GB2440195A

GB2440195A - Optimising superframe periods to reduce delay to delay-sensitive traffic

Info

Publication number: GB2440195A
Application number: GB0614455A
Authority: GB
Inventors: Timothy Adrian Lewis
Original assignee: Toshiba Research Europe Ltd
Current assignee: Toshiba Europe Ltd
Priority date: 2006-07-20
Filing date: 2006-07-20
Publication date: 2008-01-23
Anticipated expiration: 2026-07-20
Also published as: GB0614455D0; GB2440195B

Abstract

A communication controller, e.g. Access Point, provides flexible control of an IEEE802.11 WLAN superframe. In particular, the communication controller optimises the proportion of the superframe reserved as the Contention-Free Period (CFP), during which the communication controller controls access to the communications medium, and/or the Contention Period (CP), during which access to the communications medium is allowed on a contention basis, to reduce the delay to delay-sensitive traffic. During network operation, an adaptive mechanism is used by which polled traffic with a tight delay constraint can maximise its performance without adversely affecting the bandwidth available to contention traffic. However, instead of simply increasing the length of the CFP until the delay constraints are met, embodiments of the invention use the idea that strategic skipping of the contention period can provide a sufficient short term boost to the polled traffic to improve packet delays. The superframe parameters may be repeatedly re-optimised in response to changes in the network delay performance.

Description

I

Method for Reconfiguration of Polling in a Wireless LAN A Wireless Communication System This invention relates to apparatus, methods, processor control code and signals for establishment of wireless communications in a network. More particularly, but not exclusively, it relates to aspects of the technology supported by the IEEE8O2. 11 standards, especially concerning WLAN.

The IEEE8O2. 11 standard (in particular, the digest entitled "IEEE Wireless LAN Edition -A compilation based on IEEE Std 802.1 1TM- 1999 (R2003) and its amendments") describes and defines technology for the implementation of wireless network based communication between stations (STA). In the context of the standard, a network is referred to as a Basic Service Set (BSS). Two topologies of BSS are supported. Firstly, an "infrastructure" topology defines one of the stations of the BSS as an Access Point (AP). An access point is provided to act as a bridge into a wider LAN or the internet. Secondly, in an "independent" topology, a group of stations form a BSS on an ad hoc basis.

The 802.11 standard defines technology in two layers of the OSI model, namely the MAC (Media Access Control) layer and the PHY (Physical) layer. The MAC layer presents a MAC Service Access Point (SAP) to higher layers of the OSI model, accepting MAC Service Data Units (MSDUs). The MAC is capable of fragmenting a received MSDU if this allows for more robust transmission of data. After the MAC has processed a (possibly fragmented) MSDU, it presents MAC Protocol Data Units (MPDUs) to the PHY SAP.

In order to manage access to the communications medium, the standard provides a Distributed Coordination Function (DCF). Using the DCF, each station in the BSS competes for access to the resource on an equitable basis with all others, using Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA). As described in section 9.1.1 of the above referenced standard, a station using the DCF monitors the medium for a predetermined period of time (known as the DCF Inter-Frame Space, DIFS) to determine if another station is transmitting and, if not, backs off for a minimum specified time period (which in practice is a randomly generated number of slots, where a slot is a fixed period of time defined for a particular PHY specification) during which the medium must remain free, before attempting to commence transmission itself. This random back-off avoids the situation where two stations may detect the medium to have been free for a period of DIFS then simultaneously transmit and collide. There is still a small probability that two stations will choose the same random back-off period, but this is considerably less than if they were allowed to transmit straight after the DIFS.

As a feedback mechanism, the size of the range of random numbers is increased in the event of collisions occurring, decreasing that collision probability still further.

Transmissions are conducted with either a two-phase (DATAIACK) or four-phase (RTS/CTS/DATA/ACK) handshake. Jn each case, error recovery is achieved through Automatic Repeat reQuest, ARQ. The four phase handshake is particularly appropriate to networks containing hidden nodes -i.e. stations that are out of the transmission range of the transmitting station (and so would not detect an ongoing transmission) but are in transmission range of the receiving station (hence this hidden node could quite innocently start to transmit to the receiving node as well, resulting in collisions).

Prioritisation, for example of the right to send an ACK in reply over another STA's right to initiate a new exchange, is achieved through Inter Frame Spaces (IFS) -the short IFS (STFS), DIFS, PCF IFS (P1FS) and Extended/Emergency IFS (E1FS). Further description of the structure and use of the different types of IFS can be found in the above referenced version of the standard.

Use of the 802.11 standard compliant technology in an infrastructure type topology only requires provision of the DCF mechanism. However, a centralised Point Coordination function (PCF) can be optionally provided, to offer enhanced operation. This is of particular use for traffic with particular Quality of Service requirements, such as loss, delay or jitter.

The PCF offers contention-free access to a particular, nominated station. This station is implemented with a Point Coordinator (PC) functionality. Typically the Point Coordinator (PC) resides on the AP.

In order to allow PCF and DCF to coexist, the standard specifies use of a superframe, illustrated in figure 1. The superframe comprises a beacon (B) announcing the start of a contention free period (CFP), which is then followed by a contention period (CP).

During the CFP, the Access Point has contention free access to the medium. Then, in the CP, the station in question reverts to operation in accordance with the DCF.

In most published discussion of IEEE8O2. 11, and most particularly the concept of a superframe, the superframe is typically characterised by two parameters: CFPgp, which determines the rate at which the super-frame repeats (i.e. when new beacons are issued), and CFPMAX, which determines the length of the CFP. CFPp is expressed in absolute length, in milliseconds. Typically, as described in "Self-adaptive transmission scheme of integrated services over an [EEE 802.11 WLAN," (C. Li, J. Li, and X. Cai, Electronics Letters, vol. 40, pp. 1596, 2004) CFPp is in the range 20-lOOms.

Further, CFPMAX is normally expressed as a ratio of the maximum CFP period to the CFPp period.

The polling process within the CFP is initiated by the PC sending either a simple CF-Poll packet to the first STA to be polled or, if the AP has data to send to the STA, a combined Data+CF-Poll packet. Having received a CF-Poll, the STA then has the right to send a single buffered frame. The STA can then acknowledge the Data (if any) with a CF-Ack or, if it has data to send (that will fit within the TXOP), a Data+CF-Ack frame.

As a further example of a multi-purpose transmission, the PC can then send a combined Data+CF-Ack+CF-Poll frame, acknowledging the transmission from the previous STA, polling the next one and sending it some data, in one operation. This process continues until the PC concludes the CFP with a CFP-End transmission (which can also be combined with CFP-Ack to the last polled STA if needed); the DCF-based CP then begins.

STAs associate with the AP by means of an initial handshake. A STA can indicate during association that it does not require to be polled now, or ever, enabling it to be excluded from the polling list. In PCF, the list of STAs to be polled is the entire list of STAs that have associated with the AP that have expressed a desire to be polled.

One difficulty with using the PCF of IEEE8O2.1 1 is that if a station successfully contends for access to the medium during the CP, there is no way of bounding the transmission time of that station, apart from the upper limit on MPDU-size. This can lead to unnecessary and undesirable delay of the beacon and consequently the ensuing contention free period. This clearly impacts the ability of PCF to support delay-and jitter-sensitive traffic.

The IEEE8O2. 11 e standard (IEEE (Institute of Electrical and Electronics Engineers),"IEEE Standard 802.11 e -Part 11 Amendment 8: Medium Access Control (MAC) Quality of Service Enhancements," 2005.) enhances the PCF, through provision of a Hybrid Coordination Function (HCF). The HCF offers a Transmit Opportunity (TXOP), which grants access to the medium for a defined period of time (to address the delayed-beacon issue of PCF mentioned above). A Hybrid Controller (HC), which replaces the PC in 802.1 le, also has the ability to poll in the CP as well as the CFP.

Stations are also prohibited from transmitting over a defined Target Beacon Transition Time (TBTT), again to address the delayed beacon issue.

Having received a CF-Poll, the STA then has the right to send a single buffered frame (constrained by the explicit TXOP value in the CF-Poll packet, as of 11 e). For HCF, to distinguish from PCF operation, the letters "QOS" preface all CF-PoI1ICF-Ack messages; some additional parameters are employed to link the poll to a specific traffic stream identifier. In HCF, a more dynamic registration process is used than in PCF. In the latter case, a STA sends a TXOP-request to the AP in a Q0S-Info field (QoS being a common abbreviation for Quality of Service) during a preceding TXOP. This preceding TXOP is either a polled transmission or a contended transmission; it will be evident to the reader that the very first TXOP must be obtained through contention. A TXOP can be anywhere between 32 and 8160 s in length, in 32s increments, but constrained as follows: * There is a BSS-wide upper TXOPlimit (broadcast in the beacon) * There is a TXOPlimit[AC] for each AC (subject to the BSS-wide limit) set by the HC (via information fields in the beacon), for use in the EDCA/CP phase of the superframe (note that the HC itself is not bound by this limit) * An explicit poll-specific TXOP can be granted to a specific station for a specific

exchange (via a field in the CF-Poll frame).

Traffic Streams, TS, are also set up in advance via the management plane, to give the HC advance notice of the requirements of the applications in the BSS. TSs are characterised by a data structure called a TSpec, or Traffic Specification, which includes minimum, nominative and peak (where appropriate) values for MSDU size, service interval, data rate, delay and required raw PHY rate.

Further information is available in the standards, specifically section 9.3 of IEEE8O2. 11 2003 referenced above, and in corresponding sections of the IEEE8O2.lle Standard referenced above.

When using PCF (or HCF.in a backwards-compatibility mode), there is a need to determine an optimum set of parameters for the superframe, to obtain the best network performance, for a particular network configuration. When HCF is deployed using the fully-flexible dynamic TXOP allocation scheme this superframe is only notionally required in the sense that it defines a split between polled and contending traffic, which can be used to constrain the number of "polled" TXOPs that are granted during any beacon-period.

A series of publications by Changle Li and colleagues attempt to optimise parameters for the superframe by the discrete selection of pre-defined PCF parameters through pre-defined look-up tables. These look-up tables are indexed by the number of active A/V stations and the maximum allowable delay of the applications and the improvements that result. The look-up values have been found empirically. The publications are: "Self-adaptive transmission scheme of integrated services over an iEEE 802.11 WLAN," (C. Li, J. Li, and X. Cai, Electronics Letters, vol. 40, PP. 1596, 2004); "Performance Analysis of IEEE 802.11 WLAN to Support Voice Services," (C. Li, J. Li, and X. Cai, Proceedings of the 18th International Conference on Advanced Information Networking and Application (A1NA 04), vol. 2, pp. 343 -346, 2004); "A study of self-adaptive transmission for integrated voice and data services over an IEEE 802.11 WLAN," (C. Li, M. Li, and X. Cai, 15th IEEE International Symposium on Personal, Indoor and Mobile Radio Communications, 2004. PIIMIRC 2004., vol. 3, pp. 1922 -1926, 2004); and "Performance evaluation of IEEE 802.11 WLAN -high speed packet wireless data network for supporting voice service," (C. Li, J. Li, and X. Cai, presented at 2004 IEEE Wireless Communications and Networking Conference, 2004. WCNC., 2004).

EP1427150 relates to a method and controller for transmitting multimedia data over WLAN. This document discloses a system where a resource request in the CP results in an allocation in the CFP, in order to allow the system to efficiently facilitate the real time transmission of large multimedia data such as high definition TV.

An alternative solution for improving the selection of superframe parameters to achieve a balance between contention based access and access controller based access is disclosed by the present applicant in GBO5 13533.0. Optimum lengths of the contention free period and contention period are determined by performing a constrained optimisation. The optimisation may be constrained by performance parameters of the system. The optimisation may be a non-linear optimisation, and may employ a barrier method. In preferred embodiments, a superframe controller is provided, including means to determine a performance characteristic, means to determine on the basis of non-linear optimisation superframe characterisation parameters and means for deploying the superframe in accordance with the determined superframe characterisation parameters across the network. Features of this system may be used in embodiments of the present invention, and this is discussed in more detail below.

The following documents discuss further ways of reducing network delays to priority traffic: EP1579631 discusses introducing a contention procedure into the PCF. When the point co-ordinator (PC) poiis a mobile terminal who has no frame to send, the terminal does not respond to the poll and will adjust its NAV (network allocation vector) and set it to a special value that is defined as SPIFS (S-PCF Interframe space). If data is ready on this mobile terminal when it is still in the same contention free period (CFP) and its poli period is over, the terminal can contend for accessing the medium. If it detects the medium to be idle for SPIFS period during the CFP (contention free period), then the mobile terminal gains the control of medium and starts to send the ready data. So the mobile terminal can have another opportunity to gain the control of medium during this CFP instead of the next CFP, and this can reduce the transmission delay.

US69901 16 discusses increasing the overall network throughput over a wireless local area network with switching mode. The dynamic switching between the Distributed Coordination function (DCF) and the Point Coordination function (PCF) access modes is determined according to the load conditions over the WLAN. Stations and access points within a WLAN monitor conditions within the network to determine which access mechanism is most appropriate for the current load conditions. Some factors to consider in determining the load conditions are the number of transmissions, number of receptions and number of collisions.

US20040 160930 discusses a method of transmitting multimedia data over WLAN, guaranteeing the full length of a multimedia point coordination function (mPCF) period.

The rILPCF period length is dynamically allotted, by receiving a request for multimedia resource from a predetermined device connected to the WLAN during a distributed coordination function (DCF) period, when a right to use a channel is distributed through contention. The requested multimedia resource is unilaterally transmitted to the predetermined device during the mPCF period, when the right to use a channel is distributed in a centralized manner.

JP2004187021 discloses a PCF control method for an access point (AP) device, to prevent establishing a longer CP (Contention Period) time than required. The number of transmitting frames which are registered in a priority cue of an AP 1 are counted in a priority queue count. The number of STA (Stations) having the API as a CF-Pollable STA are counted in a CF-Pollable STA count. The CFP time and the CFP period (which defines a period from the CFP start to the next CFP start) are calculated based on the count value from the priority queue count and the CF-PolIable STA count. The calculated values are notified to a MAC protocol processing part at a timing just before the CFP start, thus providing CFP control parameter information.

The following papers discuss the issue of period length reconfiguration in response to devices Q0S requirements.

"Performance evaluation of IEEE 802.11 WLAN -high speed packet wireless data network for supporting voice service," (C. Li, J. Li, and X. Cai, presented at 2004 IEEE Wireless Communications and Networking Conference, 2004. WCNC., 2004) discusses adjustment of a superframe configuration process by empirical means.

"A Medium Access Control Protocol for Delay Q0S Provisioning in Wireless Network", (Li, J., D.-f. Zhao, and B. Hu, IEEE 6th CAS Symposium on Emerging Technologies: Mobile and Wireless Communications. 2004: Shanghai, China) considers ways of adjusting the polling order to offer temporary favour to particular devices.

Ziouva, E. and T. Antonakopoulos, Improved 1EEE8O2. 11 PCF Performance using silence detection and cyclic shift on stations polling, in lEE Proceedings of Communications 2003 relates to detecting periods of silence from certain devices in order to deal with bursty traffic sources.

The challenge with the superframe configuration is selecting the configuration parameters to meet the delay requirements of the polled traffic and the throughput requirements of the contention traffic. Provided the number of contending devices remains approximately constant, the throughput scales linearly with the size of CFPMAX, the smaller the better. Conversely, reducing CFPMAJ( will tend over long term to increase the average delay experienced by the polled traffic -but the relationship is non-linear. In addition varying CFPgp has little effect on throughput, but can affect the distribution of polled packet delays in a wildly unpredictable manner.

Embodiments of the present invention provide further improvements for configuring superframe parameters to improve the service on the network, with a procedure by which an access point can detect failing delay requirements and respond with a reconfiguration process that can rectify it.

One aspect of the present invention provides a communication controller for a communications network with one or more communications stations across a communications medium, the communications network effecting communication by allocating first time periods during which the communication controller is operable to control access to the communications medium and second time periods during which the access to the communications medium is allowed on a contention basis, wherein the communication controller comprises: detection means to detect poorly met delay requirements of delay sensitive traffic, and reconfiguration means to respond to the detection of said poorly met delay requirements by reconfiguration of the length of the first and/or second time periods to reduce the delay to delay sensitive traffic. The communication controller may poii at least one other station during the first time period, and the poll may include a request for delay information. The detection means may receive delay information from one or more station on the network, which may include one or more polled stations.

The controller may receive delay information in as part of the header of a packet from at least one station on the network. The packet header may be a DATA or ACK packet header. The controller may send a delay information request to a polled station in a POLL packet, preferably as part of the poii packet header. The delay information may be provided by one or more flags in the packet header. The flag may be a single bit flag or a multi-bit flag, e.g. a one-bit flag may be used to represent whether or not a delay threshold has been met.

In alternative embodiments, the delay information may be in the body of the packet rather than in the header, or in a combination of the body and the header. The delay information may be provided as a data field or other parameter, or in any suitable alternative form, instead of as a flag.

The delay information may relate to delay timings. For example, this may relate to an end-to-end delay of packets on the network. For example, the actual value of an end-to-end delay may be sent to the controller, or an estimate of end-to-end delay, or information indicating whether a measured or estimated end-to-end delay falls within certain limits. The information relating to an end-to-end delay may be obtained using a net-work packet queuing delay and a polling delay. These may be measured or estimated by the polled station.

The delay information may relate to packet queuing delays. For example, it may comprise delay times or alternative indicators of a delay. For example, the number of packets in the transmission queue of a station on the network may be used as a delay indicator. This may be taken at the end of the first time period.

The delay information may alternatively or additionally relate to the number of packets in the transmission queue of a station on the network. This may be determined at the end of the first time period. For example, this may comprises whether a station still has packets left in its queue after the end of the polling period. In some embodiments, a flag may be set according to whether the packet queue is empty.

The delay information may provide an indication of whether a delay exceeds a threshold. The threshold may be a predetermined fixed threshold or a dynamically adjusted threshold. The delay threshold may be set according to current delay requirements of one or more communication stations on the network. The threshold criteria may indicate whether the delay is an acceptable or unacceptable delay. A flag may be used to indicate failure of the delay threshold condition. The delay threshold may be determined as whether a predetermined proportion of packets have a delay time below a predetermined value.

Delay information may be received from a plurality of polled stations, and used to calculate an overall performance statistic. This may include all stations on the network that are polled by the controller. A single evaluation of the overall network delay may be made by the controller during each superframe, or the network delay may be repeatedly assessed.

The reconfiguration may comprise extending the length of a current first time period and/or reducing the length of the next second time period. The controller may be configured to trigger a catastrophic reconfiguration if there is a predetermined number of failures in a contention free period, e.g. more than one failure, or more than two failures.

The reconfiguration means may be configured to store values of the first and second time periods, to reconfigure the first and second time periods on a temporary basis, and then to revert to the stored values if delay requirements are being met. The reconfiguration may involve reducing the length of a next contention period to a minimum value. In some embodiments, this minimum value may be zero. The reconfiguration may involve reducing the length of a next contention period to a value selected to eliminate the delay backlog in a single superfrarne. The next superframe may continue normally, with the controller continuing to monitor traffic and trigger multiple reconfigurations if necessary.

The controller may be configured to make a semi-permanent change to the first and second time periods if the delays remain unacceptable after a predetermined time or number of cycles. The controller may use a look up table to determine new semi-permanent values of the first and second time periods, said look up table being configured according to a network model.

Further embodiments of the invention comprise a computer program product comprising computer executable instructions which, when executed by a general purpose computer provided with means for establishing wireless communications, cause said computer to be configured either to operate as an apparatus in accordance the invention, or to perform a method according to the invention. The computer program product may comprise a storage medium storing information defining the computer executable instructions. .The computer program product may comprise a computer receivable signal bearing data defining the computer executable instructions.

A further aspect of the invention comprises a station configured to provide delay information to the communication controller. The station comprises a communication apparatus for use on a communications network, the network being controlled by a communications controller according to the invention, wherein the communication apparatus is preferably configured to measure or estimate a network delay, and to send information on the network delay to the communication controller in response to a request from the controller during the first time period.

In preferred embodiments, the increase of CFPMAX is only over that particular superframe. In alternative embodiments, the increase of CFPMAX may be over two or more superframes. The increase may be dynamically variable, depending on the network parameters and the particular requirements at the time.

The present inventors have found that a large short term increase in CFPMAX is more effective than a smaller but semi-permanent increase of CFPMAX because with a semi-permanent increase of CFPMAX, many superframes are required to reduce the average delay. Bursty traffic such as VoIP may only fail delay requirements for infrequent short periods. Ideally, CFPMAX is chosen with long term requirements in mind. Increasing CFPMAX has more long term effect than occasionally skipping the CP. If CFP is large enough for multiple pollings per station, one extra full CFPgp period should be enough to empty packet queues.

Preferred embodiments of the invention use mathematically sound algorithms such as those described in GBO5 13533.0, validated through simulation as the basis for the configuration. However, the present invention also encompasses embodiments using empirical results from simulation-based studies, such as those described by Li in the above listed publications.

Using a reconfigurable mechanism means the system can be fine-tuned automatically without requiring user intervention or maintenance.

To allow the network controller to perform the optimisation, a lookup table (LUT) may be provided. This LUT may provide a range of values of each variable parameter, and the optimised superframe configuration times corresponding to each possible combination of variable parameters. For example, the prior art by Li et al described above provides a lookup table where CFPMAX and CFPyp values can be obtained by looking up a table of values indexed by the current number of active stations (Nv) and the current maximum allowable delay.

Even in embodiments of the invention, where the superframe configuration is calculated rather than found empirically, it may still be advantageous to use iookup tables in order to avoid any delay to the reconfiguration process that is caused by the time taken to calculate the new values.

In other embodiments of the invention, the network model parameters used to perform the optimisation process may also be reconfigured. The network model parameters may include the mean data MSDU size, data rate, A/V MSDU size, data beacon size, slot time, SIFS (short interframe space) time, PIFS (PCF interframe space) time, DIFS (DCF interframe space) time, MAC header length, PLCP (Physical layer convergence protocol) header size, control rate and ACK (acknowledgement) size. Such network model parameters will depend on which version of the IEEE8O2. 11(n) protocol is being used, and on the individual network setup.

A set of network model parameters may be used to calculate a single look up table (LUT). In a simple embodiment of the invention, there may be only a single LUT provided. However, in more complex embodiments, multiple LUTs may be provided, by using multiple sets of network model parameters. The LUT which best matches the current network configuration may beselected, and used for the optimisation of superframe parameters. In further embodiments, reconfigurable LUTs may be provided, where particular parameters used to calculated the LUT can be changed, the LUT may be reconfigured dynamically, and then the reconfigured LUT may be used in the optimisation process.

In some embodiments, quality of services checks may be performed at periodic intervals, and in other embodiments, such checks may be performed at non-periodic intervals. Preferably such checks are carried out for every five to ten superframes. In some embodiments, the quality of service check rate may be a factor of past network performance.

Embodiments of the present invention may be used in a wide range of applications.

Using a reconfigurable mechanism means the system can be fine-tuned automatically without requiring user intervention or maintenance. For example, embodiments of the invention can be used in a general-case re-configurable system for commodity chipsets or in multi-purpose APs. Further embodiments of the invention can provide low silicon usage with a single LUT on a chipset, and target a specific known app1icatio (e.g. in the Toshiba TransCube A/V product) with a single LUT in a product.

In comparison to IEEE8O2. 11, the newer 802.11 e standard aims to support delay sensitive traffic via a similar scheme that uses time division TXOPs rather than a strictly polled system. However the split into delay sensitive and delay insensitive periods is the same, and many aspects of the present invention may be used in 802.1 le systems, or in other variations of the 802.11 standard.

Further preferred features of these aspects of the invention will now be set forth by the following description of specific embodiments of the invention, provided by way of example only, with reference to the accompanying drawings in which: Figure 1 is a schematic timing diagram showing the definition of a superframe in the context of the 802.11 standard; Figure 2 is a schematic diagram of a wireless communications system in accordance with a specific embodiment of the invention; Figure 3 is a schematic diagram of an access point of the system illustrated in figure 2; Figure 4 is a flow diagram showing a process of quality control and superframe reconfiguration in an embodiment of the invention; Figure 5 is a flow diagram showing a short term reconfiguration technique in an embodiment of the invention; Figure 6 is a graph showing the delay versus contention traffic bandwidth for (i) a prior art system, and (ii) a system according to an embodiment of the invention using a short term reconfiguration technique; Figure 7 is a schematic diagram showing implementation options for the lookup table in embodiments of the invention; Figure 8 is a graph showing a fixed naïve superframe end to end delay results; Figure 9 is a graph showing an adaptive superframe end to end delay results; Figure 10 is a graph showing a fixed, perfect a priori knowledge superframe end to end delay results with additional load; and Figure 11 is a graph showing an adaptive superframe end to end delay results with additional load.

Figure 12 is a flow diagram of a process of characterising a superframe in an access controller of the access point illustrated in figure 3; Figure 13 is a flow diagram of a constraint definition sub-process of the process illustrated in figure 10; Figure 14 is a graph of performance of the process of the specific embodiment for different starting criteria; Figure 15 is a flow diagram setting out an exemplary optimisation process.

A communications network used in an embodiment of the invention is illustrated in figure 2. The communications network is an IEEE8O2. 11 compliant wireless network, with an access point 100 and network node A 102, node B 104 arid node C 106. The access point 100 is connected to an external network (such as the Internet), in this example by means of a broadband modem. It will be appreciated that other alternative arrangements can be made for the access point to establish connection to an external source of streaming packet-based data.

The access point 100 establishes wireless communication, in accordance with the 802.lle Standard, with nodes A 102, B 104 & C 106. The access point 100 is thus configured to route data between the external network and the respective nodes 102, 104, 106.

In this embodiment, by way of example only, Node A 102 is a portable laptop computer, Node B 104 is a desktop computer and Node C 106 is a multimedia device (e.g. set-top box) operable in conjunction with a television or hi-fl system. Each of these nodes 102, 104, 106 is equipped with an 802.llg wireless LAN network adaptor and configured to communicate with the access point 100.

Figure 3 shows the structure of access point 100 in an embodiment of the invention. In this embodiment, the access point 100 is equipped with a broadband modem 202 as previously described, to establish connection to the Internet. The broadband modem 202 is connected to a general purpose bus 204, which in turn connects to the components of the access point including working memory (combining RAM and ROM function, as required) 206, a processor 208, a wireless network access controller 230 and a mass storage device 216. The access controller 230 is, in turn, connected to an antenna 212. The working memory 206 may include a software application 230 for controlling the operation of the access point 100.

User operable input devices 220 are further provided, in communication with the processor 208. The user operable input devices 220 comprise any means by which an input action can be interpreted and converted into data signals.

Audio/video output devices 222 are further connected to the general-purpose bus 204, for the output of information to a user. Audio/video output devices 222 include any device capable of presenting information to a user, for example, a speaker and a video display unit.

The operation of the access point 100 will now be described. On the basis of execution of an application residing at a network node, data is retrieved from the external network via a connection established through the broadband modem 202. The manner in which the modem establishes connection is not relevant to the present invention, and can be of a conventional nature.

The retrieved data is then stored in the RAM 206. The access controller 230, determines, on an ongoing basis, access to the network by the access point 100 and by the other nodes 102, 104, 106. The access controller 230 is operable to define the superframe, as previously described in figure 1, but also to determine the length of the superframe and the proportion thereof which is reserved as the CFP.

Figure 4 is a flow chart showing an overview of a control process for reducing the network delay to time-critical traffic, and for configuring network parameters to optimise performance. Two separate processes are operating here. A short term reconfiguration is implemented to avoid short term problems caused by bursty traffic, and a permanent reconfiguration may be implemented if the short term measures are proving to be insufficient, or if the network parameters or network load undergo a significant change.

The process starts at step S300. At step S301, the access point 100 determines the model assumptions for setting up a model of the network. At this point, the access point 100 may calculate one or more lookup tables based on the network model assumptions, or alternatively, these lookup tables may be provided in advance.

Then, at step S302, the access point 100 selects the best lookup table (LUT) for this particular network. In some embodiments, only a single LUT option may be available.

However, in other embodiments, several LUT options may be available, and the access point 100 may select the one that is expected to give the best quality results, or at least select a LUT that is expected to meet the network requirements at the time.

After selection of a particular LUT option, this LUT is used to obtain superframe parameters, such as the maximum length of the contention free period, CFPMAX, and the superframe repetition rate, CPFp.

At step S303, communication between stations on the network begins, using the superframe structure of figure 1 with the superframe parameters determined at step S302. During network operation, Quality Control checks and Quality of Service (Q0S) checks are performed, as now described.

At step S304, a quality control check begins, that is performed during each contention free period. This process involves measurement of network delays at step S305, and this is discussed in more detail below with reference to figure 5. If the network delays are acceptable, at step S306, then the process continues to step S307, where the network delay checks end if the current contention free period has now finished. If the contention free period has finished, the process continues at step S309. If the CFP has not finished, the process goes back to step S305 and the steps of measuring a network delay and determining whether the delay meets requirements, are repeated.

If, at step S306, the network delays were found to be unacceptable, then the process continues to step S307. At step S307, if the delay problems have been persisting over several superframes, the process continues to step S3 12 for a long term reconfiguration.

However, if at step S307, the delay problems have not persisted over several superframes, then the process continues at step S3 10. At S3 10, a short term reconfiguration of the superframe is implemented to reduce the delay to polled traffic.

This short term reconfiguration is for a single superframe only, in this embodiment, and is discussed in more detail with reference to figure 5. After the short term reconfiguration, the process continues at step S309.

At step S309, another Q0S check may be performed. In this embodiment, this Q0S check is only performed every once every five to ten superframes, although different intervals may be used in alternative embodiments. At these time intervals, the current configuration of the superframe is re-evaluated against the system's current requirement. In this embodiment, this Q0S check involves checking to see if there has been any changes in network model assumptions, checking to see if there is any change in variable network parameters, and measuring or estimating a network delay and comparing this to the system requirements. However, in alternative embodiments, not all of these checks are performed.

The first part of this 5-10 superframe Q0S check begins at step S311, where a check is performed to see if there have been any changes in network model assumptions, e.g. traffic profiles, data rates, etc. This information be determined by the AP, or it may simply be received and used by the AP. If there are changes to the model assumptions, the process continues at step S3 12, where a new lookup table is selected if appropriate, or a iookup table is reconfigured accordingly, if appropriate. The process then continues at step S3 14, which is described in more detail below.

Optionally, step S3 11, which checks for changes in network model assumptions, can trigger an on-line generation of new superframe configurations for the new set of constraints in the more complex implementation options. There is a spectrum of increasing complexity implementations here. One LUT may be implemented, which still gives reconfigurability in the dimensions of station-count and delay. Alternatively, multiple LUTs may be used, where each LUT embodies a different set of fundamental model assumptions. A further possibility is the use of reconfigurable LUTs, where one or more sets of data can be regenerated dynamically. This latter embodiment would support general purpose products, e.g. where both video and telephony functions may be required at different times. Software-defined radio type installations could also be supported using re-configurable LUTs, where a new PHY could be installed.

Figure 7 shows the different types of LUT arrangement that may be used during the reconfiguration. The left hand side of figure 7 shows a single look up table. This is particularly suitable for use in systems where conditions are not expected to change drastically, and where it is important to use a minimum amount of silicon area. The central part of figure 7 shows multiple look up tables, where one of these look up tables is selected and paged in. If network conditions substantially change, then a different look up table may be selected in its place, and paged in. The right hand side of figure 7 shows multiple look up tables with dynamic optimisation. The parameters within a selected look up table may be modified dynamically, but if an acceptable optimisation is not possible, a different LUT may be selected and optimised instead.

A further possible embodiment is the use of multi-dimensional LUTs, similar to multiple LUTs with more degrees of freedom. For example, an n-dimensional LUT could cover other variables such as traffic profile or PHY data rates. This would require additional memory, but may be useful in some circumstances.

If there are no changes to the network model assumptions at step S3 11, the process goes on to step S313. Here, a further part of the Q0S check is performed, comprising the following: * Has the number of associated stations changed? * Has the estimated number of contending stations changed? * Have the required delay thresholds changed? Again, in some embodiments, not all of these steps need be performed.

If there has been a change in the above listed parameters, then the process continues at step S314, where the current LUT is used to reconfigure the superframe parameters.

Once a suitable new CFPMAX and new CFP are obtained at step s3 14, using the current LUT, the new superframe configuration is used for network transmissions. In some embodiments, the network delay is then measured, for use in long term configuration of the superframe. In other embodiments of the invention, the superframe can be configured even without a feedback path offered by network delay evaluation, by simply using the QoS checks for changes in system status, e.g. as described in S3 11 and S313.

The network delay may be measured or estimated, for example, by measurement of an End-to-End delay. The End-to-End delay experienced by traffic may be determined based on header time-stamps. Clocks on different stations on the network may be synchronised, albeit with course granularity, to a time-value broadcast in the beacon by the AP. This end-to-end delay measure provides a means of determining how well Q0S promises are being kept. During measurement of the network delay, some smoothing may be performed to ignore transient peaks that are not representative.

An alternative possibility of obtaining a measure of network delay is by measurement of how long a packet has been sitting in a queue local to a network node. The packet is time-stamped when the application presents it, then the time of transmission is also noted, and the time difference is sent to the AP. Using this method, no synchronisation between nodes on the network is required at all.

The performance requirements of the system arc then compared to the measured network delay. If the delay requirements are not being met, the system may proceed to step S312, where the AP 100 may select a different LUT or reconfigure the current LUT to attempt to solve the problem. The process then continues at S314, where the superframe parameters are selected from the LUT, after which the network delay is again measured. Alternatively, if delay requirements are not being met, the reconfiguration process may firstly attempt to satisfy the requirements with the LUT currently in use. Failing that, it may page in alternative LUTs if they are available and suitable. Otherwise, it may choose the least bad configuration available that gets closest to the target performance.

Different types of network delay measurements or estimates may be made for the process in step S305 and for the 5-10 superfrarne Q0S check. Alternatively, in some embodiments, no additional delay measurements are made for the 5-10 superfrarne Q0S check.

The superframe may be temporarily reconfigured, rather than permanently reconfigured, as an emergency measure to attempt to clear whatever backlog has formed.

If the network delay is found to meet the given requirements, then no further reconfiguration is needed, and the process returns to step S304, and continues to perform QoS checks at predetermined intervals.

The results of the Q0S check may be stored in memory, to allow a user to monitor the network behaviour at a later time.

Figure 5 is a flowchart showing a mechanism for collecting delay feedback from each polled station. The process begins at step S400. At step S401, the CFP begins, and the process described in the remainder of the flowchart is continued until the end of the CFP. During the CFP the AP sends a POLL packet to each device in turn, and receives either a DATA packet in return if the station has data ready to send, or an ACK to acknowledge receipt of the POLL packet. This provides ample opportunity to include small amounts of additional information in the headers of these packets. At step S402, a polled packet is sent by the AP to the first polled device. Then, at step S403, the AP receives an ACK or DATA packet back, including a flag indicating the delay status. The delay status is determined by the polled station, for example, as discussed above.

At step S404, the AP determines whether the flag in the ACK!DATA packet header indicates an unacceptable delay. If the delay is unacceptable, the process continues to step S405, where a counter is updated to count the number of unacceptable delay reports in this particular CFP.

If the delay is found to be acceptable at step S404, the process continues to step S406. A POLL packet is sent to the next polled device at step S406, and then an ACK or DATA packet is received back at step S407, with a flag indicating the delay status.

At step S408, the AP determines whether the flag in the ACKJDATA packet header indicates an unacceptable delay. If the delay is unacceptable, the process continues to step S409, where the delay counter, which counts the number of unacceptable delay reports in this particular CFP, is updated.

At step S4 10, if the number of reported delays on the delay counter is not above a predetermined threshold, then the process continues at step S406, where the next device is polled by the AP. However, if the delay counter indicates a delay that is above a predetermined threshold, then the process continues at step S4 11.

At step S4 11, a short term reconfiguration of the superframe is implemented, by extending the length of the current CFP, and reducing the length of the next CP. The process then continues at step S412, where an assessment is made of whether a short term reconfiguration has been needed for the last few CFPs. If not, the process continues at step S406. However, if a short term reconfiguration has been required for several recent superframes, then the process continues at step S4 13, where the long term superframe parameters are re-calculated, taking into account the recent network delays.

After this recalculation, the process continues at step S4 13, where the AP continues to poll the polled devices in turn.

It is typically desirable to minimise the total end-to-end delay of packets sent by delay sensitive applications. The end-to-end delay is the time from the packet first being introduced into the network stack, to the time at which the same packet is removed at the recipient end. For this system of polled stations the main point of variation in this delay occurs in the packet reaching the front of the queue and being polled by the AP.

This delay is can be readily measured internally to one device. Moreover it should be relatively straightforward to place limits on this delay given actual application end-to-end delay requirements.

Thus the current delay on a per-packet basis can be measured in the device. The delay requirement can be quoted as "X% of the packets must have a delay less than D" where X is the tolerance level (for example 95% might be a useful value) and D is the ideal delay. Once the requirement is failed a flag in the header of the DATA packet can indicate this to the AP.

In an alternative embodiment of the invention, a nonempty queue trigger to reconfiguration may be when the end of the polling period is reached and a device still has packets left in its queue. This may be signalled to the AP by a mechanism identical to that in prior art document US69901 16, where each device flags a bit in the header to signify whether its packet queue is empty on each packet it sends to the AP.

The AP may decide, given the frequency of failure and number of devices affected, when to trigger a catastrophic reconfiguration. In one embodiment, the trigger comprises more than one failure occurring in a single CFP.

The catastrophic reconfiguration takes the following form: The CP that would normally follow this one is reduced to its minimum possible value -CPmin. The current CFP extends until the beginning of this small CP placed at the end of the superframe. The next superframe continues as normal, i.e., a CFP followed by the CP. The third step permits the AP to continue to monitor the traffic, triggering multiple reconfigurations as necessary.

A simple adaptive scheme might choose to simply increase CFPMAX slightly on a semi-permanent basis when polled traffic delays are not being met. In practice this would be less effective. However, some of the benefits of the present invention would still be obtained.

The simple adaptive scheme may take many superframes to bring the average delay back down to the required level. Bursty natured traffic, such as voice-over-IP, (VOIP) may only fail delay requirements infrequently and for short periods. The adaptive scheme would need to respond quickly to drop CFPMAX once the delay became too low.

Ideally the CFPMAX should be carefully chosen with long term requirements of both traffic types taken into consideration. Over a long period a slightly high CFPMAX will adversely affect the contention throughput more than an occasional skipping of the CP would. Provided the CFPpjp is large enough to permit multiple pollings per station one extra full CFPp period should be enough to empty the packet queues of all the devices, and put the polled traffic back in a strong position to meet its delay requirements. This idea should be compatible with the lie HCF.

It requires some support in the client devices (maintaining current delay measurements) but otherwise conforms to PCF standard -the AP is free to alter the size of the polled and contention periods on a frequent basis, subject to the minimum sizes mentioned earlier.

Figure 6 gives a comparison between standard behaviour and the mechanism provided by an embodiment of the invention. The simulation results are obtained for a scenario with twelve devices each with a typical level of VOIP traffic. Using a CFPjup of l5Oms and varying the CFPMAX produces a range of average packet delays for polled traffic and levels of bandwidth available to contention traffic. There is a trade-off between increased support for contention traffic and reduced performance of polled traffic.

The upper curve in the graph shows the trade-off given by the standard PCF mechanism. The lower curve in the graph is obtained by applying the reconfiguration procedure outlined in this JAR using a trigger that forces reconfiguration when more than three devices have packets pending at the end of the polling period. This curve lies below the standard one, implying that for the same packet delay a larger level of contention traffic throughput is permitted, and conversely for a fixed contention level a smaller packet delay can be achieved.

Embodiments of the invention may be used with an A/V aware base station. Some aspects may be implemented as enhancements in wireless chipsets found in home A/V equipment.

An example is now given, of a conference room scenario of 41 stations. This example is a simplified case of one of the comparison criteria usage models in IEEE8O2.1 1TGn-UM 2004 (IEEE (Institute of Electrical and Electronics Engineers),"IEEE 802.11 Working Group, Task Group N: IEEE 802.11-03/0802: Usage Models," 2004).

Without the present invention, a person skilled in the art would implement the system to have a fixed superframe configuration arbitrarily chosen by the access point. At design time, a particular configuration of traffic would be assumed, and used in producing the design. For example, in the conference room scenario of4l stations, a lOOms superfranie repetition rate and a 10% polling/contention-free window may be selected on this basis. This configuration may have been arrived at after extensive simulation of the assumed applications and scenarios of the product, and trading off the requirements for each. If the system was them deployed in the conference room scenario, an average end-to-end delay measurement of 6Oms may be obtained.

However, in embodiments of the present invention, the system may be started in the same superframe configuration of 0.1/1 OOms, but may seif-optimise to a different configuration, e.g. 0.3/3Oms, which results in a greatly improved average end-to-end delay result (25ms) for the delay sensitive traffic. There is a corresponding marginal increase in the end-to-end delay for the background traffic, but this is acceptable because the background traffic has no specific delay requirements.

Even if the access point somehow had a priori knowledge of the perfect superframe configuration, or the designer got it perfectly right for this scenario, the present invention still has the clear advantage of its reconfiguration ability.

In figure 8, a graph of end-to-end delay against time is shown for a fixed naïve superframe. The top graph of figure 8 shows the delay-sensitive traffic results, and the bottom graph of figure 8 shows the background traffic results. Note that the y-axis units on the top and bottom graphs are different. The end-to-end delay is about 60 units for delay-sensitive traffic.

Figure 9 shows a graph of end-to-end delay versus time for an adaptive superframe according to an embodiment of the invention. The top graph of figure 9 shows the delay-sensitive traffic results, and the bottom graph of figure 9 shows the background traffic results. The end-to-end delay is about 22 units for delay-sensitive traffic, which is a marked improvement on the system of figure 8.

Figure 10 shows a graph of end-to-end delay versus time for a fixed, perfect a priori knowledge superframe, with additional load. The top graph of figure 10 shows the delay-sensitive traffic results, and the bottom graph of figure 10 shows the background traffic results. The fixed superframe structure has the "perfect" 0.3/3Oms configuration for the initial network conditions, and gives an optimal end-to-end delay ofjust over 20 units. With normal traffic flow, this system behaves in the same way as the adaptive system according to an embodiment of the invention. However, in a deviation from the normal traffic flow, e.g. if a sudden glut of delay sensitive traffic is introduced after a period of time, then the system can no longer cope, and the end-to-end delay for the delay sensitive traffic spirals out of control, to over 70 units. Also, as a result of the fixed configuration, the delay insensitive backgroundtraffic is provided with a low-delay service, which provides no advantage, as this traffic has no delaS' requirements.

Figure 11 shows a graph of end-to-end delay versus time for an adaptive superframe according to an embodiment of the invention, with additional load. The top graph of figure 11 shows the delay-sensitive traffic results, and the bottom graph of figure 11 shows the background traffic results. Where required, e.g. if a glut of delay sensitive traffic appears, the superframe can quickly be reconfigured, bringing the end-to-end delay of the delay sensitive traffic back into the target range ofjust over 20 units. As a trade-off, the background traffic will see an increased delay, but this is irrelevant as the

background traffic has no delay requirements.

Embodiments of the invention may be used with alternative standards, in addition to IEEE8O2. 11. For example, the standard IEEE8O2. lie includes a MAC protocol for use with the IEEE8O2. 11 defined technology.

It is advantageous, although not essential, if the present invention is used with a method of optimisation such as that described in GBO5 13533.0 by the present applicant.

In such an embodiment, certain assumptions are made in order to develop the objective function by which the access controller determines an optimal specification for the superframe. This example is now discussed with reference to an IEEE8O2. 1 lb compliant system as used in the Li et a!. publications referred to above.

It will be clear to the reader of these documents that these are viable parameters for typical systems presently available, and allow a fair comparison of performance of the specific embodiment of the invention with the arrangements offered in those documents. Clearly, it will be evident to the skilled person that an actual implementation will use the parameters appropriate to the implementation in question.

For example, if an aspect of the present invention is being implemented in an 802.lln system then parameters suitable for such technology would be used instead.

For clarity, the general example, with algebraic representations of the parameters, will be set out in due course. The assumptions from the cited prior art set out above, and hence the example parameters used for the remainder of this section, include: * An IEEE8O2.llb system with a data-rate of 2Mbps; * Standardised polled-traffic payload sizes of 200bytes; * Ten data stations each generating frames with a mean size of 1000 bytes at a rate of 7.5 frames per second; * No bit-errors, interference, capture-effect or hidden terminals; and * RTS/CTS handshaking and power-saving disabled throughout.

Some specific parameters used in the present example are given in Table 1: Table 1: Model Parameters Parameter Value Parameter Value Slot 0.O2ms PIFS O.O3ms SIFS O.Olms DIPS O.O5ms PLCP Header! MAC Header 28 bytes 0. 192ms Preamble Mean data MSDU 1000 bytes A/V MSDU 200 bytes Data rate 2Mbps Control rate 1Mbps Beacon 160 bytes ACK 14 bytes From these parameters, the characteristics of a standard data exchange in each phase of the superframe can be identified, as illustrated in step S1-2 of Figure 12.

Using these standard exchanges with their implicit overheads, utility functions expressing the utilisation or wastage of each phase are derived by calling a procedure in step S 1-4. This captures both the inherent efficiency of the exchange and the efficiency of the fit of the exchanges within the superframe structure. This procedure is illustrated as a two part process in figure 13.

Firstly, in step S2-2, the overhead incurred during the polling phase, V, the number of stations to be polled, is expressed in terms of CFPM,, CFP and N, as shown in equation 1: v [o.o2N +{(CFPMAX *CFPpp)_2228Np}/ (1) /(rFp *CFP / UAX kEP Then, in step S2-4, the utilisation of the contention period, L(NCTX), is similarly expressed in terms of CFPMAX, CFPp and NCTX, the number of stations transmitting during the contention period, as equation 2 shows: L(N) -i 1(0674Nrx) +{((l -CFPMAX)CFPREP)-4.978N}/ 1 (2) -[ /((1 -CFPMAX)CFPREP)i These functions give the upper bounds on performance in different configurations, for example in the graph illustrated in figure 12, V is considered for different values of CFPMAX.

Further constraints could, in alternative versions of the specific embodiment, be expressed, such as capturing the minimal sizes of the CFP and CP as specified to the standard (as specified in IEEE8O2.1 1 2003, specifically at paragraph 9.3.3.3.

Specifically, one constraint on the characterisation of the superframe is that the CFP has to be at least large enough to contain polled exchange (CFPo11+Data, CFPo11+Ack) comprising the largest payload possible in each direction, plus a Beacon and a CF-End.

Similarly, the CP has to be large enough to contain an exchange (plus ACK) of the largest payload possible. The largest payload is 2312 bytes in 802.11 b (employing WIEP, as stated in the standard).

Although the standard does not specify clearly the data rates for which transmission is to be enabled, in order to meet the objective of ensuring sufficient space for an exchange by any station, the data rate must be the lowest data rate possible -in this case, 1Mbps. Similarly, the standard does not clearly state whether time must be allowed for inter-frame spacing (IFS); however it would be evident to the skilled person that that the IFS must be included to complete the exchanges successfully. For the purposes of defining the largest possible size of the CFP duration, the standard includes SIPS (and two slot durations) in the calculation of the CP size, which suggests that this is an appropriate arrangement.

Hence, the CFP must be at least: CFPMIN P1FS + Beacon + Data + SIPS + Data + SIPS + CF-End (3) = 0.03 + 1.696 + 18.912 + 0.01 + 18.912 + 0.01 + 0.352 = 39.922 ms And the CP must be at least: CPMIN = DIPS + Data + SIPS + ACK (4) =0.05+0+18.912+0.01+0.304 = 19.276 ms It should be noted that, in the standard, it is stated that this is true "when operating with a CW of aCWmin". The back-off mechanism draws a random value from the range [0, CWJ, where CW is the exponentially increasing value (increasing on each failed transmission) that is initialised at aCWmin, which is 31 for the 802.1 lb PITY (paragraphs 9.2.4 and 18.3.3 of the standard).

These additional constraints suggest that a number of other operational restrictions would, in some specific embodiments of the invention be appropriate, including that CFPp =6Oms, and that combinations such as (CFPp=1OOms, CFPM=O.95) are invalid (as such a combination would leave a contention period duration of just 5ms).

Once all desired constraints have been derived in accordance with step S 1-4, the process continues in step SI -6, by constructing an objective function in accordance with optimisation practice. The derived utilisation functions L and V can be used directly in the following objective function: Obj(x) = (1-L(Nc))2 + (V(Np))2 (5) Then, taking into account the objective function of equation 5, the function of equation 6 below is optimised in step Si -8 to derive the apparently most desirable combination of CFPMAX and CFPp. The manner in which this is achieved will now be described.

In the following, CFPMAX = x, with the assumption that CFPp (=y) is equal to the delay D. Using the data exchange characteristics L and V derived in step S1-2 gives: fo(x,y)=[l_1MsHs)J+[l_ NP(CbCa)]2 (6) On establishment of the optimisation, the optimisation problem is a minimisation of f0(x,y) , subject to: CFP -xy =0 CP -(i -x)y = 0 0 =x =1 0 =y =D Considering the 802.1 lb constraints set out in the four publications by Li et al. above, the parameters have the following values M5 = 0.674 standardised data exchange overhead, ms H =4.978 standardised data exchange, ms N =10 number of data stations Cb=2.228 polled exchange duration, ms C =0.02 polled exchange overhead, ms CFPI = 39.922 CFPMIN at 1Mbps, ms (from equation 3) CP = 19.276 CPM[N at 1Mbps, ms, (from equation 4) D E (75, 87.5, 100, 112.5, ... 200} delay values under consideration N E{2,4,...,20} . numbers of polling stations under consideration = 0.0075 packet rate of the contending traffic, used to estimate the effective number of contending stations The "Barrier Method" is then used to solve the optimisation problem stated above. In simple terms, the barrier method uses an "indicator function", formed by modifying the inequality constraints, to indicate whether the solution is feasible or not. Slight variations are applied to a parameter of the indicator function(s) is (are) successively and iteratively, with each iteration performing Newton's Method for that set of values.

In further detail, the barrier method is applied to this multivariate optimisation problem which seeks global minima of the objective function Due to the constraints on this optimisation, the choice of solution is restricted to a subset of feasible solutions defined by g(x) = 0 and h(x) =0, where both g and h can be multi-valued functions. In the present non-linear case, provided thatf is twice continuously differentiable, solutions can be found using a mixture of analytic and numerical methods.

The optimisation function is re-expressed in vector form (as it is multivariate), and the constraints form the barriers after which the method is named. Gradients and Hessians (second derivatives) of the optimisation function and the barrier functions are taken.

Starting points (xO and yO) are found to initialise the Barrier Method and optimal methods for x and y can then be found.

Referring to figure 15, the barrier method is outlined qualitatively in steps si to s7, and detailed using pseudo-code in Table 1 below.

In step si, the loss function is expressed as in equation 5. In step s2, the constraints of the problem are chosen, as discussed above. In step s3, parameters t, p and inner and outer tolerances of the algorithm e., e are initialised. Typical values might for example be t = 0.5, p =2, e,, = 0.0001, and e = 0.0001.

In step s4, a starting vector x that satisfies the constraints is chosen. In step s5, Newton's method is run until inner tolerance e is met. Once inner tolerance e, is met, outer tolerance c,, is evaluated; if the outer tolerance is also met, then in step s6 the new near-optimal vector x is taken as the output of the process. However, if outer tolerance e, is not met, then in step s7 the logarithmic barrier accuracy parameter t is increased by a factor.&, and the Newtonian process re-started using the last vector x.

Accuracy parameter t provides a trade-off between convergence performance and the number of iterations required for convergence. As t increases it provides a better approximation to an indicator function (see below), but at the cost of slower convergence.

It will be clear to a person skilled in the art however that, particularly in the case of pre-computed sequences, the barrier method may be initialised with a relatively high value of t and eliminate step s7.

Although the barrier method is appropriate to convex problems and the skilled person will appreciate that the present problem is not convex, the barrier method is still appropriately used by suitable selection of constraints.

given strictly feasible x, t > 0, p > 1, o > 0, > 0 repeat 1. Newton's method (x. ej > 0) a. = -V2f (x)' Vf(x) A2 = -Vf (x)H Ix b. quit if A2/2 < rj return x' x c*. line search (determine 9) d. x:= X + /3Lx 2. x x 3. quit if p/t < o 4. t:= Table I: Pseudo code for barrier method optimisation for the cost function of equation 3.

For a given set of constraints (specifically the aforementioned 802.1 lb parameters and the assumptions in the papers by Li et al.) ranges of values are then produced.

The access controller 230 is operable to apply the above described processes to operating parameters during operation of the network. In that way, changes to the operating parameters can be responded to by adjustment, if optimal to do so, of either or both of the superframe repetition period and the proportion of the superframe reserved for the CFP.

The access controller 230 of the described specific embodiment is shown as implemented as a discrete unit, but it will be appreciated that it could be integrated with one or more other functional units of the access point. Further, it would be possible for the function of the access controller to be distributed. The non-linear optimising determination functionality could be provided in one node, while the superframe definition functionality could be provided by entirely separate means.

The access controller is shown as implemented by hardware, and particularly by integrated circuit (IC) devices. However, it will be appreciated by the skilled person that some or all of the function of the access controller could optionally be implemented in a general purpose hardware unit configured by a suitable software product. Such a software product could be introduced by means of optical (or other) storage media, by means of a signal such as a download, or by any other way of enabling software to become stored on a computer apparatus. In such circumstances, the suitable computer apparatus could be IEEE8O2.1 1 compliant, or even IEEE8O2.lle compliant, with further functionality in accordance with a specific embodiment of the invention to be provided by means of the software configuration.

Whereas the described embodiment illustrates operation of the optimisation process by the access controller in a dynamic environment, an alternative embodiment could provide a memory resource, such as a look up table, storing values of CFPMAX and CFP for all possible configurations of the parameters. The currently most appropriate values of CFPMAX and CFPpjp could then be looked up by the access controller 230 and applied to the system.

For the avoidance of doubt, the present invention is not limited to applications involving 802.1 le technology -it can equally be applied to 802.11 technology involving PCF.

Further, it can be applied to any other communications technology in which a balance is struck between a period wherein access is centrally controlled and a period in which access is contended for. Although a PCF solution has been described above, embodiments of the invention also give backward compatibility with legacy equipment.

A PCF solution is likely to be of lower complexity than a full 802.lle HFC solution, and thus to use less compute power and less silicon. Mechanisms described in the above PCF solution may also be used in an extension of these embodiments to a full HCF-style TXOP grant.

The present invention further includes a general purpose computer, including wireless communications means for transmitting and receiving wireless communications signals, which is configured by computer executable instructions to operate as either an access controller or a superframe controller as described above. The computer executable instructions can be introduced as a computer program product, storing information defining such computer executable instructions. The product can comprise a storage medium, such as an optical or magnetic disk, or a signal, such as an internet based download.

Further variations, modifications and additional features will be apparent to the skilled man considering the above disclosure and no statement above is intended to limit the scope of protection sought for the invention, which is to be determined by reference to the appended claims, interpreted in the light of, but not specifically limited to, the above

description of specific embodiments.

This invention is particularly useful for A/V aware base stations.

The communication controller may comprise means for repeatedly determining proposed optimum lengths of said first andlor second time periods using a current set of network parameters; and configuration means for repeatedly reconfiguring the length of said first andlor second time periods according to said proposed optimum lengths. Said current set of network parameters may include a current estimated or measured network delay time.

The network delay time may be determined according to network traffic queuing times.

The network delay time may be determined according to end-to-end network traffic delay times.

The current set of network parameters may include current delay requirements of communication stations on the network. The current set of network parameters may include at least one of a current number of contention-free stations on the network, a current estimated number of contending stations on the network, required delay thresholds, traffic profiles, data rates and other network model assumptions.

The reconfiguration may include reducing the contention-free traffic delay and increasing the contentious traffic delay in response to a measured or estimated bottleneck in the contention-free traffic.

The communication controller may be adapted to use a look up table to determine said first and second time periods, said look up table being configured according to a network model. The communication controller may be adapted to select one of a plurality of lookup tables according to a current set of network parameters. The communication controller may be adapted to reconfigure the lookup table during operation of the network in response to actual or expected changes in network parameters. The communication controller may be configured to use measured end-to-end delay values to trigger an emergency recovery mechanism where the first and second time periods are temporarily configured to prioritise contention-free traffic more.

The communication controller may be adapted to perform a registration process with stations requiring contention free services, and to maintain a record of the number of stations currently requiring delay-sensitive services. The communication controller may be further adapted to perform a de-registration process with stations that no longer require these services.

The communications controller may be adapted to obtain an estimate of the number of non-contending stations using the measured network traffic flow.

The means for determining optimum lengths of the first and second periods may comprise means for performing a constrained optimisation. The means for determining optimum lengths may comprise means for performing an optimisation constrained by performance parameters of the system. The means for determining optimum lengths may comprise means for performing a non-linear optimisation. The means for determining optimum lengths may comprise means for performing an optimisation employing a barrier method. The means for determining said proposed optimum lengths may comprise storage means storing predetermined data calculated on the basis of a prior optimisation, for reference with regard to one or more look-up parameters.

The communication controller may be IEEE8O2. 11 compliant. The communication controller may be adapted to re-configure a fixed PCF superframe structure.

Claims

CLAIMS: 1. A communication controller for a communications network with

one or more communications stations across a communications medium, the communications network effecting communication by allocating first time periods during which the communication controller is operable to control access to the communications medium and second time periods during which access to the communications medium is allowed on a contention basis, wherein the communication controller comprises: detection means to detect poorly met delay requirements of delay sensitive traffic, and reconfiguration means to respond to the detection of said poorly met delay requirements by reconfiguration of the length of the first and/or second time periods to reduce the delay to delay sensitive traffic.

2. The communication controller of claim 1, wherein the detection means comprises means for receiving delay information from at least one other station on the network.

3. The communication controller of claim 2, wherein the detection means comprises means for receiving said delay information in a packet header from at least one other station.

4. The communication controller of claim 3, wherein said detection means comprises means for receiving the delay information by detecting one or more flags in the packet header.

5. The communication controller of any one of claims 2 to 4, comprising means to poii at least one other station during the first time period, the poll including a request for delay information.

6. The communication controller of any one of claims ito 5, wherein the delay information comprises information relating to an end-to-end delay of packets on the network.

7. The communication controller of claim 6, wherein information relating to an end-to-end delay is obtained using a network packet queuing delay and a polling delay.

8. The communication controller of any one of claims 1 to 7, wherein the delay information comprises information relating to network packet queuing delays.

9. The communication controller of any one of claims 1 to 8, wherein the delay information indicates whether a delay exceeds a threshold.

10. The communication controller of claim 8, wherein the threshold is determined as whether a predetermined proportion of packets have a delay time below a predetermined value.

11. The communication controller of any previous claim, wherein delay information is received from a plurality of polled stations, and used to calculate an overall performance statistic.

12 The communication controller of any previous claim, wherein the reconfiguration comprises extending the length of a current first time period and/or reducing the length of the next second time period.

13. The communication controller of any previous claim, wherein the reconfiguration means is configured to store values of the first and second time periods, to reconfigure the first and second time periods on a temporary basis, and then to revert to the stored values if delay requirements are being met.

14. The communication controller of claim 13, wherein the reconfiguration comprises reducing the length of a next contention period to a minimum value.

15. The communication controller of claim 13 or claim 14, wherein the next superframe continues normally, with the controller continuing to monitor traffic and trigger multiple reconfigurations if necessary.

16. The communication controller of claim 1, configured to trigger a catastrophic reconfiguration if there is more than one failure in a contention free period.

17. The communication controller of claim 13, configured to make a semi-permanent change to the first and second time periods if the delays remain unacceptable after a predetermined time or number of cycles.

18. The communication controller of any previous claim, adapted to use a look up table to determine new semi-permanent values of the first and second time periods, said look up table being configured according to a network model.

19. A method of controlling access to a communications medium, comprising establishing a data frame structure for successive definition in the communications medium, the data frame structure defining a first period during which access to the communications medium is controlled centrally and a second period during which access to the communications medium is allowed on a contention basis, further comprising detecting poorly met delay requirements of delay sensitive traffic, and responding to the detection of said poorly met delay requirements by reconfiguration of the length of the first and/or second time periods to reduce the delay to delay sensitive traffic.

20. A computer program product comprising computer executable instructions which, when executed by a general purpose computer provided with means for establishing wireless communications, cause said computer to be configured either to operate as an apparatus in accordance with any of claims 1 to 18, or to perform the method of claim 19.