CN106130661A - Broadband wireless transmitter recognition methods based on Hammerstein Wiener model - Google Patents

Abstract

The invention discloses a kind of broadband wireless transmitter recognition methods based on Hammerstein Wiener model: the first step: broadband wireless transmitter models；Second step: use improvement adaptive niche technology genetic algorithm (or mesh adaption directly searches the Model Distinguish algorithms such as (mads) algorithm, adaptive niche technology genetic algorithm) to carry out Hammerstein Wiener identification of Model Parameters, obtain the model parameter vector estimated；3rd step: broadband wireless transmitter identification.The present invention has stronger global optimizing and local optimal searching ability, has the advantage that robustness is good and identification precision is high simultaneously, and its effectiveness is verified in being applied to broadband wireless transmitter identification.

Description

Broadband wireless transmitter recognition methods based on Hammerstein-Wiener model

Technical field

The invention belongs to communication technical field, be specifically related to a kind of broadband based on Hammerstein-Wiener model without Line transmitter recognition methods.

Background technology

The fast development of radio communication service brings new challenge, disabled user to management and the detection of radio-frequency spectrum Intercept and disturb, the illegal invasion etc. of counterfeit user is required for carrying out wireless device the identification certification of individual identity.Due to The problem of manufacturing process, the transmitter even from same production line, same model also can also exist subtle difference.Cause This, use new mode to be identified the most gradually highlighting its importance to the individual fine feature difference of transmitting set hardware. Existing research is concentrated mainly on the identification technology to transmitting set, and the research to broadband wireless transmitter is considerably less.

At present, transmitting set is identified main employing method based on radio-frequency fingerprint feature, the fingerprint characteristic of extraction There are wavelet coefficient, components and parts nonlinear model shape parameter, instantaneous envelope feature etc., wherein based on components and parts nonlinear model shape parameter Recognition methods is based primarily upon Hammerstein model and low order volterra series model.But, above-mentioned model is applied to width Poor performance during Nonlinear Modeling with transmitting set, Taringou F. and Hammi O. points out, Hammerstein- Wiener model has more excellent performance of modeling to broadband wireless transmitter.

In terms of system identification, existing Hammerstein-Wiener identification Method mainly has mesh adaption straight Connecing searching algorithm (mads algorithm) or some traditional algorithms, such as method of least square and iterative method etc., these methods also exist identification Precision is low, and poor robustness or convergence can not be by weak points such as Strict Proofs.Not enough for these, some intelligent algorithms, as Adaptive weighted particle swarm optimization algorithm (AWPSO), neutral net scheduling algorithm are applied to Hammerstein-Wiener model Identification, achieve preferable effect.But these intelligent algorithms, the problem that there is also local optimal searching scarce capacity.

Summary of the invention

The problems referred to above existed for prior art, the present invention proposes a kind of based on Hammerstein-Wiener model Broadband wireless transmitter recognition methods.On the discrimination method of Hammerstein-Wiener model, propose in Anguillar japonica algorithm Mark degree and the thought of neighbouring study be incorporated into adaptive niche technology genetic algorithm, construct the self adaptation your pupil of a kind of improvement Border genetic algorithm, can overcome the shortcoming that local optimal searching ability is weak.It is then based on this Revised genetic algorithum and carrys out identification Hammerstein-wiener model, obtains the estimation of model parameter.Again with estimate the model parameter vector that obtains be characterized to Amount, applies direct Euclidean distance relative method to realize broadband wireless transmitter identification.

The present invention adopts the following technical scheme that:

Broadband wireless transmitter recognition methods based on Hammerstein-Wiener model, specifically comprises the following steps that

The first step: broadband wireless transmitter models:

Hammerstein-Wiener model structure (as shown in Figure 1).The input of Hammerstein-Wiener model with The relation of output is

$\begin{array}{c}y\left(n\right)=\underset{j=1}{\overset{M^{\′}}{\mathrm{\Σ}}}{B}_{2j-1}\left|\underset{k=1}{\overset{N}{\mathrm{\Σ}}}{h}_k\underset{i=1}{\overset{M}{\mathrm{\Σ}}}{b}_{2i-1}\right|d\left(n-k\right){|}^{2i-2}d\left(n-k\right){|}^{2j-2}\\ .\underset{k=1}{\overset{N}{\mathrm{\Σ}}}{h}_k\underset{i=1}{\overset{M}{\mathrm{\Σ}}}{b}_{2i-1}|d\left(n-k\right){|}^{2i-2}d\left(n-k\right)+w\left(n\right)\end{array}---\left(1\right)$

Wherein, w (n)～N (0, σ²) it is additive white Gaussian noise.M, N and M' are the model orders of each submodule；B_kAnd b_k It is the coefficient of non-linear submodule, h_kIt it is the coefficient of linear submodule；Y (n) and d (n) is respectively output and the input of model.False If not considering the impact of wireless channel, in broadband wireless receiver, receiving signal and being equivalent to output signal y in formula (1) (n), input d (n) of model can be obtained by demodulation, input in model and output data for receiver all it is known that Such that it is able to model is carried out identification according to these data.

Without loss of generality, it can be assumed that model order M, N and M' are respectively 7,3 and 5, then can be by all ginsengs in model Number is configured to parameter vector, is written as θ=[b₁,b₃,b₅,b₇,h₀,h₁,h₂,B₁,B₃,B₅].。

Second step: use improve adaptive niche technology genetic algorithm or mesh adaption direct search (mads) algorithm or from Adapt to the Model Distinguish algorithms such as niche genetic algorithm and carry out Hammerstein-Wiener identification of Model Parameters, obtain estimation Model parameter vector.The concrete steps of the adaptive niche technology genetic algorithm improved see below, and this algorithm has the stronger overall situation and seeks Excellent and local optimal searching ability, has the advantage that robustness is good and identification precision is high simultaneously.

3rd step: broadband wireless transmitter identification:

It is characterized vector with model parameter, realizes broadband wireless transmitter identification with Euclidean distance method intuitively.Suppose there is 2 The broadband wireless transmitter of the same manufacturer production of platform is to be identified, is assumed to No. 1 transmitter and No. 2 transmitters, the field of identification respectively Scape figure is shown in Fig. 2, then decision rule can be written as:

$|\widehat{\mathrm{\&theta;}}-{\mathrm{\&theta;}}_1|\underset{H_1}{\overset{H_0}{\begin{array}{c}>\\ <\end{array}}}|\widehat{\mathrm{\&theta;}}-{\mathrm{\&theta;}}_2|---\left(2\right)$

Wherein, H₀What expression receiver received is No. 2 broadband wireless transmitter signals, H₁Represent what receiver received It it is No. 1 broadband wireless transmitter signal.Represent the parameter vector of current identification, θ₁It is No. 1 transmitter parameter vector, θ₂It it is No. 2 Transmitter parameter vector.

The Model Distinguish step of the adaptive niche technology genetic algorithm of improvement be given below:

The first step: coding: use real coding.I.e. participate in the individual each genic value of heredity with in a certain scope Real number represents.The gene of the optimal solution individuality finally meeting condition is exactly model parameter predictive value.

Second step: gradient optimizing: object function is as follows

$J=\frac{1}{K}\underset{k=1}{\overset{K}{\&Sigma;}}|y(n-k)-\hat{y}(n-k){|}^2---\left(3\right)$

Wherein, K is signal length, y (k) withIt is respectively true output and the estimation output of modular form (1).For gradient.Orderη is referred to as step-length, is a positive number.Gradient optimizing criterion is:

$\begin{array}{cc}if& J\left(X_1^t-\left|\mathrm{\&Delta;}X\right|\right)<J\left(X_1^t\right)then{X}_1^{t+1}=X_1^t-|\mathrm{\&eta;}\&dtri;J|\\ if& J\left(X_2^t+\left|\mathrm{\&Delta;}X\right|\right)<J\left(X_2^t\right)then{X}_2^{t+1}=X_2^{t+1}+|\mathrm{\&eta;}\&dtri;J|\end{array}---\left(4\right)$

Wherein,It is gene range limit,It it is gene range lower limit.Stop when object function is less than a certain thresholding Optimize.

Preserve optimum individual: calculate the fitness function value in initial population respectively, according to arranging from big to small, before memory N number of preservation.Wherein, F=1/J is fitness function.

3rd step: select: select M individuality from population with ratio selection algorithm according to fitness.

Self adaptation is intersected:WithIt is that the parent after selecting operation is individual respectively,WithIt is that self adaptation is intersected respectively After offspring individual.Self adaptation cross reference formula is as follows:

$\left\{\begin{array}{c}{x}_i^{t+1}=P_c{x}_i^t+(1-P_c)x_j^t\\ x_j^{t+1}=P_c{x}_j^t+(1-P_c)x_i^t\end{array}\right.---\left(5\right)$

Wherein, P_cRepresent crossover probability, meet:

$P_c=\left\{\begin{array}{cc}{P}_{c1}-\frac{\left(P_{c1}-P_{c2}\right)\left(f^{\&prime;}-f_{avg}\right)}{f_{\max }-f_{avg}}& f^{\&prime;}\&GreaterEqual;f_{av}\\ P_{c1}& f^{\&prime;}<f_{avg}\end{array}\right.---\left(6\right)$

Wherein, P_c1Take 0.8, P_c2Take 0.6, f_maxRepresent maximum adaptation degree in population, f_avgRepresent in population and averagely adapt to Degree, f' represents fitness bigger in two intersection individualities.

TSP question: the colony carrying out making a variation is carried out such as lower variation:

$P_m=\left\{\begin{array}{cc}{P}_{m1}-\frac{\left(P_{m1}-P_{m2}\right)\left(f_{\max }-f\right)}{f_{\max }-\stackrel{\&OverBar;}{f}}& f\&GreaterEqual;\stackrel{\&OverBar;}{f}\\ P_{m1}& f<\stackrel{\&OverBar;}{f}\end{array}\right.---\left(7\right)$

Wherein, P_m1And P_m2Represent minimum and maximum mutation probability, P respectively_m1Take 0.05, P_m2Take 0.001,Represent per generation Average fitness value after colony's intersection.

4th step: with degree of mark divided rank, compares the distance between two individual i and j: if d_ij＞ L, it is believed that not at one In microhabitat, carry out condition judgement.If d_ij＜ L, then carry out the 5th step.

5th step: judge the size of mark degree and the two body fitness again, if meet S ＜ F simultaneously, it is believed that be close Excellent individual, carry out neighbouring learning manipulation, otherwise carry out microhabitat and eliminate operation.

6th step: microhabitat is eliminated: the individual individuality remembered with second step of M the 3rd step obtained is incorporated in Together, one is obtained containing new colony individual for M+N；Individual to this M+N, obtain two individual x according to the following formula_iAnd x_jBetween sea Prescribed distance:

$\left|\right|x_i-x_j\left|\right|=\sqrt{\underset{k=1}{\overset{Chromlen}{\&Sigma;}}{(x_{ik}-x_{jk})}^2}---\left(8\right)$

Wherein, i=1,2..., M+N-1；J=i+1, i+2....M+N；Chromlen represents chromosome length.

When | | x_i-x_j| | during ＜ L, relatively individual x_iWith individual x_jFitness size, and to wherein fitness relatively low Body sentences penalty function: F_min(x_i,x_j)=Penalty, is allowed to ideal adaptation degree lower and be eliminated.According to this M+N individuality New fitness carries out descending to each individuality, and memory top n is individual.

Neighbouring study: if F is [X (b_i,h_j,B_j)] ＞ F [X (b_j,h_j,B_j)], exchange base because of section, F [X (b after exchange_j,h_j, B_j)]=F [X (b_i,h_j,B_j)].In like manner, commutative h_jAnd B_jGene section.

7th step: condition is adjudicated: judge whether meet required precision or reach maximum iteration time, as reached then program knot Bundle.Otherwise return to the 3rd step.

It is worthy of note, the knowledge based on Hammerstein-Wiener model broadband wireless transmitter that the present invention proposes Other method, wherein the identification algorithm to Hammerstein-Wiener model is not limited to aforesaid improvement adaptive niche technology something lost Propagation algorithm, other Hammerstein-Wiener Model Distinguish algorithm, as mesh adaption direct search (mads) algorithm or from Adapt to the Model Distinguish algorithms such as niche genetic algorithm and equally obtain the estimation of model parameter, simply identification performance exists Difference, thus on the recognition performance of broadband wireless transmitter, there is also corresponding difference.

Present invention broadband wireless based on Hammerstein-Wiener model transmitter recognition methods can extensively be applied In broadband wireless transmitter identification based on Hammerstein-Wiener model, overall step can be summarized as: the first step: by width Band transmitting set is modeled as Hammerstein-Wiener model, derives the Explicit functions of required problem.Second step: logical Cross the thought proposed the mark degree in Anguillar japonica algorithm and neighbouring study and be incorporated into adaptive niche technology genetic algorithm, construct one Plant the adaptive niche technology genetic algorithm improved.This algorithm is used to carry out Hammerstein-Wiener identification of Model Parameters, To the model parameter vector estimated.3rd step: be characterized vector with model parameter, with Euclidean distance method intuitively realize broadband without Line transmitter identification.The present invention has stronger global optimizing and local optimal searching ability, has that robustness is good and identification essence simultaneously Spending high advantage, its effectiveness is verified in being applied to broadband wireless transmitter identification.

Accompanying drawing explanation

Fig. 1 is Hammerstein-Wiener illustraton of model.

Fig. 2 is broadband wireless transmitter identification scene graph.

Fig. 3 is the adaptive niche technology genetic algorithmic steps figure improved.

Fig. 4 is to improve adaptive niche technology genetic algorithm, adaptive niche technology genetic algorithm and adaptive mesh search to calculate The relative error figure of method.

Fig. 5 is that discrimination is with signal to noise ratio change curve.

Detailed description of the invention

Below in conjunction with accompanying drawing, the invention will be further described.

The present invention is directed to broadband wireless transmitter identification problem conduct a research, first broadband wireless transmitter is modeled as Hammerstein-Wiener model, by introducing the mark degree in Anguillar japonica algorithm and the thought of neighbouring study, proposes one and changes The adaptive niche technology genetic algorithm entered, and this algorithm is applied to the identification of model parameter；Then it is characterized with model parameter Vector, realizes broadband wireless transmitter identification with Euclidean distance method intuitively.It it is once the detailed step of a kind of method for optimizing.

The present embodiment broadband wireless based on Hammerstein-Wiener model transmitter recognition methods, concrete steps are such as Under:

The first step: broadband wireless transmitter models:

Hammerstein-Wiener model structure (as shown in Figure 1).The input of Hammerstein-Wiener model with The relation of output is

$\begin{array}{c}y\left(n\right)=\underset{j=1}{\overset{M^{\&prime;}}{\mathrm{\&Sigma;}}}{B}_{2j-1}\left|\underset{k=1}{\overset{N}{\mathrm{\&Sigma;}}}{h}_k\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}{b}_{2i-1}\right|d\left(n-k\right){|}^{2i-2}d\left(n-k\right){|}^{2j-2}\\ .\underset{k=1}{\overset{N}{\mathrm{\&Sigma;}}}{h}_k\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}{b}_{2i-1}|d\left(n-k\right){|}^{2i-2}d\left(n-k\right)+w\left(n\right)\end{array}---\left(1\right)$

Wherein, w (n)～N (0, σ²) it is additive white Gaussian noise.M, N and M' are the model orders of each submodule；B_kAnd b_k It is the coefficient of non-linear submodule, h_kIt it is the coefficient of linear submodule；Y (n) and d (n) is respectively output and the input of model.False If not considering the impact of wireless channel, in broadband wireless receiver, receiving signal and being equivalent to output signal y in formula (1) (n), input d (n) of model can be obtained by demodulation, input in model and output data for receiver all it is known that Such that it is able to model is carried out identification according to these data.

Without loss of generality, it can be assumed that model order M, N and M' are respectively 7,3 and 5, then can be by all ginsengs in model Number is configured to parameter vector, is written as θ=[b₁,b₃,b₅,b₇,h₀,h₁,h₂,B₁,B₃,B₅].。

Second step: use the adaptive niche technology genetic algorithm improved to carry out Hammerstein-Wiener model parameter and distinguish Know, obtain the model parameter vector estimated.The concrete steps of the adaptive niche technology genetic algorithm improved see below, and this algorithm has Stronger global optimizing and local optimal searching ability, have the advantage that robustness is good and identification precision is high simultaneously.

3rd step: broadband wireless transmitter identification:

It is characterized vector with model parameter, realizes broadband wireless transmitter identification with Euclidean distance method intuitively.Suppose there is 2 The broadband wireless transmitter of the same manufacturer production of platform is to be identified, is assumed to No. 1 transmitter and No. 2 transmitters, the field of identification respectively Scape figure is shown in Fig. 2, then decision rule can be written as:

$|\widehat{\mathrm{\&theta;}}-{\mathrm{\&theta;}}_1|\underset{H_1}{\overset{H_0}{\begin{array}{c}>\\ <\end{array}}}|\widehat{\mathrm{\&theta;}}-{\mathrm{\&theta;}}_2|---\left(2\right)$

Wherein, H₀What expression receiver received is No. 2 broadband wireless transmitter signals, H₁Represent what receiver received It it is No. 1 broadband wireless transmitter signal.Represent the parameter vector of current identification, θ₁It is No. 1 transmitter parameter vector, θ₂It it is No. 2 Transmitter parameter vector.

The Model Distinguish step of the adaptive niche technology genetic algorithm of improvement be given below:

The first step: coding: use real coding.I.e. participate in the individual each genic value of heredity with in a certain scope Real number represents.The gene of the optimal solution individuality finally meeting condition is exactly model parameter predictive value.

Second step: gradient optimizing: object function is as follows

$J=\frac{1}{K}\underset{k=1}{\overset{K}{\&Sigma;}}|y(n-k)-\hat{y}(n-k){|}^2---\left(3\right)$

Wherein, K is signal length, y (k) withIt is respectively true output and the estimation output of modular form (1).For gradient.Orderη is referred to as step-length, is a positive number.Gradient optimizing criterion is:

$\begin{array}{cc}if& J\left(X_1^t-\left|\mathrm{\&Delta;}X\right|\right)<J\left(X_1^t\right)then{X}_1^{t+1}=X_1^t-|\mathrm{\&eta;}\&dtri;J|\\ if& J\left(X_2^t+\left|\mathrm{\&Delta;}X\right|\right)<J\left(X_2^t\right)then{X}_2^{t+1}=X_2^{t+1}+|\mathrm{\&eta;}\&dtri;J|\end{array}---\left(4\right)$

Wherein,It is gene range limit,It it is gene range lower limit.Stop when object function is less than a certain thresholding Optimize.

Preserve optimum individual: calculate the fitness function value in initial population respectively, according to arranging from big to small, before memory N number of preservation.Wherein, F=1/J is fitness function.

3rd step: select: select M individuality from population with ratio selection algorithm according to fitness.

Self adaptation is intersected:WithIt is that the parent after selecting operation is individual respectively,WithIt is that self adaptation is intersected respectively After offspring individual.Self adaptation cross reference formula is as follows:

$\left\{\begin{array}{c}{x}_i^{t+1}=P_c{x}_i^t+(1-P_c)x_j^t\\ x_j^{t+1}=P_c{x}_j^t+(1-P_c)x_i^t\end{array}\right.---\left(5\right)$

Wherein, P_cRepresent crossover probability, meet:

$P_c=\left\{\begin{array}{cc}{P}_{c1}-\frac{\left(P_{c1}-P_{c2}\right)\left(f^{\&prime;}-f_{avg}\right)}{f_{\max }-f_{avg}}& f^{\&prime;}\&GreaterEqual;f_{av}\\ P_{c1}& f^{\&prime;}<f_{avg}\end{array}\right.---\left(6\right)$

Wherein, P_c1Take 0.8, P_c2Take 0.6, f_maxRepresent maximum adaptation degree in population, f_avgRepresent in population and averagely adapt to Degree, f' represents fitness bigger in two intersection individualities.

TSP question: the colony carrying out making a variation is carried out such as lower variation:

$P_m=\left\{\begin{array}{cc}{P}_{m1}-\frac{\left(P_{m1}-P_{m2}\right)\left(f_{\max }-f\right)}{f_{\max }-\stackrel{\&OverBar;}{f}}& f\&GreaterEqual;\stackrel{\&OverBar;}{f}\\ P_{m1}& f<\stackrel{\&OverBar;}{f}\end{array}\right.---\left(7\right)$

Wherein, P_m1And P_m2Represent minimum and maximum mutation probability, P respectively_m1Take 0.05, P_m2Take 0.001,Represent per generation Average fitness value after colony's intersection.

4th step: with degree of mark divided rank: compare the distance between two individual i and j: if d_ij＞ L, it is believed that not at one In microhabitat, carry out condition judgement.If d_ij＜ L, carries out the 5th step.

5th step: judge the size of mark degree and the two body fitness again, if meet S ＜ F simultaneously, it is believed that be close Excellent individual, carry out neighbouring learning manipulation, otherwise carry out microhabitat and eliminate operation.

6th step: microhabitat is eliminated: the individual individuality remembered with second step of M the 3rd step obtained is incorporated in Together, one is obtained containing new colony individual for M+N；Individual to this M+N, obtain two individual x according to the following formula_iAnd x_jBetween sea Prescribed distance:

$\left|\right|x_i-x_j\left|\right|=\sqrt{\underset{k=1}{\overset{Chromlen}{\&Sigma;}}{(x_{ik}-x_{jk})}^2}---\left(8\right)$

Wherein, i=1,2..., M+N-1；J=i+1, i+2....M+N；Chromlen represents chromosome length.

When | | x_i-x_j| | during ＜ L, relatively individual x_iWith individual x_jFitness size, and to wherein fitness relatively low Body sentences penalty function: F_min(x_i,x_j)=Penalty, is allowed to ideal adaptation degree lower and be eliminated.According to this M+N individuality New fitness carries out descending to each individuality, and memory top n is individual.

Neighbouring study: if F is [X (b_i,h_j,B_j)] ＞ F [X (b_j,h_j,B_j)], exchange base because of section, F [X (b after exchange_j,h_j, B_j)]=F [X (b_i,h_j,B_j)].In like manner, commutative h_jAnd B_jGene section.

7th step: condition is adjudicated: judge whether meet required precision or reach maximum iteration time, as reached then program knot Bundle.Otherwise return to the 3rd step.

Effectiveness and the superiority of verification method is come below by way of emulation experiment.Adaptive niche technology genetic algorithm will be improved Carry out recognition performance with mads algorithm, adaptive niche technology genetic algorithm to compare.2 broadband wireless transmitters in Fig. 2 Model parameter is arranged as shown in table 1.

Table 1 (broadband wireless transmitter parameter arranges table)

On time complexity, table 2 shows that mads algorithm is minimum, improves adaptive niche technology genetic algorithm secondly, adaptive Answer niche genetic algorithm maximum.

Table 2 (Algorithms T-cbmplexity table)

Algorithm	Mads algorithm	Adaptive niche technology genetic algorithm	Revised genetic algorithum
				Average operating time (s)	152.14	483.25	321.36

On identification precision, Fig. 4 shows to improve adaptive niche technology genetic algorithm and adaptive niche technology genetic algorithm Identification effect is substantially good than mads algorithm, and the relative error of mads algorithm is maximum, and up to 1.1, average relative error is 0.2, from The relative error adapting to niche genetic algorithm is 0.25 to the maximum, and average relative error is 0.05.And improve adaptive niche technology Genetic algorithm relative error is 0.09 to the maximum, and average relative error is 0.031.

On broadband wireless transmitter recognition performance, Fig. 5 shows that SNR, when scope [2dB, 30dB] is interior, improves self adaptation Niche genetic algorithm in three kinds of algorithms preferably, high by 2.85% than adaptive niche technology genetic algorithm, and mads algorithm Identification effect is worst.

Claims (4)

1. a broadband wireless transmitter recognition methods based on Hammerstein-Wiener model, is characterized in that by following step Rapid:

The first step: broadband wireless transmitter models；

Second step: use and improve adaptive niche technology genetic algorithm, mesh adaption Direct search algorithm or adaptive niche technology Genetic algorithm carries out Hammerstein-Wiener identification of Model Parameters, obtains the model parameter vector estimated；

3rd step: broadband wireless transmitter identification.

2. broadband wireless transmitter recognition methods based on Hammerstein-Wiener model as claimed in claim 1, it is special Levy and be: the first step: the input of Hammerstein-Wiener model with the relation of output is

$\begin{array}{c}y\left(n\right)=\underset{j=1}{\overset{M^{\&prime;}}{\&Sigma;}}{B}_{2j-1}{\left|\underset{k=1}{\overset{N}{\&Sigma;}}{h}_k\underset{i=1}{\overset{M}{\&Sigma;}}{b}_{2i-1}{\left|d(n-k)\right|}^{2i-2}d(n-k)\right|}^{2j-2}\\ .\underset{k=1}{\overset{N}{\&Sigma;}}{h}_k\underset{i=1}{\overset{M}{\&Sigma;}}{b}_{2i-1}{\left|d(n-k)\right|}^{2i-2}d(n-k)+w\left(n\right)\end{array}---\left(1\right)$

Wherein, w (n)～N (0, σ²) it is additive white Gaussian noise；M, N and M' are the model orders of each submodule；B_kAnd b_kRight and wrong The coefficient of linear submodule, h_kIt it is the coefficient of linear submodule；Y (n) and d (n) is respectively output and the input of model.

3. broadband wireless transmitter recognition methods based on Hammerstein-Wiener model as claimed in claim 2, it is special Levy and be:

3rd step: be characterized vector with model parameter, realizes broadband wireless transmitter identification with Euclidean distance method intuitively；Assuming that 2 broadband wireless transmitters are to be identified, respectively No. 1 transmitter and No. 2 transmitters, then decision rule is written as:

$\left|\widehat{\mathrm{\&theta;}}-{\mathrm{\&theta;}}_1\right|\underset{H_1}{\overset{H_0}{\begin{array}{c}>\\ <\end{array}}}\left|\widehat{\mathrm{\&theta;}}-{\mathrm{\&theta;}}_2\right|---\left(2\right)$

Wherein, H₀What expression receiver received is No. 2 broadband wireless transmitter signals, H₁What expression receiver received is No. 1 Broadband wireless transmitter signal；Represent the parameter vector of current identification, θ₁It is No. 1 transmitter parameter vector, θ₂It is No. 2 transmittings Machine parameter vector.

4. broadband wireless transmitter recognition methods based on Hammerstein-Wiener model as claimed in claim 3, it is special Levy and be:

Second step: the described Model Distinguish step improving adaptive niche technology genetic algorithm is as follows:

One, coding: use real coding；

Two, gradient optimizing: object function is as follows

$J=\frac{1}{K}\underset{k=1}{\overset{K}{\&Sigma;}}{|y(n-k)-\hat{y}(n-k)|}^2---\left(3\right)$

Wherein, K is signal length, y (k) withIt is respectively true output and the estimation output of modular form (1)；For gradient；Making Δ X=η J, η is referred to as step-length, is a positive number；Gradient optimizing criterion is:

$\begin{array}{cc}if& J(X_1^t-|\mathrm{\&Delta;}X\left|\right)<J\left(X_1^t\right)then{X}_1^{t+1}=X_1^t-\left|\mathrm{\&eta;}\&dtri;J\right|\\ if& J(X_2^t+|\mathrm{\&Delta;}X\left|\right)<J\left(X_2^t\right)then{X}_2^{t+1}=X_2^{t+1}+\left|\mathrm{\&eta;}\&dtri;J\right|\end{array}---\left(4\right)$

Wherein,It is gene range limit,It it is gene range lower limit；Stop when object function is less than a certain thresholding optimizing；

Preserve optimum individual: calculate the fitness function value in initial population respectively, according to arranging from big to small, remember top n Preserve；Wherein, F=1/J is fitness function；

Three, select: from population, select M individuality according to fitness with ratio selection algorithm；

Self adaptation is intersected:WithIt is that the parent after selecting operation is individual respectively,WithIt is after self adaptation is intersected respectively Offspring individual；Self adaptation cross reference formula is as follows:

$\left\{\begin{array}{c}{x}_i^{t+1}=P_c{x}_i^t+(1-P_c)x_j^t\\ x_j^{t+1}=P_c{x}_j^t+(1-P_c)x_i^t\end{array}\right.---\left(5\right)$

Wherein, P_cRepresent crossover probability, meet:

Wherein, P_c1Take 0.8, P_c2Take 0.6, f_maxRepresent maximum adaptation degree in population, f_avgRepresent average fitness in population, f' generation Fitness bigger in two, table intersection individuality；

TSP question: the colony carrying out making a variation is carried out such as lower variation:

$P_m=\left\{\begin{array}{cc}{P}_{m1}-\frac{(P_{m1}-P_{m2})(f_{\max }-f)}{f_{\max }-\stackrel{\&OverBar;}{f}}& f\&GreaterEqual;\stackrel{\&OverBar;}{f}\\ P_{m1}& f<\stackrel{\&OverBar;}{f}\end{array}\right.---\left(7\right)$

Wherein, P_m1And P_m2Represent minimum and maximum mutation probability, P respectively_m1Take 0.05, P_m2Take 0.001,Represent often for colony Average fitness value after intersection.

Four, with degree of mark divided rank, the distance between two individual i and j is compared: if d_ij＞ L, it is believed that not a microhabitat In, carry out the 7th step condition judgement；If d_ij＜ L, carries out the 5th step；

Five, the size of mark degree and the two body fitness is judged, if meet S ＜ F simultaneously, it is believed that be close outstanding Body, carries out the 6th step adjacent to learning manipulation, otherwise carries out the 6th step microhabitat and eliminates operation；

Six, microhabitat is eliminated: the individual individuality remembered with second step of M the 3rd step obtained combines, and obtains One containing new colony individual for M+N；Individual to this M+N, obtain two individual x according to the following formula_iAnd x_jBetween Hamming distances:

$||x_i-x_j||=\sqrt{\underset{k=1}{\overset{Chromlen}{\&Sigma;}}{(x_{ik}-x_{jk})}^2}---\left(8\right)$

Wherein, i=1,2..., M+N-1；J=i+1, i+2....M+N；Chromlen represents chromosome length；

When | | x_i-x_j| | during ＜ L, relatively individual x_iWith individual x_jFitness size, and to relatively low individual of wherein fitness at With penalty function: F_min(x_i,x_j)=Penalty, is allowed to ideal adaptation degree lower and be eliminated；According to this M+N individual newly fitting Response carries out descending to each individuality, and memory top n is individual；

Neighbouring study: if F is [X (b_i,h_j,B_j)] ＞ F [X (b_j,h_j,B_j)], exchange base because of section, F [X (b after exchange_j,h_j,B_j)]=F [X(b_i,h_j,B_j)]；In like manner, commutative h_jAnd B_jGene section；

Seven, condition judgement: judge whether meet required precision or reach maximum iteration time, if reaching, EP (end of program)；Otherwise Return to three.