CN102036255A - Packet sending method based on prediction in communication channel - Google Patents

Abstract

The invention provides a packet sending method capable of predicting the arrival time of user business packet accurately so as to effectively resolve the packet collision in a communication channel. The packet sending method is as follows: in order to resolve the collision better in the packet sending process, establishing a mathematical model described by time for an arrival time sequence of the user business packet by user business modeling, facilitating the prediction on the arrival time sequence of the user business packet in future, and when a plurality of packets need to be sent and generate collision, and the packets can be decomposed according to the predicted result. The key points in the invention are as follows: an assignment interval for packet transmission in the current time slot is determined by predicting the arrival time of next user business packet, and decides the packet which is sent in the current time slot and arrives in the time duration of the assignment interval. In the invention, the adopted fractal autoregressive integrated moving average (FARIMA) model can be well matched with the current wireless network environment, and by combination with the whole packet transmission method, the packet time delay can be effectively reduced and the network throughput rate can be improved.

Description

In the communication channel based on the prediction packet transmission method

Technical field

The invention belongs to the field of using radio network technique,, particularly relate to the multiple access technique of wireless sharing channel as wireless sensor network, wireless self-organization network, WLAN (wireless local area network), satellite network etc.

Background technology

Wireless channel is a shared transmission medium.Therefore, when a plurality of users communicate by letter simultaneously, must cause collision on channel, thereby reduce the percent of pass of system.How to avoid data on broadcast channel collision and the utilance that improves link be the core problem that multiple access inserts.At present, multiple access inserts algorithm and mainly contains: fixed allocation, distribution according to need and 3 types of contentions at random.Because the distributed nature of network and provisional, and professional sudden, make the multiple access access protocol of fixed allocation, demand assigned multiple access access protocol all can not effectively play a role.The contention multiple access inserts algorithm effective busy channel resource under certain condition at random, and the technology of the time delay that reduces to transfer is widely used.There is a common feature in the system that adopts contention multiple access at random to insert algorithm, and promptly the busy channel resource at random of the user in the system sends information block, and when the information block conflict that sends, decomposition then need conflict.Advantages such as this system has the centralized control of not needing, and the terminal station increase and decrease is easy, and easy and simple to handle and transfer delay is little.Under light load condition, the probability that information block clashes in the system is less, but each station root a tree name need be effectively utilized channel.But along with increasing the weight of of load, conflict can be on the increase, because the information block of conflict needs to retransmit, this can make, and time delay increases, throughput descends, therefore adopt reasonably conflict decomposition algorithm, just become to improve the key issue of contention multiple access systematic function at random.

The basic thought that conflict is decomposed is: if there is grouping to bump in the system, then allows newly arrived customer service be grouped in outside the system and wait for, after all successful end of transmission of grouping that participates in collision, allow new transmitted in packets again.At present, classical conflict decomposition algorithm has tree-like conflict decomposition algorithm and FCFS (First-Come First-Serve Splitting Algorithm, First Come First Served conflict decomposition algorithm) conflict decomposition algorithm.

Existing conflict decomposition method has:

1, tree-like conflict is decomposed: mainly based on the binary tree decomposition algorithm, the grouping of participating in collision in the binary tree decomposition algorithm enters left side collection or right collection with the selection at random of a certain probability, and the right height that collects of the priority ratio that guarantees left side collection, promptly has only after left side collection decomposes successfully the decomposition that just can enter right collection.Referring to document J I Capetanakis.Tree Algorithms for PacketBroadcast Channels[J], IEEE Transactions on Information Theory, September1979, Vol.25, No.5, the tree-like conflict of Page (s) .505-515. is decomposed when treating the conflict message, fully based at random mode, when the packet count that clashes was many, the slot efficiency of system can reduce.

2, the FCFS conflict is decomposed: the grouping that bumps is frozen in the collision window, and the collision grouping is successively decomposed by generation time then, guarantees the grouping successfully transmission earlier that arrives earlier.Referring to document D Bertsekas, R Gallager.DataNetworks[M], 2nd Edition, Prentice-Hall, USA, 1992, Page (s): 229-238.FCFS conflict decomposition algorithm is according to dispatching the time of advent of message, can reduce idle time slot, but in colliding grouping, exist when producing closer at interval grouping, with time condition decomposition needs are repeatedly decomposed, particularly when customer service has self-similarity, professional time bursts is very strong, and the efficient of FCFS will significantly descend.

3, HSA (Hybrid Splitting Algorithm, mixed type conflict decomposition algorithm): the advantage of having inherited tree-like decomposition algorithm and FCFS conflict decomposition algorithm, not only considered the generation time of collision grouping, make the grouping that produces earlier obtain service earlier, simultaneously when existence produces more approaching at interval grouping in the system, adopt tree-like decomposition algorithm, make whole decomposable process no longer only be confined to the generation time that divides into groups, decompose required total timeslot number thereby reduced.Referring to document Min Sheng, Jiandong Liand Fan Jiang.Hybrid splitting algorithm for wireless MAC, IEEE Communications Letters, vol.9, Issue 5, May 2005Page (s): the key of 468-470.HSA algorithm is to calculate the optimum time point that switches to FCFS conflict decomposition algorithm from tree-like decomposition algorithm, obtaining of this optimum time point need supposition customer service model be Poisson process, therefore when real customer service model does not satisfy this hypothesis, can not obtain optimum time point, the performance of algorithm will be consistent with tree-like decomposition algorithm.

There is following shortcoming in above-mentioned conflict decomposition algorithm: the simple mode of decomposing at random that relies on is decomposed in tree-like conflict, and when a large amount of packet collisions took place, algorithm performance significantly descended; FCFS too relies on time conditions to decompose the collision message, and when the time bursts of customer service was very strong, efficiency of algorithm descended; The HSA limitation is too strong, and needing supposition customer service model is Poisson process, and when the real user business did not satisfy this hypothesis, its range of application was limited.

Summary of the invention

Technical problem to be solved by this invention is, provides the accurate predictive user traffic packets of a kind of energy to reach the time, thus the packet transmission method that the packet collisions in the communication channel is effectively decomposed.

The present invention solves the problems of the technologies described above the technical scheme that is adopted to be, based on the packet transmission method of prediction, may further comprise the steps in the communication channel:

A based on the prediction the grouping forwarding step:

The A1 system has sent the customer service grouping that finishes and reach before in time T (k);

The A2 system sets up FARIMA (the autoregression mark is integrated moving average) model by the time series that reaches that time T (k) has been sent before the customer service grouping that finishes;

A3 is by the time of advent of the next customer service grouping of described FARIMA model prediction

The A4 system is transmitted in the customer service grouping that arrives in [T (k), T (k)+a (k)] in current time slots k, wherein the assignment interval is

As in the current time slots k packet collisions taking place, then enter conflict and decompose the grouping forwarding step; Otherwise, return steps A 1;

The grouping forwarding step is decomposed in the B conflict:

Packet collisions, execution in step B2 as taking place in the grouping transmit status in the B1 systems inspection time slot k; As send success or transmission is grouped into sky, and and indicating device is left side collection, execution in step B3 be a left side collection as sky and indicating device, and execution in step B3, as send success and indicating device collects execution in step B4 for right;

B2 is provided with k=k+1, in current time slots k, continue to send the first half that successfully do not send grouping last time, T (k)=T (k-1) is set, α (k)=α (k-1)/2, and indicating device is set is left side collection, system is transmitted in customer service grouping that arrives in [T (k), T (k)+a (k)] in current time slots k, afterwards, return step B1;

B3 is provided with k=k+1, in current time slots k, continue to send successfully sent last time grouping back half, T (k)=T (k-1)+α (k-1) is set, α (k)=α (k-1), and indicating device is set is right collection, system is transmitted in customer service grouping that arrives in [T (k), T (k)+a (k)] in current time slots k, afterwards, return step B1;

B4 is provided with k=k+1, and indicating device is set is right collection, and returns steps A 1.

The present invention is for the decomposition that conflicts better in the grouping process of transmitting, by the customer service modeling time series that reaches of customer service grouping is set up a Mathematical Modeling of describing with the time, predict the convenient time of advent that the customer service in future is divided into groups, when a plurality of groupings need send and clash, can be according to the quick decomposition grouping that predicts the outcome.

Crucial part of the present invention is by predicting the time of advent of next customer service grouping

Determine the assignment interval that grouping sends in the current time slots

The assignment interval has determined to be sent in the current time slots grouping that reaches in a (k) time span.The interval a of this assignment (k) is the time of advent with next customer service grouping

And the value that changes, in order to guarantee

Predicted accurate, the present invention adopts the Mathematical Modeling that is operated under self similarity and the long correlation situation---FARIMA (fractional autoregressive integrated moving average, mark autoregression summation moving average) model.Traditional conflict is decomposed, all be that operating mechanism and the choosing of parameter to algorithm considered in MAC layer (MAC sublayer), do not consider the different requirements of customer service, user's business model is all assumed Poisson to be arrived, customer service in this and the current wireless network present self-similarity and when long correlation be inconsistent, good inadequately to the practical business match, can cause the practicality of conflict decomposition algorithm to descend.Unlike the prior art, the FARIMA model that the present invention adopts can mate current wireless network environment well, utilize FARIMA model and active user's traffic packets to carry out match, again in conjunction with whole group transfer approach of the present invention, can effectively reduce packet delay, improve network throughput.

The present invention adopt FARIMA (q) mathematic(al) representation of model is for p, d:

$\mathrm{\Φ}\left(B\right){\▿}^d{X}_t=\mathrm{\Θ}\left(B\right)a_t$

Wherein, X _tFor the customer service grouping arrives time series, t represents the moment that traffic packets arrives, and d is a difference order, satisfies d ∈ (0.5,0.5), and p is the autoregression exponent number, and q is the moving average exponent number, and p, q are nonnegative integers, a _tBe that a zero-mean and variance are σ ²Wiener-Hopf equation, and have in addition:

Φ(B)＝1-φ ₁B-φ ₂B ²...-φ _pB ^p

Θ(B)＝1-θ ₁B-θ ₂B ²-...-θ _qB ^q

Wherein, Φ (B) and Θ (B) are that the complex variable multinomial does not have public solution, and Φ (B) { does not have in B:|B|≤1} and separates at unit circle in addition.B is that the back is to mobile operator, i.e. BX _t=X _T-1Definition

Be difference operator,

Be the mark difference operator, its binomial expansion is:

${\▿}^d={(1-B)}^d=\underset{n=0}{\overset{\∞}{\mathrm{\Σ}}}\left|\begin{array}{c}d\\ n\end{array}\right|{(-B)}^n$

Wherein, $\left|\begin{array}{c}d\\ n\end{array}\right|=\mathrm{\Γ}(d+1)/\left[\mathrm{\Γ}(n+1)\mathrm{\Γ}(d-n+1)\right]$

Wherein, n is a temporary variable, and Γ represents gamma function, is defined as:

$\mathrm{\Γ}\left(x\right)={\∫}_0^{\∞}{e}^{-t}{t}^{x-1}\mathrm{dt}=(x-1)\mathrm{\Γ}(x-1),x>0,$ So can obtain:

${\▿}^d={(1-B)}^d={\mathrm{\Σ}}_{n=0}^{\∞}g\left(n\right)B^n,$ Wherein g (n) is defined as:

g(0)≡1，g(1)＝-d，g(n)＝g(n-1)*(n-1-d)/n

When 0＜d＜0.5, (q) model presents correlation when long to FARIMA, and it describes the Parameter H urst Parameter H=0.5+d of autocorrelation for p, d; When d=0, FARIMA (p, d, q) be degenerated to ARMA (p, q).

Because correlation is difficult to calculate parameter p wherein, q, θ when long ₁, θ ₂..., θ _q, φ ₁, φ ₂..., φ _p, σ ²Therefore, need (p, q) model (autoregressive and moving average, autoregressive moving average) obtains parameter p, q, θ by calculating the ARMA that presents short-term correlation earlier ₁, θ ₂..., θ _q, φ ₁, φ ₂..., φ _p, σ ²Value.

Concrete, the concrete grammar that system sets up autoregression mark integration moving average model by the time series that reaches that time T (k) has been sent before the customer service grouping that finishes in the steps A 2 is:

A2-1 has sent the customer service grouping X that finishes _tRemove equal Value Operations, promptly carry out X _t-μ obtains the business datum sequence X of a zero-mean this moment _t-μ, wherein μ=E[X _t] be the expectation of traffic sequence;

A2-2 adopts Hurst (Hirst) Parameter H of Rescaled Adjusted Range Statistics (heavily marking range method) algorithm estimated sequence, obtains parameter d=H-0.5;

A2-3 obtains an autoregressive moving-average model ARMA (p, q) sequence W _t,

A2-4 utilizes AIC (Akaike Information Criterion, red pond amount of information) criterion to sequence W _tDecide rank, obtain p, the value of q;

A2-5 utilizes approximate maximal possibility estimation to obtain sequence W _tAll parameter θ ₁, θ ₂..., θ _q, φ ₁, φ ₂..., φ _p, σ ²

A2-6 is d, p, q, θ ₁, θ ₂..., θ _q, φ ₁, φ ₂..., φ _p, σ ²(q) mathematic(al) representation of model promptly obtains FARIMA (p, d, q) model of customer service for p, d to bring FARIMA into.

Concrete, pass through FARIMA (p, d, q) time of advent of the next customer service grouping of model prediction in the steps A 3

Method be:

Utilization obtains

Can obtain

${\hat{X}}_t\left(1\right)=-{\mathrm{\Σ}}_{m=1}^{\∞}g\left(k\right){\hat{X}}_t(1-m)+{\mathrm{\Σ}}_{i=1}^p{\mathrm{\φ}}_i{\hat{W}}_{t+1-i}+{\mathrm{\Σ}}_{j=1}^q{\mathrm{\θ}}_j{a}_{t+1-j}$

Wherein, m, i, j are temporary variable;

${\hat{W}}_t\left(1\right)=E\left(W_{t+1}\right)$

$=E({\mathrm{\φ}}_1{W}_t+{\mathrm{\φ}}_2{W}_{t-1}+L+{\mathrm{\φ}}_p{W}_{t-p+1}+a_{t+1}-{\mathrm{\θ}}_1{a}_t-{\mathrm{\θ}}_2{a}_{t-1}-L-{\mathrm{\θ}}_q{a}_{t-q+1})$

$={\mathrm{\φ}}_1{W}_t+{\mathrm{\φ}}_2{W}_{t-1}+L+{\mathrm{\φ}}_p{W}_{t-p+1}-{\mathrm{\θ}}_1{a}_t-{\mathrm{\θ}}_2{a}_{t-1}-L-{\mathrm{\θ}}_q{a}_{t-q+1}$

The invention has the beneficial effects as follows, the present invention makes prediction by the actual user's business to future, and instruct the conflict in later stage to decompose with predicting the outcome, it is interval dynamically to adjust distribution according to the conflict situations of transmitting Packet Service, make the conflict decomposition efficiency improve, and adopt the FARIMA model of self-similarity to carry out match to practical business, match appropriateness degree is higher than employing Poisson model simulation practical business, makes the accuracy that predicts the outcome improve.

Description of drawings

The interval schematic diagram of assignment when Fig. 1 is k=4000 for current transmission time slot among the embodiment;

The interval schematic diagram of assignment when Fig. 2 is k=4001 for current transmission time slot among the embodiment;

The interval schematic diagram of assignment when Fig. 3 is k=4002 for current transmission time slot among the embodiment;

The interval schematic diagram of assignment when Fig. 4 is k=4003 for current transmission time slot among the embodiment;

The interval schematic diagram of assignment when Fig. 5 is k=4004 for current transmission time slot among the embodiment;

Fig. 6 adopts the described method of present embodiment and adopts FCFS to send the throughput simulation result comparison diagram of grouping;

Fig. 7 is for adopting the average delay simulation result comparison diagram of described method of present embodiment and FCFS;

Fig. 8 decomposes cycle simulation result comparison diagram for the average conflict of adopting described method of present embodiment and FCFS.

Embodiment

Provide the implementation method of this patent in the concrete wireless self-networking below:

Each node all can be to business model in the wireless self-networking system of single-hop, business model adopts one section anonymity grouping track of collecting from SIGCOMM ' 04 (coming from http://www.crawdad.org/data.php) meeting, traffic packets is sent according to this section track, earlier system is moved a period of time, make things convenient for node past professional first modeling during this period of time.After the modeling success, system begins normal operation and finishes up to whole transmission of this section grouping track, and process of transmitting is as follows:

The A1 system has sent the customer service grouping that finishes and reach before in time T (k);

The A2 system sets up FARIMA (the autoregression mark is integrated moving average) model by the time series that reaches that time T (k) has been sent before the customer service grouping that finishes;

A3 is by the time of advent of the next customer service grouping of described FARIMA model prediction

The A4 system is transmitted in the customer service grouping that arrives in [T (k), T (k)+a (k)] in current time slots k, the assignment interval is

As in the current time slots k packet collisions taking place, then enter conflict and decompose the grouping forwarding step; Otherwise, return steps A 1;

The grouping forwarding step is decomposed in conflict:

Packet collisions, execution in step B2 as taking place in the grouping transmit status in the B1 systems inspection time slot k; As send success or transmission is grouped into sky, and and indicating device is left side collection, execution in step B3 be a left side collection as sky and indicating device, and execution in step B3, as send success and indicating device collects execution in step B4 for right;

B2 is provided with k=k+1, in current time slots k, continue to send the first half that successfully do not send grouping last time, T (k)=T (k-1) is set, α (k)=α (k-1)/2, and indicating device is set is left side collection, system is transmitted in customer service grouping that arrives in [T (k), T (k)+a (k)] in current time slots k, afterwards, return step B1;

B3 is provided with k=k+1, in current time slots k, continue to send successfully sent last time grouping back half, T (k)=T (k-1)+α (k-1) is set, α (k)=α (k-1), and indicating device is set is right collection, system is transmitted in customer service grouping that arrives in [T (k), T (k)+a (k)] in current time slots k, afterwards, return step B1;

B4 is provided with k=k+1, and indicating device is set is right collection, and returns steps A 1.

As Fig. 1, system has sent the customer service grouping that finishes and reach at time 2000s (T (k)=2000), by the 2000.5s time of advent of the next customer service grouping of the FARIMA model prediction of setting up

The assignment interval is

So, system is transmitted in the customer service grouping that arrives in [2000,2000.5] s in current time slots k (k=4000), and wherein the assignment interval is 0.5s, as packet collisions takes place in the current time slots k (k=4000), then enters conflict and decomposes the grouping forwarding step:

Packet collisions has taken place in systems inspection in time slot k (k=4000), and indicating device is a left side collection, k=k+1 then is set, in current time slots k (k=4001), continues to send the first half that successfully do not send grouping last time, T (k)=T (k-1) is set, be T (k)=2000, α (k)=0.5/2, and indicating device is set is left side collection, system is transmitted in [2000 in current time slots k, 2000.25] the interior customer service grouping that arrives of s, as shown in Figure 2;

Afterwards, packet collisions has taken place in systems inspection in time slot k (k=4001), k=k+1 then is set, in current time slots k (k=4002), continues to send the first half that successfully do not send grouping last time, T (k)=T (k-1) is set, be T (k)=2000, α (k)=0.25/2, and indicating device is set is left side collection, system is transmitted in [2000 in current time slots k, 2000.125] the interior customer service grouping that arrives of s, as shown in Figure 3;

Afterwards, not packet collisions has taken place in systems inspection in time slot k (k=4002), [2000,2000.125] the customer service grouping that arrives in the s is successfully sent finishes, k=k+1 then is set, in current time slots k (k=4003), continue to send successfully sent last time grouping back half, T (k)=T (k-1)+α (k-1) is set, i.e. T (k)=2000.125, α (k)=α (k-1), promptly, α (k)=0.25/2, and indicating device is set is right collection, system is transmitted in [2000.125 in current time slots k, 2000.25] the interior customer service grouping that arrives of s, as shown in Figure 4;

Systems inspection is transmitted in [2000.125 in time slot k (k=4003), 2000.25] the customer service grouping that arrives in the s is successfully sent finishes, and indicating device is right collection, then continue to be provided with indicating device and be right collection, and set up the 2000.5s time of advent of the next customer service grouping of FARIMA model prediction by the customer service grouping that in time 2000.25s, reaches

So, system is transmitted in the customer service grouping that arrives in [2000.25,2000.5] s in current time slots k (k=4004), and wherein the assignment interval is 0.25s,, as shown in Figure 5.As packet collisions takes place in the current time slots k (k=4004), then enter conflict again and decompose the grouping forwarding step, then continue by determining new assignment interval the time of advent of predicting next customer service grouping as successfully sending.

The concrete grammar that system sets up the FARIMA model by the time series that reaches that time T (k) has been sent before the customer service grouping that finishes in the steps A 2 is:

A2-1 has sent the customer service grouping X that finishes _tRemove equal Value Operations, promptly carry out X _t-μ obtains the business datum sequence X of a zero-mean this moment _t-μ, wherein μ=E[X _t] be the expectation of traffic sequence;

A2-2 adopts Hurst (Hirst) Parameter H of Rescaled Adjusted Range Statistics algorithm estimated sequence, obtains parameter d=H-0.5;

A2-3 obtains an autoregressive moving-average model ARMA (p, q) sequence W _t,

A2-4 utilizes AIC (Akaike Information Criterion) criterion to sequence W _tDecide rank, obtain p, the value of q;

A2-5 utilizes approximate maximal possibility estimation to obtain sequence W _tAll parameter θ ₁, θ ₂..., θ _q, φ ₁, φ ₂..., φ _p, σ ²

A2-6 is d, p, q, θ ₁, θ ₂..., θ _q, φ ₁, φ ₂..., φ _p, σ ²(q) mathematic(al) representation of model promptly obtains FARIMA (p, d, q) model of customer service for p, d to bring FARIMA into.

Concrete, pass through FARIMA (p, d, q) time of advent of the next customer service grouping of model prediction in the steps A 3

Method be:

Utilization obtains

Can obtain

${\hat{X}}_t\left(1\right)=-{\mathrm{\Σ}}_{m=1}^{\∞}g\left(m\right){\hat{X}}_t(1-m)+{\mathrm{\Σ}}_{i=1}^p{\mathrm{\φ}}_i{\hat{W}}_{t+1-i}+{\mathrm{\Σ}}_{j=1}^q{\mathrm{\θ}}_j{a}_{t+1-j}$

Wherein, m, i, j are temporary variable;

${\hat{W}}_t\left(1\right)=E\left(W_{t+1}\right)$

$=E({\mathrm{\&phi;}}_1{W}_t+{\mathrm{\&phi;}}_2{W}_{t-1}+L+{\mathrm{\&phi;}}_p{W}_{t-\mathrm{<figure-callout\; id="1"\; label="p\; +"\; filenames="HDA0000035850390000013.png,HDA0000035850390000021.png"\; state="\{\{state\}\}">p</figure-callout>}+1}+a_{t+1}-{\mathrm{\&theta;}}_1{a}_t-{\mathrm{\&theta;}}_2{a}_{t-1}-L-{\mathrm{\&theta;}}_q{a}_{t-q+1})$

$={\mathrm{\&phi;}}_1{W}_t+{\mathrm{\&phi;}}_2{W}_{t-1}+L+{\mathrm{\&phi;}}_p{W}_{t-p+1}-{\mathrm{\&theta;}}_1{a}_t-{\mathrm{\&theta;}}_2{a}_{t-1}-L-{\mathrm{\&theta;}}_q{a}_{t-\mathrm{<figure-callout\; id="1"\; label="q\; +"\; filenames="HDA0000035850390000013.png,HDA0000035850390000021.png"\; state="\{\{state\}\}">q</figure-callout>}+1}$

(as the m in the steps A 3) adopts bigger numerical to replace when relating to infinitely-great summation in the practical operation, and infinitely-great summation is changed into summation to bigger numerical.

From computer artificial result as can be seen, the conflict decomposition of employing present embodiment has under the Business Stream of self similarity and can efficient quick decomposition conflict divide into groups, system throughput, and average delay and average packet conflict decomposition cycle all obviously are better than the FCFS algorithm:

As shown in Figure 6, the throughput simulation result contrast of present embodiment and FCFS algorithm: Prediction_CRA refers to present embodiment, FCFS refers to the FCFS algorithm, Sig04_ver01, Sig04_ver02 and Sig04_ver03 represent three sections Business Streams of collecting from SIGCOMM ' 04 meeting, and every section all is to comprise 10000 messages.Abscissa is the arrival rate of message, and unit is message number/time slot, and ordinate is a throughput, and unit is message number/time slot.As can be seen from Figure 6, the maximum throughput rate of FCFS has only 0.37, and the maximum throughput rate of present embodiment is 0.4.

As shown in Figure 7, the average delay simulation result contrast of present embodiment and FCFS algorithm: Prediction_CRA refers to present embodiment, FCFS refers to FCFS, Sig04_ver01, Sig04_ver02 and Sig04_ver03 represent three sections Business Streams of collecting from SIGCOMM ' 04 meeting, and every section all is to comprise 10000 messages.Abscissa is the arrival rate of message, and unit is message number/time slot, and ordinate is an average delay, and unit is a time slot, and scalable manner is logarithmic coordinates.As can be seen from Figure 7, the average delay of present embodiment is littler by about 5% than FCFS algorithm.

As shown in Figure 8, cycle simulation result contrast is decomposed in the average conflict of present embodiment and FCFS algorithm: Prediction_CRA refers to present embodiment, FCFS refers to FCFS, Sig04_ver01, Sig04_ver02 and Sig04_ver03 represent three sections Business Streams of collecting from SIGCOMM ' 04 meeting, and every section all is to comprise 10000 messages.Abscissa is the arrival rate of message, and unit is message number/time slot, and ordinate is the decomposition cycle of on average conflicting, and unit is a number of times.Under the Business Stream background of the Sig04_ver03 of Fig. 8, the average conflict decomposition cycle of present embodiment has only about 70% of FCFS algorithm.

Claims (4)

1. In the communication channel based on the prediction packet transmission method, it is characterized in that, may further comprise the steps:

   A based on the prediction the grouping forwarding step:

   The A1 system has sent the customer service grouping that finishes and reach before in time T (k);

   The A2 system sets up the FARIMA model by the time series that reaches that time T (k) has been sent before the customer service grouping that finishes;

   A3 is by the time of advent of the next customer service grouping of described FARIMA model prediction

   

   The A4 system is transmitted in the customer service grouping that arrives in [T (k), T (k)+a (k)] in current time slots k, wherein the assignment interval is

   

   As in the current time slots k packet collisions taking place, then enter conflict and decompose the grouping forwarding step; Otherwise, return steps A 1;

   The grouping forwarding step is decomposed in the B conflict:

   Packet collisions, execution in step B2 as taking place in the grouping transmit status in the B1 systems inspection time slot k; As send success or transmission is grouped into sky, and and indicating device is left side collection, execution in step B3 be a left side collection as sky and indicating device, and execution in step B3, as send success and indicating device collects execution in step B4 for right;

   B2 is provided with k=k+1, in current time slots k, continue to send the first half that successfully do not send grouping last time, T (k)=T (k-1) is set, α (k)=α (k-1)/2, and indicating device is set is left side collection, system is transmitted in customer service grouping that arrives in [T (k), T (k)+a (k)] in current time slots k, afterwards, return step B1;

   B3 is provided with k=k+1, in current time slots k, continue to send successfully sent last time grouping back half, T (k)=T (k-1)+α (k-1) is set, α (k)=α (k-1), and indicating device is set is right collection, system is transmitted in customer service grouping that arrives in [T (k), T (k)+a (k)] in current time slots k, afterwards, return step B1;

   B4 is provided with k=k+1, and indicating device is set is right collection, and returns steps A 1.
2. According to claim 1 in the communication channel based on the packet transmission method of prediction, it is characterized in that the concrete grammar that system sets up autoregression mark integration moving average model by the time series that reaches that time T (k) has been sent before the customer service grouping that finishes in the steps A 2 is:

   A2-1 has sent the customer service grouping X that finishes _tRemove equal Value Operations, promptly carry out X _t-μ obtains the business datum sequence X of a zero-mean this moment _t-μ, wherein μ=E[X _t] be the expectation of traffic sequence;

   A2-2 adopts the Hirst Parameter H of heavily marking the range method estimated sequence, obtains parameter d=H-0.5;

   A2-3 obtains an autoregressive moving-average model ARMA (p, q) sequence W _t,

   A2-4 utilizes red pond amount of information criterion to sequence W _tDecide rank, obtain p, the value of q;

   A2-5 utilizes approximate maximal possibility estimation to obtain sequence W _tAll parameter θ ₁, θ ₂..., θ _q, φ ₁, φ ₂..., φ _p, σ ²

   A2-6 is d, p, q, θ ₁, θ ₂..., θ _q, φ ₁, φ ₂..., φ _p, σ ²(q) model promptly obtains FARIMA (p, d, q) model of customer service for p, d to bring FARIMA into.
3. As in the communication channel as described in the claim 2 based on the packet transmission method of prediction, it is characterized in that, described FARIMA (q) model is for p, d:

   $\mathrm{\&Phi;}\left(B\right){\&dtri;}^d{X}_t=\mathrm{\&Theta;}\left(B\right)a_t$

   Wherein, X _tFor the customer service grouping arrives time series, t represents the moment that traffic packets arrives, and d is a difference order, satisfies d ∈ (0.5,0.5), and p is the autoregression exponent number, and q is the moving average exponent number, and p, q are nonnegative integers, a _tBe that a zero-mean and variance are σ ²Wiener-Hopf equation, and:

   Φ(B)＝1-φ ₁B-φ ₂B ²...-φ _pB ^p

   Θ(B)＝1-θ ₁B-θ ₂B ²-...-θ _qB ^q

   Wherein, Φ (B) and Θ (B) are that the complex variable multinomial does not have public solution, and Φ (B) { does not have in B:|B|≤1} and separates at unit circle in addition; B is that the back is to mobile operator, i.e. BX _t=X _T-1Definition

   

   Be difference operator,

   

   Be the mark difference operator, its binomial expansion is:

   ${\&dtri;}^d={(1-B)}^d=\underset{n=0}{\overset{\&infin;}{\mathrm{\&Sigma;}}}\left|\begin{array}{c}d\\ n\end{array}\right|{(-B)}^n$

   Wherein, $\left|\begin{array}{c}d\\ n\end{array}\right|=\mathrm{\&Gamma;}(d+1)/\left[\mathrm{\&Gamma;}(n+1)\mathrm{\&Gamma;}(d-n+1)\right];$

   Wherein n is a temporary variable, and Γ represents gamma function, is defined as:

   $\mathrm{\&Gamma;}\left(x\right)={\&Integral;}_0^{\&infin;}{e}^{-t}{t}^{x-1}\mathrm{dt}=(x-1)\mathrm{\&Gamma;}(x-1),x>0,$ So:

   ${\&dtri;}^d={(1-B)}^d={\mathrm{\&Sigma;}}_{n=0}^{\&infin;}g\left(n\right)B^n,$ Wherein g (n) is defined as:

   g(0)≡1，g(1)＝-d，g(n)＝g(n-1)*(n-1-d)/n。
4. As in the communication channel as described in the claim 3 based on the packet transmission method of prediction, it is characterized in that, in the steps A 3 by FARIMA (p, d, q) time of advent of the next customer service grouping of model prediction

   

   Method be:

   Utilization obtains

   

   Can obtain

   

   ${\hat{X}}_t\left(1\right)=-{\mathrm{\&Sigma;}}_{m=1}^{\&infin;}g\left(k\right){\hat{X}}_t(1-m)+{\mathrm{\&Sigma;}}_{i=1}^p{\mathrm{\&phi;}}_i{\hat{W}}_{t+1-i}+{\mathrm{\&Sigma;}}_{j=1}^q{\mathrm{\&theta;}}_j{a}_{t+1-j}$

   Wherein, m, i, j are temporary variable;

   ${\hat{W}}_t\left(1\right)=E\left(W_{t+1}\right)$

   $=E({\mathrm{\&phi;}}_1{W}_t+{\mathrm{\&phi;}}_2{W}_{t-1}+L+{\mathrm{\&phi;}}_p{W}_{t-p+1}+a_{t+1}-{\mathrm{\&theta;}}_1{a}_t-{\mathrm{\&theta;}}_2{a}_{t-1}-L-{\mathrm{\&theta;}}_q{a}_{t-q+1})$

   $={\mathrm{\&phi;}}_1{W}_t+{\mathrm{\&phi;}}_2{W}_{t-1}+L+{\mathrm{\&phi;}}_p{W}_{t-p+1}-{\mathrm{\&theta;}}_1{a}_t-{\mathrm{\&theta;}}_2{a}_{t-1}-L-{\mathrm{\&theta;}}_q{a}_{t-q+1}$

Images