hop neighbor of SUi, d _ijrepresent the distance between cognitive user SUi and one hop neighbor user SUj, d _comrepresent between two SU can proper communication ultimate range;

2.2 each SUi, i ∈ 1 ..., N} receives after all hop neighbor users' local class barycenter,, upgrade local class barycenter, obtain:

$c_i^{t+1}\left(k\right)={(a_i^{t+1}\left(k\right)+2\mathrm{\η}|S_i\left|\right)}^{-1}\{a_i^{t+1}\left(k\right)O_i-2{\mathrm{\λ}}_i^t\left(k\right)+\mathrm{\η}\underset{j\∈S_i}{\mathrm{\Σ}}[c_i^t\left(k\right)+c_J^t\left(k\right)\left]\right\},\∀i\∈\{1,...,N\},k\∈\left\{\mathrm{1,2}\right\}$

Wherein, η >0; | S _i| the number of element in a hop neighbor S set i of expression SUi;

be a local Lagrange multiplier dynamically updating, its update rule is according to following step 2.4;

2.3 each SUi ∈ 1 ..., N} upgrades local class ownership coefficient, obtains:

$a_i^{t+1}\left(k\right)=\frac{\left|\right|O_i-c_i^{t+1}\left(k\right)|{|}^{-\frac{2}{p-1}}}{{\mathrm{\Σ}}_{k=1}^K\left|\right|O_i-c_i^{t+1}\left(k\right)|{|}^{-\frac{2}{p-1}}},\∀i\∈\{1,...,N\},\∀k\∈\left\{\mathrm{1,2}\right\}$

Wherein, p>1;

2.4 each SUi ∈ 1 ..., N} upgrades local Lagrange multiplier, obtains:

${\mathrm{\λ}}_i^{t+1}\left(k\right)={\mathrm{\λ}}_i^t\left(k\right)+\frac{\mathrm{\η}}{2}\underset{j\∈S_i}{\mathrm{\Σ}}[c_j^{t+1}\left(k\right)-c_i^{t+1}\left(k\right)],\∀i\∈\{1,...,N\},k\∈\left\{\mathrm{1,2}\right\}$

Step 2.1-2.4 iteration is carried out, through iteration repeatedly, if condition with

meet, iteration stops simultaneously; Wherein, ε _a, ε _cand ε _λbe the positive number close to 0, in reality, often get ε _a=ε _c=ε _λ∈ [10 ^-6, 10 ^-3], if end condition does not meet, rebound step 2.1, if condition meets, iteration stops;

After 2.5 iteration stop, each SUi, i ∈ 1 ..., N} carries out the judgement of local class ownership according to following rule:

Each SUi, i ∈ 1 ..., definite SBS class or the non-SBS class of entering of N};

When

sUi enters the class of k=1, and will

be set to 1,

be set to 0;

Otherwise work as

sUi enters the class of k=2, and will

be set to 0,

be set to 1;

Through the class ownership judgement of step 2.5, all SU that belong to SBS class spontaneously flock together, and form SBS user's collection:

Wherein,

represent barycenter larger class, i.e. SBS class,

represent that SUi belongs to SBS class

s _sBSrepresent all SBS classes that belong to the set of SU;

Step 3. realizes SBS distribution within class formula cooperation frequency spectrum based on common recognition theory and detects:

In step 2, form on the basis of SBS class to the SUs self-organizing of optimal perceived performance, first each SU in SBS class carries out local energy perception, then mutual local energy measured value and between neighbours SUs, carry out data fusion based on common recognition agreement, through iteration repeatedly, all SUs in final SBS class reach the common recognition to spectrum energy measured value, based on this common recognition, each SU carries out this locality judgement, obtains frequency spectrum state-detection result and be frequency spectrum idle or take:

Concrete by carrying out following distributed iterative method realization: based on t=0,1,2 ... this locality common recognition variable that inferior iteration obtains

each SUi, i ∈ 1 ..., N} carries out iteration the t+1 time:

Initialization: SBS class is optimal perceived performance cognitive user S set _sBSinterior each SUi ∈ S _sBScarry out local energy detection, obtain Perspective of Energy measured value E _iand its initial local common recognition variable is made as

3.2 each SUi ∈ S _sBSwith the one hop neighbor variable of knowing together alternately

be each SUi ∈ S _sBSby its common recognition variate-value

be broadcast to a hop neighbor user $\mathrm{SUj}\∈S_i\stackrel{\mathrm{\Δ}}{=}\left\{j\right|d_{\mathrm{ij}}\≤d_{\mathrm{com}},\∀j\∈\{1,...,N\}\},$ Receive the common recognition variable from a hop neighbor cognitive user SU simultaneously

3.3 each SUi, i ∈ 1 ..., and N} receives after all hop neighbor users' common recognition variable,, each SUi ∈ S _sBScarry out information fusion according to following common recognition agreement:

$x_i^{t+1}=x_i^t+s_x\underset{j\∈S_i}{\mathrm{\Σ}}(x_j^t-x_i^t)$

Wherein s _x>0 is iteration step length, conventionally gets

If condition

meet, iteration stops, the whole network asymptotic reaching of on average knowing together, and consensus value is asymptotic is

wherein ε _xbe the positive number close to 0, in reality, often get ε _a=ε _c=ε _λ∈ [10 ^-6, 10 ^-3], if condition does not meet, rebound step 3.2, if condition meets, iteration stops;

Once 3.4 iteration stop, each SU obtains final common recognition variate-value x ^*, carry out following local judgement,

Wherein λ is the decision threshold of frequency spectrum detection, and λ detects performance working point (P corresponding to one _fa, P _d), P _farefer to false alarm probability, i.e. the actual spectrum free time, court verdict is the probability that frequency spectrum takies; P _dbe detection probability, actual spectrum takies originally, the probability that court verdict also takies for frequency spectrum;

Step 4. is broadcasted testing result, completes following work:

Each SUi ∈ S in SBS class _sBStesting result d is broadcast in the neighbours that neighbours SU outside class is SUi and belongs to the user of non-SBS class, thereby make the user of non-SBS class upgrade the cognition for frequency spectrum free time/seizure condition, realize the whole network SU sensing results is reached common understanding.

3. the distributed cooperation frequency spectrum sensing method based on without supervision clustering in cognitive self-organizing network according to claim 2, is characterized in that in described step 1, in the time that frequency spectrum is idle, and E _i(m) only comprise noise energy; In the time that frequency spectrum takies, E _i(m) be the energy that not only comprises noise, also comprise the energy of PU signal.

4. the distributed cooperation frequency spectrum sensing method based on without supervision clustering in cognitive self-organizing network according to claim 2, is characterized in that described p=2.

Images