CN105281854A - Local maximum efficacy invariance test spectrum sensing method based on non-circular signal - Google Patents

Abstract

The invention discloses a local maximum efficacy invariance test spectrum sensing method based on a non-circular signal in a multi-antenna scenario. The method comprises the steps as follows: A, calculating a sampling covariance matrix according to a received signal of the receiving end of a cognitive user, and constructing a corresponding augmented sampling covariance matrix; B, calculating the F norm of the augmented sampling covariance matrix, and constructing the test statistical magnitude; C comparing the test statistical magnitude with a preset threshold, judging that the spectrum is occupied if the statistical magnitude is greater than the threshold, or judging that the spectrum is idle, thus completing detection. By means of the characteristics of the non-circular signal, the correlation structure of the signal is utilized fully. In a multi-antenna scenario, high detection accuracy can be guaranteed, the number of required sampling points is small, and the detection efficiency is high.

Description

A kind of local maximum effect invariant test frequency spectrum sensing method based on not rounded signal

Technical field

The present invention relates to communication technical field, specifically relate to a kind of local maximum effect invariant test frequency spectrum sensing method based on not rounded signal.

Background technology

As a link very important in cognitive radio networks, frequency spectrum perception technology has been subject to the extensive concern of Chinese scholars.By frequency spectrum perception technology, cognitive user under the prerequisite of not interfere with primary users, can find frequency spectrum cavity-pocket, thus utilizes unappropriated frequency spectrum resource, greatly improve the availability of frequency spectrum.But often there is false-alarm and undetected situation in perception, the reduction causing the availability of frequency spectrum of false-alarm, undetected generation then can bring interference to primary user.On the other hand, because the access situation of primary user changes at any time, so must ensure to complete perception within the time short as far as possible.Therefore study accuracy of detection and perception efficiency higher, and be easy to realize frequency spectrum perception algorithm have very important significance.

Traditional frequency spectrum perception algorithm comprises matched filtering detection, cyclostationary characteristic detection and energy measuring etc., thus matched filtering is subject to many limitations in actual applications owing to needing whole prior informations of known primary user, it is too high that cyclostationary characteristic detects then computation complexity, energy measuring is owing to being simple and easy to realize, be one the most conventional in practical application, but affect seriously by noise power.Multi-antenna technology, owing to having good anti-fading characteristic and diversity gain etc., receives the concern of Chinese scholars in recent years, has also been widely applied to frequency spectrum perception field.Its basic thought is, when primary user takies frequency range, the signal that each antenna of cognitive user receiving terminal receives derives from same cognitive user, therefore has very strong correlation each other; And when the frequency range free time, be noise due to what receive, so there is not correlation.Utilize this dependency structure, do not need known noise power just can design good detector.MME (Maximum-MinimumEigenvalue) detection method utilizes the eigenvalue of maximum of Received signal strength sample covariance matrix and minimal eigenvalue than constructing test statistics.Subsequently, under the framework of broad sense maximum likelihood ratio inspection (GLRT), there is AGM (ArithmetictoGeometricMean) algorithm.In practical application, due to reasons such as the non-demarcation of antenna, cause receiving terminal each antenna place noise power inconsistent (non-uniform noise), in such a scenario, the test statistics that GLRT detects can be realized by the Hadamard ratio of calculating sampling covariance matrix, is called Hadamard algorithm.Also for ease of the impact overcoming non-uniform noise, VD (Volume-basedDetection) algorithm is by the determinant structure test statistics of sample covariance matrix.Be not difficult to find, said method is all utilize the sample covariance matrix of signal to design corresponding detection algorithm to construct test statistics, but, for not rounded signal very common in communication system, complete second-order statistics, except covariance matrix, also comprises conjugation covariance matrix, therefore only utilizes the covariance matrix of signal, do not utilize the statistical property of signal completely, therefore detection perform is not high.Based on this, on the basis of traditional Hadamard algorithm, occurred the NC-HDM algorithm for not rounded signal, the method proposes equally on GLRT framework.But, need to utilize the maximal possibility estimation of unknown parameter to calculate likelihood ratio because GLRT detects, thus structure test statistics, therefore less at sampling number, when channel circumstance is poor, evaluated error can affect its detection perform greatly.

Summary of the invention

In order to address this problem, the present invention proposes a kind of NC-based on not rounded signal local maximum effect invariant test detection algorithm.Be specially a kind of local maximum effect invariant test frequency spectrum sensing method based on not rounded signal, comprise the following steps:

A, Received signal strength according to cognitive user receiving terminal, calculating sampling covariance matrix, and build corresponding augmentation sample covariance matrix;

B, for augmentation sample covariance matrix, calculate the F norm of its matrix, thus build test statistics T;

C, by making comparisons with the thresholding preset, if statistic is greater than thresholding, then think that frequency spectrum is occupied, otherwise, then think that frequency spectrum is idle, thus complete detection.

On the basis of technique scheme, in steps A, the Received signal strength of a kth sampling instant cognitive user receiving terminal has following form

$\left.\begin{array}{c}{H}_0:x\left(k\right)=w\left(k\right)\\ H_1:x\left(k\right)=Hs\left(k\right)+w\left(k\right)\end{array}\right\}$

Wherein, H ₀for the hypothesis of frequency spectrum free time, H ₁for the hypothesis that frequency spectrum is taken by primary user, x (k) is a kth signal vector that sampling instant cognitive user receiving terminal receives, the noise signal vector that w (k) is a kth sampling instant, the signal vector that s (k) launches for a kth sampling instant primary user transmitting terminal, k=0,1 ..., K-1, K are sampling number; Particularly,

x(k)＝[x ₁(k),...,x _n(k),...,x _N(k)] ^T

w(k)＝[w ₁(k),...,w _n(k),...,w _N(k)] ^T

s(k)＝[s ₁(k),...,s _m(k),...,s _M(k)] ^T

X _n(k) and w _nk () represents the signal that cognitive user n-th reception antenna receives and noise respectively; s _mk () represents the signal that m root antenna place launches; H is that N × M ties up fading channel matrix.Primary user's signal s _mk () is not rounded signal, noise is circle Gaussian noise, namely wherein be the noise power of n-th reception antenna place the unknown, in practice, due to reasons such as the non-demarcation of antenna, the noise power at different reception antenna place may be inconsistent.Meanwhile, statistical iteration each other between noise, and separate with primary user's signal.

On the basis of technique scheme, the sample covariance matrix in steps A has following form

$S=\frac{1}{K}{\mathrm{XX}}^H$

Wherein, S is sample covariance matrix, X=[x (0) ..., x (k) ..., x (K-1)] be sampled signal, subscript () ^hrepresent conjugate transpose.

On the basis of technique scheme, the augmentation sample covariance matrix in steps A has following form

$\underset{\‾}{S}=\frac{1}{K}{\underset{\‾}{XX}}^H=\left[\begin{array}{cc}S& \tilde{S}\\ {\tilde{S}}^{*}& S^{*}\end{array}\right]$

Wherein, x=[ x(0) ..., x(k) ..., x(K-1)], x(k)=[x (k) ^t, x (k) ^h] ^t, sfor augmentation sample covariance matrix, for complementary sample covariance matrix, subscript () ^*represent and get conjugation.

On the basis of technique scheme, the test statistics in step B has following form

$T=\left|\right|\hat{C}|{|}_F^2$

Wherein, and F=diag ( s) be the diagonal matrix be made up of the diagonal entry of augmentation sample covariance matrix, || || _fthe F norm of representing matrix.

On the basis of technique scheme, step C needs to be realized by following discrimination model

$T=\left|\right|\hat{C}|{{|}_F^2}_{H_0}^{H_1}\mathrm{\γ}$

Wherein, γ is decision threshold, even T > γ, then think and suppose H ₁set up, frequency spectrum is occupied, otherwise, then think and suppose H ₀set up, frequency spectrum is idle, thus completes detection.

Compared with prior art, advantage of the present invention is as follows:

The present invention, by the characteristic by not rounded signal, takes full advantage of the dependency structure of Received signal strength, thus can be very little at sampling number, under the scene that signal to noise ratio is very low, reaches the detection perform more excellent than GLRT.

Accompanying drawing explanation

Fig. 1 is the test statistics experience profiles versus figure of different cognitive method;

Fig. 2 is each cognitive method performance comparison curve chart under AWGN environment;

Fig. 3 is each detection method performance comparison curve chart under MIMO-Rayleigh environment;

Fig. 4 is the ROC curve comparison figure of each detector under non-uniform noise environment.

Embodiment

Below in conjunction with drawings and the specific embodiments, the present invention is described in further detail.

The embodiment of the present invention provides in a kind of baseband pool shares virtual resource collocation method, in this embodiment, for the MIMO frequency spectrum perception system be made up of single primary user and single cognitive user, wherein the antenna number of the transmitting terminal configuration of primary user is M, and the antenna number of cognitive user receiving terminal configuration is N.

It comprises the following steps:

S1, Received signal strength according to cognitive user receiving terminal, calculating sampling covariance matrix, and build corresponding augmentation sample covariance matrix;

For the Received signal strength of a kth sampling instant cognitive user receiving terminal, meet following form,

$\left.\begin{array}{c}{H}_0:x\left(k\right)=w\left(k\right)\\ H_1:x\left(k\right)=Hs\left(k\right)+w\left(k\right)\end{array}\right\}$

Wherein, H ₀for the hypothesis of frequency spectrum free time, H ₁for the hypothesis that frequency spectrum is taken by primary user, k=0,1 ..., K-1, K are sampling number;

x(k)＝[x ₁(k),...,x _n(k),...,x _N(k)] ^T

w(k)＝[w ₁(k),...,w _n(k),...,w _N(k)] ^T

s(k)＝[s ₁(k),...,s _m(k),...,s _M(k)] ^T

X _n(k) and w _nk () represents the signal that cognitive user n-th reception antenna receives and noise respectively; s _mk () represents the signal that m root antenna place launches; H is that N × M ties up fading channel matrix.Primary user's signal s _mk () is not rounded signal, noise is circle Gaussian noise, namely wherein be the noise power of n-th reception antenna place the unknown, in practice, due to reasons such as the non-demarcation of antenna, the noise power at different reception antenna place may be inconsistent.Meanwhile, statistical iteration each other between noise, and separate with primary user's signal.

For Received signal strength x (k), corresponding augmentation vector is x(k)=[x (k) ^t, x (k) ^h] ^t, correspondingly, for sampled signal X=[x (0) ..., x (k) ..., x (K-1)], corresponding augmentation form is x=[ x(0) ..., x(k) ..., x(K-1)].

On this basis, sample covariance matrix can be defined as

$S=\frac{1}{K}{\mathrm{XX}}^H$

Complementary sample covariance matrix can be defined as

$\tilde{S}=\frac{1}{K}{\mathrm{XX}}^T$

Accordingly, augmentation sample covariance matrix can be defined as

$\underset{\‾}{S}=\frac{1}{K}{\underset{\‾}{XX}}^H=\left[\begin{array}{cc}S& \tilde{S}\\ {\tilde{S}}^{*}& S^{*}\end{array}\right]$

S2, for augmentation sample covariance matrix, calculate the F norm of its matrix, thus build test statistics T;

Utilize local maximum effect invariant test, it is as follows for can building test statistics:

$T=\left|\right|\hat{C}|{|}_F^2$

Wherein, and F=diag ( s) be the diagonal matrix be made up of the diagonal entry of augmentation sample covariance matrix, || || _fthe F norm of representing matrix.

S3, by making comparisons with the thresholding preset, if statistic is greater than thresholding, then think that frequency spectrum is occupied, otherwise, then think that frequency spectrum is idle, thus complete detection.

For decision threshold γ, following formula is utilized to realize frequency spectrum perception

$T=\left|\right|\hat{C}|{{|}_F^2}_{H_0}^{H_1}\mathrm{\γ}$

Wherein, γ is decision threshold, even T > γ, then think and suppose H ₁set up, frequency spectrum is occupied, otherwise, then think and suppose H ₀set up, frequency spectrum is idle, thus completes detection.

Below by way of experiment simulation, the present invention is described in detail.

Simulation parameter is arranged:

Suppose that primary user's signal is bpsk signal, consider non-uniform noise scene, for the situation of N=4, suppose that each antenna end noise power is for [117072] dB, for the situation of N=6, each antenna end noise power is [-12 ,-03,26,-08,24 ,-27] dB, often organizes simulation result and obtains by 10000 Monte-Carlo Simulation experiments.As a comparison, also investigated desirable energy measuring utilize the energy measuring of noise power estimation value traditional Hadamard method, and based on not rounded signal the detection perform of method.

In FIG, give under channel circumstance, method and the present invention propose the experienced probability distribution of local maximum effect invariant test method test statistics.Number of transmit antennas M=1 in emulation, reception antenna number N=4, sampling number K=30, signal to noise ratio snr=-5dB.As can be seen from result, the present invention proposes the test statistics of the local maximum effect invariant test method probability distribution overlapping area under two kinds of hypothesis much smaller than method, therefore under same Parameter Conditions, its detection perform is better.

Fig. 2 gives under channel circumstance, the detection perform correlation curve figure of each detection method, supposes number of transmit antennas M=1 in emulation, reception antenna number N=4, sampling number K=15, as can be seen from result, when sampling number is very little, proposes maximum effect invariant test method in local is better than additive method, and performance is best.

Fig. 3 gives under channel circumstance, the detection perform correlation curve figure of each detection method, supposes number of transmit antennas M=1 in emulation, reception antenna number N=6, sampling number K=15, as can be seen from result, equally when sampling number is very little, proposes maximum effect invariant test method in local is better than additive method, and performance is best.

Fig. 4 gives each detector under non-uniform noise (each reception antenna end noise power is unequal) curve, supposes primary amount M=1 in emulation, cognitive user reception antenna number sampling number K=30, signal to noise ratio for non-uniform noise, suppose that the noise power of each reception antenna end is for [117072] dB.As can be seen from result, proposition maximum effect invariant test method in local effectivelyly can overcome the impact of non-uniform noise.

The present invention, by the characteristic by not rounded signal, takes full advantage of the dependency structure of Received signal strength, thus can be very little at sampling number, under the scene that signal to noise ratio is very low, reaches the detection perform more excellent than GLRT.

Those skilled in the art can carry out various modifications and variations to the embodiment of the present invention, if these amendments and modification are within the scope of the claims in the present invention and equivalent technologies thereof, then these revise and modification also within protection scope of the present invention.

The prior art that the content do not described in detail in specification is known to the skilled person.

Claims (6)

1., based on the constant detection frequency spectrum sensing method of local maximum effect of not rounded signal, it is characterized in that comprising the following steps:

A, Received signal strength according to cognitive user receiving terminal, calculating sampling covariance matrix, and build augmentation sample covariance matrix corresponding to described sample covariance matrix;

B, calculate the F norm of described augmentation sample covariance matrix, build test statistics T;

C, judge whether described test statistics T is greater than default threshold value, if T is greater than thresholding T > γ, then think that frequency spectrum is occupied, otherwise, then think that frequency spectrum is idle, thus complete detection.

2., as claimed in claim 1 based on the constant detection frequency spectrum sensing method of local maximum effect of not rounded signal, it is characterized in that: the Received signal strength described in steps A has following form

$\left.\begin{array}{c}{H}_0:x\left(k\right)=w\left(k\right)\\ H_1:x\left(k\right)=Hs\left(k\right)+w\left(k\right)\end{array}\right\}$

Wherein, H ₀for the hypothesis of frequency spectrum free time, H ₁for the hypothesis that frequency spectrum is taken by primary user, x (k) is a kth signal vector that sampling instant cognitive user receiving terminal receives, the noise signal vector that w (k) is a kth sampling instant, the signal vector that s (k) launches for a kth sampling instant primary user transmitting terminal, k=0,1 ..., K-1, K are sampling number; Particularly,

x(k)＝[x ₁(k),...,x _n(k),...,x _N(k)] ^T

w(k)＝[w ₁(k),...,w _n(k),...,w _N(k)] ^T

s(k)＝[s ₁(k),...,s _m(k),...,s _M(k)] ^T

X _n(k) and w _n(k) (n=1,2 ..., N) represent the signal that cognitive user n-th reception antenna receives and noise respectively; s _m(k) (m=1,2 ..., M) represent the signal that m root antenna place launches; Subscript () ^trepresent transposition; M and N represents the antenna number of primary user's transmitting terminal and the configuration of cognitive user receiving terminal respectively; H is that N × M ties up fading channel matrix.Primary user's signal s _mk () is not rounded signal, noise is circle Gaussian noise, namely wherein be the noise power of n-th reception antenna place the unknown, in practice, due to reasons such as the non-demarcation of antenna, the noise power at different reception antenna place may be inconsistent.Meanwhile, statistical iteration each other between noise, and separate with primary user's signal.

3., as claimed in claim 1 based on the constant detection frequency spectrum sensing method of local maximum effect of not rounded signal, it is characterized in that: the sample covariance matrix in steps A has following form

$S=\frac{1}{K}{\mathrm{XX}}^H$

Wherein, S is sample covariance matrix, X=[x (0) ..., x (k) ..., x (K-1)] be sampled signal, subscript () ^hrepresent conjugate transpose.

4., as claimed in claim 1 based on the constant detection frequency spectrum sensing method of local maximum effect of not rounded signal, it is characterized in that: the augmentation sample covariance matrix in steps A has following form

$\underset{\‾}{S}=\frac{1}{K}\underset{\‾}{X}{\underset{\‾}{X}}^H=\left[\begin{array}{cc}S& \tilde{S}\\ {\tilde{S}}^{*}& S^{*}\end{array}\right]$

Wherein, x=[ x(0) ..., x(k) ..., x(K-1)], x(k)=[x (k) ^t, x (k) ^h] ^t, sfor augmentation sample covariance matrix, for complementary sample covariance matrix, subscript () ^*represent and get conjugation.

5., as claimed in claim 1 based on the constant detection frequency spectrum sensing method of local maximum effect of not rounded signal, it is characterized in that: the test statistics in step B has following form

$T=\left|\right|\hat{C}|{|}_F^2$

Wherein, and F=diag ( s) be the diagonal matrix be made up of the diagonal entry of augmentation sample covariance matrix, || || _fthe F norm of representing matrix.

6., as claimed in claim 1 based on the constant detection frequency spectrum sensing method of local maximum effect of not rounded signal, it is characterized in that: step C needs are realized by following discrimination model

$T=\left|\right|\hat{C}|{{|}_F^2}_{H_0}^{H_1}\mathrm{\γ}$

Wherein, γ is decision threshold, even T > γ, then think and suppose H ₁set up, frequency spectrum is occupied, otherwise, then think and suppose H ₀set up, frequency spectrum is idle, thus completes detection.