CN110730469B - Method for predicting bandwidth based on extended Kalman wireless network and congestion control thereof - Google Patents

Method for predicting bandwidth based on extended Kalman wireless network and congestion control thereof Download PDF

Info

Publication number
CN110730469B
CN110730469B CN201910829517.4A CN201910829517A CN110730469B CN 110730469 B CN110730469 B CN 110730469B CN 201910829517 A CN201910829517 A CN 201910829517A CN 110730469 B CN110730469 B CN 110730469B
Authority
CN
China
Prior art keywords
wireless network
congestion
bandwidth
time delay
network link
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201910829517.4A
Other languages
Chinese (zh)
Other versions
CN110730469A (en
Inventor
王辉
洪波
唐俊勇
于孟晨
徐江涛
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xian Technological University
Original Assignee
Xian Technological University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Xian Technological University filed Critical Xian Technological University
Priority to CN201910829517.4A priority Critical patent/CN110730469B/en
Publication of CN110730469A publication Critical patent/CN110730469A/en
Application granted granted Critical
Publication of CN110730469B publication Critical patent/CN110730469B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W24/00—Supervisory, monitoring or testing arrangements
    • H04W24/02—Arrangements for optimising operational condition
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W24/00—Supervisory, monitoring or testing arrangements
    • H04W24/06—Testing, supervising or monitoring using simulated traffic
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W28/00—Network traffic management; Network resource management
    • H04W28/02—Traffic management, e.g. flow control or congestion control
    • H04W28/0289—Congestion control

Abstract

The invention discloses a bandwidth prediction method based on an extended Kalman wireless network and congestion control thereof, wherein the bandwidth prediction method comprises the following steps: constructing a state prediction model of the wireless network link bandwidth; constructing an observation model of the wireless network link bandwidth; obtaining a predicted estimated bandwidth according to the state prediction model; and correcting the predicted estimated bandwidth according to the observation model to obtain the final predicted bandwidth. According to the characteristics of the wireless network, bandwidth estimation based on the extended Kalman is introduced into the wireless network, and the available bandwidth of the wireless network is predicted by constructing a state prediction model and an observation model of the wireless network link bandwidth.

Description

Method for predicting bandwidth based on extended Kalman wireless network and congestion control thereof
Technical Field
The invention belongs to the technical field of wireless communication, and particularly relates to a bandwidth prediction method based on an extended Kalman wireless network and congestion control thereof.
Background
Due to the characteristics of the wireless network in topology, retransmission and service, the traditional TCP New Reno congestion control algorithm cannot fully guarantee the service quality of the wireless network. Aiming at the characteristics of large time delay, asymmetric bandwidth, high bit error rate and more short-term flows of a wireless network, network congestion is no longer the only reason for data packet loss, and when a transmission channel is affected by the outside to cause signal attenuation or interference, noise packet loss can be caused. When the TCP New Reno runs in a network with relatively poor link quality, packet loss types cannot be distinguished, once packet loss occurs, a congestion control algorithm is frequently called, a congestion window and a slow start threshold value are reduced, and the mode greatly restricts and influences the performance of a wireless network.
Currently, bandwidth is widely paid attention as a key parameter in terms of network traffic, congestion control, quality of Service (Qos) guarantee, and the like. Therefore, bandwidth estimation is introduced into network congestion control, the available bandwidth of the network is estimated in real time, the data volume to be sent is determined, and the influence caused by the fact that the traditional TCP algorithm judges the network congestion according to the data packet loss to blindly limit the sending rate can be effectively avoided. TCP Westwood is a strategy for realizing wireless network congestion control according to bandwidth estimation, when noise packet loss frequently occurs due to poor link environment of a wireless network, congestion control parameters can change rapidly, and meanwhile, due to sensitive parameter selection of TCPW, once bandwidth estimation is inaccurate, the congestion of the wireless network cannot be accurately judged, so that the throughput of the wireless network is seriously reduced. However, the accuracy of the bandwidth estimation directly determines the quality of the wireless network traffic service. If the available bandwidth is overestimated, more traffic data is accessed to the network, affecting the quality of service of the current traffic and causing network congestion. Conversely, if the available bandwidth estimate is too low, the capacity of the system is underutilized, which may reduce channel utilization and thus affect overall system throughput. Therefore, continuing congestion control based on inaccurate bandwidth estimation will inevitably cause congestion oscillation or waste of resources. In order to solve the above problem, kalman filtering is considered to be introduced to perform bandwidth prediction. The Kalman filtering is a discrete-time linear filter, can effectively solve the problem of discrete data linear filtering, can estimate and correct the current state information of the system, and can perform iterative prediction on the future state of the system. However, the conventional kalman filter is based on a linear system, and most of the actual network is a nonlinear system, so that it has a great challenge to be widely applied to the network. A common tool for solving the problem of nonlinear system state estimation is Extended Kalman Filtering (EKF), which linearizes nonlinear dynamics and assumes that the process and noise obey gaussian distribution, thereby achieving effective prediction of bandwidth.
However, when the bandwidth estimation is performed by using the extended kalman filter, the accuracy of the bandwidth estimation is emphasized too much in the conventional EKF method, which results in relatively complicated calculation, slow convergence and insufficient real-time performance.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides a bandwidth prediction method based on an extended Kalman wireless network and congestion control thereof.
The embodiment of the invention provides a bandwidth prediction method based on an extended Kalman wireless network, which comprises the following steps:
constructing a state prediction model of the wireless network link bandwidth;
constructing an observation model of the wireless network link bandwidth;
obtaining a predicted estimated bandwidth according to the state prediction model;
and correcting the predicted estimated bandwidth according to the observation model to obtain the final predicted bandwidth.
In one embodiment of the invention, the constructing of the state prediction model of the wireless network link bandwidth comprises the following steps:
sending a group of detection packets at a sending end of a wireless network link, and recording input time intervals between adjacent detection packets;
recording link transmission time delay of each detection packet between the wireless network link sending end and the wireless network link receiving end;
acquiring background flow according to a preset bottleneck bandwidth of a wireless network link and the utilization rate of the wireless network link;
constructing an output time interval function of the adjacent detection packets at the receiving end of the wireless network link according to the input time interval, the link transmission delay, the background flow and the preset bottleneck bandwidth of the wireless network link;
and constructing a state prediction model of the wireless network link according to the output time interval function and the preset available bandwidth function.
In one embodiment of the present invention, constructing the observation model of the wireless network link includes:
and constructing an observation model of the wireless network link according to the output time interval function.
In an embodiment of the present invention, obtaining a predicted estimated bandwidth according to the state prediction model includes:
obtaining a first Jacobian matrix according to the state prediction model;
and updating the state prediction model according to the state prediction model and the first Jacobian matrix, and obtaining the predicted estimated bandwidth of the wireless network link according to the updated state prediction model.
In an embodiment of the present invention, the obtaining a final predicted bandwidth by performing a correction process on the predicted estimated bandwidth according to the observation model includes:
obtaining a second Jacobian matrix according to the observation model;
updating an observation model according to the observation model and the second Jacobian matrix, and acquiring an output time interval of a wireless network link according to the updated observation model;
and correcting the predicted estimated bandwidth according to the output time interval and the second Jacobian matrix to obtain a final predicted bandwidth.
In one embodiment of the present invention, the final predicted bandwidth is:
Figure GDA0002278004810000031
wherein, K k The method is expressed in terms of the kalman gain,
Figure GDA0002278004810000041
represents a second Jacobian matrix, < >>
Figure GDA0002278004810000042
Represents a predicted estimate bandwidth, <' > or>
Figure GDA0002278004810000043
Representing an output time interval.
Another embodiment of the present invention provides a congestion control method for a wireless network, where the congestion control method includes:
acquiring a first round-trip time delay of each detection packet in the wireless network link, a second round-trip time delay of each detection packet and the size of a congestion window of the wireless network link;
smoothing the first time delay by using a mean value filtering mode to obtain a smooth first time delay;
obtaining a first sending rate according to the size of the congestion window and the smooth first time delay;
constructing a message backlog sensing factor according to the smooth first time delay, the second time delay and the first sending rate;
and controlling the wireless network congestion according to the message backlog sensing factor and a preset congestion threshold value.
In an embodiment of the present invention, before performing the wireless network congestion control according to the message backlog sensing factor and a preset congestion threshold, the method further includes:
obtaining a second sending rate according to the size of the congestion window and the second time delay;
and comparing the first sending rate with the second sending rate, if the first sending rate is greater than the second sending rate, performing wireless network congestion control according to the message backlog sensing factor and the preset congestion threshold, and if the first sending rate is less than or equal to the second sending rate, not performing wireless network congestion control according to the message backlog sensing factor and the preset congestion threshold.
In an embodiment of the present invention, performing congestion control on a wireless network according to the message backlog sensing factor and a preset congestion threshold includes:
and comparing the message backlog sensing factor with the preset congestion threshold, if the message backlog sensing factor is smaller than or equal to the preset congestion threshold, updating the slow start threshold and the congestion window according to the message backlog sensing factor and the congestion window size to control the wireless network congestion, and if the message backlog sensing factor is larger than the preset congestion threshold, updating the slow start threshold and the congestion window size according to a new final predicted bandwidth, the second time delay and a preset maximum message length to control the wireless network congestion.
In an embodiment of the present invention, the new final predicted bandwidth is calculated by any one of the above methods based on the extended kalman bandwidth prediction.
Compared with the prior art, the invention has the beneficial effects that:
according to the characteristics of the wireless network, bandwidth estimation based on the extended Kalman is introduced into the wireless network, and the available bandwidth of the wireless network is predicted by constructing a state prediction model and an observation model of the wireless network link bandwidth.
The present invention will be described in further detail with reference to the accompanying drawings and examples.
Drawings
Fig. 1 is a schematic flowchart of a bandwidth prediction method based on an extended kalman wireless network according to an embodiment of the present invention;
fig. 2 is a schematic structural diagram of bandwidth prediction performed by a probe packet in an extended kalman wireless network bandwidth prediction method according to an embodiment of the present invention;
fig. 3 is a flowchart illustrating a method for controlling congestion of a wireless network according to an embodiment of the present invention;
fig. 4 is a flowchart illustrating another method for controlling congestion of a wireless network according to an embodiment of the present invention;
fig. 5 is a schematic diagram of a network topology structure based on an extended kalman wireless network bandwidth prediction method according to an embodiment of the present invention;
fig. 6 is a schematic diagram illustrating a comparison of congestion window changes in the CSEKB and TCPW methods according to an embodiment of the present invention;
fig. 7 is a schematic diagram illustrating throughput comparison between CSEKB and TCPW methods according to an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to specific examples, but the embodiments of the present invention are not limited thereto.
Example one
At present, in a wireless network environment, due to the influences of channel states, link competition, various interference factors and the like, the available bandwidth of a wireless network changes significantly dynamically and randomly, and the transmission performance of the wireless network is greatly reduced. Therefore, in a wireless network environment, accurate estimation of available bandwidth is of great importance. The existing bandwidth estimation method comprises a TCPW (TCP Westwood, TCPW for short) BE bandwidth estimation method, a Kalman filtering bandwidth estimation method and an extended Kalman filtering bandwidth estimation method, wherein the TCPW BE bandwidth estimation method adopts a filter weight value which changes timely, so that the stability of bandwidth estimation is improved, but the network bandwidth is estimated excessively by an algorithm, and the accuracy is insufficient, so that the TCPW BE occupies excessive bandwidth and is not friendly to TCP Reno; the Kalman filtering bandwidth estimation method realizes estimation of different bandwidths and is applied to different scenes, but the Kalman filtering based bandwidth estimation methods are all based on a linear system, and most of actual networks are nonlinear systems, so that the wide application of the Kalman filtering based bandwidth estimation methods to the networks has huge challenges; when the bandwidth estimation is carried out by the extended Kalman filtering, the effective prediction of the bandwidth is realized, but the accuracy of the bandwidth estimation is emphasized too much, so that the calculation is relatively complex, the convergence is slow, and the instantaneity is insufficient.
Based on the above existing problems, please refer to fig. 1, where fig. 1 is a schematic flow chart of a method for predicting a bandwidth based on an extended kalman wireless network according to an embodiment of the present invention. The embodiment of the invention provides a bandwidth prediction method based on an extended Kalman wireless network, which comprises the following steps:
step 1, constructing a state prediction model of wireless network link bandwidth;
step 2, constructing an observation model of the wireless network link bandwidth;
step 3, obtaining a predicted estimated bandwidth according to the state prediction model;
and 4, correcting the predicted estimated bandwidth according to the observation model to obtain the final predicted bandwidth.
Specifically, the present embodiment specifically combines the characteristics of the wireless network, constructs a state prediction model structure and an observation model of the wireless network link bandwidth of the present embodiment, performs prediction estimation on the current available bandwidth by using the constructed state prediction model to obtain a predicted estimated bandwidth, and then corrects the predicted estimated bandwidth according to the observation model to obtain a final predicted bandwidth.
The method is low in calculation complexity, capable of predicting the available bandwidth of the wireless network in real time and good in convergence and accuracy. The available bandwidth prediction model comprises a state prediction model structure and an observation model.
Further, in step 1 of this embodiment, a state prediction model of the wireless network link bandwidth is constructed.
Specifically, in an actual network, since a wireless network link is not completely idle, the present embodiment defines a part of the capacity used in the wireless network link as background Traffic (Cross Traffic), and a part of the capacity not used as Available Bandwidth (Available Bandwidth). Although background traffic shares the capacity of the wireless network link with available bandwidth, the wireless network link may only have "used" and "idle" states at time t instant. Therefore, for better definition of background traffic and available bandwidth, in the present embodiment, ρ is defined assuming that the wireless network link is in a stable state, i.e., the background traffic is kept constant during the time (t, t + τ) i (t) is an instantaneous utilization state of the wireless network link i at time t, idle is 0, used is 1, and instantaneous utilization state ρ i The (t) is specifically:
Figure GDA0002278004810000071
the time interval (t, t + tau) can be obtained according to the formula (1)Instantaneous utilization state ρ of wireless network link i at each moment i (t), the average utilization rate ρ of the wireless network link i between time (t, t + τ) i τ The (t) is specifically:
Figure GDA0002278004810000072
thus, when the capacity of the wireless network link i is C i And a capacity of C i With average utilization rate p between times (t, t + tau) i τ (t) when used, then background flow
Figure GDA0002278004810000073
The method specifically comprises the following steps: />
Figure GDA0002278004810000074
And available bandwidth of wireless network link
Figure GDA0002278004810000081
The method specifically comprises the following steps:
Figure GDA0002278004810000082
having the available bandwidth of one wireless network link, path P (l) of n wireless network links 1 ,l 2 ,…,l n ) The available bandwidth from the transmitting end to the receiving end between times (t, t + τ) is specifically:
Figure GDA0002278004810000083
referring to fig. 2, fig. 2 is a schematic structural diagram of performing bandwidth prediction based on a probe packet in an extended kalman wireless network bandwidth prediction method according to an embodiment of the present invention. In order to accurately and quickly predict the available bandwidth of the network, the embodiment designs an EKF-based wireless network available bandwidth prediction model, and realizes calculation of the available bandwidth so as to improve the accuracy of subsequent wireless network congestion control. The model assumes that there is only one bottleneck link in the link path of the wireless network to be tested, i.e. the narrow link and the tight link are the same link. By using a packet pair technology, according to the change of the background flow to the inter-packet interval in the packet pair, a function between the input time interval of the packet pair and the background flow is obtained, then the bottleneck link bandwidth is used for subtracting the background flow to obtain the available bandwidth of the wireless network link path, and the available bandwidth A (t) is specifically designed as follows:
Figure GDA0002278004810000084
wherein, g in G, the time interval is input, specifically representing the time difference between the probe packet P1 and the probe packet P2 in the probe packet pair at the transmitting end of the wireless network link out For outputting the time interval, specifically representing the time difference between the probe packet P1 and the probe packet P2 in the probe packet pair at the receiving end of the wireless network link, C bott Representing the bottleneck link bandwidth in the wireless network link. Input time interval g of probe packet in The output time interval g of the detection packet can be set at the sending end of the wireless network link out It can be calculated at the receiving end of the wireless network link by recording the arrival time of each probe packet.
Based on the design of the available bandwidth of the formula (6), step 1 in this embodiment specifically includes step 1.1, step 1.2, step 1.3, step 1.4, step 1.5, and step 1.6:
step 1.1, sending a group of detection packets at a sending end of the wireless network link, and recording input time intervals between adjacent detection packets.
Specifically, in this embodiment, a group of probe packets is sent at the sending end of the wireless network link, and the packet pair technique is used to record the input time difference between adjacent probe packets for the input time interval.
And step 1.2, recording link transmission time delay of each detection packet between the sending end of the wireless network link and the receiving end of the wireless network link.
Specifically, the link transmission delay of the probe packet between the sending end of the wireless network link and the receiving end of the wireless network link in this embodiment is g B Time delay g of link transmission B The method comprises the following specific steps:
Figure GDA0002278004810000091
wherein L is p Indicating the length of each probe packet.
And step 1.3, acquiring background flow according to the preset bottleneck bandwidth of the wireless network link and the utilization rate of the wireless network link.
Specifically, the present embodiment obtains the propagation delay g in the link according to the above formula (2) B The utilization rate of the wireless network link is obtained according to the formula (3) to obtain the current background flow B of the wireless network link f Wherein, the capacity C in the formula (3) i Is the bottleneck link bandwidth C bott 。
Due to the actual wireless network link, background traffic B f With the abruptness and the inability of a single probe packet to capture the average of background traffic, this embodiment uses several equal input time intervals g in To measure the average value of background traffic.
And step 1.4, constructing an output time interval function of the adjacent detection packet at a receiving end of the wireless network link according to the input time interval, the link transmission delay, the background flow and the preset bottleneck bandwidth of the wireless network link.
Specifically, in this embodiment, when two adjacent detection packets P1 and P2 in a group of detection packets input by a sending end of a wireless network link successively arrive at a router, and the router has a detection packet queue, an output time interval function of a detection packet at a receiving end of the wireless network link specifically is as follows:
Figure GDA0002278004810000092
in this embodiment, the link transmission delay g in the formula (7) B Substituting into formula (8) to obtain the output time interval function g of the wireless network link receiving end out Comprises the following steps:
Figure GDA0002278004810000101
and step 1.5, establishing a state prediction model of the wireless network link according to the output time interval function and a preset available bandwidth function.
Specifically, in the present embodiment, the preset available bandwidth function is specifically designed as formula (6), and the output time interval function g of formula (9) is set out Substituting into the preset available bandwidth function in formula (6), the constructed available bandwidth a (t) of the wireless network link in this embodiment is:
Figure GDA0002278004810000102
further, constructing a state prediction model of the wireless network link bandwidth according to the formula (10) is designed as follows:
A k ＝f(A k-1 ,g in ,L p ,B f )+w k-1 (11)
wherein the content of the first and second substances,
Figure GDA0002278004810000103
representing the available bandwidth of the wireless network link at time k; />
Figure GDA0002278004810000104
A non-linear state transfer function representing the transfer of a wireless network link from time k-1 to time k, i.e. the available bandwidth function at time k-1, w k-1 Representing noise in the state process, specifically zero-mean independent white Gaussian noise, the noise error corresponding to the noise is represented by a covariance matrix Q k-1 Given, in particular, the present embodiment combines the covariance matrix Q k-1 Is set as a diagonal element of10 -2 The off-diagonal entries are set to zero.
This embodiment sends a set of probe packets with "increasing input time interval" at the sending end. By recording the input time interval and the output time interval, the transmission is stopped when the input time interval and the output time interval are equal.
Further, in step 2 of this embodiment, an observation model of the wireless network link bandwidth is constructed.
Specifically, the present embodiment directly uses the output time interval function g obtained in step 1.4 out As an observation variable, specifically, the observation model of the wireless network link bandwidth is specifically:
Figure GDA0002278004810000105
wherein the content of the first and second substances,
Figure GDA0002278004810000106
a non-linear observation function representing the transition of the wireless network link from the time k-1 to the time k, namely an output time interval function, v k Representing noise in observation process, specifically independent white Gaussian noise with zero mean, and the noise error corresponding to the noise is represented by covariance matrix R k Given, in particular, the present embodiment combines the covariance matrix R k Is set to 10 -2 The off-diagonal entries are set to zero.
Further, in step 3 of this embodiment, a predicted estimated bandwidth is obtained according to the state prediction model.
Specifically, in this embodiment, first, the available bandwidth in the wireless network link is estimated, specifically, the predicted estimated bandwidth is obtained through a state prediction model, and step 3 specifically includes step 3.1 and step 3.2:
and 3.1, obtaining a first Jacobian matrix according to the state prediction model.
Specifically, the present embodiment applies to the input time interval g in the state prediction model of equation (11) in Probe packet length L p And background flowB f Respectively obtaining a first Jacobian matrix by partial derivation
Figure GDA0002278004810000111
The first Jacobian matrix->
Figure GDA0002278004810000112
The method specifically comprises the following steps:
Figure GDA0002278004810000113
and 3.2, updating the state prediction model according to the state prediction model and the first Jacobian matrix, and obtaining the predicted estimated bandwidth of the wireless network link according to the updated state prediction model.
Specifically, in this embodiment, a Taylor expansion is used for the state prediction model of the formula (11), and an updated state prediction model is obtained by combining the first jacobian matrix, where the updated state prediction model specifically includes:
Figure GDA0002278004810000114
wherein the content of the first and second substances,
Figure GDA0002278004810000115
represents the optimal predicted bandwidth, A, of the wireless network link at time k-1 k-1 Represents an estimate of available bandwidth at time k-1 for the wireless network link, and->
Figure GDA0002278004810000116
The first Jacobian matrix is represented, and H.O.T.1 represents other order Taylor expansion of the state prediction model.
It should be noted that the Taylor expansion of high-order terms h.o.t.1 is negligible, and therefore the extended kalman filter is also referred to as a first-order filter.
Further, the predicted estimated bandwidth of the wireless network link is obtained through the state prediction model updated by the formula (14), and thus, the method is shown in the specificationExample Bandwidth estimate A at the last k-1 time k-1 Based on the predicted estimated bandwidth of the next k time
Figure GDA0002278004810000117
Prediction estimation bandwidth>
Figure GDA0002278004810000118
The method specifically comprises the following steps:
Figure GDA0002278004810000121
further, in step 4 of this embodiment, the predicted estimated bandwidth is corrected according to the observation model to obtain a final predicted bandwidth.
Specifically, step 4 in this embodiment specifically includes step 4.1, step 4.2, and step 4.3:
and 4.1, obtaining a second Jacobian matrix according to the observation model.
Specifically, the present embodiment is directed to the input time interval g in the observation model of equation (12) in Probe packet length L p And background flow rate B f Respectively obtaining a second Jacobian matrix by partial derivation
Figure GDA0002278004810000122
A first Jacobian matrix>
Figure GDA0002278004810000123
The method specifically comprises the following steps:
Figure GDA0002278004810000124
and 4.2, updating the observation model according to the observation model and the second Jacobian matrix, and acquiring the output time interval of the wireless network link according to the updated observation model.
Specifically, in this embodiment, taylor expansion is used for the observation model of formula (12), and an updated observation model is obtained by combining the second jacobian matrix, where the updated observation model specifically includes:
Figure GDA0002278004810000125
/>
wherein the content of the first and second substances,
Figure GDA0002278004810000126
representing the predicted estimated bandwidth of the wireless network link at time k, A k An available bandwidth estimate for the wireless network link at time k, based on the estimated value of the available bandwidth at time k>
Figure GDA0002278004810000127
Representing the second jacobian matrix and h.o.t.2 representing other order Taylor expansions of the observation model.
It should be noted that the Taylor expansion of high-order terms h.o.t.2 is negligible, and therefore the extended kalman filter is also referred to as a first-order filter.
And 4.3, correcting the predicted estimated bandwidth according to the output time interval and the second Jacobian matrix to obtain the final predicted bandwidth.
Specifically, the actual observed value of the wireless network link at the current k time, that is, the output time interval, is obtained through the observation model updated by the formula (17)
Figure GDA0002278004810000128
And then utilizes an output time interval of the current wireless network bandwidth status>
Figure GDA0002278004810000129
To modify the predicted estimated bandwidth ≥ based on equation (15) in step 3.2>
Figure GDA00022780048100001210
So as to obtain the optimal predicted bandwidth of the wireless network link at the current k moment, and the optimal predicted bandwidth is->
Figure GDA0002278004810000131
The method specifically comprises the following steps:
Figure GDA0002278004810000132
wherein, K k Representing the kalman gain of the wireless network link at the current time k,
Figure GDA0002278004810000133
represents a second Jacobian matrix, < >>
Figure GDA0002278004810000134
Represents a predicted estimated bandwidth, at the current time k, of the wireless network link, greater or lesser than>
Figure GDA0002278004810000135
Representing the output time interval of the wireless network link at the current time k.
Kalman gain K of the wireless network link at the current K moment in the embodiment k The method specifically comprises the following steps:
Figure GDA0002278004810000136
wherein the content of the first and second substances,
Figure GDA0002278004810000137
represents a second Jacobian matrix->
Figure GDA0002278004810000138
Transpose of (P) k A first prediction error covariance matrix representing the current time k, a first prediction error covariance matrix P k The method specifically comprises the following steps:
Figure GDA0002278004810000139
wherein, K k-1 Kalman gain, S, representing the current k-1 moment of a wireless network link k A second prediction error covariance matrix representing the current k time instant, the second prediction errorCovariance matrix S k The method comprises the following specific steps:
Figure GDA00022780048100001310
wherein the content of the first and second substances,
Figure GDA00022780048100001311
represents a first Jack-ratio matrix->
Figure GDA00022780048100001312
Transpose of (P), P k-1 Is the first prediction error covariance matrix at the current time instant k-1.
The present embodiment utilizes actual measurements of the output time interval of the wireless network link at the current time k
Figure GDA00022780048100001313
To modify the predicted estimated bandwidth ≥ based on the formula (15) of step 3>
Figure GDA00022780048100001314
Thereby obtaining the optimal estimated value which is closer to the real value of the bandwidth of the wireless network>
Figure GDA00022780048100001315
And then continuously recurrently and iteratively carrying out bandwidth prediction estimation and correction on the wireless network, and finally realizing the final predicted bandwidth A from the transmitting end to the receiving end of the wireless network link τ And (4) predicting. When actually used, the latest optimal prediction bandwidth is taken>
Figure GDA00022780048100001316
The method is used as a new initial value for predicting and estimating the bandwidth, and any nonzero value is input as the initial value of the first prediction error covariance matrix and the initial value of the second prediction error covariance matrix, because the EKF-based wireless network bandwidth prediction method can automatically run in a circulating mode, the final predicted bandwidth A from the transmitting end to the receiving end of the wireless network link is realized τ And (4) predicting.
In the network communication process of this embodiment, the size of the actual available bandwidth of each TCP connection depends on the physical bandwidth of the bottleneck link on the wireless network transmission link and the number of wireless network links competing for the same link connection. In the embodiment, the bandwidth is predicted and corrected by an available bandwidth prediction model based on the EKF, and the optimal estimated value of the available bandwidth is obtained continuously and circularly. The change condition of the wireless network link information is calculated through the formula, and the filter guidance is provided for the next filtering of the filter, so that the bandwidth prediction speed is increased.
In summary, by utilizing the characteristic that the EKF can reduce the bandwidth estimation error to the minimum under the appropriate condition, the EKF is applied to the wireless network, and the EKF-based available bandwidth prediction method suitable for the wireless network is provided.
Example two
On the basis of the first embodiment, please refer to fig. 3, and fig. 3 is a flowchart illustrating a method for controlling congestion of a wireless network according to an embodiment of the present invention. The embodiment of the invention provides a wireless network congestion control method, which comprises the following steps:
step 1, obtaining a first round-trip time delay of each detection packet in a wireless network link, a second round-trip time delay of each detection packet and a congestion window size of the wireless network link.
And 2, smoothing the round-trip first time delay of the detection packet by using a mean value filtering mode to obtain a smooth first time delay.
And 3, obtaining a first sending rate according to the size of the congestion window and the smooth first time delay.
And 4, constructing a message backlog sensing factor according to the smooth first time delay, the second time delay and the first sending rate.
And 5, controlling the wireless network congestion according to the message backlog sensing factor and a preset congestion threshold value.
Specifically, in this embodiment, after receiving an ACK response message by a sending end of a wireless network link, a congestion window size of the wireless network link at a current time, a first round-trip delay of each probe packet in the wireless network link, and a second round-trip delay of each probe packet are recorded, where the first delay is a sum of round-trip delays of the current time corresponding to all the probe packets in the wireless network link, and the second delay is a minimum round-trip delay of the current time corresponding to a certain probe packet, and the first delay is smoothed to obtain a first smooth delay, and a first sending rate is obtained according to the congestion window size and the smooth delay, and a message backlog sensing factor is constructed according to the smooth first delay, the second delay, and the first sending rate, and finally, wireless network congestion control is performed according to the message backlog sensing factor and a preset congestion threshold.
In the embodiment, by designing the message backlog sensing factor, when the detection packet detects that the data packet is lost, the message backlog sensing factor is calculated, the packet loss type is refined by using the preset congestion threshold, and a basis is provided for setting a new congestion window parameter which is determined as a noise packet loss condition, so that the parameter change better conforms to the actual condition of the wireless network link, the waste of bandwidth caused by the overlarge congestion window is reduced, and the transmission performance of the wireless network link is further improved.
Further, in this embodiment, step 1 obtains a first time delay of round trip of each probe packet in the wireless network link, a second time delay of round trip of each probe packet, and a congestion window size of the wireless network link.
Specifically, the first time delay obtained in this embodiment is a sum Rtt of round-trip time delays at the current time corresponding to all probe packets in the wireless network link all The second time delay is the minimum round-trip time delay Rtt of the current time corresponding to a certain detection packet min The congestion window size of the wireless network link is cwnd.
Further, in step 2 of this embodiment, the first time delay is smoothed by using a mean filtering manner to obtain a smoothed first time delay.
In particular, to eliminate noiseIn the embodiment, the mean filtering mode is adopted to carry out round-trip first time delay Rtt all Performing smoothing, specifically smoothing the first time delay Rtt smo The method is defined as an average value of round trip delays of all detection packets in a wireless network link, and specifically comprises the following steps:
Rtt smo ＝Rtt all /n (22)
further, in step 3 of this embodiment, a first sending rate is obtained according to the size of the congestion window and the smooth first time delay.
Specifically, in the present embodiment, the smooth first time delay Rtt is obtained from the formula (22) in step 2 smo And calculating a first sending rate in a wireless network link according to the congestion window size cwnd obtained in the step 1, wherein the first sending rate is the actual sending rate V act Actual transmission rate V act The method specifically comprises the following steps:
V act ＝cwnd/Rtt smo (23)
further, in this embodiment, step 4 obtains a message backlog sensing factor according to the smoothed first delay, the second delay, and the first sending rate.
Specifically, some message data to be sent are easily backlogged in the cache of the sending end of the bottleneck link of the wireless network, and the backlogged message data in the caches can cause the first time delay Rtt of the wireless network all To be larger, i.e. smooth, the first time delay Rtt smo The larger the size, the embodiment defines the message backlog sensing factor F np Message data volume representing backlog, message backlog sensing factor F np The method specifically comprises the following steps:
F np ＝(Rtt smo -Rtt min )×V act (24)
the embodiment designs the message backlog sensing factor F np Judging the backlog condition of the message data volume in the cache of the current wireless network bottleneck link sending end, and further carrying out wireless network congestion control according to the backlog condition.
Further, in this embodiment, step 5 performs wireless network congestion control according to the message backlog sensing factor and a preset congestion threshold.
Specifically, before performing the wireless network congestion control according to the message backlog sensing factor and the preset congestion threshold, the embodiment further includes:
obtaining a second sending rate according to the size of the congestion window and the second time delay;
and comparing the first sending rate with the second sending rate, if the first sending rate is greater than the second sending rate, performing wireless network congestion control according to the message backlog sensing factor and a preset congestion threshold, and if the first sending rate is less than or equal to the second sending rate, not performing wireless network congestion control according to the message backlog sensing factor and the preset congestion threshold.
In particular, the present embodiment depends on the congestion window size cwnd and the second time delay Rtt min Obtaining a second sending rate, wherein the second sending rate is the optimal sending rate V opti Optimum transmission rate V opti The method specifically comprises the following steps:
V opti ＝cwnd/Rtt min (25)
when the actual transmission rate V is act >Optimum transmission rate V opti In time, it is shown that some message data to be sent are backlogged in the cache of the sending end of the bottleneck link of the wireless network, so that the actual sending rate V is act The ideal optimum transmission rate V is not achieved opti . It is due to the accumulated message data in the buffer that the first time delay Rtt is smoothed smo Therefore, the embodiment designs the message backlog sensing factor F np To determine the congestion status of the current network.
And further, performing wireless network congestion control according to the message backlog sensing factor and a preset congestion threshold.
Specifically, please refer to fig. 4, where fig. 4 is a flowchart illustrating another method for controlling congestion of a wireless network according to an embodiment of the present invention. When the actual transmission rate V is act >Optimum transmission rate V opti Time of day, i.e. actual transmission rate V act Not reaching the optimal transmission rate V opti Then need to calculate the message backlog sensing factor F np . The students of Tuan Anh and the like sort and analyze data characteristics of unacknowledged message data volume which are accumulated in a cache area and have noise packet loss characteristics when packet loss is transmitted by researching an actual wireless network link, and obtain preset congestion threshold values f under different bandwidth conditions. By comparing messages to backlog a perception factor F np And a preset congestion threshold value F, wherein the purpose of distinguishing packet loss types is achieved, specifically, the message backlog sensing factor is compared with the preset congestion threshold value, and if the message backlog sensing factor F is detected np If the packet loss is less than or equal to the preset congestion threshold F, determining that a noise packet is lost, the packet loss does not cause congestion of the wireless network, the wireless network is still in a normal working state, at this time, bandwidth prediction does not need to be carried out again according to the extended kalman bandwidth prediction method in the first embodiment, and only parameter fine adjustment is needed according to multiplicative reduction conditions, namely, according to the message backlog perception factor F np Updating a slow start threshold ssthresh by the congestion window size cwnd, wherein ssthresh = cwnd, and according to the message backlog perception factor F np The congestion window size cwnd updates the congestion window size cwnd, and cwnd = cwnd-F np To control the wireless network congestion, if the message backlog sensing factor F np If the value is greater than the preset congestion threshold value f, it is determined that congestion packet loss occurs, and the packet loss further affects the current congestion state of the wireless network, and at this time, bandwidth prediction needs to be performed again according to the extended kalman bandwidth prediction method in embodiment one to obtain a new final predicted bandwidth a τ According to the new final predicted bandwidth A τ The second time delay Rtt min Updating the slow start threshold ssthresh according to the preset Maximum message Size (MSS), and
Figure GDA0002278004810000181
at this time, the congestion window size cwnd needs to be designed according to the actual situation of the network to control the congestion of the wireless network. The congestion window size cwnd needs to be designed according to the actual situation of the network, and comprises the following steps: fast recovery when network state is inAnd in the repeated stage, the value of the congestion window cwnd is the current slow start threshold ssthresh, and when the network state is in the non-fast recovery stage, the value of the congestion window cwnd is a fixed value 1.
In the traditional TCPW method, when a multiplicative congestion reduction mechanism is triggered by a wireless network link, whether noise packet loss or congestion packet loss occurs is not distinguished, and congestion in the network is directly judged, so that the throughput of the wireless network is reduced, and the performance of the wireless network is reduced. In the Congestion Control method (CSEKB for short) based on the Extended Kalman Wireless network provided in this embodiment, the parameter settings are not changed in the slow start stage and the additive growth stage, but after the Wireless network is congested, a message backlog sensing factor F is designed np And whether congestion packet loss is achieved or not is distinguished according to the packet loss type, so that the slow start threshold value entering and the congestion window size setting are refined. As message backlog sensing factor F np When the congestion window size cwnd is smaller than the preset congestion threshold f and enters a fast recovery or overtime retransmission stage, the parameter is finely adjusted, so that the congestion window size cwnd can be increased quickly, the real time of congestion arriving is advanced, and at the moment, the situation that part of message data accumulated due to noise packet loss still exists in the cache of the sending end of the bottleneck link of the wireless network is not confirmed is judged, the congestion window size cwnd is limited, a sufficient space of a slow start threshold ssthresh is reserved, the message data can cope with the noise packet loss, and the high efficiency of the transmission quality of the wireless network is ensured.
In summary, in this embodiment, the congestion state of the current wireless network link is determined according to the designed message backlog sensing factor, the bandwidth prediction of the network in the congestion state is accurately and efficiently performed according to the extended kalman wireless network bandwidth prediction-based method, and the predicted bandwidth value is used as a basis for congestion avoidance, so that the size of the congestion window is accurately adjusted, and the maximization of the network performance is realized.
In order to verify the effectiveness of the wireless network congestion control method provided by the present application, the following simulation experiments are further described:
simulation scene:
as the NS3 (Network simulator 3, NS3 for short) Network simulator widely draws the success technology and experience of the existing excellent open source Network simulator, the Network simulator has good development environment, abundant modules and open sources of codes, can provide high-performance Network simulation which is closer to a real Network. Therefore, NS3 is installed on the Linux Ubuntu 16.04 system as a simulation experiment environment, and a highly controllable reusable simulation platform is constructed to verify the performance of the bandwidth prediction-based wireless network congestion control algorithm. The application is mainly based on the comparison of two methods of CSEKB and TCPW.
Referring to fig. 5, fig. 5 is a schematic diagram of a network topology based on an extended kalman wireless network bandwidth prediction method according to an embodiment of the present invention. FIG. 5 shows a dumbbell network topology, node s, of the simulation study of the present application 1 ,s 2 ,…s n Representing the transmitting end, node r, of a wireless network 1 ,r 2 Router being a backbone, node d 1 ,d 2 ,…d n Representing the receiving end of the wireless network. Assuming that the bottleneck link bandwidth randomly changes according to uniform distribution, the end-to-end aggregated FTP data stream traversing the topology comprises two types, namely a long-life cycle stream and a short-life cycle stream. The simulation experiment set up 10 long-life FTP streams traversing the bottleneck link from two directions, and 20 short-life FTP streams traversing the bottleneck link from source side to destination side from left to right.
The simulation parameters of the present application are specifically set as shown in table 1.
TABLE 1 simulation parameters specific settings
Figure GDA0002278004810000191
It should be noted that the simulation parameters in table 1 are designed for parameters of one wireless network link, and other wireless network links may be designed with the same parameters or with different parameters, and the parameters are only for illustrating the effectiveness of the method of the present application and are not limited to the design of such parameters.
Referring to fig. 6, fig. 6 is a schematic diagram illustrating a comparison of congestion window changes in the CSEKB and TCPW methods according to an embodiment of the present invention. In fig. 6, the abscissa represents time and the ordinate represents the congestion window size. The CSEKB method and the TCPW method are respectively used to transmit the files of the wireless network, and the congestion window changes are compared, as can be seen from fig. 5, when the packet backlog sensing factor F is introduced np Compared with the traditional TCPW method, the CSEKB method of the application has the following obvious advantages: in the TCPW method, once packet loss is detected, the congestion window size cwnd is frequently reduced, and when a multiplicative reduction mechanism is started, the TCPW method directly reduces the congestion window to the minimum value; the CSEKB method of the application utilizes a message backlog sensing factor F np The packet loss types are distinguished, the congestion window size cwnd can be continuously kept on the increasing trend when the noise packet loss is judged, and a proper new congestion window parameter is set according to the final predicted bandwidth obtained by the method based on the bandwidth prediction method of the extended Kalman wireless network, so that the CSEKB method can be used for keeping the congestion window size cwnd at a higher level finally as long as the packet loss blindly reduces the value of the congestion window. Therefore, the CSEKB can effectively sense the bandwidth change of the wireless network, and the congestion control in the wireless network is realized by using the bandwidth prediction method based on the extended Kalman wireless network.
Referring to fig. 7, fig. 7 is a schematic diagram illustrating throughput comparison between CSEKB and TCPW methods according to an embodiment of the present invention. In fig. 7, the abscissa represents time, and the ordinate represents throughput. In the same simulation environment, 50-200 MB files are transmitted by utilizing FTP application, and meanwhile, real-time throughput data in a network are acquired by using Wireshark analysis software, as can be seen from fig. 7, the network throughput is always kept at a higher level and can be kept relatively stable when the files are transmitted by the CSEKB method of the application, however, under the condition of using the traditional TCPW method, the network throughput is relatively higher but unstable when the files are transmitted, so that the overall throughput in the network transmission process is reduced, and the CSEKB method introduces a message backlog perception factor F np Judgment of F by comparison np And a preset congestion threshold f to reset the appropriateThe new congestion window parameter is set by the bandwidth prediction method based on the extended Kalman wireless network, so that frequent rapid decrease of the congestion window and reduction of throughput are avoided. Compared with the traditional TCPW method, the CSEKB method has more stable throughput and is embodied in the remarkable enhancement of the throughput performance in the whole data transmission process.
The foregoing is a more detailed description of the invention in connection with specific preferred embodiments and it is not intended that the invention be limited to these specific details. For those skilled in the art to which the invention pertains, numerous simple deductions or substitutions may be made without departing from the spirit of the invention, which shall be deemed to belong to the scope of the invention.

Claims (2)

1. A method for controlling congestion in a wireless network, comprising:
acquiring a first round-trip time delay of each detection packet in a wireless network link, a second round-trip time delay of each detection packet and the size of a congestion window of the wireless network link; the first time delay is the sum of round-trip time delays of the current time corresponding to all detection packets in a wireless network link, and the second time delay is the minimum round-trip time delay of the current time corresponding to a certain detection packet;
smoothing the first time delay by using a mean value filtering mode to obtain a smooth first time delay;
obtaining a first sending rate according to the size of the congestion window and the smooth first time delay;
constructing a message backlog sensing factor according to the smooth first time delay, the second time delay and the first sending rate;
performing wireless network congestion control according to the message backlog sensing factor and a preset congestion threshold value;
the step of controlling the wireless network congestion according to the message backlog sensing factor and a preset congestion threshold value comprises the following steps:
and comparing the message backlog sensing factor with the preset congestion threshold, if the message backlog sensing factor is smaller than or equal to the preset congestion threshold, updating the slow start threshold and the congestion window according to the message backlog sensing factor and the congestion window size to control the wireless network congestion, and if the message backlog sensing factor is larger than the preset congestion threshold, updating the slow start threshold and the congestion window size according to the new final predicted bandwidth, the second time delay and the preset maximum message length to control the wireless network congestion.
2. The method according to claim 1, wherein before performing the wireless network congestion control according to the message backlog sensing factor and the preset congestion threshold, the method further comprises:
obtaining a second sending rate according to the size of the congestion window and the second time delay;
and comparing the first sending rate with the second sending rate, if the first sending rate is greater than the second sending rate, performing wireless network congestion control according to the message backlog sensing factor and the preset congestion threshold, and if the first sending rate is less than or equal to the second sending rate, not performing wireless network congestion control according to the message backlog sensing factor and the preset congestion threshold.
CN201910829517.4A 2019-09-03 2019-09-03 Method for predicting bandwidth based on extended Kalman wireless network and congestion control thereof Active CN110730469B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910829517.4A CN110730469B (en) 2019-09-03 2019-09-03 Method for predicting bandwidth based on extended Kalman wireless network and congestion control thereof

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910829517.4A CN110730469B (en) 2019-09-03 2019-09-03 Method for predicting bandwidth based on extended Kalman wireless network and congestion control thereof

Publications (2)

Publication Number Publication Date
CN110730469A CN110730469A (en) 2020-01-24
CN110730469B true CN110730469B (en) 2023-03-24

Family

ID=69217839

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910829517.4A Active CN110730469B (en) 2019-09-03 2019-09-03 Method for predicting bandwidth based on extended Kalman wireless network and congestion control thereof

Country Status (1)

Country Link
CN (1) CN110730469B (en)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111836297B (en) * 2020-06-30 2022-07-08 珠海云洲智能科技股份有限公司 Communication link parameter prediction method and device and terminal equipment
CN114143271B (en) * 2021-11-30 2024-04-02 百果园技术(新加坡)有限公司 Bandwidth estimation method and device based on congestion detection
CN114422443B (en) * 2022-01-24 2023-08-15 西安电子科技大学 Satellite network TCP congestion control method based on bandwidth estimation and congestion prediction
CN114710807A (en) * 2022-04-28 2022-07-05 云南师范大学 5G millimeter wave communication-oriented Kalman filtering dynamic congestion control method
CN116528375B (en) * 2023-06-28 2023-09-29 浙江大华技术股份有限公司 Bandwidth prediction method, bandwidth prediction device, and computer-readable storage medium

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104093169A (en) * 2014-06-10 2014-10-08 北京邮电大学 MAC-TCP cross-layer optimization congestion control method and system
CN106059950A (en) * 2016-05-25 2016-10-26 四川大学 Adaptive network congestion control method based on SCPS-TP

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9419900B2 (en) * 2013-12-31 2016-08-16 International Business Machines Corporation Multi-bit indicator set according to feedback based on an equilibrium length of a queue

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104093169A (en) * 2014-06-10 2014-10-08 北京邮电大学 MAC-TCP cross-layer optimization congestion control method and system
CN106059950A (en) * 2016-05-25 2016-10-26 四川大学 Adaptive network congestion control method based on SCPS-TP

Also Published As

Publication number Publication date
CN110730469A (en) 2020-01-24

Similar Documents

Publication Publication Date Title
CN110730469B (en) Method for predicting bandwidth based on extended Kalman wireless network and congestion control thereof
US10567300B2 (en) Layer 3 fair rate congestion control notification
US8130650B2 (en) Retro flow control for arriving traffic in computer networks
US8385201B2 (en) End-system dynamic rate limiting of background traffic
US11558302B2 (en) Data transmission method and apparatus
KR102350504B1 (en) Apparatus and method for controlling downlink throughput in communication system
US8369216B2 (en) System and method of monitoring packets in flight for optimizing packet traffic in a network
US10873529B2 (en) Method and apparatus for low latency data center network
Xu et al. Hybrid congestion control for high-speed networks
Ziegler et al. A quantitative Model for the Parameter Setting of RED with TCP traffic
Barrera et al. Statistical approach for congestion control in gateway routers
US8289851B2 (en) Lightweight bandwidth-management scheme for elastic traffic
US10904150B1 (en) Distributed dynamic load balancing in network systems
Chen et al. Fluid-flow Analysis of TCP Westwood with RED
US7336611B1 (en) Rate-based multi-level active queue management with drop precedence differentiation
Lim et al. Robust end-to-end loss differentiation scheme for transport control protocol over wired/wireless networks
JP3853784B2 (en) Data communication management method
Dijkstra et al. Modeling active queue management algorithms using stochastic Petri nets
Lu et al. EQF: An explicit queue-length feedback for TCP congestion control in datacenter networks
Fridovich-Keil et al. A model predictive control approach to flow pacing for TCP
Meng et al. Research on TCPW improvement in hybrid network
Zhang et al. Adaptive fast TCP
CN110650491B (en) Forward error correction FEC parameter analysis method for vehicle-mounted ad hoc network communication
Manikandan et al. Active queue management based congestion control protocol for wireless networks
Herfeh et al. Active queue management in TCP networks based on minimum variance adaptive control

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant