CN105634634A - Asynchronous channel perception method with unknown timing - Google Patents

Abstract

The invention discloses an asynchronous channel perception method with unknown timing. The asynchronous channel sensing method specifically comprises the following steps: a first step, predicating an existence probability and a perception time difference of a certain PU at a k moment; a second step, updating the existence probability predicted value and the perception time difference via an observation value; a third step, comparing the updated existence probability predicted value with a decision threshold to obtain a state value of the PU; a fourth step, calculating the perception time difference of the PU at the k moment according to the state value; a fifth step, calculating a state value and a weight of a born particle at the k moment according to the perception time difference at a k-1 moment; and a sixth step, judging whether the k moment is the last moment, if so, recording and storing the perception time differences and state values of all moments; and otherwise, iteratively predicting the existence probability and the perception time difference of the PU at a k+1 moment through a particle set at the k moment. The asynchronous channel perception method has the advantages of being suitable for the dynamic spectrum sharing of a heterogeneous wireless network, avoiding the frequent signaling interaction between the PU and SU, reducing the configuration complexity of a CR system and saving the time and the energy cost.

Description

Asynchronous channel sensing method with unknown timing

Technical Field

The invention belongs to the field of communication, and particularly relates to an asynchronous channel sensing method with unknown timing.

Background

With the rapid development of wireless communication, spectrum has become extremely scarce as an unrecoverable resource; in order to solve the problem, Cognitive Radio (CR) technology has been widely focused and studied. The cognitive radio technology aims to repeatedly utilize and share frequency spectrum resources from multiple dimensions of time, space, frequency and the like on the premise of not influencing the normal work of authorized users (PUs).

Since cognitive radio users (SUs) cannot interfere with authorized users, spectrum sensing becomes a first prerequisite in the CR technology; spectrum sensing refers to that an SU acquires occupation information of a specific frequency band through signal detection and processing.

Classical spectrum sensing techniques include: energy Detection (ED), Matched Filter Detection (MFD), and cyclostationary feature detection (CD), in addition to the above-mentioned conventional detection methods, detection methods such as wavelet analysis, compressed sensing, and covariance matrix have been developed in recent years. It is noted that the existing spectrum sensing schemes assume synchronization between the SU sensing time and the PU transmission time.

In many practical applications, such as when LTE-U or other heterogeneous networks perform LBT operations, due to the difficulty in achieving cooperative timing between SU and PU, only channel noise is contained in some sensing time slots of SU, and PU signals are contained in other sensing time slots, resulting in mutual deviation between PU transmission time and SU sensing time.

The unknown sensing time difference can significantly affect the statistical characteristics of SU received signals, and particularly for a system sensitive to synchronous timing information, the spectrum sensing performance of the system can be seriously deteriorated, so that the actual performance of the existing spectrum sensing algorithm is seriously deteriorated, and the actual application requirements are difficult to meet.

Disclosure of Invention

The invention provides an asynchronous channel sensing method with unknown timing, which aims at the problem that in a heterogeneous network, the cooperation timing between an authorized user and a cognitive user can not be carried out usually, so that the sensing time difference exists between an authorized user transmitter and a cognitive user receiver;

the method comprises the following specific steps:

step one, aiming at a certain authorized user PU distributed with spectrum resources, predicting the existence probability q of the PU at the kth moment_k|k-1And a perceived time difference;

q is used as a prediction value of the existence probability of PU at the k moment_k|k-1Expressed, the formula is as follows:

q_k|k-1＝p_b×(1-q_k-1|k-1)+p_s×q_k-1|k-1

wherein q is_k-1|k-1The existence probability of the PU at the k-1 moment is obtained; p is a radical of_bThe birth probability is that the authorized user is in an idle state at the k-1 moment and jumps to a working state at the k moment; p is a radical of_sThe survival probability is that the authorized user is in the working state at the k-1 th moment and the k-th moment is also in the working state.

PU perception time difference t at k-th moment_kApproximating with a set of discrete particle values, the particles are divided into two parts: particles that are always present and birth particles.

Aiming at the perception time difference of the PU at the kth moment, the predicted value t of the particle_k|k-1The device comprises two parts: predicted value t of sensing time difference particle state value_k|k-1 ⁽ⁱ⁾And the predicted value of the particle weight_k|k-1 ⁽ⁱ⁾。

Firstly, the predicted value t of the perceived time difference particle state value of the PU at the kth moment_k|k-1 ⁽ⁱ⁾Generating according to the importance function:

t_k|k-1 ⁽ⁱ⁾～π_t(t_k|k-1 ⁽ⁱ⁾|t_k-1|k-1 ⁽ⁱ⁾)

wherein, i is 1,2., N + B; t is t_k-1|k-1 ⁽ⁱ⁾Is the state value of the perception time difference corresponding to the ith particle at the moment of k-1, pi_t(t_k|k-1 ⁽ⁱ⁾|t_k-1|k-1 ⁽ⁱ⁾) A transition probability distribution that is a perceptual time difference;

then, the predicted value of the weight of the sensing time difference particles_k|k-1 ⁽ⁱ⁾The device comprises two parts: the predicted value of the existing particle weight and the predicted value of the birth particle weight;

${{\mathrm{\ϵ}}_{k|k-1}}^{\left(i\right)}=\left\{\begin{array}{c}\frac{p_s\·q_{k-1|k-1}}{q_{k|k-1}}\·{{\mathrm{\ϵ}}_{k-1|k-1}}^{\left(i\right)};i=1,2,\mathrm{...},N\\ \frac{p_b\·(1-q_{k-1|k-1})}{q_{k|k-1}}\·\frac{1}{B};i=N+1,N+2,...,N+B\end{array}\right.$

1,2, N; n represents the number of particles that are always present; i ═ N +1, N +2,. and N + B; b represents the number of the birth particles.

Step two, the observation value z received by the PU at the k-th moment_kCalculating an updated value q of the presence probability prediction value_k|kAnd an update value of the perception time difference prediction value;

step 201, calculating an updated value q of the predicted value of the existence probability of the PU at the k-th moment_k|k；

The formula is as follows:

$q_{k|k}=\frac{I_k\·q_{k|k-1}}{1-q_{k|k-1}+I_k\·q_{k|k-1}}$

I_krepresenting the likelihood ratio of the PU;

the function of the likelihood is represented by,indicating that the authorized user PU does not exist; y is_k＝{t_k|k-1 ⁽ⁱ⁾Denotes that the authorized user PU is present;

μ₀and mu₁Is the mean value of the likelihood function; mu.s₀＝Mσ_u ²；μ₁＝Mσ_u ²+(M-t_k|k-1 ⁽ⁱ⁾)·α²；σ₀ ²And σ₁ ²Is the variance; sigma₀ ²＝2Mσ_u ⁴；σ₁ ²＝2Mσ_u ⁴+4(M-t_k|k-1 ⁽ⁱ⁾)α²·σ_u ²(ii) a M represents the sampling times of each perception moment; sigma_u ²Indicating the channel noise variance α is the channel fading factor.

Step 202, calculating the updated value of the particle weight of the PU sensing time difference at the kth moment by using the likelihood function as a number¹ _k|k ⁽ⁱ⁾(ii) a The formula is as follows:

i＝1,2...N+B；

step 203, updating the particle weight¹ _k|k ⁽ⁱ⁾Normalization is carried out to obtain a normalized updated value_k|k ⁽ⁱ⁾：

${{\mathrm{\ϵ}}_{k|k}}^{\left(i\right)}={\mathrm{\ϵ}}^{\′}{{}_{k|k}}^{\left(i\right)}/\underset{i=1}{\overset{N+B}{\Σ}}{\mathrm{\ϵ}}^{\′}{{}_{k|k}}^{\left(i\right)}$

i＝1,2...N+B；

Update value of particle weight'_k|k ⁽ⁱ⁾Normalization was performed to include particles that were always present and particles that were born.

Step 204, selecting the existing particle weight values to normalizeUpdated value after change_k|k ⁽ⁱ⁾Resampling the predicted value of the particle state value to obtain an updated value t of the predicted value of the particle state value of the perception time difference_k|k ⁽ⁱ⁾；

Normalizing the updated value of each always existing particle weight_k|k ⁽ⁱ⁾N, ═ 1,2.. N; with randomly generated n⁽ⁱ⁾In contrast, n⁽ⁱ⁾∈[0,1]If, if_k|k ⁽ⁱ⁾≥n⁽ⁱ⁾Then, the state value of the particle is saved, let t_k|k ⁽ⁱ⁾＝t_k|k-1 ⁽ⁱ⁾(ii) a Otherwise, let t_k|k ⁽ⁱ⁾Equal to the last particle state value saved.

Step 205, normalizing the updated value of each always existing particle weight_k|k ⁽ⁱ⁾Reset to

The set of particles that always exist after resampling is t_k|k ⁽ⁱ⁾,_k|k ⁽ⁱ⁾Therein ofi＝1,2...N。

Step three, updating the value q of the existence probability predicted value obtained in the step two_k|kComparing with the judgment threshold gamma to obtain the state value of the current authorized user PU

The decision threshold gamma is selected as the median of the existing probabilities of the PU.

The calculation is as follows:

${\hat{x}}_k=\left\{\begin{array}{cc}0,& q_{k|k}<\mathrm{\&gamma;}\\ 1,& q_{k|k}\&GreaterEqual;\mathrm{\&gamma;}\end{array}\right.$

when updating the value q_k|kWhen the value is less than the judgment threshold gamma, the state value of the PU of the authorized userIs 0; otherwise, authorizing the state value of the user PUIs 1.

Step four, calculating the perception time difference of the current authorized user PU at the moment k according to the state value of the current authorized user PU

When in useTime, observed value z_kTherein is onlyObserving the noise by using the difference in perceived time at time k-1 as the difference in perceived time at time k, i.e.

When in useThen, the PU is used for sensing the updated value of the time difference particle at the k moment, and the sensing time difference value at the k moment is calculatedNamely:

${\hat{t}}_k=\underset{i=1}{\overset{N}{\&Sigma;}}{t_{k|k}}^{\left(i\right)}{{\mathrm{\&epsiv;}}_{k|k}}^{\left(i\right)}$

step five, according to the perception time difference of the k-1 momentCalculating the state value t of the birth particle at the moment k_k|k ⁽ⁱ⁾And normalizing the updated value of each birth particle weight_k|k ⁽ⁱ⁾Reset to

${t_{k|k}}^{\left(i\right)}~{\mathrm{\&pi;}}_t\left({t_{k|k}}^{\left(i\right)}|{\hat{t}}_{k-1}\right);$

i＝N+1,N+2,...,N+B。

The updated value after each birth particle weight value is normalized_k|k ⁽ⁱ⁾Reset to

${{\mathrm{\&epsiv;}}_{k|k}}^{\left(i\right)}=\frac{1}{B}$

i＝N+1,N+2,...,N+B；

The state value t of the birth particle at the moment k_k|k ⁽ⁱ⁾Sum weight_k|k ⁽ⁱ⁾The following were used:

$\left\{\begin{array}{c}{t_{k|k}}^{\left(i\right)}~{\mathrm{\&pi;}}_t\left({t_{k|k}}^{\left(i\right)}|{\hat{t}}_{k-1}\right)\\ {{\mathrm{\&epsiv;}}_{k|k}}^{\left(i\right)}=\frac{1}{B}\end{array}\right.$

i＝N+1,N+2,...,N+B。

step six, judging whether the moment k is the last moment, if so, recording and storing the perception time difference of all the momentsAnd the status value of the authorized user PUOtherwise, entering the step seven;

step seven, repeating the step one, and using the particle set { t at the moment k obtained in the step two and the step five_k|k ⁽ⁱ⁾,_k|k ⁽ⁱ⁾N + B, iteratively predicting the existence probability and the perception time difference of the PU at the k +1 th moment.

The invention has the advantages that:

1) the asynchronous channel sensing method with unknown timing is suitable for spectrum sensing under the situation that dynamic unknown time offset exists between authorized users and cognitive users, is particularly suitable for heterogeneous wireless network dynamic spectrum sharing, avoids complicated signaling interaction between a PU and an SU, reduces the configuration complexity of a CR system, effectively saves time and energy expenses, and improves the CR realizability to the maximum extent.

2) The asynchronous channel perception method with unknown timing takes accumulated energy as system observability, and the energy observability which can be realized at low complexity is easy for system integration and design; meanwhile, the device has good expansibility, and other observed quantities can be conveniently taken into the device;

3) the asynchronous channel sensing method with unknown timing carries out joint estimation aiming at the state of an authorized user and unknown time deviation, and can effectively eliminate the information uncertainty of a received signal by estimating and tracking the sensing time difference of dynamic change, thereby obviously improving the spectrum sensing performance and having positive theoretical and practical significance for the dynamic spectrum sharing of a future wireless network.

4) The asynchronous channel sensing method with unknown timing can realize maximum posterior probability and maximum likelihood, and the prior art can not estimate unknown timing deviation by utilizing related likelihood distribution when a PU signal does not exist; when the timing deviation changes, the statistical characteristics of the receiving channel cannot be determined, and the spectrum detection is difficult to realize; in contrast, the method is based on Bayesian statistical reasoning and random finite set, and can solve the problems of mixed detection and estimation.

5) The asynchronous channel sensing method with unknown timing fully utilizes prior transition probability information of the working state of an authorized user, effectively overcomes the non-stable and non-Gaussian characteristics presented by an observation signal (accumulated energy) by adopting the edge particle filtering technology, and avoids the problem of overhigh calculation complexity when the traditional estimation scheme deals with high-dimensional detection.

Drawings

Fig. 1 is a block diagram of a signal processing apparatus at a spectrum sensing receiving end.

FIG. 2 is a diagram illustrating a prediction process of the existence probability at the kth time;

FIG. 3 is a flow chart of a method for asynchronous channel sensing with unknown timing;

FIG. 4 is a flow chart of a method of updating the presence probability prediction value and the perception time difference prediction value of the present invention;

fig. 5 shows the spectrum sensing detection accuracy of the new method under different sensing time difference ranges.

Fig. 6 shows relative errors of estimated perceptual time differences for different perceptual time difference ranges.

Fig. 7 is a graph comparing the performance of the new method spectrum sensing with the performance of the conventional ED.

Detailed Description

The present invention will be described in further detail below with reference to the accompanying drawings.

The invention provides a novel spectrum detection method and a novel spectrum detection device aiming at the spectrum detection problem under the asynchronous sensing scene of unknown time offset, and realizes the joint estimation of the authorized user state and the unknown sensing time deviation. The invention provides a unified dynamic state space model based on a Bayesian statistical estimation theory, provides an iterative estimation scheme based on a maximum posterior probability criterion and a Bernoulli random finite set by utilizing a random finite set, and obtains related complex distribution in a numerical approximation mode through particle filtering by utilizing an importance sampling principle. By accurately acquiring the sensing time difference, mixed spectrum sensing and unknown time offset estimation are realized. The invention can effectively eliminate the information uncertainty of the received signal, thereby obviously improving the spectrum sensing performance, and can be applied to further improve the spectrum efficiency in a future wireless communication network.

As shown in fig. 1, an asynchronous channel sensing apparatus with unknown timing comprises two modules: the device comprises a PU state estimation module and a perception time difference estimation module; the PU state estimation module is simultaneously connected with the antenna and the perception time difference estimation module.

And at a receiving end, joint estimation is realized by utilizing a PU state estimation module and a perception time difference estimation module.

A PU state estimation module: and sequentially updating and estimating the posterior probability of the working state of the authorized user by utilizing a Particle Filtering (PF) technology according to the observation signal of the receiving terminal at the current moment and the output value of the last moment perception time difference estimation module.

A perceptual time difference estimation module: and carrying out iterative estimation on the related state, namely the perception time difference when the PU signal exists by utilizing the thought of a random finite set according to the observation signal of the receiving end at the current moment and the output value of the master user state estimation module at the current moment.

The PU state estimation module receives a signal sent by the PU through the SU receiver, judges whether the PU uses the frequency spectrum resource at the moment in a mode of energy detection by considering the predicted sensing time difference obtained by the sensing time difference module, and transmits the result obtained by judgment to the sensing time difference estimation module.

The perception time difference estimation module judges whether the perception time difference needs to be estimated according to the PU state judgment result obtained by the last module. If the PU exists at a certain moment (namely the PU uses the frequency spectrum resources), the sensing time difference estimation module works, and the sensing time difference at the moment is estimated by the sensing time difference estimation module; if the PU does not exist at a certain moment (i.e. the PU is not using the spectrum resource), the sensing time difference estimation module does not work, and the estimation value of the sensing time difference is equal to the last moment. And obtaining a predicted value of the perception time difference at the next moment according to the currently estimated perception time difference, and transmitting the predicted value to the PU state estimation module for the next iteration.

An asynchronous channel sensing method with unknown timing comprises the steps of firstly, establishing a spectrum sensing dynamic state space detection model, taking the sensing time difference between the state of an unknown authorized user and dynamic change as two hidden states to be estimated, and taking the energy cumulant easy to integrate as the observable of a system.

The established spectrum sensing dynamic state space model is as follows:

x_k＝H(x_k-1)(1)

t_k＝T(t_k-1)(2)

z_k＝G(x_k,t_k,u_k,m)(3)

where equations (1) and (2) are referred to as state equations, and equation (3) represents a measurement equation.

x_kTransferring the state value of the authorized user PU at the kth moment according to a prior state transfer function H (-), wherein two hypothesis tests exist in the spectrum sensing, namely the existence of an authorized user signal and the existence of an authorized user signal are respectively detected by using H₀And H₁And (4) showing. When the authorized user signal is absent, i.e. the authorized band is idle, x_k0; when there is an authorized user signal, the obtained x_k＝1。

t_kAnd transferring the perception time difference of the authorized user transmitter and the cognitive user receiver at the kth moment according to a prior state transfer function T (-).

The observation signal is obtained by a conventional energy detection method, namely z_kThe value of the sum of the energies of the sampled signals within the observation time window of a certain length is calculated as follows:

$z_k=\left\{\begin{array}{cc}\underset{m=1}{\overset{M}{\&Sigma;}}{u_{k,m}}^2,& H_0\\ \underset{m=1}{\overset{t_k}{\&Sigma;}}{u_{k,m}}^2+\underset{m=t_k+1}{\overset{M}{\&Sigma;}}{\left(s_{k,m}\&times;x_k+u_{k,m}\right)}^2,& H_1\end{array}\right.---\left(4\right)$

wherein M is the number of sampling points in each sensing period Ts, M is Ts × fs, fs is the sampling frequency, u is the sampling frequency_k,mThe m-th sampling value of the channel noise at the time k is subjected to mean value of 0 and variance of sigma_u ²(ii) a gaussian distribution of; suppose that a user is authorized to operate in a state x within a sensing period_kIs not changed, and s_k,mSignals are sent on behalf of the PU, and without loss of generality, the PU can be assumed to adopt BPSK modulation signals, i.e. s_k,m∈ { -1, +1}, it should be noted that the method of the present invention can be extended to any complex modulated signal.

For convenience of explanation, z is used_0:k＝{z₀,z₁,...,z_kDenotes the observed value sequence from time 0 to time k. Estimation based on MAP criteria

$\left\{{\hat{x}}_k,{\hat{t}}_k\right\}=\arg \max p\left(x_k,t_k|z_{0:k}\right)$

Wherein,the transition is made according to a hypothetical first order Markov chain, with a prior distribution of p (x) at a known initial state₀) And p (t)₀) In the case of (2), the estimated value is obtained by two steps of prediction and update according to the MAP criterion.

In order to approximate the complex distribution in the estimation process, a particle filtering technique based on sequential importance sampling is used. Particle Filter (PF) is based on Monte Carlo importance sampling thought, and integral operation is converted into particle summation operation, and along with the increase of the number of particles, the probability density function of the particles gradually approaches to the required posterior probability density function, so that the Bayesian estimation effect is achieved. In a specific implementation, particle filtering employs a set of filters with weights ω⁽ⁱ⁾Of discrete particles x⁽ⁱ⁾To approximate a complex a posteriori distribution p (x), i.e.: p (x) ═ Σ_iω⁽ⁱ⁾(x-x⁽ⁱ⁾). Wherein the discrete particles x⁽ⁱ⁾And its weight omega⁽ⁱ⁾Sequential updates are made with new observations. The particle filtering mainly comprises the following 5 steps: (a) initializing particles; (b) sampling the order importance; (c) updating the weight value of the particle according to the new observed value; (d) resampling is carried out; (e) an estimate of the state is obtained based on the particles and the corresponding weights.

Based on the dynamic state space system, the invention designs and provides an iterative estimation algorithm based on the maximum posterior probability by utilizing Bayesian statistical inference and Bernoulli random finite set, and can carry out joint estimation on the state of an unknown authorized user and the dynamic perception time difference; finally, in order to further reduce the algorithm complexity, the importance sampling principle and the particle filtering are utilized, the involved complex distribution is approximated in a numerical mode, and the mixed spectrum sensing and unknown time offset estimation is realized.

As shown in fig. 3, the specific steps are as follows:

step one, aiming at a certain authorized user PU distributed with spectrum resources, predicting the existence probability q of the PU at the kth moment_k|k-1And a perceived time difference;

the predicted value of the existence probability of the PU at the kth moment consists of two parts, namely q_k|k-1Expressed, as shown in FIG. 2, the formula is as follows:

q_k|k-1＝p_b×(1-q_k-1|k-1)+p_s×q_k-1|k-1

wherein q is_k-1|k-1The existence probability of the PU at the k-1 moment is obtained; p is a radical of_bThe birth probability is that the authorized user is in an idle state at the k-1 moment and jumps to a working state at the k moment; p is a radical of_sThe survival probability is that the authorized user is in the working state at the k-1 th moment and the k-th moment is also in the working state.

PU perception time difference t at k-1_k-1A discrete set of particle values is used for approximation, as follows:

$i_{k-1}\underset{\&OverBar;}{\underset{\&OverBar;}{\mathrm{\&Delta;}}}\underset{i=1}{\overset{N^{\&prime;}}{\&Sigma;}}{t_{k-1}}^{\left(i\right)}{{\mathrm{\&epsiv;}}_{k-1}}^{\left(i\right)}$

wherein i represents the ith particle; n' represents the number of particles, t_k-1 ⁽ⁱ⁾Representing the state value of the ith particle at time k-1,_k-1 ⁽ⁱ⁾representing the weight of the ith particle at time k-1.

T is used for a predicted value of a PU (polyurethane) sensing time difference particle at the kth moment_k|k-1It also includes two parts: predicted value t for sensing time difference particle state value_k|k-1 ⁽ⁱ⁾The predicted value of the weight of the known particle_k|k-1 ⁽ⁱ⁾。

State value x for authorized user PU at time k_kAnd the perceived time difference t between the authorized user transmitter and the cognitive user receiver at time k_kThe invention will x_kAnd t_kThese two hidden states are represented by a Bernoulli Random Finite Set (BRFS) Y.

The bernoulli random finite set is not only a random variable of the elements in the set, but also a random variable of the number of the elements in the set (called the cardinality of the bernoulli random finite set, denoted by | Y |), and for an asynchronous perception scene, | Y | ∈ {0,1}, that is:

by a random variable x_kTo represent the cardinality of a finite set of bernoulli randomness; when x is_kWhen 1, the base number indicating the bernoulli random finite set is 1, and corresponds to | Y_k1, |; when x is_kWhen 0, the cardinality representing the bernoulli random finite set is 0, i.e. | Y_k|＝0。

Accordingly, the Bernoulli random finite set probability density is defined as follows:

where q represents the probability of an authorized user being present, p (t)_k) Representing a perceived time difference of t_kProbability of time.

From the above formula, Y at each time_kPossibly getOr t_kThus, the particles are divided into two parts during the implementation of the asynchronous perception algorithm: particles that are always present and birth particles.

Firstly, a predicted value t of a PU sensing time difference particle state value at the kth moment_k|k-1 ⁽ⁱ⁾Generating according to the importance function:

t_k|k-1 ⁽ⁱ⁾～π_t(t_k|k-1 ⁽ⁱ⁾|t_k-1|k-1 ⁽ⁱ⁾)，i＝1，2，…，N+B

wherein, t_k|k-1 ⁽ⁱ⁾Is the state value of the perception time difference corresponding to the ith particle at the moment k_t(t_k|k-1 ⁽ⁱ⁾|t_k-1|k-1 ⁽ⁱ⁾) To sense the time difference t_kThe transition probability distribution of (2).

Then, the predicted value of the weight of the sensing time difference particles_k|k-1 ⁽ⁱ⁾The device comprises two parts: the predicted value of the existing particle weight and the predicted value of the birth particle weight;

${{\mathrm{\&epsiv;}}_{k|k-1}}^{\left(i\right)}=\left\{\begin{array}{c}\frac{p_s\&CenterDot;q_{k-1|k-1}}{q_{k|k-1}}\&CenterDot;{{\mathrm{\&epsiv;}}_{k-1|k-1}}^{\left(i\right)};i=1,2,\mathrm{...},N\\ \frac{p_b\&CenterDot;(1-q_{k-1|k-1})}{q_{k|k-1}}\&CenterDot;\frac{1}{B};i=N+1,N+2,...,N+B\end{array}\right.$

1,2, N; n represents the number of particles that are always present; i ═ N +1, N +2,. and N + B; b represents the number of the birth particles.

Step two, the observation value z received by the PU at the k-th moment_kCalculating an updated value q of the predicted value of the existence probability_k|kAnd an update value of the perception time difference prediction value;

as shown in fig. 4, the specific steps are as follows:

step 201, calculating an updated value q of the predicted value of the existence probability of the PU at the k-th moment_k|k；

The formula is as follows:

$q_{k|k}=\frac{I_k\&CenterDot;q_{k|k-1}}{1-q_{k|k-1}+I_k\&CenterDot;q_{k|k-1}}$

I_krepresenting the likelihood ratio of the PU;

as can be seen from equation (4) and the central limit theorem, when the number of sampling points in each sensing period Ts is sufficiently large, the observed quantity z is observed_kThe approximation follows a gaussian distribution, and the likelihood function is calculated as follows:

z_kis the observed value at the k-th moment;the function of the likelihood is represented by,indicating that the authorized user PU does not exist; y is_k＝{t_k|k-1 ⁽ⁱ⁾Denotes that the authorized user PU is present;

μ₀and mu₁Is the mean value of the likelihood function; mu.s₀＝Mσ_u ²；μ₁＝Mσ_u ²+(M-t_k|k-1 ⁽ⁱ⁾)α²；σ₀ ²And σ₁ ²Is the variance; sigma₀ ²＝2Mσ_u ⁴；σ₁ ²＝2Mσ_u ⁴+4(M-t_k|k-1 ⁽ⁱ⁾)·α²·σ_u ²(ii) a M represents the number of sampling points at each sensing moment; sigma_u ²Representing the channel noise variance, α is the signal-to-fading factor.

Step 202, calculating the updated value of the particle weight of the PU sensing time difference at the kth moment by using the likelihood function¹ _k|k ⁽ⁱ⁾；

The formula is as follows:

i＝1,2...N+B；

step 203, normalizing the updated value of the particle weight to obtain a normalized updated value_k|k ⁽ⁱ⁾：

${{\mathrm{\&epsiv;}}_{k|k}}^{\left(i\right)}={\mathrm{\&epsiv;}}^{\&prime;}{{}_{k|k}}^{\left(i\right)}/\underset{i=1}{\overset{N+B}{\&Sigma;}}{\mathrm{\&epsiv;}}^{\&prime;}{{}_{k|k}}^{\left(i\right)}$

N + B1, 2.; update value of particle weight'_k|k ⁽ⁱ⁾Normalization was performed to include particles that were always present and particles that were born.

Step 204, normalizing the updated value according to the existing particle weight_k|k ⁽ⁱ⁾Resampling the predicted value of the particle state value to obtain an updated value t of the predicted value of the particle state value with the existing perception time difference_k|k ⁽ⁱ⁾；

And resampling the particle state value according to the updated weight value. The resampling mainly aims at overcoming the phenomenon of particle degradation, and the main idea is to eliminate particles with low weight value on the basis of importance sampling and copy reserved weightHigh-value particles, thereby achieving the purpose of inhibiting the degradation of the particles; obtaining the updated value t of the particle state_k|k ⁽ⁱ⁾。

The method specifically comprises the following steps: judging the updated value of each particle weight value after normalization_k|k ⁽ⁱ⁾And generating a random number n⁽ⁱ⁾Size of (1), n⁽ⁱ⁾∈[0,1]If, if_k|k ⁽ⁱ⁾≥n⁽ⁱ⁾Then the predicted value t of the particle state value is calculated_k|k-1 ⁽ⁱ⁾Stored as the updated value t of the particle state_k|k ⁽ⁱ⁾I.e. t_k|k ⁽ⁱ⁾＝t_k|k-1 ⁽ⁱ⁾(ii) a Otherwise_k|k ⁽ⁱ⁾＜n⁽ⁱ⁾The predicted value t of the particle state value is not saved_k|k-1 ⁽ⁱ⁾(ii) a Taking the last saved particle state value as the updated value t of the particle state_k|k ⁽ⁱ⁾；

The last saved particle state value may be t_k|k-1 ^(i-1),t_k|k-1 ^(i-2),...,t_k|k-1 ⁽¹⁾To a certain value.

Step 205, normalizing the updated value of each always existing particle weight_k|k ⁽ⁱ⁾Reset to

The resampled set of particles is { t }_k|k ⁽ⁱ⁾,_k|k ⁽ⁱ⁾Therein ofi＝1,2...N。

Step three, updating the value q of the existence probability predicted value obtained in the step two_k|kComparing with the judgment threshold gamma to obtain the state value of the current authorized user PU

The decision threshold γ is an intermediate value of the PU existence probability, and γ is 0.5 in this embodiment.

The calculation is as follows:

${\hat{x}}_k=\left\{\begin{array}{cc}0,& q_{k|k}<\mathrm{\&gamma;}\\ 1,& q_{k|k}\&GreaterEqual;\mathrm{\&gamma;}\end{array}\right.$

when updating the value q_k|kWhen the value is less than the judgment threshold gamma, the state value of the PU of the authorized userIs 0; otherwise, authorizing the state value of the user PUIs 1.

Step four, calculating the current grant according to the state value of the PU of the current authorized userPerceptual time difference of right user PU at k moment

When in useTime, observed value z_kIn which only the observation noise is used, the perception time difference value at the k-1 moment is used as the perception time difference value at the k moment, i.e. the

When in useThen, the PU is used for sensing the updated value of the time difference particle at the k moment, and the sensing time difference value at the k moment is calculatedNamely:

${\hat{t}}_k=\underset{i=1}{\overset{N}{\&Sigma;}}{t_{k|k}}^{\left(i\right)}{{\mathrm{\&epsiv;}}_{k|k}}^{\left(i\right)}$

step five, according to k-1 hourTemporal perceived time differenceExtracting the state value t of the birth particle at the moment k_k|k ⁽ⁱ⁾And normalizing the updated value of each birth particle weight_k|k ⁽ⁱ⁾Reset to

${t_{k|k}}^{\left(i\right)}~{\mathrm{\&pi;}}_t\left({t_{k|k}}^{\left(i\right)}|{\hat{t}}_{k-1}\right);$

i＝N+1,N+2,...,N+B。

The updated value after each birth particle weight value is normalized_k|k ⁽ⁱ⁾Reset to

${{\mathrm{\&epsiv;}}_{k|k}}^{\left(i\right)}=\frac{1}{B}$

i＝N+1,N+2,...,N+B；

The state value t of the birth particle at the moment k_k|k ⁽ⁱ⁾Sum weight_k|k ⁽ⁱ⁾The following were used:

$\left\{\begin{array}{c}{t_{k|k}}^{\left(i\right)}~{\mathrm{\&pi;}}_t\left({t_{k|k}}^{\left(i\right)}|{\hat{t}}_{k-1}\right)\\ {{\mathrm{\&epsiv;}}_{k|k}}^{\left(i\right)}=\frac{1}{B}\end{array}\right.$

i＝N+1,N+2,...,N+B。

step six, judging whether the moment k is the last moment, if so, recording and storing the perception time difference of all the momentsAnd the status value of the authorized user PUOtherwise, entering the step seven;

step seven, repeating the step one, and using the particle set { t at the moment k obtained in the step two and the step five_k|k ⁽ⁱ⁾,_k|k ⁽ⁱ⁾N + B, iteratively predicting the existence probability and the perception time difference of the PU at the k +1 th moment.

Example 1:

the asynchronous channel sensing method of the invention is simulated, and the birth probability p of the authorized user is set_b0.2, survival probability p_sThe number of sampling points M is 100, and the total time is 4000, when it is 0.8. M_tRange representing difference in perceived timeCircle, i.e. sense time difference t_k∈[0,M_t]Respectively take M_tIs 20, 60, 100; the simulation result of the spectrum sensing performance is shown in fig. 5, fig. 5 is a simulation diagram of the total correct detection probability under three sensing time difference ranges, the abscissa is the actual signal-to-noise ratio snr, and the ordinate is the overall correct detection probability P_D. From the simulation results, it can be seen that, in the case of the fixed sampling point number M being 100, the unknown sensing time difference M_tThe larger the dynamic range of (2), the worse the spectrum sensing performance; and meanwhile, the higher the perceived signal-to-noise ratio is, the better the obtained performance is. For example: when snr takes-6 dB, M_tP in the 20 state_D＝0.9299，M_tP in 60 state_D＝0.8722，M_tP in the 100 state_D0.7959; it can be seen that the influence of the sensing time difference on the sensing performance is great, and therefore, it is very necessary to consider the unknown time difference in the sensing process.

The method of the invention fully considers the influence of unknown time difference in spectrum sensing, estimates and tracks the unknown sensing time difference while executing spectrum detection, the obtained unknown time difference estimation performance is shown as figure 6, figure 6 is a relative error simulation diagram of the estimated sensing time difference under three sensing time difference ranges, the abscissa is the actual signal-to-noise ratio snr, and the ordinate is the relative error of the sensing time difference_t(ii) a It can be seen that the unknown time differential state range M is low in signal to noise ratio_tThe smaller, the resulting pair t_kThe better the estimation effect; the estimation error will decrease significantly as the signal-to-noise ratio increases gradually. For example, when the snr increases to 14dB, the estimation error of the unknown time difference decreases to 0.0155. Therefore, the asynchronous sensing method provided by the invention can accurately estimate the unknown dynamic sensing time difference.

A simulation result graph comparing the asynchronous sensing method and the conventional spectrum sensing method is shown in fig. 7, where the abscissa is the actual snr and the ordinate is the overall correct detection probability P_D. It is found from fig. 7 that when there is a dynamically unknown sensing time difference, the spectrum sensing performance will be seriously affected. Considering that the overall detection probability is 0.95, for the traditional energy perception algorithm which does not perform any processing, the required signal-to-noise ratio is 10 dB; assuming that a conventional energy detection scheme can obtain the expectation of an unknown time difference, and taking into account the time difference t when performing the energy decision_kThe dynamic variation characteristic can further improve the perception performance by about 1.8 dB. In contrast, the Bayesian asynchronous sensing scheme provided by the invention can eliminate the uncertainty of the observed signal and the information in the statistical characteristics thereof to the maximum extent by jointly estimating the unknown time difference, thereby remarkably improving the spectrum sensing performance. Experiments show that in order to achieve the overall detection probability of 0.95, the required sensing signal-to-noise ratio is only 2dB, and compared with the traditional energy sensing algorithm which does not perform any processing, the sensing performance can be improved by about 8 dB.

The asynchronous sensing method and the device can accurately estimate and track the time offset of dynamic change and obtain the sensing time difference of unknown change, thereby determining the time-varying statistical characteristic of the received signal and eliminating the information uncertainty of the received signal in the spectrum sensing process to the maximum extent, thereby obviously improving the spectrum sensing performance. In addition, the system model and the joint spectrum sensing algorithm can be conveniently expanded to other application scenes, the estimated and obtained time offset information has important significance for design and performance improvement of the CR communication system, and the CR communication system has good application potential in a future dynamic spectrum sharing network.

Claims (3)

1. An asynchronous channel sensing method with unknown timing is characterized by comprising the following specific steps:

step one, aiming at a certain authorized user PU distributed with spectrum resources, predicting the existence probability q of the PU at the kth moment_k|k-1And a perceived time difference;

predicted value q of PU existence probability at k moment_k|k-1The formula is as follows:

q_k|k-1＝|p_b×(1-q_k-1|k-1)+p_s×q_k-1|k-1

wherein q is_k-1|k-1The existence probability of the PU at the k-1 moment is obtained; p is a radical of_bTo the birth probability, p_sIs the survival probability;

the perception time difference of the PU at the kth moment is represented by a predicted value of a particle, and the perception time difference comprises two parts: predicted value t of sensing time difference particle state value_k|k-1 ⁽ⁱ⁾And the predicted value of the particle weight_k|k-1 ⁽ⁱ⁾；

Firstly, a predicted value t of a sensing time difference particle state value_k|k-1 ⁽ⁱ⁾Generating according to the importance function:

t_k|k-1 ⁽ⁱ⁾～π_t(t_k|k-1 ⁽ⁱ⁾|t_k-1|k-1 ⁽ⁱ⁾)

wherein, i is 1,2., N + B; t is t_k-1|k-1 ⁽ⁱ⁾Is the state value of the perception time difference corresponding to the ith particle at the moment of k-1, pi_t(t_k|k-1 ⁽ⁱ⁾|t_k-1|k-1 ⁽ⁱ⁾) A transition probability distribution that is a perceptual time difference;

then, a predicted value of the perceived time difference particle weight is obtained_k|k-1 ⁽ⁱ⁾The device comprises two parts: the predicted value of the existing particle weight and the predicted value of the birth particle weight;

${{\mathrm{\&epsiv;}}_{k|k-1}}^{\left(i\right)}=\left\{\begin{array}{c}\frac{p_s\&CenterDot;q_{k-1|k-1}}{q_{k|k-1}}\&CenterDot;{{\mathrm{\&epsiv;}}_{k-1|k-1}}^{\left(i\right)};i=1,2,...,N\\ \frac{p_b\&CenterDot;\left(1-q_{k-1|k-1}\right)}{q_{k|k-1}}\&CenterDot;\frac{1}{B};i=N+1,N+2,...,N+B\end{array}\right.$

1,2, N; n represents the number of particles that are always present; i ═ N +1, N +2,. and N + B; b represents the number of the birth particles;

step two, the observation value z received by PU at the moment k_kCalculating an updated value q of the presence probability prediction value_k|kAnd an update value of the perception time difference prediction value;

step 201, calculating an updated value q of the predicted value of the existence probability of the PU at the moment k_k|k；

The formula is as follows:

$q_{k|k}=\frac{I_k\&CenterDot;q_{k|k-1}}{1-q_{k|k-1}+I_k\&CenterDot;q_{k|k-1}}$

I_krepresenting the likelihood ratio of the PU;

step 202, calculating an updated value 'of the particle weight of the PU sensing time difference at the k-th moment by using a likelihood function'_k|k ⁽ⁱ⁾；

The formula is as follows:

i＝1,2...N+B；

step 203, update value of particle weight'_k|k ⁽ⁱ⁾Normalizing to obtain normalized value_k|k ⁽ⁱ⁾：

${{\mathrm{\&epsiv;}}_{k|k}}^{\left(i\right)}={\mathrm{\&epsiv;}}^{\&prime;}{{}_{k|k}}^{\left(i\right)}/\underset{i=1}{\overset{N+B}{\mathrm{\&Sigma;}}}{\mathrm{\&epsiv;}}^{\&prime;}{{}_{k|k}}^{\left(i\right)}$

i＝1,2...N+B；

Step 204, selecting the existing updated value after the normalization of the weight value of the particles_k|k ⁽ⁱ⁾Resampling the predicted value of the particle state value to obtain an updated value t of the predicted value of the particle state value of the perception time difference_k|k ⁽ⁱ⁾；

Normalizing the updated value of each always existing particle weight_k|k ⁽ⁱ⁾N, ═ 1,2.. N; with randomly generated n⁽ⁱ⁾In contrast, n⁽ⁱ⁾∈[0,1]If, if_k|k ⁽ⁱ⁾≥n⁽ⁱ⁾Then, the state value of the particle is saved, let t_k|k ⁽ⁱ⁾＝t_k|k-1 ⁽ⁱ⁾(ii) a Otherwise, let t_k|k ⁽ⁱ⁾Equal to the last saved particle state value;

step 205, normalizing the updated value of each always existing particle weight_k|k ⁽ⁱ⁾Reset to

The set of particles that always exist after resampling is t_k|k ⁽ⁱ⁾,_k|k ⁽ⁱ⁾Therein of

Step three, updating the value q of the existence probability predicted value obtained in the step two_k|kComparing with the judgment threshold gamma to obtain the state value of the current authorized user PU

Selecting a middle value of PU existence probability by the decision threshold gamma;

the calculation is as follows:

${\hat{x}}_k=\left\{\begin{array}{cc}0,& q_{k|k}<\mathrm{\&gamma;}\\ 1,& q_{k|k}\&GreaterEqual;\mathrm{\&gamma;}\end{array}\right.$

when updating the value q_k|kWhen the value is less than the judgment threshold gamma, the state value of the PU of the authorized userIs 0; otherwise, authorizing the state value of the user PUIs 1;

step four, calculating the perception time difference of the current authorized user PU at the moment k according to the state value of the current authorized user PU

When in useTime, observed value z_kIn which only the observation noise is used, the perception time difference value at the k-1 moment is used as the perception time difference value at the k moment, i.e. the

When in useThen, the PU is used for sensing the updated value of the time difference particle at the k moment, and the sensing time difference value at the k moment is calculatedNamely:

${\hat{t}}_k=\underset{i=1}{\overset{N}{\&Sigma;}}{t_{k|k}}^{\left(i\right)}{{\mathrm{\&epsiv;}}_{k|k}}^{\left(i\right)}$

step five, according to the perception time difference of the k-1 momentCalculating the state value t of the birth particle at the moment k_k|k ⁽ⁱ⁾And normalizing the updated value of each birth particle weight_k|k ⁽ⁱ⁾Reset to

${t_{k|k}}^{\left(i\right)}~{\mathrm{\&pi;}}_t\left({t_{k|k}}^{\left(i\right)}\right|{\hat{t}}_{k-1});$

i＝N+1,N+2,...,N+B；

The updated value after each birth particle weight value is normalized_k|k ⁽ⁱ⁾Reset to

${{\mathrm{\&epsiv;}}_{k|k}}^{\left(i\right)}=\frac{1}{B}$

i＝N+1,N+2,...,N+B；

Step six, judging whether the moment k is the last moment, if so, recording and storing the perception time difference of all the momentsAnd the status value of the authorized user PUOtherwise, entering the step seven;

step seven, repeating the step one, and using the particle set { t at the moment k obtained in the step two and the step five_k|k ⁽ⁱ⁾,_k|k ⁽ⁱ⁾N + B, iteratively predicting the existence probability and the perception time difference of the PU at the k +1 th moment.

2. The method as claimed in claim 1, wherein in step 201, the likelihood ratio I of the PU is determined by the unknown timing_k:

The function of the likelihood is represented by,indicating that the authorized user PU does not exist;indicating the existence of an authorized user PU;

μ₀and mu₁Is the mean value of the likelihood function; mu.s₀＝Mσ_u ²；μ₁＝Mσ_u ²+(M-t_k|k-1 ⁽ⁱ⁾)·α²；σ₀ ²And σ₁ ²Is the variance; sigma₀ ²＝2Mσ_u ⁴；σ₁ ²＝2Mσ_u ⁴+4(M-t_k|k-1 ⁽ⁱ⁾)·α²·σ_u ²(ii) a M represents the sampling times of each perception moment; sigma_u ²Representing the channel noise variance, α is the channel fading factor.

3. The method as claimed in claim 1, wherein in step 203, the updated value of the particle weight value'_k|k ⁽ⁱ⁾Normalization is carried out, including updating values of the always existing particle weight values'_k|k ⁽ⁱ⁾Carry out normalization and update value of birth particle weight'_k|k ⁽ⁱ⁾And (6) carrying out normalization.