CN102035586A - Energy efficient distributed relay selection algorithm in wireless cooperative relay network - Google Patents

Abstract

The invention provides an energy efficient distributed relay option algorithm in a wireless cooperative relay network. The relay network comprises a source node, a target node and a plurality of relay nodes. A method for selecting relay nodes comprises the following steps of: respectively determining the channel states and the energy states of all relay nodes at any time slot according to channel state models and energy state models of the plurality of relay nodes; calculating the system rewards of each relay node through a preset system reward model on the basis of the channel states and the energy states; and selecting a relay node for each time slot, wherein the selected relay nodes ensure that the obtained system reward in the whole data transmission process is maximized.

Description

The distributed relay selection algorithm of energy efficient in a kind of wireless cooperation junction network

Technical field

The application relates to wireless communication technology field, particularly, relates to the distributed relay system of selection of energy efficient in the wireless cooperation junction network.

Background technology

Space diversity (spatial diversity) technology is generally had an optimistic view of by industry as the effective technology of the adverse effect that a kind of reduction multipath fading (multipath fading) produces, and also therefore is suggested by wireless cooperation relaying (cooperative relaying) network that forms the distributed space diversity gain of virtual antenna array (virtual antenna arrays) acquisition.

In the wireless cooperation junction network, listening to the node that source node sends to the information of destination node is not considered as the information that listens to disturbing, but being transmitted (relaying), this information gives destination node, therefore destination node is received from the information after a plurality of independent declines (independently faded) of source node and via node, has been formed a Virtual space diversity transmission system.The trunking plan that has proposed comprise and amplify transmitting (amplify and forward, AF), decoding transmit (decode and forward, DF) and distributed Space Time Coding (space time coded, STC).Yet,, make that the enforcement of distributed space time coding scheme still is a very big difficult problem because the antenna number of system synchronization problem and the collaborative relaying of participation is uncertain.

For this reason, some documents have proposed the relay selection method in the collaborative junction network.Yet these methods have supposed that all channel fading is very slow, and current time remains unchanged at next Frame time slot by the channel status of Signalling exchange estimation, and the channel status of current time is simply as the channel estimating of next time slot.Yet, because the random time varying characteristic of mobile radio channel, this channel constant (channel changeless) supposes often not to be suitable for real network, and the relaying of selecting according to the channel status of current time is not to be optimum relaying for next Frame.In addition, these methods do not have to consider effectively to save the consumption of energy, via node do not adopt yet Adaptive Modulation and Coding (adaptive modulation and coding, AMC).

Summary of the invention

In order to solve above one or more problems of the prior art, the application has proposed a kind of via node method of selecting in the wireless cooperation junction network.

Can comprise according to the disclosed via node method of selecting in the wireless cooperation junction network of the application: channel status model and energy state model according to a plurality of via nodes in the wireless cooperation junction network are determined channel status and the energy state of each via node at any time slot respectively; Based on described channel status and energy state, by system's recompense of predetermined system's each described via node of recompense Model Calculation; And being via node of each time slot selection, selected via node makes the system's recompense maximization that obtains in the whole data transmission procedure.

The method that the application proposes is selected the via node of the use of each time slot under the situation of the dump energy of the time-varying characteristics of having considered channel and via node.According to the application's method, do not need center control nodes in the junction network, this method is moved with distributed way, and each via node can add or leave candidate collection at any time.In addition, the application's method considers that the DF trunking plan has the advantage of digital processing system, the noise scale-up problem of having avoided the AF scheme to bring.

Description of drawings

Fig. 1 is collaborative junction network model schematic diagram;

Fig. 2 is the exemplary process diagram according to the relay selection method of the application's execution mode;

Fig. 3 is the system's recompense data comparison diagram that the simulation result of the application's illustrative embodiments and existing relay selection method is compared acquisition;

Fig. 4 is the average error rate data comparison diagram that the simulation result of the application's illustrative embodiments and existing relay selection method is compared acquisition;

Fig. 5 is the spectrum efficiency data comparison diagram that the simulation result of the application's illustrative embodiments and existing relay selection method is compared acquisition; And

Fig. 6 is that the application's illustrative embodiments and existing relay selection method is at different threshold value N _ThThe network lifetime data comparison diagram of following acquisition.

Embodiment

For the application's scheme clearly is described, be elaborated below in conjunction with the illustrative embodiments of accompanying drawing to the application.

At first, introduce the distributed collaboration junction network with reference to Fig. 1.

In this application, consider a kind of equity (peer to peer) distributed collaboration junction network, each node all has the ability to transmit (relaying) packet into other nodes in the network, and its network model as shown in Figure 1.In collaborative trunking traffic, the transmission course of each frame data can be divided into two sub-slots: first sub-slots, and source node sends Frame to destination node, and destination node and via node receive data simultaneously; Second sub-slots, via node is transmitted this Frame, and destination node merges first sub-slots and makes cascading judgement from Frame and second sub-slots of source node from the same Frame of via node.

In this network, source node and destination node for example can use the request transmission/clear to send (Request-To-Send/Clear-To-Send, RTS/CTS) message avoid conflict and estimate channel signal to noise ratio (signal-to-noise ratio, SNR).In the network, the radio channel state available signal-to-noise ratio is represented, source node is to via node (source-to-relay, S2R) error rate of link changes with the SNR of this link, via node is to destination node (relay-to-destination, R2D) (modulation and coding scheme, MCS) SNR with this link adjusts the modulation coding mode of link adaptively.Destination node can be by digital folding, as (hybrid automatic repeat request HARQ), realizes the cascading judgement of Frame to mix automatic repeat requests.

Fig. 2 shows the flow chart of selecting the method 200 of via node according to the application's illustrative embodiments in the wireless cooperation junction network.As shown in Figure 2, at step S201, determine channel status and the energy state of described each via node respectively at any time slot according to the channel status model and the energy state model of a plurality of via nodes.At step S202, based on described channel status and energy state, by system's recompense of predetermined system's each described via node of recompense Model Calculation.The energy that this system's recompense also can be further need consume based on the via node forwarding information and calculating. at step S203, be system's recompense maximization that each time slot is selected a via node, selected via node to make to obtain in the whole data transmission procedure.

Below above each step is described in detail.

Suppose all relayings support K kind modulation coding modes, the spectrum efficiency of various modulation system correspondences is η ₀, η ₁..., η _K-1, corresponding minimum decoding SNR is

The data transmission period of source node and destination node is divided into T time slot, each time slot

By Frame of via node transmission of selecting.N available relaying formed candidate relay set Each via node

At time slot

Candidate state use

Expression, promptly selected if via node n is activated (active) at time slot t, a then _n(t)=1, if be not activated (passive) then a _n(t)=0.

Step S201

In the application's a execution mode, the channel status model can be a single order finite state markov channel model.The energy state model can be a Markov-chain model.By current channel status and predetermined channel status transfer matrix, can determine the channel status of each via node at any time slot.By current energy state and predetermined energy state transfer matrix, can determine the energy state of each via node at any time slot.Above-mentioned channel status transfer matrix and energy state transfer matrix can obtain by history observation and training.Above-mentioned channel status can comprise the channel status of source node to the channel status of via node and via node to destination node.

As indicated above, at first sub-slots of collaborative trunking traffic, it is MC that source node adopts MCS _S2DSend Frame to destination node, destination node and via node are intercepted this Frame simultaneously.Via node can be estimated the channel status of current time S2R link by intercepting the RTS signal that source node sends.

The mistake of S2R link decoding will be propagated to destination node, so the error rate of S2R link is the key factor of relay selection.The state of S2R channel adopts single order finite state markov channel model to describe.Given modulation coding mode (MC _S2D), the error rate of each time slot of S2R link is by the channel status decision of this time slot.SNR (the γ of S2R link _S2R) be divided into L grade, a state of the corresponding Markov chain of each grade.The state of next time slot channel changes according to current state of living in and markov state transition probability.Use C={C ₀, C ₁..., C _L-1The limited state space of expression, use

The S2R link of expression via node n At t constantly from state g _nTransfer to state h _nProbability, then the channel status transition probability matrix of S2R link is expressed as:

${\mathrm{\Φ}}_n\left(t\right)={\left[{\mathrm{\φ}}_{g_n{h}_n}\right]}_{L\×L}---\left(1\right)$

Wherein, ${\mathrm{\φ}}_{g_n{h}_n}\left(t\right)=\mathrm{Pr}({\mathrm{\γ}}_{{S2R}_n}(t+1)=h_n|{\mathrm{\γ}}_{{S2R}_n}\left(t\right)=g_n),$

According to the current channel condition and the channel transfer matrix of S2R link, can determine the channel status of S2R link at any time slot.

At second sub-slots of collaborative trunking traffic, it is MC that selecteed via node adopts MCS _R2DTransmit Frame to destination node, destination node merges the Frame of receiving twice.Via node can be estimated the channel status of current time R2D link by intercepting the cts signal that destination node sends.

The MCS that the R2D link adopts has determined the spectrum efficiency of system, so the MCS of R2D link is the key factor of relay selection.The state of R2D channel adopts single order finite state markov channel model to describe.Given target error rate, the MCS (MC of each time slot of R2D link _R2D) determine that by the channel status of this time slot the minimum decoding SNR of K kind modulation system correspondence is

SNR (the γ of R2D link _R2D) be divided into K grade:

${\mathrm{\&gamma;}}_{R2D}=\left\{\begin{array}{cc}{\mathrm{<figure-callout\; id="0"\; label="D"\; filenames="H2009101788410E0000022.png,H2009101788410E0000031.png"\; state="\{\{state\}\}">D</figure-callout>}}_0,& \mathrm{if}{\mathrm{\&gamma;}}_0^{*}\&le;{\mathrm{\&gamma;}}_{R2D}<{\mathrm{\&gamma;}}_1^{*}\\ {\mathrm{<figure-callout\; id="1"\; label="D"\; filenames="H2009101788410E0000031.png,H2009101788410E0000032.png"\; state="\{\{state\}\}">D</figure-callout>}}_1,& \mathrm{if}{\mathrm{\&gamma;}}_1^{*}\&le;{\mathrm{\&gamma;}}_{R2D}<{\mathrm{\&gamma;}}_2^{*}\\ .& .\\ .& .\\ .& .\\ D_{K-1},& \mathrm{if}{\mathrm{\&gamma;}}_{K-1}^{*}\&le;{\mathrm{\&gamma;}}_{R2D}\end{array}\right.$

A state of the corresponding Markov chain of each grade.The state of next time slot channel changes according to current state of living in and markov state transition probability.With

Represent limited state space, use The R2D link of expression via node n

At t constantly from state u _nTransfer to state v _nProbability, then the channel status transition probability matrix of R2D link is expressed as:

${\mathrm{\&Psi;}}_n\left(t\right)={\left[{\mathrm{\&psi;}}_{u_n{v}_n}\left(t\right)\right]}_{K\&times;K}---\left(2\right)$

Wherein, ${\mathrm{\&psi;}}_{u_n{v}_n}\left(t\right)=\mathrm{Pr}({\mathrm{\&gamma;}}_{{R2D}_n}(t+1)=v_n|{\mathrm{\&gamma;}}_{{R2D}_n}\left(t\right)=u_n),$

According to the current channel condition and the channel transfer matrix of R2D link, can determine the channel status of R2D link at any time slot.In one embodiment, (relay-to-destination, R2D) MCS of link adjusts with the SNR of this link via node adaptively to destination node.

Because most wireless mobile apparatus are battery-powered, finite energy.For prolonging network lifetime, the electric weight of battery should be used efficiently.Should consider when selecting relaying to transmit the energy that packets need consumes, also will consider each via node balancing energy consumption.Via node can obtain the energy state of current time self by electric quantity monitoring.Owing to the electric weight of mobile device battery is understood because any business (multimedia application or wireless transmission) that operates on the equipment reduce, the energy state of next time slot can not accurately be predicted, therefore can regard the dump energy of via node n as a stochastic variable e _nBe the simplification problem, continuous energy value e _nBe divided into H grade, with ε={ ε ₀, ε ₁..., ε _H-1The limited state space of expression.With

Dump energy (the e of expression via node n _n) after t takes to move a constantly from state f _nTransfer to state y _nProbability, then the state transition probability matrix of dump energy is expressed as:

${\mathrm{\&Theta;}}_n^a\left(t\right)={\left[{\mathrm{\&theta;}}_{f_n{y}_n}^a\left(t\right)\right]}_{H\&times;H}---\left(3\right)$

Wherein, ${\mathrm{\&theta;}}_{f_n{y}_n}^a\left(t\right)=\mathrm{Pr}(e_n(t+1)=y_n|e_n\left(t\right)=f_n,a_n\left(t\right)=a),$ f _n，y _n∈ε，

According to the state-transition matrix of via node current dump energy and dump energy, can determine the dump energy of via node at any time slot.

The energy state of next time slot of via node changes according to current state of living in and markov state transition probability.In one embodiment, can think in the energy model that after each transfer of data, dump energy all can reduce fixed size.This model is a kind of special case of Markov model:

${\mathrm{\&theta;}}_{f_n{y}_n}^1\left(t\right)=\left\{\begin{array}{cc}1,& \mathrm{if}{y}_n\mathrm{is\; the\; lower\; energy\; state\; next\; to}{\mathrm{<figure-callout\; id="0"\; label="f\; n"\; filenames="H2009101788410E0000022.png,H2009101788410E0000031.png"\; state="\{\{state\}\}">f</figure-callout>}}_n\\ 0,& \mathrm{otherwise}.\end{array}\right.---\left(4\right)$

In the real system, above channel status transition probability matrix and energy state transfer matrix all can obtain by historical observation and training.

In one embodiment, via node n is at the state of the time slot t channel status by the S2R link

The channel status of R2D link

With energy state e _n(t) common decision, that is:

$i_n\left(t\right)=[{\mathrm{\&gamma;}}_{{S2R}_n}\left(t\right),{\mathrm{\&gamma;}}_{{R2D}_n}\left(t\right),e_n\left(t\right)]---\left(5\right)$

In the real network,

With e _n(t) be mutual independent random variables.Therefore, the state of via node will shift in the markov mode, and the limited state space of via node n is used

Expression has

State transition probability matrix is expressed as:

Wherein, With

By (1) formula, (2) formula and (3) formula define, G=L * K * H respectively.Matrix

Each element

The state of expression via node n at t constantly from i _nTransfer to j _n, and

Step S202

According to the channel status and the energy state that are associated with each via node, can calculate system's recompense of each via node by predetermined system's recompense model.In one embodiment, the optimization aim of optimum relay selection strategy is the error propagation that alleviates the S2R link, improves the spectrum efficiency of R2D link, reduces and transmits the energy that packets need consumes, the energy consumption of balanced each via node.

According to this optimization aim, can adopt Restless Multi-armed Bandit model as system's recompense (system reward) model, determine the system recompense of each node at each time slot:

$R_{i_n\left(t\right)}^{a_n\left(t\right)}=a_n\left(t\right)\&CenterDot;R({\mathrm{\&omega;}}_p\&CenterDot;P_b({\mathrm{MC}}_{S2D},{\mathrm{\&gamma;}}_{{S2R}_n}),{\mathrm{\&omega;}}_{\mathrm{\&eta;}}\&CenterDot;{\mathrm{\&eta;}}_k({\mathrm{MC}}_{R2D},{\mathrm{\&gamma;}}_{{R2D}_n}),{\mathrm{\&omega;}}_J\&CenterDot;J(P_t,l,r_k),{\mathrm{\&omega;}}_e\&CenterDot;e_n\left(t\right))---\left(7\right)$

Wherein, | ω _p|+| ω _η|+| ω _J|+| ω _E|=1, ω _pAnd ω _JBe negative value, the weights of the corresponding error rate and energy consumption, ω _ηAnd ω _eFor on the occasion of, the weights of corresponding spectrum efficiency and dump energy; P _bBe the error rate of S2R link, by modulation coding mode (MC _S2D) and channel status

Decision; η _kBe the spectrum efficiency of R2D link, by modulation coding mode (MC _R2D) determine, with the channel state

Adjust adaptively; J transmits the energy that packets need consumes, by transmitting power P _t, packet size l and data rate r _kDecision.Can system's recompense obtain to be decided by that via node n is at time slot t whether be activated (selection), i.e. a _n(t)=1 still be a _n(t)=0.

Step S203

According to concrete optimization aim, can select the via node of each time slot use by above definite system's recompense.In the execution mode that uses Restless Multi-armed Bandit model, optimum relay selection problem in the mobile time-varying network is modeled as the problem that Restless Multi-armed Bandit in the stochastic control theory asks optimal solution.For example, can relax (linear programming relaxation) and original duplicate key exploratory method (primal-dual index heuristic) by linear programming finds the solution.

When relaying node n is a in the candidate state of time slot t _n(t) time, obtain instantaneous recompense

Because via node state process over time is a random process, maximum instantaneous system recompense does not also mean that the obtainable total system recompense of whole data transmission procedure maximum.According to Restless Multi-armed Bandit model,, need to introduce desired value bounded and the convergence of time-based discount (time-discounted) factor-beta (0＜β＜1) to guarantee total recompense if T is that a very big value or convergence are infinite.If T is less, β=1 can be set.Via node in the candidate state of each time slot according to Markov policy Change,

The set of expression Markov policy.The target of optimum relay selection strategy is the desired value of the obtainable total recompense of the whole data transmission procedure of maximization, represents with following formula:

In each above-mentioned Markov policy, for each time slot, having a unique candidate state is 1 via node, that is to say have a via node to be activated for forwarding information at each time slot.By to the finding the solution of following formula, can determine the selected via node of each time slot.

Below, concrete solution procedure is introduced.Because Restless Multi-armed Bandit model is that (Markov decision chain MDC), therefore can be converted into linear programming (linear programming, LP) expression formula to a kind of Markovian decision chain with time-based discount characteristic.At first introduce performance measurement variable (performance measure):

$x_{i_n}^{a_n}\left(u\right)=E_u\left[\underset{t=0}{\overset{T-1}{\mathrm{\&Sigma;}}}\left(I_{i_n}^{a_n}\left(t\right){\mathrm{\&beta;}}^t\right)\right]---\left(9\right)$

Wherein,

$I_{i_n}^{a_n}\left(t\right)=\left\{\begin{array}{cc}1,& \mathrm{if}{\mathrm{action\; a}}_n\mathrm{is\; tacken\; at\; time\; t\; by\; relay\; n\; in\; state}{i}_n.\\ 0,& \mathrm{otherwise}.\end{array}\right.$

The performance measurement variable

Expression via node n is according to Markov policy u, when state is i _nThe time candidate state changed into a _nThe desired value of total discount time.Represent by performance vectors (performance vector) with X

At all Markov policies

The corresponding performance zones (performance region) in following expansion back:

Because Restless Multi-armed Bandit model is a kind of Markovian decision chain with time-based discount characteristic, what D.Berstimas and J.Nino-Mora delivered is entitled as " Restless bandits; linear programming relaxations; and a primal dual index heuristic " (referring to Operations Research, 2000, vol.48, no.1, pp.80-90) article has proved that performance zones X is polyhedron (polytope), and X is called restless bandit polytope

Thereby Restless Multi-armed Bandit problem can be converted into the LP expression formula:

For via node n, the definition polyhedron

Be restless bandit polytope

At variable

The accurate projection in space.And,

It also is the performance zones of the single order MDC of corresponding via node n.With

Expression via node n initial condition is i _nProbability,

Expression initial condition probability vector, then

Complete expression formula is:

Therefore, then the single order of LP (first-order) relaxes (relaxation) expression formula and is:

subject?to

(primal-dual index heuristic) finds the solution by original duplicate key exploratory method, and single order relaxes linear programming (LP ¹) dual (dual) expression formula be:

subject?to

λ≥0. (14)

With With

The expression single order relaxes (LP ¹) and optimum original (primal) and dual (dual) of dual (dual) separate (solution pair) in pairs.With

Corresponding optimum reduction cost (reduced cost) coefficient of expression:

Optimum reduction cost

With

Can be interpreted as variable respectively With

Unit of the every increase of value, single order relaxes linear programming (LP ¹) the changing down (rate of decrease) of desired value (objective-value), optimum reduction cost

With

It all is nonnegative value.

The input parameter that exploratory method (heuristic) is found the solution comprises via node current state vector (i ₁, i ₂..., i _N), optimum original (primal) separates

Corresponding optimum reduction cost

The output result who obtains is the candidate state (a of each via node ^*(i ₁), a ^*(i ₂) ..., a ^*(i _N)).Exploratory method is found the solution two stages formations by original (primal) and dual (dual).In original (primal) stage, original (primal) variable of be activated (active)

Strict for positive via node is exactly optimum relaying, if having a plurality of or the zero via node

Strictness then can't be determined optimum relaying in original (primal) stage for just, needs select optimum relaying by dual (dual) stage.In dual (dual) stage, if the neither one via node

Strict for just, according to right

Explanation: the optimum of be activated (active) reduction cost Big more, variable

LP of unit of every increase ¹The changing down of desired value big more.Exploratory method selects to have the via node of minimum active reduction cost as optimal solution.If a plurality of via nodes are arranged

Strict for just, according to right

Explanation: the optimum of be not activated (passive) reduction cost Big more, variable

LP of unit of every increase ¹The changing down of desired value big more.Exploratory method selects to have the via node of maximum passive reduction cost as optimal solution.

What D.Berstimas and J.Nino-Mora delivered is entitled as " Restless bandits; linear programming relaxations; and a primal dual index heuristic " (referring to Operations Research, 2000, vol.48, no.1, article pp.80-90) is converted into priority index exploratory method (priority-index heuristic) to original dual exploratory method.The full content of this article is incorporated this paper by reference into.In this article, be i with state _nVia node corresponding index value (index) be defined as:

${\mathrm{\&delta;}}_{i_n}={\stackrel{\&OverBar;}{\&Element;}}_{i_n}^1-{\stackrel{\&OverBar;}{\&Element;}}_{i_{\mathrm{<figure-callout\; id="0"\; label="n"\; filenames="H2009101788410E0000022.png,H2009101788410E0000031.png"\; state="\{\{state\}\}">n</figure-callout>}}}^0.---\left(16\right)$

The priority index judges that (priority-index rule) is exactly the via node that activates (selection) index value minimum.

Method according to the application, the all corresponding priority index value (priority-index) of each state of via node, the parameter that input is relevant: state transition probability matrix, recompense, the time-based discount factor, with the initial condition probability vector, just can calculate the priority index value of every kind of state.Therefore use the calculating of this method can be divided into off-line (off-line) and online (on-line) two stages.The off-line stage: before transfer of data begins, set up concordance list (index table), every kind of corresponding priority index value of state.The on-line stage: before each frame data transmission, each via node is observed current state, and inquiry current state corresponding index value is also informed source node.Source node is selected the relay station of the via node of index value minimum as this Frame.

Before transfer of data begins, each via node need be shared relevant parameter, these parameters can be added in existing broadcasting of network or the handshake message, and the communication of off-line and calculating only need be carried out once before actual data transfer, so expense can be very not big.The on-line stage only need be compared the index value of each via node, if there is via node to leave candidate collection, does not then send the index value of oneself.

Below, by emulation, with relay selection method of the present invention (Optimal Relay Selection Scheme) and system of selection at random of the prior art (Random Selection Scheme) with suppose that the result of the constant system of selection of channel status (Existing Channel Changeless Method) (that is, the channel status of current time is simply as the channel estimating of next time slot) compares.

The channel quality of S2R link is divided into bad (s0) and two states of good (s1), and corresponding target error rate is respectively P _b=10 ^-3And P _b=10 ^-5The suppose relay node is supported BPSK, QPSK and three kinds of modulation systems of 8PSK, and corresponding spectrum efficiency is respectively η ₁=1, η ₂=2 and η ₃=3bit/s/Hz.The channel quality of R2D link is divided into good-for-BPSK (d0), good-for-QPSK (d1) and three states of good-for-8PSK (d2).The channel status transition probability matrix of S2R link and R2D link is respectively:

$\mathrm{\&Phi;}=\left(\begin{array}{cc}0.7& 0.3\\ 0.3& 0.7\end{array}\right),$ $\mathrm{\&Psi;}=\left(\begin{array}{ccc}0.6& 0.3& 0.1\\ 0.2& 0.6& 0.2\\ 0.1& 0.3& 0.6\end{array}\right).$

The dump energy of via node is divided into zero electric weight (e0), low electric weight (e1) and high electric weight (e2), and the state transition probability matrix after not being activated and being activated is respectively:

${\mathrm{\&Theta;}}^0=\left(\begin{array}{ccc}1.00& 0.00& 0.00\\ 0.01& 0.99& 0.00\\ 0.00& 0.01& 0.99\end{array}\right),$ ${\mathrm{\&Theta;}}^1=\left(\begin{array}{ccc}1.00& 0.00& 0.00\\ 0.08& 0.92& 0.00\\ 0.00& 0.08& 0.92\end{array}\right).$

The state transition probability matrix of via node

With

Can obtain by (6) formula.System's payoff function is defined by (7) formula.For the recompense that makes each optimization aim has comparativity, get lg (P _b/ 10 ^-3) as the recompense of the error rate, when the channel status of S2R link obtains recompense 0 during for s0, when the channel status of S2R link obtains recompense-2 ω during for s1 _pAssumed transmit power P _tFixing with data packet length l size, then transmit energy J and data rate r that packets need consumes _kOr spectrum efficiency η _kBe inversely proportional to, when the channel status of R2D link obtains recompense ω during for d0 _η+ ω _J, when the channel status of R2D link obtains recompense 2 ω during for d1 _η+ 1/2 ω _J, when the channel status of R2D link obtains recompense 3 ω during for d2 _η+ 1/3 ω _JWhen the dump energy state obtains recompense 0 during for e0, when the dump energy state obtains recompense ω during for e1 _e, when the dump energy state obtains recompense 2 ω during for e2 _eWhen the relaying node is not activated (selection) or dump energy state when being e0, the whole system recompense is 0, so via node in the be not activated instantaneous recompense of (selection) of each state is: R ⁰=0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0}

The instantaneous recompense of via node when each state is activated (selection) is:

R ¹＝{0，ω _η+ω _J+ω _e，ω _η+ω _J+2ω _e，0，2ω _η+1/2ω _J+ω _e，

2ω _η+1/2ω _J+2ω _e，0，3ω _η+1/3ω _J+ω _e，3ω _η+1/3ω _J+2ω _e，

0，-2ω _p+ω _η+ω _J+ω _e，-2ω _p+ω _η+ω _J+2ω _e，

0，-2ω _p+2ω _η+1/2ω _J+ω _e，-2ω _p+2ω _η+1/2ω _J+2ω _e，

0，-2ω _p+3ω _η+1/3ω _J+ω _e，-2ω _p+3ω _η+1/3ω _J+2ω _e}

Weights ω _p, ω _η, ω _JAnd ω _eCan adjust according to network or user preference.In emulation, establish ω _p=-0.3, ω _η=0.4, ω _J=-0.1, ω _e=0.2, discount factor β=0.8.Each emulation continues 2000 seconds, at the emulation initial time 20 candidate relay is arranged.

At first, compare the three kinds of obtainable system of relay selection method recompenses, Fig. 3 shows system's recompense contrast schematic diagram.Wherein the curve of the top is system's recompense that this method obtains, and middle curve is system's recompense that existing system of selection obtains, and the curve of below is system's recompense that system of selection at random obtains.Can see that this method is better than existing system of selection.Along with the time increases, each curve is all descending, and this is that candidate relay is fewer and feweri because of increasing via node depleted of energy, has not had candidate relay node, the system recompense of making to drop near zero in the time of 2000 seconds substantially.Change the size of each weights, can obtain similar result.

Fig. 4 shows the average error rate that three kinds of relay selection method obtain, wherein the curve of below is the average error rate that this method obtains, middle curve is the average error rate that existing system of selection obtains, and the curve of the top is the average error rate that system of selection at random obtains.As can be seen from the figure, this method can be selected optimum relaying exactly for next Frame, makes average error rate minimum; Because whether the time-varying characteristics of wireless channel, existing relay selection method can not guarantee selected relaying optimum at next time slot; The system of selection performance is the poorest at random.Can see that before 800 seconds, candidate relay is more, this method can access 10 ^-5Average error rate, existing relay selection method obtains 10 ^-4Average error rate.The average error rate of system of selection acquisition initially just begins to increase from emulation at random, and adopt the system of selection of energy efficient after 800 seconds, just to increase, this is because system of selection does not consider to transmit energy consumption and balancing energy at random, makes some via node depleted of energy soon.

Fig. 5 shows the spectrum efficiency that three kinds of relay selection method obtain, wherein the curve of the top is the spectrum efficiency that this method obtains, middle curve is the spectrum efficiency that existing system of selection obtains, and the curve of below is the spectrum efficiency that system of selection at random obtains.As can be seen from the figure, this method can be selected optimum relaying exactly for next Frame, obtains the highest spectrum efficiency; Existing relay selection method is with the best relay of the current time transfer of data as next time slot, because the time-varying characteristics of wireless channel can not guarantee whether selected relaying is optimum at next time slot; The system of selection performance is the poorest at random, and spectrum efficiency initially just begins to reduce from emulation, and adopts the system of selection of energy efficient just to reduce after 800 seconds.Along with increasing via node depleted of energy, candidate relay is fewer and feweri, has not had the candidate relay node in the time of 2000 seconds substantially, makes spectrum efficiency drop near zero.

Network towards different application has different definition to network lifetime.A definition commonly used is: when the interstitial content of depleted of energy reaches threshold value N _ThMake network can not reach target capabilities.When network lifetime is most important to network, ω can be set _e=0.5, ω _J=-0.5, other weights are set to 0.At this moment, the energy of via node dump energy and the consumption of forwarding packets need is exactly system's recompense.The via node number of depleted of energy is along with the time increases more and more.Fig. 6 shows three kinds of relay selection method at different threshold value N _ThThe network lifetime datagram of following acquisition.Consistent with expection, the network lifetime of three kinds of relay selection method acquisitions is all with threshold value N _ThIncrease and increase, this method always obtains the longest network lifetime.As we can see from the figure, the via node of first depleted of energy appears in system of selection about 250 seconds at random, the via node of first depleted of energy appears in existing system of selection about 330 seconds, the via node of first depleted of energy appears in this method about 450 seconds.

More than the application's method is described in detail, the explanation of above execution mode just is used to help to understand the application's core concept, and not as the qualification to the application.Those skilled in the art can carry out suitable modification and equivalent variations to above-mentioned embodiment in the application's spirit and scope.

Claims (10)

1. select the via node method for one kind in the wireless cooperation junction network, described junction network comprises source node, destination node and a plurality of via node, and described method comprises:

Channel status model and energy state model according to described a plurality of via nodes are determined channel status and the energy state of described each via node at any time slot respectively;

Based on described channel status and energy state, by system's recompense of predetermined system's each described via node of recompense Model Calculation; And

At each time slot is system's recompense maximization that transfer of data between described source node and the destination node is selected a via node, selected via node to make to obtain in the whole described data transmission procedure.

2. the method for claim 1, wherein described system recompense model is a Restless Multi-armed Bandit model.

3. the method for claim 1, wherein described channel status model is a single order finite state markov channel model.

4. the method for claim 1, wherein described energy state model is a Markov-chain model.

5. the method for claim 1, wherein the channel status of described each via node is determined by current channel status and predetermined channel status transfer matrix.

6. the method for claim 1, wherein the energy state of described each via node is determined by current energy state and predetermined energy state transfer matrix.

7. the method for claim 1, wherein described channel status comprises that described source node arrives the channel status of described each via node and the channel status that described each via node arrives described destination node.

8. the method for claim 1, wherein selected via node is adjusted adaptively to described destination node according to channel status and is transmitted the employed modulating-coding pattern of described information.

9. the energy that the method for claim 1, wherein described system recompense also need consume based on described each via node forwarding information and calculating.

10. as claim 5 or 6 described methods, wherein, described channel status transfer matrix and described energy state transfer matrix obtain by history observation and training.

Images