WO2006092583A1

WO2006092583A1 - Shared medium network

Info

Publication number: WO2006092583A1
Application number: PCT/GB2006/000712
Authority: WO
Inventors: Vasilios Anastasios Siris; Costas Courcoubetis
Original assignee: British Telecommunications Public Limited Company
Priority date: 2005-03-01
Filing date: 2006-03-01
Publication date: 2006-09-08

Abstract

A shared medium access network in which the amount of time for which each transceiver is granted access to the shared medium varies inversely to the transmission rate of the transceiver. In some embodiments, a congestion price is calculated which results in the cost of transmission for a given time period by each wireless transceiver being greater for greater transmission rates by the transceiver.

Description

SHARED MEDIUM NETWORK

The present invention relates to a shared medium network and to methods of operation of a shared medium network.

Shared medium networks, in which transceivers send and receive information across a shared medium by modulating electromagnetic carriei ;s are well-known. There are examples of both shared medium networks with fixed transceivers (e.g. Ethernet local area networks) and with portable or mobile transceivers (e.g. wireless local area networks, mobile phone networks). In most countries, each wireless network technology is assigned a band of frequencies within which it must operate. Network operators are painfully aware of the value of a right to use such a band of frequencies following the spending of billions of pounds on licences to use spectrum made available for third- generation mobile phone networks.

Thus, especially in the case of wireless networks, there is a clear need to maximise the utilisation of such spectral resource. Various methods are used to divide the available spectral resource between users. All cellular mobile phone systems re-use frequencies in cells which are sufficiently remote from one another. In addition to such geographical re- use, the US second-generation mobile phone network (and all third-generation mobile phone networks), use Code Division Multiple Access technology to divide the local frequency band between users. The second-generation mobile phone technology used predominantly in the rest of the world, GSM, divides the frequency bands available locally using a combination of frequency division modulation (different phones in the same cells transmit signals in different 25kHz-wide frequency bands) and time division modulation (each phone uses one of eight available timeslots in each 25kHz wide frequency band).

A problem with all these forms of allocation of spectral resource is that bandwidth available in channels which are not being used is wasted.

This wasting of bandwidth was recognised some time ago. An early network which overcame it was developed by Norman Abramson at the University of Hawaii. He proposed that each mobile terminal should transmit using the entire frequency band made available to the network whenever it had data to send. This idea was embodied in the ALOHA system. If the terminal's transmission of a frame interferes with another transmission (in an ALOHA system, the terminal can tell if its transmission is interfered with by listening to the transmission itself), the terminal waits a random length of time before re-sending the frame. Systems in which multiple users share a common channel in such a way that collisions can occur between the transmissions from different transceivers are known as contention systems. This can be seen as a disordered form of time-division multiplexing.

Various improvements in contention systems have been presented over the years. One improvement was to introduce some order into ALOHA's time-division multiplexing by having the terminals send frames in system-wide time-slots. This means that frames are not destroyed owing to a slight overlap between them. It improves the efficiency of utilisation of the available spectral resource further.

The next step from there was to add so-called 'carrier-sense'. When carrier-sense is used, terminals listen to find whether another terminal is transmitting when they have data to send. If another terminal is transmitting, the terminal waits a random length of time before again checking whether another terminal is transmitting. However, the problem of transmissions colliding still exists. If, on sensing a collision, the sending wireless devices terminate the transmission of the current frame, the scheme is known as Carrier Sense Multiple Access with Collision Detection (CSMA/CD) - otherwise it is known as Carrier

Sense Multiple Access with Collision Avoidance (CSMA/CA).

IEEE 802.11 wireless local area networks can switch between a contention mode and a non-contention mode. In the former mode, carrier sense is used (an idea which mirrors the use of carrier-sense in local area networks which operate in accordance with IEEE's 802.3 standard - which share the bandwidth of a transmission line). In the latter mode, a mechanism more like that used in token ring networks (another IEEE standard - this time 802.5) - each terminal is told by the access point when it can send.

Another feature of 802.11 networks is that they are multi-rate networks - various schemes are used to allow the transceivers to transmit data at different rates - 802.11a transceivers vary the number of modulated carrier signals - each carrier signal being at a different frequency with the frequency range assigned to the 802.11a network. 802.11b wireless devices can transmit at four different rates, namely 1, 2, 5.5 and 11 Mbits/s. One variant of the 802.11 standard currently being worked on is the 802.11e standard which introduces the idea of differentiating between users in accordance with the quality of service they require from the wireless network.. In order to provide a different quality of service to different users, it is necessary to divide the available spectral resource unevenly between the users.

In 802.11e's contention mode, the mechanism for doing this is to vary the contention window - the length of time a user waits after attempting to send a message, but finding the carrier busy. As explained above, the length of contention window is random - however, it is chosen between a lower bound and an upper bound. It is the lower bound that is set in networks operating in accordance with the 802.11e standard.

In a paper entitled 'Providing Throughput Guarantees in IEEE 802.11 wireless LANs', presented at the 18th International Teletraffic Congress (ITC-18) in North Holland, 2003, A. Banchs and others calculated the throughput that a wireless device operating in an IEEE 802.11e network would receive if its probability of transmission within- a given timeslot was weighted by a value Wj.

However, that work was based on an assumption that each device would send at a uniform rate. This assumption leads to a result which does not allocate the wireless spectral resources available between wireless devices as efficiently as the present invention. Although envisaged in work on 802.11e wireless LANs, the present invention has broad utility - being useful in any network whose transceivers share a shared access medium and can operate at a plurality of rates.

US patent application 2003/0081628 recognises that 802.11 networks which allow user devices to operate at different bit-rates are operated most efficiently if slower rate users are given less time to occupy the shared channel than faster rate users. This is achieved in two ways - either by having all users send packets of fixed duration, irrespective of the rate of the user device, or by making the contention window for slower rate devices proportionally longer - this providing another method of limiting the time for which the slower rate devices occupy the shared channel. In one embodiment, these measures are only incrementally applied as congestion builds up in the shared medium. Neither of the above proposals can provide a fair allocation of resources in a multi-rate network which provides different levels of quality of service to different users.

According to a first aspect of the present invention there is provided a communications network comprising:

a shared access transmission medium;

a plurality of network devices arranged in operation to send digital signals using said shared access transmission medium, said digital signals being sent at a bit-rate selected from one of a plurality of available bit-rates;

said plurality of network devices being arranged in operation to occupy said shared access transmission medium for less time when they are operating at a lower bit-rate.

By reducing the time for which a device occupies a shared access medium responsive to that device operating at a lower bit-rate, more efficient allocation of that shared access medium between devices is achieved.

In some embodiments of the present invention, said network further comprises a congestion monitor operable to broadcast, on said shared access medium, a signal indicative of the level of congestion in said shared access medium;

one or more of said network devices being arranged in operation to react to a rise in the level of congestion indicated by reducing the time for which they occupy said shared access medium, said reduction being greater when said one or more network devices are operating at a lower bit-rate.

Often, only one of the wireless devices is connected to a network (to which all the other devices require connection) via a transmission line. The congestion monitor might form a component of that device, or might alternatively be a separate device.

Preferably, said signal indicative of the level of congestion comprises two components, one relating to the cost of congestion independently of the transmission rate of a wireless device, the other relating to the cost of congestion which is dependent on the transmission rate of a wireless device. Such a formulation leads the network to an efficient operation.

Where the network is a packet network, congestion can be signalled by marking packets sent using the shared access medium. This has the advantage that this can easily be combined with a system in one or more connected packet networks which similarly label packets in response to congestion in that network.

In many embodiments, said one or more network devices alter one or more operating parameters in dependence upon the bit-rate at which said one or more network devices is operating. That operating parameter might be the contention window. As mentioned above, the contention window is often chosen as a random value between two defined extremes. The operating parameter might comprise either of these extremes or an average value to which a random variation is to be added / subtracted. Instead of the contention window, where the time available for access to the shared medium is organised into slots, then the probability of transmission within a slot might be used.

In preferred embodiments, said plurality of devices are arranged to update said operating parameter in. response to a predetermined change in the number of devices using said shared access medium. This has the advantage that the amount of processing and signalling required in order to achieve the benefits of the present invention is reduced.

According to a second aspect of the present invention there is provided a method of operating a shared access medium network comprising a plurality of transceivers in communication via said shared access medium, said method comprising the steps of:

altering the rate of transmission of one of more of said transceivers; and

concomitantly altering operating parameters of said one or more transceivers so as to decrease the time for which the transmissions of said transceivers occupy said shared access medium on the rate of transmission of said one or more transceivers falling.

At this point, it will be useful to explain how the present inventor applied micro-economic theory in order to arrive at a result which underlies the efficient use of spectral resource in the wireless network specifically described below. The present inventor began by working out an expression for the average throughput Xj of a wireless device in a multi-rate wireless network as shown in Equation (1) beiow.

Equation (1)

In equation (1), P₁ is the probability of transmission in a given timeslot by wireless device i, P is the probability of any of the wireless devices associated with the network transmitting within a given timeslot, Tj^suc is the duration of a successful transmission by the jth wireless device, T^∞I is the duration of a collision between two wireless device transmissions (note that this is independent of which wireless device is transmitting here - a situation which arises in multi-rate networks which use Request To Send and Clear To Send packets to get around the hidden station problem), P* is the probability that any wireless device other than device k transmits in a given timeslot, and L is the average length of a transmission (in Mbits when the throughput X| is in Mbit/s for example).

Equation (1) derived by the inventor captures a known property of 802.11 networks - namely, a wireless device with a small transmission rate leads to decreased throughput not only for that wireless device, but also for all other wireless devices in the same network, independently of their transmission rate. Equation (1) shows that this is because a small transmission rate for some wireless device j results in a large value for T_j ^suc which in turn results in a small throughput for all wireless devices using the network.

The inventor then goes on to apply micro-economic theory. He starts by assuming that users behave elastically and attribute a value (in economic terms a 'utility') to different amounts of throughput. Utility is here referred to by the function Uj(Xj). As can be seen from Equation (1) above, X₁ is itself a function of p_r, the probability that wireless device i gets to transmit in each timeslot. The next step in applying micro-economic theory is to allocate resources between users (wireless devices) so as to maximise the social welfare of all users. In other words, the aggregate utility of all users should be maximised.

So, the problem is to maximise ∑UhcJ over the interval {pi >= 0, i e N} where N is the

set of wireless devices associated with the network. The solution to that problem requires:

_ ^dΣi U,^(χ) ₌ dτ Ur_t(W_r) ₊ γ ar U/J.(VvCJ.)J _{= Q VJ e N} Equation (2)

⁸P₁ ⁹P₁ ^{J« 5}P; ^{' •}

Substituting the expression for throughput in Equation (1) into Equation (2) gives the result:

Equation (3)

The inventor then approximates E[T] by:

E[T] = (1 - P) Y^ _PjT*^uc 4- P²T^co1 + I - P . j

Equation (4)

Next, the inventor defines the following user problem:

maximise U₁(Xi) - faV^uc +X₂)Pi over p, >=0 Equation (5)

This gives the necessary condition for the user maximisation problem as

Equation (6)

The user problem (i.e. Equation 6) and the global optimisation problem (Equation 3) coincide when:

Equation (7)

Equation (8)

In equation (5), the congestion price is in terms of the transmission probability p,. Substituting the expression for throughput - i.e. equation (1) - into equation (5) and combining the result with equations (7) and (8) above, gives the following expression for the user problem:

maximize #^"»(#«) - (βiTf^uc + p&)xi over pi > 0 ,

•-_, . Equation (9)

In this case, for the user and global problems to coincide, the following expressions apply for the two components of the congestion price:

Equation (10)

Equation (11)

The congestion price in equation (9) contains two components: The first component P₁Tj sue contains the factor μ-i which depends on the level of congestion in the wireless channel, and the duration of a successful transmission Tj^suc. The second component μ₂ is related to the level of congestion in the wireless channel. The interpretation of the above is that the congestion cost for a wireless device depends, in addition to its throughput, also on the duration of the successful transmission interval, hence on the wireless device's transmission rate. For wireless devices with the same throughput, the wireless device with the lower transmission rate, which will have a longer successful transmission interval, will encounter a higher congestion cost; the reason being that a higher successful transmission interval results in the wireless channel being occupied for a longer time. The relative importance of the successful transmission interval is determined by the ratio μ^.

In cases where the variation of utility with throughput is known or approximated in advance by a differentiable function, then the above equations can be taken further to give a useful value for various operating parameters of transceivers in the network.

For example, if the utility function of each of the transceivers is taken to be:

Ui(Xi) = W) In Xj Equation (12) (where the W₁ is a weight or a willingness-to-pay factor), the sum ∑£/' D . can be

approximated as:

Equation (13)

when the following approximation for the throughput Xj for wireless device j is used:

Equation (14)

The expression for the sum ∑ZT_jp . in equation 12 can be substituted into the

expressions for the optimal congestion price components in Equations (10) and (11) to give the following expressions for the optimal congestion price components in the case that each transceiver has the utility function given in equation (12):

Equation (15)

Equation (16) Given the way in which the congestion price is calculated in terms of its two components (Equation (9)), differentiating the utility function in Equation (12) leads to the conclusion:

^'

Expanding that result using equations (15) and (16) above and placing the resulting expression for throughput into Equation (1) - which gives the throughput of an 802.11 transceiver - leads to an expression for the optimal probability of transmission for each wireless transceiver - namely:

Equation (17)

The last expression can be used to compute the optimal transmission probabilities for the wireless devices, based on the .willingness-to-pay factors, the successful transmission duration for each wireless device T?"⁰, and the collision duration T⁰⁰'. The optimal minimum contention window CW^_j can then be computed.

There now follows, by way of example, a description of specific embodiments of the present invention, in which embodiments the above results are used in order to allocate the available spectral resource more efficiently between users than has hitherto been the case. The description is to be read in conjunction with the attached drawings in which:

Figure 1 is an illustrative diagram of an 802.11e wireless network having an access point in communication with three wireless transceivers;

Figure 2 is a timing diagram of the transmissions of the wireless transceivers and access point;

Figure 3 is a table of transceiver sending rates maintained by an electronic memory in the access point; Figure 4 is a flow-chart showing additional processing, carried out by the access point once per second in accordance with a first embodiment of the present invention;

Figure 5 is a table stored in each of the wireless devices in a second embodiment which sets out the value placed by a user on the different levels of throughput obtainable from the wireless network;

Figure 6 is a flow-chart showing additional processing carried out by the access point in accordance with a second embodiment of the present invention;

Figure 7 shows processing carried out by each of the wireless devices in reaction to receiving two congestion price components from the access point.

Figure 1 shows an 802.11b wireless LAN installed in a building 10 to provide a 'hot-spot' of the type now commonly found in airport lounges, within office buildings and at cafes and the like. Although based on 802.11b technology, the wireless LAN additionally allows the setting of different contention windows for the different wireless devices.

The LAN is provided using an access point 12 located inside the building 10, that access point 12 being connected via a broadband Digital Subscriber Link (DSL) connection 13 to a DSLAM 20 (DSL Access Multiplexer) operated by a local internet service provider. The DSLAM 20 is in turn connected via a wide area network 15 to a server computer 22 providing a link to the Internet 24. The devices outside the building 10 are configured and connected in a conventional manner which will be familiar to those that provide Internet connectivity. In particular, the server computer 22 is responsible for generating charging data for and authenticating users of the wireless LAN.

Within the building 10, three wireless devices are connected to the Internet 24 via a radio link from them to the access point 12. Two of these wireless devices are laptop PC- compatible personal computers equipped with a BT Voyager 1060 Laptop Adapter available from British Telecommunications pic. The software in those personal computers is modified to cause them to carry out the additional functionality described below. The third wireless device is a mobile communicator - such as a Nokia 9500 communicator which operates in accordance with the 802.11b standard, but additionally has enhanced operating software in order to incorporate the advanced functionality of the embodiments described below.

The access point is also compatible with the 802.11b standard, and has enhanced operating software as will be described below.

Those skilled in the art of designing such access points and wireless devices will have little difficulty in altering their products to accord with the specifications that follow.

Figure 2 shows the length of collision and successful transmission intervals in the network illustrated in Figure 1. Each of the three or four-letter codes used in Figure 2 represent a time period for which the signals associated with the corresponding feature occupy the spectral resource (the spectral resource in this case being the 22MHz frequency band that this particular 802.11b LAN adopts when started-up).

The wireless LAN in Figure 1 , being 802.11b compatible, uses a carrier-sense technique to reduce the chance of collisions between transmissions from different wireless devices. This involves listening to signals in the 22MHz channel for a period known as the interframe spacing. Providing the channel is quiet for that long, the wireless device can transmit.

Before transmission of a data frame is begun, a wireless device must wait for a length of time known as the Distributed Co-ordination Function Interframe Spacing (DCF-IFS or DIFS). This is 50 microseconds long. However, once the transmission of a data frame has started, a Short Interframe spacing (SIFS) - 10 microsends long - is used instead.

RTS (Request to Send) and CTS (Clear to Send) have been mentioned above. The first is a short signal sent out by a wireless device in order to alert nearby wireless devices to the fact that it is about to transmit. The information includes an indication of how long the frame to be sent is. Any wireless device receiving this frame should avoid transmitting for long enough to allow the frame to be sent. The transmission of an RTS frame lasts for 160 microseconds. The intended receiver responds to the RTS signal by transmitting a Clear to Send signal - copying the data in the RTS signal concerning the length of the frame to be sent - again any wireless device that hears the Clear to Send signal is expected to avoid transmission for long enough to allow the transmission of the frame? The transmission of the CTS signal lasts for 112 microseconds.

Any transmission includes a physical layer overhead, PHY. In the present embodiment, each wireless device and the access point use a long Physical Layer Convergence Protocol preamble which occupy the available frequency band-for 192 microseconds.

The sending of data is acknowledged by the receiver using an acknowledgement packet (ACK) which occupies the available bandwidth for 112 microseconds.

All the above signals are sent at a 1 Mbits^"1 data rate, irrespective of the rate the wireless device is adopting to send data to the access point. In the 802.11b network illustrated in Figure 1 , the data rate can take one of four values, namely 1 , 2, 5.5 and 11 Mbits^"1. As those skilled in the art will know, the transmission rate is varied in dependence on the condition of the radio link between the wireless device and the access point. The variation in the condition of the radio link can be caused by a variety of factors. The throughput ■ depends in part of the data transmission rate, but also on the contention window value used by the wireless device.

Assuming a data frame length of 1500 bytes, and the use of the RTS/CTS procedure, then the duration of the collision and success intervals shown in Figure 2 is as given in table 1 below:

It will be realised from, the above discussion of Figure 2, that the variation in success intervals with transmission rate is entirely due to the time taken to transmit the user data (this being the only part of the transmission which takes place at whatever rate the wireless device chooses to operate at). The access point 12 stores in its memory a rate table setting out, for each wireless device (14,16,18) associated with it, the transmission rate at which the wireless device is sending data, and the corresponding successful transmission interval. This table is shown in Figure 3. Also stored in the table is a relative weight which indicates the value the user of each wireless device places on usage of the wireless network. This data might be obtained from the user of each device at the time they connect to the wireless LAN for example.

In the first embodiment, the access point carries out the additional processing shown in Figure 4. This processing is triggered once per second (step 40).

The additional processing begins with the access point 12 estimating the current aggregate transmission probability in the wireless LAN. The access point 12 is of course aware of the length of its contention window - and from that can calculate its transmission probability in accordance with the equation:

P_*. =

Equation (18)

where CW_min,_AP is the length of the access point's contention window (expressed in units equal to the slot length - i.e. in units 20 microseconds long).

Having calculated its transmission probability, p_Ap, the access point finds the throughput of its transmissions (R_Ap), and the total throughput (R) of the wireless LAN - and then calculates the aggregate probability of transmission (P) by any of the devices 14,16,18 using the fact that aggregate throughput should relate to aggregate transmission probability in the same way that its throughput relates to its transmission probability - i.e.

/ R Equation (19)

The calculation of the aggregate transmission probability using equation (19) forms step 42. The calculated aggregate transmission probability P is stored by the access point 12. In step 43, the weighting factors W_j of all wireless devices currently using the wireless LAN are summed to arrive at a total weight. As mentioned above, these weights might be supplied by the wireless devices at the time they connect to the network. Alternatively, the weights might be set by the application running on' the device which requires the services of the wireless LAN at the time the application is started up.

In step 44, the access point 12 calculates the average time interval (an average of the collision interval, the success interval and the idle interval weighted by the probability of those intervals occurring) for the wireless network as it is currently operating. Initially default values of the transmission probability (as seen in the fourth column of Figure 3) for the wireless devices 14,16, 18 are used. The calculated average time interval E[T] is stored by the .access point 12.

The access point then initialises a counter variable i to zero (step 46). Thereafter, the access point 12 repeats a loop of instructions (steps 48 to 56) for each of the wireless devices 14,16,18 associated with the access point 12.

In step 50, the access point 12 reads from its rate table (Figure 3) to find the successful transmission interval for each of the wireless devices (14,16, 18).

In step 51 , the access point 12 reads from its rate table (Figure 3) the weight Wj the user of device i attaches to quality of service from the wireless network. The access point then divides this value by the total weighting found in step 43 to arrive at a relative weighting for this device. In this particular embodiment, average throughput is the quality of service parameter which differs between devices. However, in other embodiments different quality of services measures might be used.

In step 52, the access point 12 uses equation (17) to calculate an updated transmission probability pi for the ith wireless device.

The contention window for the wireless device currently being considered is then calculated using equation (18), stored in the rate table (the fifth column of Figure 3) and transmitted to the wireless device in a so-called 'beacon' frame. On receiving that beacon frame, the wireless device will also store the contention window and use that value in determining how long to wait after finding the 22MHz channel busy before sensing the channel once again.

The loop of instructions then finishes with a test (step 56) to establish whether all the wireless devices have been updated in the current frame.

It will be seen (from the application of equation (17)) how the transmission probability values calculated for the wireless devices in this first embodiment will be lower for wireless devices operating at lower bit-rates (and hence that its contention window will be correspondingly longer). However, if the user of the wireless device has indicated that he or she attaches a relatively high value to throughput obtained through the wireless LAN, then this will tend to result in the quality of service being improved for that user. This leads to a fairer (in the sense that it takes into account the relative value the user places on quality of service from the network) allocation of the time for which the wireless devices are able to occupy the available 22MHz channel than is the case in conventional wireless networks.

A second, preferred embodiment of the present invention will now be described in relation to Figures 5 to 7. This embodiment uses the same hardware as the first embodiment, and also is based on 802.11b technology but provides the access point 12 and the wireless devices 14,16,18 with additional functionality as described below.

The first embodiment described above assumed a logarithmic form of the user utility versus throughput function as seen in equation (12). The second embodiment described below can be used in relation to any form of utility v. throughput data. It utilises congestion prices calculated by the access point 12 in accordance with equations (10) and (11) above to maximise the aggregate throughput of the wireless network.

The utility v. throughput data is stored in a utility table shown in Figure 5A. It expresses the value a user places on each of the transmission rates enabled by the 802.11b network of Figure 1. This data could be gathered from a user-edited profile, or could be set by the application (for example, a video player application might place a much higher value on the higher throughputs than a voice telephony application). ^' The additional processing carried out by the access point 12 in this second embodiment will now be described with reference to Figures 6 and 7. The process starts with the setting a value M to zero (step 60). This is in preparation for the process which loosely corresponds to the procedure known in economics as 'tatonnement' or groping - where a market's reaction to various prices is tested prior to deciding upon market-clearing price at which the level of supply equals the level of demand.

The value M is the value that is changed in this process - it corresponds to an estimate of

the value ^J /E\TΛ ' ^Altn0U9^{n tne u}'i P^{art of tnis} expression could be calculated from the table shown in Figure 5, E[T] and the pj values are less easily acquired - in the first embodiment an initial estimate was used - in the present case, tatonnement is used to

arrive at a value for the

/_VXTΛ ^wnicn maximises the throughput through the network.

The first congestion price components μi and μ₂ sent (step 64) to the wireless devices 14,16, 18 are zero (this follows immediately from M=O and equations (10) and (11) above).

Setting the congestion price to zero in this way causes each of the wireless devices to minimise their contention window. The access point 12 waits (step 66) for one second to allow the wireless network to reach an equilibrium with a congestion price of zero.

Thereafter, the access point 12 measures the aggregate throughput in the network by summing the data rates of the signals it is sending and receiving.

The 'tatonnement¹ process then starts (steps 69 to 86). Tatonnement¹ involves iteratively approaching a congestion price which causes the throughput in the wireless network to be maximised. To carry out this process, it . is necessary to move the price upwards if the price is below the optimum price, and to decrease it if the price is above the optimum price. To keep track of the direction in which the price is moving, a Boolean variable called 'Inflation' (which is TRUE when the price is rising, and FALSE when the price is falling) is used. That variable is initially set (step 69) to TRUE (which must be the case initially since the price is initially zero).

The process then enters and endless loop of instructions (step 70 to 86) (this process might be run as a thread allowing other processes necessary for the access point to operate in accordance with the 802.11b standard to run whilst this process waits for the wireless devices to react to changes in price).

The loop of instructions begins with a test, using the Boolean variable 'Inflation¹, to see if the price is to rise or fall on this iteration (step 70). If the price is to rise, then the value M is increased by a factor α (step 72). If the price is to fall, then the value M is decreased by the same factor α (step 74).

Once M has been altered, the access point 12 calculates new congestion price components using Equations (10) and (11). The calculation starts by finding the aggregate transmission probability P in the wireless network (step 90) - this can be carried out in the same way as was described above in relation to the first embodiment (step 42 in

Figure 4). Thereafter, the calculation of the congestion price components μ-_t and μ₂ is straightforward - the value T⁰⁰' being a constant 402 microseconds for the 802.11b network.

Once the access point 12 has calculated the new congestion price components, it broadcasts them to the wireless devices 14,16,18 (step 78) and then waits (step 80) for 1 second in order to allow the access devices to react to the new congestion price. The process followed by the wireless devices in reacting to the new congestion price components will be described in more detail below with reference to Figure 7.

Once the one second has elapsed, the access point 12 again measures the aggregate throughput in the wireless network (step 82 - similar to step 68). It then tests to find whether the aggregate throughput has increased (step 84). If it has, then the direction of price movement (represented by the Boolean variable 'Inflation') is unchanged, and the next iteration of the tatonnement process is started. If, on the other hand, the last price change has resulted in a decrease in throughput, then the direction of change in congestion price is changed by changing the Boolean variable 'Inflation' (step 86) prior to starting the next iteration of the tatonnement process. On receiving the new congestion price components, each wireless device carries out the process illustrated in Figure 7.

The first step is to calculate the congestion charge from the received congestion price components (step 102). This is done using the formula below, taking the value Ti^suc from a rate table which records these values for each of the transmission rates which the wireless device might be operating at (Figure 5B).

Cost = (μiTi^suc +μ₂) x, Equation (20)

The utility of different levels of throughput is then retrieved (step 104) from the utility table (Figure 5B) and the net utility (i.e. the utility less the cost of each transmission rate) is calculated for each.

The wireless device thus calculates its optimum throughput by extrapolation between the values stored. Having calculated the optimum throughput, the wireless device sets its contention window length to a value which corresponds the optimum throughput. The appropriate length of contention window is easily calculated since the throughput is proportional to the transmission probability. Thus, the constant of proportionality can be measured and then stored for use in the next calculation.

The access point calculates the congestion charge for each wireless device and aggregates the charges over the user session to arrive at a session charge which is then forwarded to the server 22 so that the charge can be debited from the user's account.

It will be seen how the second embodiment will penalise those wireless devices which have a high value for T|^suc by raising the cost of throughput for them. As a result, those devices will tend to set their contention windows to longer time periods, and thereby reduce the time for which they occupy the available bandwidth. However, this effect is counteracted in cases where the user of the wireless device is prepared to pay for the increased burden they place on the network owing to them operating at a lower rate. This leads to a more efficient division of access time to the channel than has hitherto been achieved in 802.11 networks. Other variations on the above embodiments which fall within the scope of the present invention include:

i) in a variation on the second embodiment, sending congestion marks at a rate which corresponds to the congestion price calculated in accordance with equation (20) above. In such a scheme, the owner of the wireless device could be expected to pay a price per congestion mark.

ii) the present inventor has also applied micro-economic theory to a wireless network in which devices send -data using the Transmission Control Protocol. In that case, the utility can be shown to vary with throughput in the following way:

. Equation (21)

This leads to the result that the occupancy of the shared access medium is optimally shared amongst multi-rate transceivers when the probability of transmission in any given timeslot is calculated in accordance with the expression:

Equation (22)

Those skilled in the art will have little difficulty in introducing a mechanism which calculates the round trip time (RTT) associated with a TCP connection. Once that measurement has been enabled, it will be seen how the above embodiments can be modified to calculate the quantity pi from equation (22) and adapt the behaviour of the wireless devices accordingly in order to fairly allocate the shared channel between the wireless devices. iii) in the embodiments described above, the level of usage of the network was measured at one second intervals and the contention window length of the wireless device updated accordingly. In preferred embodiments, that measurement and the associated processing

It and signalling is carried out only on a predetermined change in the number of devices using the shared access medium occuring.

iv) the above embodiments describe a wireless LAN network, however the invention is also applicable to other shared medium networks, for example networks which share the bandwidth of a fixed transmission line. An example is a network operating in accordance with IEEE's 802.3 standard.

It will be seen that the above-described embodiments provide a shared medium access network in which the amount of resource given to a user of a multi-rate shared access / medium network takes into account the bit-rate at which a wireless device operates (and tends to discourage such user's operating at a low bit-rate from hogging the access medium) whilst also taking into account the value the user places on different qualities of service which might be operated by the network.

Claims

1. A communications network comprising:

a shared access transmission medium;

a congestion monitor operable to broadcast, on said shared access medium, a signal indicative of the level of congestion in said shared access medium;

one or more stores storing utility data indicative of the value of different qualities of service from the shared access transmission medium to different wireless devices;

each of said network devices being arranged in operation to increase the time for which they occupy said shared access transmission medium in response to said utility data indicating that said wireless device places a relatively high value on quality of service provided by said shared transmission medium, but to reduce the time for which they occupy said shared access transmission medium when operating at a lower bit-rate and in response to said level of congestion increasing.

2. A network according to claim 1 in which said signal indicative of the level of congestion comprises a congestion price.

3. A network according to claim 2 in which said congestion price has two components, one relating to the cost of congestion independent of the transmission rate of a wireless device, the other relating to the cost of congestion which is dependent on the transmission rate of a wireless device.

4. A network according to claim 3 wherein said congestion cost is calculable in accordance with the formula: ^' Congestion Price = (μ-,Ti^SUC +μ₂) Xj

5. A network according to claim 2 wherein said network is a packet network.

6. A network according to claim 5 wherein said signal indicative of the level of congestion comprises a number of packets marked with a congestion indication.

7. A network according to claim 1 wherein one or more network devices alter one or more operating parameters in dependence upon the bit-rate at which said one or more network devices is operating.

8. A network according to claim 7 wherein said operating parameter comprises a contention window value.

9. A network according to claim 1 in which said network is a wireless network wherein said shared access medium is a portion of the electromagnetic spectrum.

10. A network according to claim 7 in which said plurality of devices are arranged to update said operating parameter in response to a predetermined change in the number of devices using said shared access medium.

11. A method of operating a shared access medium network comprising a plurality of transceivers in communication via said shared access medium, said method comprising the steps of:

altering the rate of transmission of one of more of said transceivers; and