GB2461516A - A method of predicting traffic in a wireless network - Google Patents

Abstract

A method of predicting traffic in a wireless network having a plurality of wireless devices, is carried out at at least one of the wireless devices in communication with at least one further wireless device, and comprises determining a measure of traffic on the network over a preceding time interval and predicting its value for a forward time interval based on the determined measure.

Description

WIRELESS COMMUNICATION METHOD AND APPARATUS

Field of the Invention

The present invention relates to wireless communication. It is particularly, but not exclusively, concerned with a method and apparatus for predicting a quality parameter of a wireless network.

Background of the Invention

Wireless communication between electronic devices is becoming increasingly in demand, particularly due to the growth of multimedia communication services, such as video streaming, video conferencing, packet data transfer and so on. Accordingly, wireless networks are widely deployed to support these services. Generally, these networks are capable of supporting communications for multiple users by sharing the available network resources. One example of such network is a wireless local area network (WLAN).

A typical arrangement of a WLAN is illustrated in Figure 1. Such a WLAN 10 includes one or more access points (APs) 14, 16 that serve one or more wireless devices 20. An AP (14 or 16) is typically a stand-alone device that is connected to an Ethernet switch 18 in a wired network 12, and it relays data between the wireless devices 20 and devices in the wired network 12. Each of the APs manages wireless traffic in the area of coverage (lOa, lob), and typically allows the devices 20 to communicate to and from a wired network 12. The Ethernet switch 18 can be connected to a modem 22 to allow the devices 20 to connect to the Internet 24.

As shown in Figure 1, two or more APs may link together to form a larger network to allow the devices 20 to roam from one WLAN lOa to another WLAN lob. The devices described herein may be mobile terminals such as personal digital assistants (PDAs), notebook computers, or fixed terminals such as desktops and workstations that are equipped with a wireless network interface.

One of the key issues in the implementation of a WLAN is its management of quality of service (QoS) such as load balancing, handover between APs, network traffic monitoring, traffic congestion reduction arid so on. This can involve, for example, gathering information concerning the traffic load of a wireless network and predicting the traffic load in a forward time interval at an AP. This can allow the AP to predict traffic patterns, and to use the predicted traffic load to perform resource management tasks.

Most of the existing studies on network traffic prediction focus on analysing real traffic traces collected from wireless networks to generate guidelines on design and deployment of similar networks. These studies mainly employ descriptive statistics to gain more insight into different aspects of wireless networks such as user mobility patterns, the relationship between number and type of users and traffic load at an AP, the number of associations, popular user applications, traffic generated by different types of applications, the distribution of user session time and so on. It will be appreciated that these types of analysis can only provide useful information for designing better wireless networks, in particular, improving network capacity planning, proper positioning of access points, and use of caching for improving response times.

WO 2005/048499 discloses a system and method for predicting traffic load in a WLAN.

In this document, the traffic prediction is performed at a wireless transmit/receive unit (WTRU) to predict traffic between a WTRU and an AP. The traffic prediction information is then sent from the WTRU to the AP where it is used in conjunction with generation of commands sent to the WTRU to control the manner of access by the WTRU to the WLAN via the AP. However, this document does not describe any details on the method of carrying out the prediction.

A number of traffic forecasting algorithms are proposed in Papadopouli, M, Raftopoulos, E. and Shen, H., "Evaluation of short-term traffic forecasting algorithms in wireless networks ", 2'd Conference on Next Generation Internet Design and Engineering, Valencia, Spain, April 2006. In this document, the traffic load at APs are modelled, and forecasting algorithms are used to predict the traffic load at APs, based on the traffic models, in different time-scales. Two main types of models are described, namely simple models which incorporate daily periodicities (such as hour of the day), and hybrid models which incorporate both periodicities and additional information such as types of traffic, and client profile. The simple models using periodicity information do not take into account fluctuations of the traffic on shorter time scales. The hybrid models, which take into account of fluctuations occurring on shorter time scales, are more complex. In addition, these models require the underlying non-stationary data to be transformed into stationary time series prior to modelling.

Summary of the Invention

In a first aspect of the present invention there is provided a method of predicting a quality parameter of a wireless network having a plurality of wireless devices, said method being performed at at least one of said plurality of wireless devices in communication with at least one further wireless device of said plurality of wireless devices, and the method comprising determining a quality parameter of the wireless network over a preceding time interval, and predicting said quality parameter of said wireless network for a forward time interval based on said determined quality parameter.

The quality parameter may include traffic load of said wireless network.

Use of historical traffic load of a wireless network to predict future traffic load can allow a wireless device to manage network resources of the wireless network proactively as opposed to the approach of reacting to subsisting network conditions -as described in the prior art. Therefore, the wireless device can adapt quickly to changes in the traffic of the wireless network.

The preceding time interval may be set immediately preceding a current time. By setting the preceding time interval immediately preceding the current time can allow the most recent traffic level to be measured. As a result, the traffic level of the wireless network in the immediate future can be predicted accurately.

The step of determining traffic load of the wireless network over the preceding time interval may include sampling the traffic load with N number of samples between the preceding time interval.

The step of predicting traffic load of said wireless network may be performed by means of an adaptive filtering algorithm.

The method can include adjusting the adaptive filtering algorithm to the environment (which can be substantially changeable such as the network traffic level) in such a way that its performance improves through continuing interaction with its surrounding.

Therefore, this can provide a more accurate prediction of the traffic load.

In one embodiment of the present invention, the adaptive filtering algorithm may be a Recursive Least Squares (RLS) algorithm. One of the benefits of using a RLS algorithm is that there is no need to invert large matrices, thereby saving computing power. Another advantage of using a RLS algorithm is that its rate of convergence is, in general, faster than other adaptive filtering algorithm, for example the Least Means Squares (LMS) algorithm. Therefore, the RLS algorithm can adapt quickly to changes in the underlying traffic and predicts the traffic load accurately in a short period of time.

It is noted that the accuracy of predicting the future traffic level of the wireless network depends on the number samples, N, being sampled between the preceding time interval.

In other words, the higher the number of samples, N, the more accurate is the predicted traffic level. Conversely, the number of arithmetic operations increases as the number of samples, N, increases, thereby increasing the computation load of the wireless device.

In this example, the number of samples, N, may advantageously be set as 5. It will be appreciated that any other values of N may also be used in the computation to predict the future traffic load.

The wireless devices may include at least one wireless access point, and at least one wireless terminal.

The method of the above aspect may be performed at the wireless access point. By performing the prediction at the wireless access point can allow the predicted traffic load to be broadcasted to the wireless terminal(s) directly such that the wireless terminal(s) can make decision according to the predicted traffic load. This is as opposed to performing the prediction at the wireless terminal(s) and sending the predicted traffic load back to the wireless access points followed by broadcasting the predicted traffic load to other wireless terminals in the wireless network. It is noted that the former approach makes efficient use of the available capacity of the network.

The predicted traffic load may be provided to a network resource management module connected to the wireless access point, so as to allow said network resource management module to perform proactive network resource management tasks.

In one embodiment of the above aspect, the quality parameter may further include handoff parameter, and channel quality between the wireless devices.

In a second aspect of the present invention, there is provided an apparatus for predicting a quality parameter of a wireless network having a plurality of wireless devices, said apparatus being coupled to at least one of said plurality of wireless devices in communication with at least one further wireless device of said plurality of wireless devices, and the apparatus comprising means for determining a quality parameter of the wireless network over a preceding time interval, and means for predicting said quality parameter of said wireless network for a forward time interval based on said determined quality parameter.

Aspects of the invention may comprise a computer program product comprising computer executable instructions operable to cause a computer to become configured to perform a method in accordance with any of the above identified aspects of the invention. The computer program product can be in the form of an optical disc or other computer readable storage medium, a mass storage device such as a flash memory, or a read only memory device such as ROM. The method may be embodied in an application specific device such as an ASIC, or in a suitably configured device such as a DSP or an FPGA. A computer program product could, alternatively, be in the form of a signal, such as a wireless signal or a physical network signal.

Brief description of the drawings

Embodiments of the present invention will now be described with reference to the accompanying drawings, wherein: Figure 1 illustrates the physical arrangement of a conirnunications network; Figure 2 illustrates a traffic load prediction process in accordance with an embodiment of the present invention; Figure 3 illustrates the process of predicting traffic load in accordance with an embodiment of the present invention; Figure 4 shows the mean of the prediction error obtained in accordance with an embodiment of the present invention; Figure 5 shows the standard deviation of the prediction error obtained in accordance with an embodiment of the present invention; Figure 6 shows the prediction error obtained in accordance with an embodiment of the present invention; Figure 7 shows the throughput obtained in accordance with an embodiment of the present invention; and Figure 8 is a schematic diagram of an example access point.

Detailed Description

Specific embodiments of the present invention will be described in further detail on the basis of the attached diagrams. It will be appreciated that this is by way of example only, and should not be viewed as presenting any limitation on the scope of protection sought.

Figure 1 has been described above in relation to the prior art example. For consistency, the invention will now be described by way of specific embodiment in relation to the arrangement illustrated in figure 1.

In figure 1, two access points (APs) 14, 16 are illustrated. Each of these APs connects wireless devices 20 together to form a WLAN lOa, lOb. The APs 14, 16 are also connected to a fixed wired network 12 via an Ethernet switch 18 such that data can be relayed between wireless devices 20 and wired devices (not shown). The Ethernet switch 18 is in turn connected to a modem 22 so as to allow the devices 20 to connect to the Internet 24.

Typically the coverage ranges of an AP ranges within a radius of 100 metres. The APs can also be link together to form a larger wireless network 10 such that the wireless devices 20 can roam from one WLAN lOa to another WLAN lOb.

As shown in figure 1, the APs 14, 16 are connected and are registered to a resource management platform 26 which monitors the registered APs via Simple Network Management Protocol (SNMP) Management Information Databases (MIBs). The resource management platform 26 analyses information received from the registered APs 14, 16 to detect and report the condition of the associated WLAN environment.

The example commences in a situation whereby the APs 14, 16 predict traffic loads of the respective WLANs 1 Oa, lOb, and convey the predicted information to the resource management platform 26 dynamically.

Essentially, the predicted information can potentially be used in different resource management tasks. For example, the predicted traffic load can be broadcast to the wireless devices through beacon messages or probe response messages. With this information, the wireless devices can, for example, make a proactive handoff decision based on the trends observed in the metric of interest (related to handoff). One example of predicting traffic is by using a small section of traffic in the past to predict traffic in the immediate future.

Figure 8 illustrates schematically an AP (14, or 16) providing an example of background to the invention. The AP 14 comprises a processor 80 operable to execute machine code instructions stored in a working memory 90 and/or retrievable from a mass storage device 82. By means of a general-purpose bus 84, user operable input devices 86 are in communication with the processor 80. The user operable input devices 86 comprise, in this example, a keyboard and a touchpad, but could include a mouse or other pointing device, a contact sensitive surface on a display unit of the device, a writing tablet, speech recognition means, haptic input means, or any other means by which a user input action can be interpreted and converted into data signals.

Audio/video output devices 88 are further connected to the general-purpose bus 84, for the output of infonnation to a user. Audio/video output devices 88 include a visual display unit, and a speaker, but can also include any other device capable of presenting information to a user.

A communications unit 94 is connected to the general-purpose bus 84, and further connected to an antenna 96. By means of the communications unit 94 and the antenna 96, the AP 14 is capable of establishing wireless communication with another device (for example, an wireless device). The communications unit 94 is operable to convert data passed thereto on the bus 84 to an RF signal carrier in accordance with a communications protocol previously established for use by a system in which the AP 14 is appropriate for use.

In the AP 14 of figure 8, the working memory 90 stores user applications 92 which, when executed by the processor 80, cause the establishment of a user interface to enable communication of data to and from a user. The applications 92 thus establish general purpose or specific computer implemented utilities and facilities that might habitually be used by a user.

The prediction of traffic loads will now be described with reference to figure 2 which illustrates processes carried out by APs of the WLANs in figure 1.

In step 50, the process commences by an initialisation process which include (1) setting a sampling period (SP) between two successive measurements of traffic load in the past, (2) setting the number of past samples, N, required to compute the prediction, (3) initialising the prediction counter, and (4) scheduling the prediction timer to t + SP, where t 0 at this stage.

In step 52, the prediction timer records the number of observations of traffic load that are currently available.

The number of observations of traffic load is compared against the number of past samples, N, required to compute the prediction (step 54). If the number of observations of traffic load is less than the required number of past samples, the prediction timer is set to "current time + SP", in step 58. However, if the number of observations of traffic load is equal to or greater than the required number of past samples, the predicted traffic load over the next time interval will be computed (step 56). Essentially, the next interval is set as "current time + SP".

The computed predicted output is subsequently provided to the resource management platform (step 62).

In this example, an iterative Recursive Least Squares (RLS) algorithm is executed to predict traffic load over the next interval (step 56). RLS algorithm has an advantage of fast convergence property compared to other adaptive algorithm. The execution of an iteration of the RLS algorithm will be described with reference to figure 3, and the steps below.

Step 70 -Compute gain vector: K(n) = -l)u(n) 1 + X'uT(n)P(n -1)u(n) Step 72 -Compute prediction error: e(n) = u(n) -uhat Step 74 -Update weight vector: w(n) w(n -1) + K(n) * e(n) Step 76 -Update P matrix: P(n) = -1)-A'K(n)uT(n)P(n -1) Step 78 -Compute the prediction: uhat = uT(n)w(n) The notations used in the RLS algorithm are as follows: n current step (current time instant) N number of past samples (history) required to compute a prediction u(n) "Nxl" input (vector) containing the traffic load values of the "N" intervals of the recent past w(n) "Nxl" weight (vector) containing weights associated with the recent past values of traffic load. Initially, all values in the weight vector are set to 0. These values adapt online as time progresses uhat Predicted value of traffic load over the next Sampling Period e(n) Prediction error (scalar) at time "n" Forgetting factor which governs the level of importance should be given to past observations P(n-.1) "NxN" matrix (inverse of the input autocorrelation matrix) K(n) "Nx 1" gain vector (updating step size) It is noted that the RLS algorithm estimates the future traffic load as a weighted sum of traffic loads observed over the last "N" intervals, and therefore is an intuitively simple algorithm. It is further noted that the estimation process involves the most basic arithmetic operations during the estimation process and hence it is simple, from an implementation perspective.

Theoretically, the RLS algorithm requires 3N(N+3)/2 arithmetic operations for every iteration. It will be appreciated that the accuracy of the RLS algorithm depends on the amount of past samples (or observations), N, required to predict the traffic load in the future.

Different values of N ranging between 1 and 10 were experimented in order to find a value of N which delivers accurate predictions. Figures 4 and 5 illustrate the mean and standard deviation, respectively, of the prediction error on a 1-minute timescale. It is noted from these figures, the prediction accuracy improves with higher value of N. Note, however, that the gain in prediction accuracy is nominal beyond 3 to 4 samples.

This indicates that a small N value is sufficient to predict traffic load in the future.

It is also noted that the prediction error of the RLS algorithm converges at 2N iterations.

In other words, the convergence is delayed as the number of N observations increases.

Thus, a small value of N is preferred in order to achieve fast convergence.

As an example, a value of N=5 is used. Therefore, the number of arithmetic operations required is 60 -resulting in a lightweight implementation. In addition, the RLS should also (theoretically) converge in 10 iterations.

The results depicted in figures 6 and 7 show that the RLS algorithm is able to react rapidly to changes. For example, at 540 minutes, the prediction error corresponding to the throughput spike is very high arid is reduced substantially in a short period of time.

It should be noted that the above method is not only applicable to predicting traffic load in a wireless network but can be used more generally with predicting quality parameters of a wireless communications network.

For instance, the handoff parameter (such as signal strength or throughput) of a wireless communications network can be predicted using the above method. The predicted value can be used to trigger a handoff when the predicted value exceeds a threshold. This provides a proactive handoff decision as opposed to the conventional approach of reacting only when the situation (signal strength or throughput) is significantly deteriorated. It will be appreciated that the band off triggering is not limited in scope to threshold based methods. It is noted that other methods may also be employed without loss of generality.

Another example for the application of the above method is that a wireless device can select a data rate to transmit data based on the channel quality as measured in the past, on a short time scale. Essentially, the wireless device (for example, a receiver device) predicts the channel quality by performing past measurements on the quality of an associated downlink channel, and providing this information to the sender (for example, a transmitter). The transmitter device can then choose an appropriate data rate (and consequently an appropriate MCS) based on the predicted channel quality.

As a further example, the feedback sent from the receiver to the sender can be conveyed using, for example a Clear to Send (CTS) packet (which is part of the RTS-CTS handshake) or an acknowledgment (ACK) packet. However, it is also noted that the present invention is not limited to these two methods of conveying feedback, and any other method of conveying feedback may be employed.

It will be appreciated that provided that the downlink and uplink channels exhibit a similar channel quality, the information rate for the prediction can also be carried out at the sender based on past measurements of the downlink channel quality. This provides a number of advantages such as allowing the wireless device to act proactively by predicting channel quality in the immediate future, and eliminating feedbacks from the receiver to the sender device to change the data rate, thereby eliminating feedback delay.

A further example for the application of the present disclosure is that a wireless device can predict the quality of the channel based on a received signal transmitted from a further wireless device, and provide the predicted information back to the further wireless device such that this further wireless device can adapt its data transmission rate based on the predicted information.

It will be understood that the invention has been described above purely by way of example, and modifications of detail can be made within the scope of the invention.

Each feature disclosed in the description and (where appropriate) the claims and drawings may be provided independently or in any appropriate combination.

Claims (20)

1. CLAIMS: 1. A method of predicting a quality parameter of a wireless network having a plurality of wireless devices, said method being performed at at least one of said plurality of wireless devices in communication with at least one further wireless device of said plurality of wireless devices, and the method comprising determining a quality parameter of the wireless network over a preceding time interval; and predicting said quality parameter of said wireless network for a forward time interval based on said determined quality parameter.
2. 2. A method according to claim 1 wherein the quality parameter includes traffic load of said wireless network.
3. 3. A method according to claim 1 or claim 2 wherein the preceding time interval is set immediately preceding a current time.
4. 4. A method according to claim 2 or claim 3 wherein the step of determining traffic load of the wireless network over the preceding time interval includes sampling the traffic load with N number of samples between the preceding time interval.
5. 5. A method according to any one of claims 2 to 4 wherein the step of predicting traffic load of said wireless network is performed by means of an adaptive filtering algorithm.
6. 6. A method according to claim 5 wherein the adaptive filtering algorithm is a Recursive Least Squares (RLS) algorithm.
7. 7. A method according to any one of the preceding claims wherein the wireless devices include at least one wireless access point, and at least one wireless terminal.
8. 8. A method according to claim 7 wherein said method of determining said quality parameter is performed at the wireless access point.
9. 9. A method according to any one of claims 2 to 8 wherein the predicted traffic load is provided to a network resource management module connected to the said wireless access point, so as to allow said network resource management module to perform proactive network resource management tasks.
10. 10. An apparatus for predicting a quality parameter of a wireless network having a plurality of wireless devices, said apparatus being coupled to at least one of said plurality of wireless devices in communication with at least one further wireless device of said plurality of wireless devices, and the apparatus comprising means for determining a quality parameter of the wireless network over a preceding time interval, and means for predicting said quality parameter of said wireless network for a forward time interval based on said determined quality parameter.
11. 11. An apparatus according to claim 10 wherein the quality parameter includes traffic load of said wireless network.
12. 12. An apparatus according to claim 10 or claim 11 wherein the preceding time interval is set immediately preceding a current time.
13. 13. An apparatus according to claim 11 or claim 12 wherein the means for determining traffic load of the wireless network over the preceding time interval includes means for sampling the traffic load with N number of samples between the preceding time interval.
14. 14. An apparatus according to any one of claims 11 to 13 wherein the means for predicting traffic load of said wireless network is performed by means of an adaptive filtering algorithm.
15. 15. An apparatus according to claim 14 wherein the adaptive filtering algorithm is a Recursive Least Squares (RLS) algorithm.
16. 16. An apparatus according to any one of claims 10 to 15 wherein the wireless devices include at least one wireless access point, and at least one wireless terminal.
17. 17. An apparatus according to claim 16 wherein said means for determining said quality parameter is coupled to the wireless access point.
18. 18. An apparatus according to any one of claims 11 to 17 wherein the predicted traffic load is provided to a network resource management module connected to the said wireless access point, so as to allow said network resource management module to perform proactive network resource management tasks.
19. 19. A storage medium storing computer executable instructions which, when executed on general purpose computer controlled communications apparatus, cause the apparatus to become configured to perform the method of any of claims 1 to 9.
20. 20. A signal carrying computer receivable information, the information defining computer executable instructions which, when executed on general purpose computer controlled communications apparatus, cause the apparatus to become configured to perform the method of any of claims 1 to 9.