CN102572872B - Admission control method based on Kalman filter - Google Patents

Abstract

The invention relates to the technical field of admission control of service transmission in the mobile communication networking method and system and discloses an admission control method based on Kalman filter, wherein the admission control method based on the Kalman filter comprises the following steps of: S1, at the moment n when a new service flow applies for reaching an ingress node, measuring to obtain the statistical property of the admitted aggregated service flow, and predicting with the statistical property of the admitted aggregated service flow, the statistical property of the new service flow and hat(P)<loss>=[n-1|n-1] to obtain hat(P)<loss>=[n|n-1], if hat(P)<loss>=[n-1|n-1] < epsilon<QoS>, admitting the new service flow, otherwise, refusing to admit the new service flow; and S2, the ingress node receiving a measurement value hat(P)<loss>[n] of the packet loss rate at the moment n carried in a feedback signaling packet sent from an egress node, and correcting the measurement value hat(P)<loss>[n] to obtain a corrected value hat(P)<loss>[n|n] of the packet loss rate at the moment n. According to the admission control method, the errors introduced into the judgment of the admission control due to the inaccurate state information sample values can be reduced, the end-to-end state information of the network can be tracked and predicted with a Kalman filter, and the admission control precision is increased.

Description

Based on the acceptance controlling method of Kalman filtering

Technical field

The Admission control field that the present invention relates to business transmission in mobile communication network-building method system, is specifically related to a kind of acceptance controlling method based on Kalman filtering.

Background technology

As a kind of important measures that ensure network service quality, receiving the target of controlling is the value volume and range of product that enters the connection of network by control, effectively utilizes Internet resources, to ensure to meet the service quality of all connections that entered network.

Mobile network is exactly the network of node motion, because the movement of node has caused the uncertainty of network topology structure, thereby has improved burstiness that business arrives and the burstiness of data flow, makes the node state can not held stationary, affects the accuracy of acceptance judging.

Traditional receiving control is divided into two classes: receiving control (the Traffic-Descriptor based Admission Control based on flow descriptor, TDAC) method and receiving control (Measurement-Based Admission Control, the MBAC) method based on measuring.CAC (the Call Admission Control specifying at first in atm network, Call Admission Control, also control referred to as receiving) all based on traffic carrying capacity descriptor, based on the CAC of traffic carrying capacity descriptor, refer to be new business submits to network itself discharge characteristic by traffic carrying capacity descriptor.These traffic carrying capacity descriptors comprise peak value information source speed, continue information source speed, Maximum Burst Size and minimum information source speed.But due to the randomness of various traffic streams on network, the multiple discharge characteristic failing to be convened for lack of a quorum after gathering is difficult to describe, and based on measure CAC (MBCAC) overcome this shortcoming, it is without the discharge model of knowing business, by offered load is measured in real time, using this as receiving the foundation of controlling, greatly improve resource utilization, but still had poor expandability and the large problem of signaling consumption.Then a kind of statistics is received and is controlled (statistical CAC, SCAC) be suggested, under SCAC framework, taking service access node as decision point, using convergence service stream flow and end-to-end link packet loss as network state information, access node is measured the current flow that has accessed convergence service stream in addition, and business output node is measured packet loss and periodically fed back to access node.Reached at the bandwidth that input node computing network is current and new service flow are by the packet loss causing, using this as judgement foundation.In a word, SCAC algorithm has good autgmentability and less signaling consumption, but its accuracy aspect needs further to improve.

The degree that current admission control algorithm is accepted according to request can be divided into again two classes: static CAC and dynamically CAC.Static CAC refers to above the CAC within the scope of traditional network definition frame, as long as new QoS (service quality) requirement connecting can be met, and can not affect the service quality of existing connection in network, and new connection always can be accepted.This class CAC strategy is conducted extensive research, now existing a large amount of achievement.In recent years, occurred again a kind of based on optimized CAC strategy, i.e. dynamic CAC, or be called " intelligence is blocked " (Intelligence Block).First this strategy defines a reward function (Revenue function), and it receives the target of controlling is to meet under the prerequisite of QoS constraint, obtains the maximum of the long-term compensation of whole network.Under this strategy, some qos requirements can be met and the receiving request that can not affect existing connection also can be rejected because after can reserving so more resource and receiving to the larger connection of network remuneration contribution.

Summary of the invention

(1) technical problem that will solve

Technical problem to be solved by this invention is: the accuracy that how to improve acceptance judging.

(2) technical scheme

For solving the problems of the technologies described above, the invention provides a kind of acceptance controlling method based on Kalman filtering, comprise the following steps:

S1, in the time of moment n, new service flow application arrives ingress, obtains the statistical property of the convergence service stream of having received by measurement, utilizes statistical property, the statistical property of new service flow and the packet loss correction value in n-1 moment of the convergence service stream of having received predict the packet loss predicted value that obtains the n moment if receive this Business Stream, otherwise refusal, wherein ε _qoSrepresent default desired packet loss maximum;

S2, the packet loss rate measurement value in the n moment that ingress reception egress feedback signaling bag carries to measured value revise the packet loss correction value that obtains the n moment

The least mean-square error correction value M[n|n in S3, renewal n moment].

Preferably, utilize following formula to calculate the packet loss predicted value in n moment:

${\hat{P}}_{\mathrm{loss}}\left[n\right|n-1]=\frac{1}{\sqrt{2\mathrm{\&pi;}}}\&CenterDot;e^{-\frac{{\left({\mathrm{\&sigma;}}_R\right[n]\&CenterDot;\sqrt{-2\ln \left({\hat{P}}_{\mathrm{loss}}\right[n-1|n-1])-\ln \left(2\mathrm{\&pi;}\right)}-\stackrel{\&OverBar;}{r}[n\left]\right)}^2}{2\left({\mathrm{\&sigma;}}_R^2\right[n]+{\mathrm{\&sigma;}}_r^2[n\left]\right)}}$

Wherein, σ _r[n] represents the variance of the convergence service stream that n moment received, represent the average discharge of n moment new service flow, σ _r(n) variance of expression n moment new service flow.

Preferably, in step S2, utilize following formula to calculate the packet loss correction value in n moment

${\hat{P}}_{\mathrm{loss}}\left[n\right|n]={\hat{P}}_{\mathrm{loss}}[n|n-1]+K\left[n\right]\left({\hat{P}}_{\mathrm{loss}}\right[n]-{\hat{P}}_{\mathrm{loss}}[n|n-1]),$

Wherein, $M\left[n\right|n-1]={\left(a\right[n-1\left]\right)}^{2}M[n-1|n-1]+{\mathrm{\&sigma;}}_u^2,$

$K\left[n\right]=\frac{M\left[n\right|n-1]}{{\mathrm{\&sigma;}}_n^2+M\left[n\right|n-1]},$

M[n|n-1] represent the minimum mean-square error forecast value in n moment, M[n-1|n-1] represent the least mean-square error correction value in n-1 moment, represent to drive noise variance, represent observation noise variance, a[n-1] represent the derivative after state transitions coefficient a () discretization;

K[n] represent the kalman gain in n moment.

Preferably, utilize K[n] and M[n|n-1] renewal M[n|n]:

M[n|n]＝(1-K[n])M[n|n-1]。

Preferably, the statistical property of described new service flow is carried by the application signaling of new service flow.

(3) beneficial effect

Acceptance controlling method of the present invention is to improve for the one based on surveying quantitative statistics admission control algorithm extensively adopting at present, and the method is followed the tracks of and predicted the state information of network end-to-end by Kalman filter, improved the accuracy of acceptance judging.And the method can also reduce the error of acceptance controlling decision being introduced due to inaccurate state information sample value.

Brief description of the drawings

Fig. 1 is SCAC algorithm block diagram;

Fig. 2 is the method flow diagram of the embodiment of the present invention;

Fig. 3 is the state diagram of the acceptance judging of ingress.

Embodiment

Under regard to a kind of acceptance controlling method based on Kalman filtering proposed by the invention, in conjunction with the accompanying drawings and embodiments describe in detail.

In acceptance controlling method of the present invention, select end-to-end delay and the packet loss parameter as qos requirement, if the packet of source business is entering egress to the transmission delay d in (be ingress and egress to) _pexceed its end-to-end delay circle D, this packet is dropped so.The probability that packet is occurred to for overtime packet loss calls packet loss, the packet loss P in t moment _loss(t) represent:

P _loss(t)＝P(d _p＞D) (1)

For a Business Stream, if its packet loss P _loss(t) less than or equal to default desired packet loss maximum ε _qoS, think this Business Stream transmission success, otherwise think that it cannot meet its qos requirement, i.e. bust this.Therefore the receiving control criterion of Business Stream is:

P(d _p＞D)≤ε _QoS (2)

First explain SCAC algorithm, SCAC (Statistical CAC, SCAC) algorithm comprises three parts: information, enter egress between can reach capacity estimation, acceptance judging, introduce respectively below with reference to Fig. 1:

1), information

In the present invention, collecting the overtime packet loss causing is necessary to bound with statistical QoS guarantee is provided.Collect packet loss end to end by egress, and periodically it is fed back to ingress.The packet loss P that ingress utilization obtains _loss, and statistical property that measure, current (or being called existing) convergence service stream of having received (comprises average discharge and variances sigma _r(t)) estimate resource situation and the state of current network, to carry out acceptance judging.

2), can reach capacity estimation for given QoS packet loss requirement, the packet loss P obtaining according to feedback _loss, and the statistical property of the existing convergence service measuring stream (t) and σ _r(t), can utilize formula (3) estimation network below current can reach end to end capacity C _achv(t):

$C_{\mathrm{achv}}\left(t\right)\&ap;\stackrel{\&OverBar;}{R}\left(t\right)+{\mathrm{\&sigma;}}_R\left(t\right)\&CenterDot;\sqrt{-2\ln \left(P_{\mathrm{loss}}\left(t\right)\right)-\ln \left(2\mathrm{\&pi;}\right)}---\left(3\right)$

3), acceptance judging

What utilization estimated above can reach capacity and the statistical property average discharge of new service flow end to end and variances sigma _r(t), further derived in the situation that receiving this new service flow, by the packet loss causing by formula (4) to formula (6) below.It is pointed out that because existing business stream in new service flow and network is separate, so just there is formula (5):

$C_{\mathrm{achv}}\left(t\right)\&ap;\stackrel{\&OverBar;}{R}\left(t^{+}\right)+{\mathrm{\&sigma;}}_{R+r}\left(t\right)\&CenterDot;\sqrt{-2\ln \left(P_{\mathrm{loss}}\left(t^{+}\right)\right)-\ln \left(2\mathrm{\&pi;}\right)}---\left(4\right)$

$\stackrel{\&OverBar;}{R}\left(t^{+}\right)=\stackrel{\&OverBar;}{R}\left(t\right)+\stackrel{\&OverBar;}{r}\left(t\right)$

${\mathrm{\&sigma;}}_{R+r}\left(t\right)={\mathrm{\&sigma;}}_R\left(t\right)+{\mathrm{\&sigma;}}_r\left(t\right)---\left(5\right)$

${\mathrm{\&epsiv;}}_{\mathrm{est}}\left(t\right)=P_{\mathrm{loss}}\left(t^{+}\right)$

$=\frac{1}{\sqrt{2\mathrm{\&pi;}}}\exp [-\frac{{(C_{\mathrm{achv}}\left(t^{-}\right)-(\stackrel{\&OverBar;}{R}\left(t\right)+\stackrel{\&OverBar;}{r}\left(t\right)))}^2}{2({\mathrm{\&sigma;}}_R^2\left(t\right)+{\mathrm{\&sigma;}}_r^2\left(t\right))}]---\left(6\right)$

Calculate the packet loss of receiving after this new service flow according to formula (6), if this packet loss calculating is less than the packet loss of qos requirement, meets and receive control criterion (2), receive this Business Stream, otherwise refuse this Business Stream.

Introduce method of the present invention below.

Utilize Kalman filter can follow the tracks of estimator, prediction is a part for Kalman filter, and the error of forecast period was reduced in the correction stage.To utilize Kalman filter that the end-to-end packet loss of network is predicted and followed the tracks of below, to reduce the impact of observation noise on it, further improve SCAC algorithm and obtain a kind of acceptance controlling method based on Kalman filtering.

In formula (1)-(6), formula (3) is for calculating the current capacity C that can reach end to end _achv(t), formula (6) is estimated the packet loss ε in the time receiving new service flow _est(t).Formula (3) substitution formula (6) can be obtained:

$\begin{array}{c}{\mathrm{\&epsiv;}}_{\mathrm{est}}\left(t\right)=P_{\mathrm{loss}}\left(t^{+}\right)\\ -\frac{{[\stackrel{\&OverBar;}{R}\left(t\right)+{\mathrm{\&sigma;}}_R\left(t\right)\&CenterDot;\sqrt{-2\ln \left(P_{\mathrm{loss}}\left(t^{-}\right)\right)-\ln \left(2\mathrm{\&pi;}\right)}-(\stackrel{\&OverBar;}{R}\left(t\right)+\stackrel{\&OverBar;}{r}\left(t\right))]}^2}{2({\mathrm{\&sigma;}}_R^2\left(t\right)+{\mathrm{\&sigma;}}_r^2\left(t\right))}\\ =\frac{1}{\sqrt{2\mathrm{\&pi;}}}e\end{array}$

$=\frac{1}{\sqrt{2\mathrm{\&pi;}}}\&CenterDot;e^{-\frac{{[{\mathrm{\&sigma;}}_R\left(t\right)\&CenterDot;\sqrt{-2\ln \left(P_{\mathrm{loss}}\left(t^{-}\right)\right)-\ln \left(2\mathrm{\&pi;}\right)}-\stackrel{\&OverBar;}{r}\left(t\right)]}^2}{2({\mathrm{\&sigma;}}_R^2\left(t\right)+{\mathrm{\&sigma;}}_r^2\left(t\right))}}---\left(7\right)$

Wherein, t ⁺the moment that new service flow arrives, P _loss(t ⁺) be the packet loss of prediction, t ^-the moment that the last egress feedback signaling bag arrives, P _loss(t ^-) be that the node of recent renewal is to packet loss observed result end to end.Due to business arrive and packet loss to observe more new capital be in the event of series of discrete time point, therefore discrete time point n for formula above (n is more than or equal to 0 integer) can be represented.Formula above represents, the arrival of new business is encouraging whole network system, changes or affect the performance of network, is in particular in the variation of packet loss here.But meanwhile, owing to there is the background noise being caused by the reason such as resource-sharing of the business of interference in network, this noise causes the fluctuation of network state, also will directly affect the performance performance of network.And SCAC method is not considered the impact of this noise.Based on above analysis, again improve and compensation type (7), obtain the following formula (8) that represents with discrete time point n, wherein u[n] represent the noise that interference business causes, and state equation using formula (8) as network:

$P_{\mathrm{loss}}\left[n\right]=\frac{1}{\sqrt{2\mathrm{\&pi;}}}\&CenterDot;e^{-\frac{{\left({\mathrm{\&sigma;}}_R\left[n\right]\&CenterDot;\sqrt{-2\ln \left(P_{\mathrm{loss}}\right[n-1\left]\right)-\ln \left(2\mathrm{\&pi;}\right)}-\stackrel{\&OverBar;}{r}\right[n\left]\right)}^2}{2\left({\mathrm{\&sigma;}}_R^2\right[n]+{\mathrm{\&sigma;}}_r^2[n\left]\right)}}+u\left[n\right]---\left(8\right)$

Egress is periodically measured current packet loss, obtains the observation sample of end-to-end packet loss because observation exists certain error, therefore set up observational equation (9), wherein w[n] expression observation error.

${\hat{P}}_{\mathrm{loss}}\left[n\right]=P_{\mathrm{loss}}\left[n\right]+w\left[n\right]---\left(9\right)$

According to Kalman filter algorithm formula, state equation and the observational equation set up by formula (8) and formula (9), P _loss[n] is quantity of state s[n], and quantity of state s[n] measured value be the actual observation sequence that the n moment contains white noise.The state equation is here nonlinear, i.e. s[n]=a (s[n-1])+u[n], a is from s[n-1] state transformation is to s[n] transfer matrix of state.So adopting the Kalman filter of expansion (if a is a constant or constant matrix, is Kalman filter in general sense, and a is a function in the present invention, so state equation is nonlinear, therefore be called the Kalman filter of expansion), the recursive calculative formula is as follows:

The packet loss predicted value in n moment is:

${\hat{P}}_{\mathrm{loss}}\left[n\right|n-1]=\frac{1}{\sqrt{2\mathrm{\&pi;}}}\&CenterDot;e^{-\frac{{\left({\mathrm{\&sigma;}}_R\right[n]\&CenterDot;\sqrt{-2\ln \left({\hat{P}}_{\mathrm{loss}}\right[n-1|n-1])-\ln \left(2\mathrm{\&pi;}\right)}-\stackrel{\&OverBar;}{r}[n\left]\right)}^2}{2\left({\mathrm{\&sigma;}}_R^2\right[n]+{\mathrm{\&sigma;}}_r^2[n\left]\right)}}---\left(10\right)$

Wherein, represent the packet loss correction value (also referred to as estimated value) in n-1 moment.

Least mean-square error (MSE) predicted value in n moment:

$M\left[n\right|n-1]={\left(a\right[n-1\left]\right)}^{2}M[n-1|n-1]+{\mathrm{\&sigma;}}_u^2---\left(11\right)$

Wherein, M[n-1|n-1] represent the least mean-square error correction value in n-1 moment, represent to drive noise variance.

The kalman gain in n moment:

$K\left[n\right]=\frac{M\left[n\right|n-1]}{{\mathrm{\&sigma;}}_n^2+M\left[n\right|n-1]}---\left(12\right)$

Wherein, it is observation noise variance.

The packet loss correction value in n moment:

${\hat{P}}_{\mathrm{loss}}\left[n\right|n]={\hat{P}}_{\mathrm{loss}}[n|n-1]+K\left[n\right]\left({\hat{P}}_{\mathrm{loss}}\right[n]-{\hat{P}}_{\mathrm{loss}}[n|n-1])---\left(13\right)$

The least mean-square error correction value in n moment:

M[n|n]＝(1-K[n])M[n|n-1] (14)

Carry out in the process of recursion utilizing formula (10)～formula (14), need the derivative a[n-1 after computing mode transfer ratio (or being called state-transition matrix) a () discretization], thus can calculate (a[n-1]) in formula (11) ².Derive computing formula as shown in the formula (15), wherein,

$\mathrm{\&alpha;}=\frac{1}{{\mathrm{\&sigma;}}_R^2\left[n\right]+{\mathrm{\&sigma;}}_r^2\left[n\right]}$

$\mathrm{\&beta;}={\mathrm{\&sigma;}}_R\left[n\right]\&CenterDot;\sqrt{-2\ln \left({\hat{P}}_{\mathrm{loss}}\right[n-1|n-1])-\ln \left(2\mathrm{\&pi;}\right)}$

$a[n-1]=\frac{\&PartialD;a(\&CenterDot;)}{\&PartialD;P_{\mathrm{loss}}[n-1]}{|}_{P_{\mathrm{loss}}[n-1]={\hat{P}}_{\mathrm{loss}}[n-1|n-1]}$

$=\frac{1}{\sqrt{2\mathrm{\&pi;}}}\&CenterDot;e^{-\frac{\mathrm{\&alpha;}}{2}{(\mathrm{\&beta;}-\stackrel{\&OverBar;}{r}[n\left]\right)}^2}\&CenterDot;\frac{\mathrm{\&alpha;}(\mathrm{\&beta;}-\stackrel{\&OverBar;}{r}[n\left]\right)\&CenterDot;{\mathrm{\&sigma;}}_R^2\left[n\right]}{\mathrm{\&beta;}\&CenterDot;{\hat{P}}_{\mathrm{loss}}[n-1|n-1]}$

Set up now state equation and observational equation, can utilize the Kalman filter of expansion that this quantity of state of network packet loss rate is predicted and followed the tracks of, and carry out acceptance controlling decision with this.

The process of whole acceptance judging is such: S1, in the time of moment n, and new service flow application arrives ingress, obtains by measurements the statistical property that convergence service flows, and the statistical property of new service flow is by applying for that signaling carries, utilization (utilize in the n-1 moment recurrence formula (13) obtains) through type (10) predicts and obtains if receive this Business Stream, otherwise refusal.S2, then ingress is received the packet loss rate measurement value that egress feedback signaling bag carries through type (11) is revised and is obtained to formula (13) s3, final updating M[n|n].

If this new service flow is rejected, so this time result of prediction is in the recursion below can not substitution, therefore can not change the state of network because in fact unaccepted Business Stream can not enter network.Thus propelling in time constantly upgrade, recursion, to business, acceptance judging is made in application.Fig. 3 is the state diagram of the acceptance judging of ingress.

Above execution mode is only for illustrating the present invention; and be not limitation of the present invention; the those of ordinary skill in relevant technologies field; without departing from the spirit and scope of the present invention; can also make a variety of changes and modification; therefore all technical schemes that are equal to also belong to category of the present invention, and scope of patent protection of the present invention should be defined by the claims.

Claims (6)

1. the acceptance controlling method based on Kalman filtering, is characterized in that, comprises the following steps:

S1, in the time of moment n, new service flow application arrives ingress, obtains the statistical property of the convergence service stream of having received by measurement, utilizes statistical property, the statistical property of new service flow and the packet loss correction value in n-1 moment of the convergence service stream of having received predict the packet loss predicted value that obtains the n moment if receive this Business Stream, otherwise refusal, wherein ε _qoSrepresent default, desired packet loss maximum;

S2, the packet loss rate measurement value in the n moment that ingress reception egress feedback signaling bag carries to measured value revise the packet loss correction value that obtains the n moment

2. the method for claim 1, is characterized in that, also comprises step S3 after step S2, upgrades the least mean-square error correction value M[n|n in n moment].

3. method as claimed in claim 2, is characterized in that, in step S1, utilizes following formula to calculate the packet loss predicted value in n moment:

${\hat{P}}_{\mathrm{loss}}\left[n\right|n-1]=\frac{1}{\sqrt{2\mathrm{\&pi;}}}\&CenterDot;e^{-\frac{{\left({\mathrm{\&sigma;}}_R\right[n]\&CenterDot;\sqrt{-2\ln \left({\hat{P}}_{\mathrm{loss}}\right[n-1|n-1])-\ln \left(2\mathrm{\&pi;}\right)}-\stackrel{\&OverBar;}{r}[n\left]\right)}^2}{2\left({\mathrm{\&sigma;}}_R^2\right[n]+{\mathrm{\&sigma;}}_r^2[n\left]\right)}}$

Wherein, σ _r[n] represents the variance of the convergence service stream that n moment received, represent the average discharge of n moment new service flow, σ _r(n) variance of expression n moment new service flow.

4. method as claimed in claim 3, is characterized in that, in step S2, utilizes following formula to calculate the packet loss correction value in n moment

${\hat{P}}_{\mathrm{loss}}\left[n\right|n]={\hat{P}}_{\mathrm{loss}}[n|n-1]+K\left[n\right]\left({\hat{P}}_{\mathrm{loss}}\right[n]-{\hat{P}}_{\mathrm{loss}}[n|n-1]),$

Wherein, $K\left[n\right]=\frac{M\left[n\right|n-1]}{{\mathrm{\&sigma;}}_n^2+M\left[n\right|n-1]},$

$M\left[n\right|n-1]={\left(a\right[n-1\left]\right)}^{2}M[n-1|n-1]+{\mathrm{\&sigma;}}_u^2,$

M[n|n-1] represent the minimum mean-square error forecast value in n moment, M[n-1|n-1] represent the least mean-square error correction value in n-1 moment, represent to drive noise variance, represent observation noise variance, a[n-1] represent the derivative after state transitions coefficient discretization;

K[n] represent the kalman gain in n moment.

5. method as claimed in claim 4, is characterized in that, in step S3, utilizes K[n] and M[n|n-1] renewal M[n|n]:

M[n|n]＝(1-K[n])M[n|n-1]。

6. the method as described in any one in claim 1～5, is characterized in that, the statistical property of described new service flow is carried by the application signaling of new service flow.