CN105915300A - RLNC-based back-off frequency spectrum prediction method in CR network - Google Patents

Abstract

The invention discloses a random-linear-network-code (RLNC)-based back-off frequency spectrum prediction method in a cognitive radio (CR) network. With utilization of frequency spectrum predictability caused by RLNC introduction into a main user channel of cognitive radio and a prediction effect on a frequency spectrum by a hidden markov model, a back-off prediction method is introduced to carry out frequency spectrum state prediction at a next time slot. With the method, defects that frequency spectrum state is switched frequently and the spectrum hole can not be sensed easily in the traditional CR network and the reliability is low of the existing frequency spectrum prediction algorithm can be overcome. With introduction of back-off prediction method, the frequency spectrum state prediction at a next time slot is carried out by using the frequency spectrum predictability caused by RLNC introduction into the cognitive radio network and the prediction effect on the frequency spectrum by the hidden markov model, so that frequency spectrum prediction reliability is improved.

Description

A kind of based on RLNC in CR network keep out of the way spectrum prediction method

Technical field

The present invention relates to the spectrum prediction field in cognitive radio networks, particularly to cognitive radio Based on random linear network encoding in network keep out of the way spectrum prediction method.

Background technology

Along with the high speed development of the wireless communication technologys such as mobile Internet and intelligent mobile terminal, channel radio The demand of frequency spectrum resource is sharply increased by communication system, and under the fixing frequency spectrum method of salary distribution, frequency spectrum provides Source is limited, and frequency spectrum imbalance between supply and demand becomes increasingly conspicuous, and therefore, the most rare of frequency spectrum resource becomes The new bottleneck of restriction Development of Wireless Communications.But, while frequency spectrum resource scarcity, fixing frequency spectrum divides Formula formula also results in the significant wastage of frequency spectrum resource.According to FCC (The Federal Communications Commission, FCC) work report shows, at fixing frequency Under spectrum distribution policy, some frequency ranges are the most crowded, and band segment is often idle, average frequency Spectrum utilization rate is only 15%～85%, causes great frequency spectrum waste.In this context, Mitola Et al. proposed cognitive radio (Cognitive Radio, CR) technology in 1999.CR technology It is the wireless communication technology of a kind of intelligence, it is possible to realize by the way of intellectual learning protection primary user While (Primary User, PU), it is achieved effective utilization of frequency spectrum resource, improve spectrum utilization Rate, thus solve the problem that frequency spectrum is in short supply, imbalance between supply and demand becomes increasingly conspicuous.Therefore, cognitive radio skill Art becomes one of study hotspot of future wireless system development

Frequency spectrum perception is the basic function of cognitive radio, is to realize spectrum management and the premise shared, There is in cognitive radio fundamental position.Owing to secondary user has relatively low priority, it is therefore necessary to Ensureing when primary user occurs, secondary user stops the use of this channel immediately and carries out channel switching, with Ensure primary user's communication not to be caused harmful interference.But frequency spectrum perception algorithm based on detection, necessarily needs To carry out the detection of frequency spectrum state the regular hour, i.e. there is detection time delay.Therefore, how to reduce or Eliminate the impact of detection time delay, it is achieved the reliable prediction of frequency spectrum state, improve perceptual performance and have important Research Significance.Based on this, Akbar, I.A et al. in 2007 at Proceedings 2007 IEEE " the Dynamic spectrum allocation in cognitive delivered on SoutheastCon Radio using hidden Markov models:Poisson distributed case " literary composition, Propose prediction formula frequency spectrum shift algorithm based on HMM, SU by prediction algorithm prediction frequency spectrum state and The time that PU signal occurs, thus in the spectrum switching of PU initiating business request advance line frequency, effectively reduce The transmission collision probability of SU and PU.Xing Xiaoshuang et al. in 2013 at IEEE " the Spectrum prediction in cognitive delivered on Wireless Communications Radio networks " literary composition systematically describe in CR based on prediction frequency spectrum perception technology, and Elaborate spectrum prediction skill based on hidden Markov model (Hidden Markov Models, HMM) The proposition of art, apply and implement process, discussing spectrum prediction algorithm based on HMM in reality Superiority in application.But, owing to, in traditional cognitive radio networks, packet transmission has Having randomness and scattered property, frequency spectrum State Transferring is frequent, and the HMM model in above research is based on hurrying The frequent transitions of not busy two states is predicted, and using frequency spectrum detecting result as HMM parameter training sequence Row, reliability is relatively low.

For conventional cognitive radio network intermediate frequency spectrum State Transferring problem frequently, Wang Shanshan Et al. delivered on IEEE Transactions on Vehicular Technology in 2012 “The Impact of Induced Spectrum Predictability Via Wireless Network Coding " one literary composition, propose network code frequency spectrum state is had shaping operation, by The PU channel of CR network introduces network code, utilizes the shaping operation of network code to make spectrum structure Change, improve spectrum prediction；2014, Anthony Fanous et al. in 2014 at IEEE " Reliable is delivered on Journal on Selected Areas in Communications Spectrum Sensing and Opportunistic Access in Network-Coded Communications " literary composition, accesses frequency spectrum perception algorithm based on network code and opportunistic and calculates Method is inquired into, and verifies that introducing network code in cognitive radio can be effectively improved perceptibility Energy and throughput.

To sum up, relatively low for frequency spectrum State Transferring frequent and existing spectrum prediction algorithm reliability Problem, the present invention combines hidden Markov model to the predicting function of frequency spectrum with at cognitive radio networks Middle introducing random linear network encoding (Random Linear Network Coding, RLNC) brings Spectrum prediction, propose in a kind of cognitive radio networks based on random linear network encoding keep out of the way frequency Spectrum Forecasting Methodology, it is intended to realize the reliable prediction of frequency spectrum state.

Summary of the invention

It is an object of the present invention to provide a kind of spectrum prediction of keeping out of the way based on RLNC in CR network Method.The present invention can overcome conventional cognitive radio (CR) network intermediate frequency spectrum State Transferring frequently, Frequency spectrum cavity-pocket is not easy to perception and the relatively low feature of existing spectrum prediction algorithm reliability, is combined in and recognizes Know and radio net introduces spectrum prediction and the hidden horse that random linear network encoding (RLNC) brings The Er Kefu model predicting function to frequency spectrum, and introduce keep out of the way Forecasting Methodology, carry out the frequency spectrum of next time slot Status predication, thus improve the reliability of spectrum prediction.

In order to realize the purpose of the present invention and solve the problem of above-mentioned existence, the invention provides one and exist Spectrum prediction method of keeping out of the way based on RLNC in CR network, the attainable scheme of this method is: utilize The spectrum prediction brought of random linear network encoding and hidden is introduced in primary user's channel of cognitive radio The Markov model predicting function to frequency spectrum, introduces and keeps out of the way Forecasting Methodology, carry out the frequency spectrum of next time slot Status predication.

Preferably, following steps are specifically included:

Step one, initialization HMM parameter model λ={ Π, A, B}: by the primary user at cognitive radio Primary user's channel spectrum state that in channel, introducing random linear network encoding obtains carries out hidden Markov mould Type modeling obtains HMM parameter model λ={ Π, A, B}；

Step 2, operation accumulation and energy detection algorithm carry out frequency spectrum perception, obtain the observation of T time slot Sequence O={o_t∈ V | t=1,2 ..., T}, the state space V={0,1}, 0 of the sequence of observations represent that time user sentences Certainly this time slot is idle, and 1 judgement is busy；By described sequence of observations O={o_t∈ V | t=1,2 ..., T} makees For training sequence；

Step 3, operation Baum-Welch algorithm carry out the training of HMM parameter Π and A, are estimated HMM parameter

Step 4, operation Viterbi algorithm obtain hidden state sequence Q={q₁,q₂,…,q_t,…,q_T}；

Step 5, utilize hidden state sequence Q={q₁,q₂,…,q_t,…,q_TSubstitute sequence of observations O, so Rear operating procedure three carries out secondary HMM parameter estimation, obtains HMM parameter lambda_Q；

Step 6, utilize HMM parameter lambda_QEach frequency spectrum shape of next time slot is calculated with the sequence of observations updated The prediction probability of state, relatively each frequency spectrum status predication probability, the frequency spectrum state that probability is high is next time slot Frequency spectrum status predication result

Step 7, according to frequency spectrum status predication resultCarry out keeping out of the way prediction or frequency spectrum detection confirms, and then Update the sequence of observations, carry out the frequency spectrum status predication of next time slot；

Step 8, storage frequency spectrum status predication result

Preferably, described Π is initial state probabilities distribution, is expressed as Π={ π₀,π₁, π₀And π₁Respectively Representing that primary user's channel original state is not busy or the probability of busy condition, A is state transition probability matrix, It is expressed as A={ai_j}_(k+1)×(k+1), a_ij=P{q_t+1=S_j|q_t=S_iRepresent that current time t time-frequency spectrum state is S_i, S is transferred to during subsequent time t+1_jTransition probability, wherein k is coded block size；Described step one HMM parameter model λ=the hidden state space of Π, A, B} is S={0,1 ..., k}, wherein 0 represents primary user's letter Road is in not busy state, and 1～k is busy condition；Obtain state transition probability matrix A and the observation of HMM Probability matrix B is as follows:

$B={\left\{b_{ij}\right\}}_{(k+1)\×2}=\left[\begin{array}{cc}1-p_f& p_f\\ p_m& 1-p_m\\ .& .\\ .& .\\ .& .\\ p_m& 1-p_m\end{array}\right]$

Wherein α is the probability that state 0 is converted to state 1, and β is the probability that state k is converted to state 0, K is positive integer, false-alarm probability p_fWith false dismissal probability p_mBe calculated as follows:

$P_m=\mathrm{Pr}\[\frac{1}{n}\underset{i=1}{\overset{n}{\Σ}}{Y}_i^2\≤\mathrm{\τ}|H_1\]=\frac{\mathrm{\γ}(\frac{n}{2},\frac{n\mathrm{\τ}}{2({\mathrm{\σ}}^2+P)})}{\mathrm{\Γ}(n/2)}$

$P_f=\mathrm{Pr}\[\frac{1}{n}\underset{i=1}{\overset{n}{\Σ}}{Y}_i^2\≥\mathrm{\τ}|H_0\]=\frac{\mathrm{\Γ}(\frac{n}{2},\frac{n\mathrm{\τ}}{2{\mathrm{\σ}}^2})}{\mathrm{\Γ}(n/2)}$

WhereinFor gamma function,WithPoint Wei not descend incomplete gamma functions and upper incomplete gamma functions.

Preferably, being implemented as of described step 3:

1) sequence of observations O={o is utilized_t∈ V | t=1,2 ..., T} as HMM training sequence, and by with Lower two formulas carry out the iterative estimate of parameter, and the parameter obtaining the r time iteration is as follows:

$a_{ij}^r=\frac{\underset{t=1}{\overset{T-1}{\Σ}}{\mathrm{\ξ}}_t(i,j)}{\underset{t=1}{\overset{T-1}{\Σ}}{\mathrm{\γ}}_t\left(i\right)}$

${\mathrm{\π}}_i^r={\mathrm{\γ}}_1\left(i\right)$

Wherein γ_tI () represents at known HMM model and sequence of observations O={o_t∈ V | t=1,2 ..., the feelings of T} Under condition, t channel spectrum state is S_iProbability；ξ_t(i, channel status S when j) representing moment t_iAt next Moment, t+1 forwarded state S to_jExpectation transition probability；Therefore γ is obtained_t(i) and and ξ_t(i, expression formula j) is as follows:

${\mathrm{\γ}}_t\left(i\right)=\underset{j=1}{\overset{k+1}{\Σ}}{\mathrm{\ξ}}_t(i,j)$

$\begin{array}{c}{\mathrm{\ξ}}_t(i,j)=P(q_t=S_i,q_{t+1}=S_j|O,\mathrm{\λ})=\frac{P(q_t=S_i,q_{t=1}=S_j,O|\mathrm{\λ})}{P\left(O\right|\mathrm{\λ})}\\ =\frac{{\mathrm{\α}}_t\left(i\right)a_{ij}{b}_j\left(o_{t+1}\right)\left({\mathrm{\β}}_{t+1}\right)\left(j\right)}{\underset{i=1}{\overset{k+1}{\Σ}}\underset{j=1}{\overset{k+1}{\Σ}}{\mathrm{\α}}_t\left(i\right)a_{ij}{b}_j\left(o_{t+1}\right){\mathrm{\β}}_{t+1}\left(j\right)}\end{array}$

Wherein α_t(i)=P (o₁o₂…o_t,q_t=S_i| λ), 1≤t≤T-1 is forward variable, i.e. at given HMM mould Under type λ, the state of moment t is S_iAnd partial sequence is o₁o₂…o_tProbability；Correspondingly, β_t(i)=P (o_t+1o_t+2…o_T,q_t=S_i| λ), 1≤t≤T-1 is backward variable；α_t(i) and β_tI the calculating of () is respectively by front Obtain to algorithm and backward algorithm；

2) when the HMM parameter that iteration obtains meets the condition of convergence, terminate iteration, obtain HMM ginseng Number

Preferably, described iteration convergence condition is to work as λ_r={ π_r,A_rAnd λ_r-1={ π_r-1,A_r-1Parameters The quadratic sum of corresponding element difference less than 10^-6Time, iteration terminates.

Preferably, being implemented as of described step 4:

1) initialize:

δ₁(i)=π_ib_i(o₁), i=1 ..., k+1

Wherein δ_tI () is Viterbi variable, when representing moment t, HMM passes through a certain status switch path Arrival state S_i, and the sequence of observations is o₁o₂…o_TMaximum of probability, be expressed as follows:

${\mathrm{\δ}}_t\left(i\right)=\underset{q_1{q}_2...q_{t-1}}{max}P(q_1{q}_2...q_{t-1},q_t=i,o_1{o}_2...o_t|\mathrm{\λ})$

Or, δ_tI () is obtained by recurrence:

${\mathrm{\δ}}_{t+1}\left(j\right)=\[\underset{i}{max}{\mathrm{\δ}}_t\left(i\right)a_{ij}\]\·b_j\left(o_{t+1}\right)$

2) loop iteration:

${\mathrm{\δ}}_t\left(j\right)=\[\underset{1\≤i\≤k+1}{max}{\mathrm{\δ}}_{t-1}\left(i\right)a_{ij}\]\·b_j\left(o_t\right)$

J=1 ..., k+1

T=2 ..., T

3) iteration terminates:

$P^{*}=\underset{1\≤i\≤k+1}{max}\[{\mathrm{\δ}}_T\left(i\right)\]$

$q_T^{*}=\underset{1\≤i\≤N}{\mathrm{argmax}}\[{\mathrm{\δ}}_T\left(i\right)\].$

Preferably, described step 6 calculates the concrete recurrence meter of prediction probability of next each state of time slot Calculation formula is as follows:

$P_{{\mathrm{\λ}}_Q}\left(s_{t+1}\right|y_t)=\underset{s_t\∈S}{\Σ}{a}_{s_t{s}_{t+1}}{P}_{{\mathrm{\λ}}_Q}\left(s_t\right|y_t)$

$P_{{\mathrm{\λ}}_Q}\left(s_t\right|y_t)=\frac{P_{{\mathrm{\λ}}_Q}\left(s_t\right|y_{t-1})b\left(y_t\right|s_t)}{\underset{s_t\∈S}{\Σ}{P}_{{\mathrm{\λ}}_Q}\left(s_t\right|y_{t-1})b\left(y_t\right|s_t)}$

${\hat{s}}_{t+1}=\underset{s_{t+1}}{\mathrm{argmax}}{P}_{{\mathrm{\λ}}_Q}\left(s_{t+1}\right|y_t)$

Wherein,Initial value be When representing moment t=1, at HMM In state s₁Time probability,It is the predicted state of next time slot.

Preferably, described step 7 is according to frequency spectrum status predication resultThen carry out keeping out of the way prediction, If frequency spectrum status predication resultThen carry out frequency spectrum detection.

Preferably, described step 7 is according to frequency spectrum status predication resultCarry out different operations:

1) prediction is kept out of the way: if frequency spectrum status predication resultBe busy, then the frequency spectrum to a upper time slot StateCarry out detection to confirm, ifFor the free time, then it is predicted again after keeping out of the way k time slot, and will afterwards The observed result sequence of k time slot be set to hurry, and add the sequence of observations of HMM, update observation Sequence, repeats step 3 and carries out the spectrum prediction of next time slot to step 7；If the frequency spectrum shape of a upper time slot StateIt is busy, then keeps out of the way 1 time slot, and observed result busy for 1 time slot is added the sequence of observations, Update the sequence of observations, repeat step 3 and carry out the spectrum prediction of next time slot to step 7；

2) frequency spectrum detection: if frequency spectrum status predication resultFor spare time, the then CUSUM of operating procedure two Frequency spectrum detection algorithm carries out frequency spectrum detection, to confirm spectrum prediction result, if the sequence of observations of this time slot O=1 is busy, and the most secondary user does not carries out data transmission, and the results of observations obtaining this time slot is busy, adds Enter the sequence of observations of HMM, update the sequence of observations, repeat step 3 and carry out next time slot to step 7 Spectrum prediction；If O=0 is not busy, then carries out data transmission, and the results of observations obtaining this time slot is In the spare time, update the sequence of observations, repeat step 3 to step 7 and enter the prediction of next time slot.

The present invention at least includes following beneficial effect: the present invention can overcome in conventional cognitive radio network Frequency spectrum State Transferring is frequently, frequency spectrum cavity-pocket is not easy to perception and existing spectrum prediction sends out method reliability relatively Low feature.Random linear network encoding is introduced, to the most scattered frequency spectrum shape in cognitive radio State carries out shaping, makes frequency spectrum state tend to regularization, it is easier to perception；Then hidden Markov is utilized Primary user's channel spectrum state after introducing random network code is modeled analyzing by model；At frequency spectrum During prediction, relative to traditional spectrum prediction algorithm based on HMM, the present invention proposes utilization and estimates The hidden state sequence of meter carries out secondary HMM parameter estimation as the sequence of observations, and based on introduce with Hidden Markov model after machine linear network encoding proposes to keep out of the way spectrum prediction algorithm, and then improves frequency Spectrum forecasting reliability.

Accompanying drawing explanation

Fig. 1 is present invention flow chart keeping out of the way spectrum prediction method based on RLNC in CR network；

Fig. 2 is CR network model of the present invention；

Fig. 3 is to present invention introduces the spectrum structure comparison diagram before and after RLNC；

Fig. 4 is present invention primary user based on RLNC channel spectrum state model；

Fig. 5 is that Baum-Welch algorithm of the present invention realizes process.

Detailed description of the invention

The present invention is described in further detail below in conjunction with the accompanying drawings, to make those skilled in the art's reference Description word can be implemented according to this.

In cognitive radio shown in Fig. 2 (CR) network model, a cognitive radio networks comprises N number of PU channel and M SU, the most each PU channel is by a PU base station (Base Station, BS) and L PU Receiving node constitutes a PU sub-network.In each PU sub-network, packet arrives with the form of data block Reaching BS, and arrival process obeys Poisson distribution, Mean Speed is v.PU transmitting procedure introduces with Machine linear network encoding, utilizes random linear network encoding that scattered random frequency spectrum state is carried out shaping, Make frequency spectrum status architecture and regularization, as shown in Figure 3.When coded block size is k, BS accumulation k Each PU receiving node it is broadcast to again according to carrying out random linear network encoding after bag.SU is calculated by frequency spectrum perception Method perception PU channel also utilizes idle frequency spectrum to transmit data.Assume that each channel time synchronizes, a primary user The time of packet transmission is a time slot.

In the following embodiments, it is assumed that PU signal S [i]～CN (0,1), noise N [i]～CN (0,1), i.e. signal to noise ratio are SNR=0dB；Coded block size is k.Data block arrives the Mean Speed of BS Experiment carries out perception, 20 samples of each slot samples to 200 time slots.

A kind of way of realization of the present invention may is that utilization introduces in primary user's channel of cognitive radio Spectrum prediction that random linear network encoding brings and the hidden Markov model predicting function to frequency spectrum, Then introducing the HMM parameter model obtained and keep out of the way Forecasting Methodology, the frequency spectrum state carrying out next time slot is pre- Survey, such that it is able to improve the reliability of spectrum prediction.

As it is shown in figure 1, the method for the present invention includes below step:

Step 101, initialization HMM parameter model λ={ Π, A, B}: by the primary user at cognitive radio Primary user's channel spectrum state that in channel, introducing random linear network encoding obtains carries out hidden Markov mould Type modeling obtains HMM parameter model λ={ Π, A, B}；

Step 102, operation accumulation and energy measuring (CUSUM) algorithm carry out frequency spectrum perception, when obtaining T The sequence of observations O={o of gap_t∈ V | t=1,2 ..., T}, state space V={0,1}, 0 table of the sequence of observations Showing that time user adjudicates this time slot is the free time, and 1 judgement is busy；By the described sequence of observations O={o_t∈ V | t=1,2 ..., T} is as training sequence；

Step 103, operation Baum-Welch algorithm carry out the training of HMM parameter Π and A, are estimated The HMM parameter of meter

Step 104, operation Viterbi algorithm obtain hidden state sequence Q={q₁,q₂,…,q_t,…,q_T}；

Step 105, utilize hidden state sequence Q={q₁,q₂,…,q_t,…,q_TSubstitute sequence of observations O, Then operating procedure 103 carries out secondary HMM parameter estimation, obtains HMM parameter lambda_Q；

Step 106, utilize HMM parameter lambda_QEach frequency spectrum of next time slot is calculated with the sequence of observations updated The prediction probability of state, relatively each frequency spectrum status predication probability, the frequency spectrum state that probability is high is next time slot Frequency spectrum status predication result

Step 107, according to frequency spectrum status predication resultCarry out keeping out of the way prediction or frequency spectrum detection confirms, enter And update the sequence of observations, carry out the frequency spectrum status predication of next time slot；

Step 108, storage frequency spectrum status predication resultSo far whole spectrum prediction process is completed, By performing the method process, it is possible to overcome conventional cognitive radio network intermediate frequency spectrum State Transferring frequency Numerous, frequency spectrum cavity-pocket is not easy to perception and the relatively low feature of existing spectrum prediction algorithm reliability, effectively Improve spectrum prediction probability, reduce false-alarm probability, there is higher predicting reliability, and then realize in fall The low order user correctness to improving spectrum prediction while the interference of primary user.

Specifically, step 101 HMM parameter model λ=Π, A, B} initialize in, described HMM It is the primary user's channel condition model after introducing random linear network encoding in primary user's channel, is expressed as λ={ Π, A, B}.Assume that each primary user's channel is by a primary user base station (Base Station, BS) and L Individual primary user's receiving node constitutes primary user's sub-network.When coded block size is k, BS accumulates k Each primary user's receiving node it is broadcast to again after carrying out random linear network encoding after individual packet.Concrete real Existing process prescription is as follows:

(1) BS accumulation k according to bag and forms data block X=[x₁,…,x_k]；

(2) by galois field GF (q) stochastic generation code coefficient vector (a length of k), data block is carried out Random linear network encoding, the most each coding bag is expressed as:Code coefficient c_ij∈GF(q)；

(3) BS broadcast code packet gives each primary user's receiving node；

(4) raw data packets X is decoded after each receiving node receives k coded data packet completely；

(5) BS continues with next data block.If not enough k is according to bag in buffer, then BS stops Transmit and wait that caching k is reprocessed according to after bag.

It is assumed that the time of a main user data bag transmission is a time slot, primary user's channel State do not change in a time slot, then need k time slot to transmit k coded data packet, the busiest Period is at least k time slot.Therefore, primary user's channel status is modeled as the markov of k+1 state The hidden state definition space of process, i.e. HMM is S={0,1 ..., k}, wherein 0 represents primary user's channel Being in not busy state, 1～k represents busy condition.In HMM parameter, Π is initial state probabilities distribution, is expressed as Π={ π₀,π₁, π₀And π₁Represent the probability that primary user's channel original state is spare time/busy condition respectively；A is shape State transition probability matrix, is expressed as A={a_ij}_(k+1)×(k+1), a_ij=P{q_t+1=S_j|q_t=S_iRepresent current time t Time state be S_i, during subsequent time t+1, transfer to S_jTransition probability.In the present invention, connect due to BS Receive k according to bag and after carrying out random linear network encoding, just start the busy condition of k time slot, it is assumed that α Be converted to the probability of state 1 for state 0, β is the probability that state k is converted to state 0, then obtain shape Transition probability matrix A is as follows for state:

B is the observation probability matrix in HMM parameter, is expressed as B={b_ij}_(k+1)×2, wherein b_ijWhen representing t Carving frequency spectrum time of day is S_iTime SU observed result be V_jProbability, i.e. b_ij=P{o_t=V_j|q_t=S_i}。 Observation probability matrix is mainly by the detection performance impact of frequency spectrum perception algorithm.False-alarm probability P_fFor working as PU Signal do not exist and SU be judged as exist probability；False dismissal probability P_mThen for existing and SU when PU signal Judge the non-existent probability of PU.It is expressed as follows thus, it is possible to obtain observation probability matrix B:

$B={\left\{b_{ij}\right\}}_{(k+1)\×2}=\left[\begin{array}{cc}1-p_f& p_f\\ p_m& 1-p_m\\ .& .\\ .& .\\ .& .\\ p_m& 1-p_m\end{array}\right]$

Frequency spectrum detection algorithm in inventive algorithm is employing energy detection algorithm under Gaussian channel, it is assumed that N is the sample number in a time slot, and detection threshold value is τ, then false dismissal probability P_mWith false-alarm probability P_fPoint It is not calculated by following formula:

$P_f=\mathrm{Pr}\[\frac{1}{n}\underset{i=1}{\overset{n}{\Σ}}{Y}_i^2\≥\mathrm{\τ}|H_0\]=\frac{\mathrm{\Γ}(\frac{n}{2},\frac{n\mathrm{\τ}}{2{\mathrm{\σ}}^2})}{\mathrm{\Γ}(n/2)}$

$P_m=\mathrm{Pr}\[\frac{1}{n}\underset{i=1}{\overset{n}{\Σ}}{Y}_i^2\≤\mathrm{\τ}|H_1\]=\frac{\mathrm{\γ}(\frac{n}{2},\frac{n\mathrm{\τ}}{2({\mathrm{\σ}}^2+P)})}{\mathrm{\Γ}(n/2)}$

WhereinFor gamma function,WithPoint Wei not descend incomplete gamma functions and upper incomplete gamma functions.

As it is shown in figure 5, step 103 can be achieved by:

1) sequence of observations O={o is utilized_t∈ V | t=1,2 ..., T} as HMM training sequence, and by with Lower two formulas carry out the iterative estimate of parameter, and the parameter obtaining the r time iteration is as follows:

$a_{ij}^r=\frac{\underset{t=1}{\overset{T-1}{\Σ}}{\mathrm{\ξ}}_t(i,j)}{\underset{t=1}{\overset{T-1}{\Σ}}{\mathrm{\γ}}_t\left(i\right)}$

${\mathrm{\π}}_i^r={\mathrm{\γ}}_1\left(i\right)$

Wherein γ_tI () represents at known HMM model and sequence of observations O={o_t∈ V | t=1,2 ..., the feelings of T} Under condition, t channel spectrum state is S_iProbability；ξ_t(i, channel status S when j) representing moment t_iAt next Moment, t+1 forwarded state S to_jExpectation transition probability；Therefore γ is obtained_t(i) and and ξ_t(i, expression formula j) is as follows:

${\mathrm{\γ}}_t\left(i\right)=\underset{j=1}{\overset{k+1}{\Σ}}{\mathrm{\ξ}}_t(i,j)$

$\begin{array}{c}{\mathrm{\ξ}}_t(i,j)=P(q_t=S_i,q_{t+1}=S_j|O,\mathrm{\λ})=\frac{P(q_t=S_i,q_{t=1}=S_j,O|\mathrm{\λ})}{P\left(O\right|\mathrm{\λ})}\\ =\frac{{\mathrm{\α}}_t\left(i\right)a_{ij}{b}_j\left(o_{t+1}\right)\left({\mathrm{\β}}_{t+1}\right)\left(j\right)}{\underset{i=1}{\overset{k+1}{\Σ}}\underset{j=1}{\overset{k+1}{\Σ}}{\mathrm{\α}}_t\left(i\right)a_{ij}{b}_j\left(o_{t+1}\right){\mathrm{\β}}_{t+1}\left(j\right)}\end{array}$

Wherein α_t(i)=P (o₁o₂…o_t,q_t=S_i| λ), 1≤t≤T-1 is forward variable, i.e. at given HMM mould Under type λ, the state of moment t is S_iAnd partial sequence is o₁o₂…o_tProbability；Correspondingly, β_t(i)=P (o_t+1o_t+2…o_T,q_t=S_i| λ), 1≤t≤T-1 is backward variable；α_t(i) and β_tI the calculating of () is respectively by front Obtain to algorithm and backward algorithm；

2) when the HMM parameter that iteration obtains meets the condition of convergence, terminate iteration, obtain HMM ginseng Number

Specifically, it is also possible to be set to work as λ by above-mentioned iteration convergence condition_r={ π_r,A_rWith λ_r-1={ π_r-1,A_r-1The quadratic sum of corresponding element difference of parameters less than 10^-6Time, iteration terminates.

Specifically, can be achieved by for step 104:

1) initialize:

δ₁(i)=π_ib_i(o₁), i=1 ..., k+1

Wherein δ_tI () is Viterbi variable, when representing moment t, HMM passes through a certain status switch path Arrival state S_i, and the sequence of observations is o₁o₂…o_TMaximum of probability, be expressed as follows:

${\mathrm{\δ}}_t\left(i\right)=\underset{q_1{q}_2...q_{t-1}}{max}P(q_1{q}_2...q_{t-1},q_t=i,o_1{o}_2...o_t|\mathrm{\λ})$

Or, δ_tI () is obtained by recurrence:

${\mathrm{\δ}}_{t+1}\left(j\right)=\[\underset{i}{max}{\mathrm{\δ}}_t\left(i\right)a_{ij}\]\·b_j\left(o_{t+1}\right)$

2) loop iteration:

${\mathrm{\δ}}_t\left(j\right)=\[\underset{1\≤i\≤k+1}{max}{\mathrm{\δ}}_{t-1}\left(i\right)a_{ij}\]\·b_j\left(o_t\right)$

J=1 ..., k+1

T=2 ..., T

3) iteration terminates:

$P^{*}=\underset{1\≤i\≤k+1}{max}\[{\mathrm{\δ}}_T\left(i\right)\]$

$q_T^{*}=\underset{1\≤i\≤N}{\mathrm{argmax}}\[{\mathrm{\δ}}_T\left(i\right)\].$

It can be the best concealment status switch that will obtain in step 104 for step 105 Q={q₁,q₂,…,q_t,…,q_TSubstitute sequence of observations O={o_t∈ V | t=1,2 ..., T} calculates as Baum-Welch The HMM parameter training sequence of method, and rerun Baum-Welch algorithm and carry out HMM parameter Quadratic estimate obtains new HMM parameter lambda_Q。

The concrete recursive calculation of prediction probability step 106 being calculated to each frequency spectrum state of next time slot is public Formula can select as follows:

$P_{{\mathrm{\λ}}_Q}\left(s_{t+1}\right|y_t)=\underset{s_t\∈S}{\Σ}{a}_{s_t{s}_{t+1}}{P}_{{\mathrm{\λ}}_Q}\left(s_t\right|y_t)$

$P_{{\mathrm{\λ}}_Q}\left(s_t\right|y_t)=\frac{P_{{\mathrm{\λ}}_Q}\left(s_t\right|y_{t-1})b\left(y_t\right|s_t)}{\underset{s_t\∈S}{\Σ}{P}_{{\mathrm{\λ}}_Q}\left(s_t\right|y_{t-1})b\left(y_t\right|s_t)}$

${\hat{s}}_{t+1}=\underset{s_{t+1}}{\mathrm{argmax}}{P}_{{\mathrm{\λ}}_Q}\left(s_{t+1}\right|y_t)$

Wherein,Initial value be When representing moment t=1, at HMM In state s₁Time probability,It is the predicted state of next time slot.

Specifically, described step 107 can be according to frequency spectrum status predication resultThen keep out of the way Prediction, if frequency spectrum status predication resultThen carry out frequency spectrum detection.

Specifically, described step 107 can be according to frequency spectrum status predication resultCarry out different operations:

1) prediction is kept out of the way: if frequency spectrum status predication resultBe busy, then the frequency spectrum to a upper time slot StateCarry out detection to confirm, ifFor the free time, then it is predicted again after keeping out of the way k time slot, and will afterwards The observed result sequence of k time slot be set to hurry, and add the sequence of observations of HMM, update observation Sequence, repeats step 103 and carries out the spectrum prediction of next time slot to step 107；If the frequency of a upper time slot Spectrum stateIt is busy, then keeps out of the way 1 time slot, and observed result busy for 1 time slot is added observation sequence Row, update the sequence of observations, repeat step 103 and carry out the spectrum prediction of next time slot to step 107；

2) frequency spectrum detection: if frequency spectrum status predication resultFor spare time, the then CUSUM of operating procedure two Frequency spectrum detection algorithm carries out frequency spectrum detection, to confirm spectrum prediction result, if the sequence of observations of this time slot O=1 is busy, and the most secondary user does not carries out data transmission, and the results of observations obtaining this time slot is busy, adds Enter the sequence of observations of HMM, update the sequence of observations, repeat under the most rapid 103 to step 107 carries out The spectrum prediction of one time slot；If O=0 is not busy, then carries out data transmission, and obtain the observation of this time slot Result is not busy, updates the sequence of observations, repeats step step 103 and enters next time slot to step 107 Prediction.

Although embodiment of the present invention are disclosed as above, but it is not restricted to description and embodiment party Listed utilization in formula, it can be applied to various applicable the field of the invention completely, to being familiar with this area Personnel for, be easily achieved other amendment, therefore without departing substantially from claim and equivalency range Under the general concept limited, the present invention is not limited to specific details and shown here as the legend with description.

Claims (9)

1. based on RLNC in CR network keep out of the way spectrum prediction method for one kind, it is characterised in that profit Be used in primary user's channel of cognitive radio introduce the spectrum prediction brought of random linear network encoding and The hidden Markov model predicting function to frequency spectrum, introduces and keeps out of the way Forecasting Methodology, carry out the frequency of next time slot Spectrum status predication.

The most according to claim 1 based on RLNC in CR network keep out of the way spectrum prediction method, It is characterized in that, specifically include following steps:

Step one, initialization HMM parameter model λ={ Π, A, B}: by the primary user at cognitive radio Primary user's channel spectrum state that in channel, introducing random linear network encoding obtains carries out hidden Markov mould Type modeling obtains HMM parameter model λ={ Π, A, B}；

Step 2, operation accumulation and energy detection algorithm carry out frequency spectrum perception, obtain the observation of T time slot Sequence O={o_t∈ V | t=1,2 ..., T}, the state space V={0,1}, 0 of the sequence of observations represent that time user sentences Certainly this time slot is idle, and 1 judgement is busy；By described sequence of observations O={o_t∈ V | t=1,2 ..., T} makees For training sequence；

Step 3, operation Baum-Welch algorithm carry out the training of HMM parameter Π and A, are estimated HMM parameter

Step 4, operation Viterbi algorithm obtain hidden state sequence Q={q₁,q₂,…,q_t,…,q_T}；

Step 5, utilize hidden state sequence Q={q₁,q₂,…,q_t,…,q_TSubstitute sequence of observations O, so Rear operating procedure three carries out secondary HMM parameter estimation, obtains HMM parameter lambda_Q；

Step 6, utilize HMM parameter lambda_QEach frequency spectrum shape of next time slot is calculated with the sequence of observations updated The prediction probability of state, relatively each frequency spectrum status predication probability, the frequency spectrum state that probability is high is next time slot Frequency spectrum status predication result

Step 7, according to frequency spectrum status predication resultCarry out keeping out of the way prediction or frequency spectrum detection confirms, and then Update the sequence of observations, carry out the frequency spectrum status predication of next time slot；

Step 8, storage frequency spectrum status predication result

The most according to claim 2 based on RLNC in CR network keep out of the way spectrum prediction method, It is characterized in that, described Π is initial state probabilities distribution, is expressed as Π={ π₀,π₁, π₀And π₁Table respectively Showing that primary user's channel original state is not busy or the probability of busy condition, A is state transition probability matrix, table It is shown as A={a_ij}_(k+1)×(k+1), a_ij=P{q_t+1=S_j|q_t=S_iRepresent that current time t time-frequency spectrum state is S_iUnder, S is transferred to during one moment t+1_jTransition probability, wherein k is coded block size；The HMM of described step one Parameter model λ=the hidden state space of Π, A, B} is S={0,1 ..., k}, wherein 0 represents primary user's channel Being in not busy state, 1～k is busy condition；The state transition probability matrix A and the observation that obtain HMM are general Rate matrix B is as follows:

$B={\left\{b_{ij}\right\}}_{(k+1)\×2}=\left[\begin{array}{cc}1-p_f& p_f\\ p_m& 1-p_m\\ \begin{array}{c}.\\ .\\ .\end{array}& \begin{array}{c}.\\ .\\ .\end{array}\\ p_m& 1-p_m\end{array}\right]$

Wherein α is the probability that state 0 is converted to state 1, and β is the probability that state k is converted to state 0, K is positive integer, false-alarm probability p_fWith false dismissal probability p_mBe calculated as follows:

$P_m=\mathrm{Pr}\[\frac{1}{n}\underset{i=1}{\overset{n}{\Σ}}{Y}_i^2\≤\mathrm{\τ}|H_1\]=\frac{\mathrm{\γ}(\frac{n}{2},\frac{n\mathrm{\τ}}{2({\mathrm{\σ}}^2+P)})}{\mathrm{\Γ}(n/2)}$

$P_f=\mathrm{Pr}\[\frac{1}{n}\underset{i=1}{\overset{n}{\Σ}}{Y}_i^2\≥\mathrm{\τ}|H_0\]=\frac{\mathrm{\Γ}(\frac{n}{2},\frac{n\mathrm{\τ}}{2{\mathrm{\σ}}^2})}{\mathrm{\Γ}(n/2)}$

WhereinFor gamma function,WithPoint Wei not descend incomplete gamma functions and upper incomplete gamma functions.

4. keep out of the way spectrum prediction side according to based on RLNC in CR network described in Claims 2 or 3 Method, it is characterised in that being implemented as of described step 3:

1) sequence of observations O={o is utilized_t∈ V | t=1,2 ..., T} as HMM training sequence, and by with Lower two formulas carry out the iterative estimate of parameter, and the parameter obtaining the r time iteration is as follows:

$a_{ij}^r=\frac{\underset{t=1}{\overset{T-1}{\Σ}}{\mathrm{\ξ}}_t(i,j)}{\underset{t=1}{\overset{T-1}{\Σ}}{\mathrm{\γ}}_t\left(i\right)}$

${\mathrm{\π}}_i^r={\mathrm{\γ}}_1\left(i\right)$

Wherein γ_tI () represents at known HMM model and sequence of observations O={o_t∈ V | t=1,2 ..., the feelings of T} Under condition, t channel spectrum state is S_iProbability；ξ_t(i, channel status S when j) representing moment t_iAt next Moment, t+1 forwarded state S to_jExpectation transition probability；Therefore γ is obtained_t(i) and and ξ_t(i, expression formula j) is as follows:

${\mathrm{\γ}}_t\left(i\right)=\underset{j=1}{\overset{k+1}{\Σ}}{\mathrm{\ξ}}_t(i,j)$

$\begin{array}{c}{\mathrm{\ξ}}_t(i,j)=P(q_t=S_i,q_{t+1}=S_j|O,\mathrm{\λ})=\frac{P(q_t=S_i,q_{t+1}=S_j,O|\mathrm{\λ})}{P\left(O\right|\mathrm{\λ})}\\ =\frac{{\mathrm{\α}}_t\left(i\right)a_{ij}{b}_j\left(o_{t+1}\right){\mathrm{\β}}_{t+1}\left(j\right)}{\underset{i=1}{\overset{k+1}{\Σ}}\underset{j=1}{\overset{k+1}{\Σ}}{\mathrm{\α}}_t\left(i\right)a_{ij}{b}_j\left(o_{t+1}\right){\mathrm{\β}}_{t+1}\left(j\right)}\end{array}$

Wherein α_t(i)=P (o₁o₂…o_t,q_t=S_i| λ), 1≤t≤T-1 is forward variable, i.e. at given HMM mould Under type λ, the state of moment t is S_iAnd partial sequence is o₁o₂…o_tProbability；Correspondingly, β_t(i)=P (o_t+1o_t+2…o_T,q_t=S_i| λ), 1≤t≤T-1 is backward variable；α_t(i) and β_tI the calculating of () is respectively by front Obtain to algorithm and backward algorithm；

2) when the HMM parameter that iteration obtains meets the condition of convergence, terminate iteration, obtain HMM ginseng Number

The most according to claim 4 based on RLNC in CR network keep out of the way spectrum prediction method, It is characterized in that, described iteration convergence condition is to work as λ_r={ π_r,A_rAnd λ_r-1={ π_r-1,A_r-1Parameters The quadratic sum of corresponding element difference is less than 10^-6Time, iteration terminates.

The most according to claim 5 based on RLNC in CR network keep out of the way spectrum prediction method, It is characterized in that, being implemented as of described step 4:

1) initialize:

δ₁(i)=π_ib_i(o₁), i=1 ..., k+1

Wherein δ_tI () is Viterbi variable, when representing moment t, HMM passes through a certain status switch path Arrival state S_i, and the sequence of observations is o₁o_2…o_TMaximum of probability, be expressed as follows:

${\mathrm{\δ}}_t\left(i\right)=\underset{q_1{q}_2...q_{t-1}}{max}P(q_1{q}_2...q_{t-1},q_t=i,o_1{o}_2...o_t|\mathrm{\λ})$

Or, δ_tI () is obtained by recurrence:

${\mathrm{\δ}}_{t+1}\left(j\right)=\[\underset{i}{max}{\mathrm{\δ}}_t\left(i\right)a_{ij}\]\·b_j\left(o_{t+1}\right)$

2) loop iteration:

${\mathrm{\δ}}_t\left(j\right)=\[\underset{1\≤i\≤k+1}{max}{\mathrm{\δ}}_{t-1}\left(i\right)a_{ij}\]\·b_j\left(o_t\right)$

J=1 ..., k+1

T=2 ..., T

3) iteration terminates:

$P^{*}=\underset{1\≤i\≤k+1}{max}\[{\mathrm{\δ}}_T\left(i\right)\]$

$q_T^{*}=\underset{1\≤i\≤N}{\mathrm{argmax}}\[{\mathrm{\δ}}_T\left(i\right)\].$

7. according to the spectrum prediction of keeping out of the way based on RLNC in CR network described in claim 2 or 6 Method, it is characterised in that described step 6 calculates specifically passing of the prediction probability of next each state of time slot Return computing formula as follows:

$P_{{\mathrm{\λ}}_Q}\left(s_{t+1}\right|y_t)=\underset{s_t\∈S}{\Σ}{a}_{s_t{s}_{t+1}}{P}_{{\mathrm{\λ}}_Q}\left(s_t\right|y_t)$

$P_{{\mathrm{\λ}}_Q}\left(s_t\right|y_t)=\frac{P_{{\mathrm{\λ}}_Q}\left(s_t\right|y_{t-1})b\left(y_t\right|s_t)}{\underset{s_t\∈S}{\Σ}{P}_{{\mathrm{\λ}}_Q}\left(s_t\right|y_{t-1})b\left(y_t\right|s_t)}$

${\hat{s}}_{t+1}=\underset{s_{t+1}}{\mathrm{argmax}}{P}_{{\mathrm{\λ}}_Q}\left(s_{t+1}\right|y_t)$

Wherein,Initial value be When representing moment t=1, at HMM In state s₁Time probability,It is the predicted state of next time slot.

The most according to claim 2 based on RLNC in CR network keep out of the way spectrum prediction method, It is characterized in that, described step 7 is according to frequency spectrum status predication resultThen carry out keeping out of the way prediction, if Frequency spectrum status predication resultThen carry out frequency spectrum detection.

The most according to claim 8 based on RLNC in CR network keep out of the way spectrum prediction method, It is characterized in that, described step 7 is according to frequency spectrum status predication resultCarry out different operations:

1) prediction is kept out of the way: if frequency spectrum status predication resultBe busy, then the frequency spectrum to a upper time slot StateCarry out detection to confirm, ifFor the free time, then it is predicted again after keeping out of the way k time slot, and will afterwards The observed result sequence of k time slot be set to hurry, and add the sequence of observations of HMM, update observation Sequence, repeats step 3 and carries out the spectrum prediction of next time slot to step 7；If the frequency spectrum shape of a upper time slot StateIt is busy, then keeps out of the way 1 time slot, and observed result busy for 1 time slot is added the sequence of observations, Update the sequence of observations, repeat step 3 and carry out the spectrum prediction of next time slot to step 7；

2) frequency spectrum detection: if frequency spectrum status predication resultFor spare time, then the CUSUM frequency of operating procedure two Spectrum detection algorithm carries out frequency spectrum detection, to confirm spectrum prediction result, if the sequence of observations O=1 of this time slot Being busy, the most secondary user does not carries out data transmission, and the results of observations obtaining this time slot is busy, adds HMM The sequence of observations, update the sequence of observations, repeat step 3 and carry out the frequency spectrum of next time slot to step 7 Prediction；If O=0 is not busy, then carry out data transmission, and the results of observations obtaining this time slot is the spare time, more The new sequence of observations, repeats step 3 to step 7 and enters the prediction of next time slot.