The achievable delay and utilisation performance of the channels of a packet switching system are governed by the underlying bandwidth assignment scheme.

Satellite Medium Access Control (MAC) protocols for data traffic have traditionally employed Demand Assigned Multiple Access (DAMA) with requests for bandwidth made on a regular basis, derived from the instantaneous queue levels at the ground terminals. Thus any terminal having more than a predetermined number of packets awaiting delivery makes regular requests for bandwidth. As bandwidth becomes available one such terminal is selected for transmission of its next packet. Such a systems is described, for example, by Mohamed and Le-Ngoc in a paper entitled "Performance Analysis of Combined/Free Demand Assigned Multiple Access (CFDAMA) Protocol for Packet Satellite Communications, (IEEE New Orleans, May 1994) A typical satellite uplink frame format is shown in Figure 1, consisting of a series of data transmission slots D, F interleaved with DAMA request slots R. A request algorithm for such a scheme is given in Figure 3. Ground terminals 71 (see Figure 7) make requests for bandwidth accompanying their uplink packet transmissions in the adjacent request slots, as and when required. At the time a packet is to be transmitted in one of the slots D (step 31) the terminal 71 determines its current packet queue size (step 32) and the number of slots already requested which have not yet been satisfied (step 33). If the queue size is greater than the number of slots already requested (because further packets have been added to the queue since the previous packet was transmitted), it then transmits a request for

further slots (step 34) based on the instantaneous ground terminal queue size and the number of outstanding slot requests (less the packet currently being transmitted). At the satellite 72, the scheduler 73 assigns slots on a frame-by-frame basis. In the first instance slots are demand-assigned to terminals based on requests queued at the scheduler in a first come first serve (FCFS) manner, with each terminal 71 being allocated a run of contiguous slots D based on the number of slots requested. In the absence of any queued requests, successive slots in the frame are allocated one-by- one on a free assigned round robin basis to all terminals in the system. To give terminals that have not requested bandwidth for a while a better chance of obtaining a free assigned slot, terminals are put to the bottom of the round-robin free assignment list subsequent to being allocated demand-assigned slots.

A geostationary satellite orbit is approximately 33,500km above the earth's surface. For most points on the earth's surface the nearest geostationary satellite is not at the zenith, so the distance is even greater-up to 40, 000km. The resulting long propagation delay in geo-stationary earth orbit (GEO) satellite links inhibits the effectiveness of such schemes. A"hop", the propagation delay for transmission of a radio signal up to a satellite and back down to the ground, is about 0.25 seconds, but varies depending on the elevation angle to the satellite. The distance to the satellite is a minimum (about 0.24 seconds) when directly overhead an earth station at the equator, and it is a maximum (about 0.28 seconds) when an earth station is located at the edge of global coverage.. Since a request for bandwidth has to be transmitted to the scheduler and the reply returned before a packet can be transmitted (which has itself then to be transmitted up to the satellite and back), each packet is delayed by at least two satellite hops (in addition to any processing delay) if the scheduler is located on the satellite, or more if it is on the ground, or distributed. In order to circumvent the long delay, DAMA is often combined with either random access (e. g.

Slotted ALOHA) or a form of free assignment of bandwidth as found in the Combined Free/Demand Assignment Multiple Access (CFDAMA) schemes discussed by Le-Ngoc et al, in"Performance of combined free/demand assignment multiple-access schemes in satellite communications", International Journal of Satellite Communications, vol.

14, no. 1, pp. 11-21, 1996. Leland et al, in"On the self-similar nature of Ethernet traffic (extended version)", IEEE/ACM Transactions on Networking, vol. 2, no. 1, pp.

1-15, 1994. show that modern computer Local Area Network (LAN) traffic exhibits a burstiness characteristic over a wide range of time scales.

The present invention presents a novel packet reservation system for data traffic, suited to handling long bursts of packets from ON-OFF type traffic sources.

According to the invention, there is provided a demand assignment process for a packet switching communications system in which a terminal requests capacity from a scheduler for the transmission of bursts of packets, and in which the terminal transmits position signals to the scheduler with one or more of the packets, the position signals being indicative of those packets'positions in a burst. The position signals may be used by the scheduler to determine the length of the burst and preferentially allocate capacity to those terminals currently in the middle of transmitting a burst, allowing transmission of further packets of the burst.

This system differs from prior art arrangements in that each terminal provides an indication of how the packets in its queue are arranged in bursts, allowing priority to be given to transmission of packets to complete a burst that has already been partially transmiited. The packets awaiting transmission that make up subsequent bursts are therefore given less significance in the allocation process.

In one embodiment the terminal transmits a final position signal with at least the last packet in each burst, thereby indicating the transition of the terminal from a mid-burst ("ON") state to a non-burst ("OFF") state and causing a flag in the scheduler to indicate the transition to the"OFF"state for the terminal in question.

The transition in the reverse direction, from"OFF"to"ON"state may simply be indicated by the arrival of a packet from a terminal currently in the"OFF"state, or may be triggered by an initial position signal transmitted with the first packet in each burst. Although in the described embodiment a simple"ON/OFF"signal with the first and last packets of each burst is used, other arrangements may be envisaged. For example an initial position signal may be used to indicate the length of the burst. As well as dispensing with the need for the final position signal, this arrangement allows the scheduler to allocate slots to a frame taking into account the expected demand for slots in one or more further frames as indicated by the requests for capacity, allowing capacity to be at least provisionally allocated several frames ahead. The use of position markers may also be used to allow a single request for capacity to be made for each burst, that request being maintained for as many frames as necessary

to transmit the complete burst, without the need to repeat the request for each frame. The scheduler can thus allocate slots to one or more frames taking into account the expected demand for slots in such frames indicated by the requests for capacity. Preferably the scheduler identifies, from the position signals transmitted by the terminals, whether any terminals are not part way through transmission of a burst, (that is to say, they have completed one burst and have not started another), and if there are any such terminals, the scheduler allocates capacity to allow those terminals to request transmission of new bursts should they require to do so. The proportions of the capacity allocated to allow such terminals to request capacity, and the capacity allocated to terminals already part way through a burst, may be varied according to the current number of terminals currently in each of those conditions.

The invention also extends to a terminal for a packet switching system comprising means for requesting capacity from a scheduler for the transmission of bursts of packets, and comprising means for transmitting position signals to the scheduler with at least some of the packets, indicative of those packets'positions in a burst. The terminal may have means for transmitting a final position signal with at least the last packet in each burst, or for transmitting an initial position signal with at least the first packet in each burst, which may include an indication of the length of the burst.

The invention also extends to a scheduler for a packet switching system comprising allocation means for allocating capacity to a plurality of terminals for the transmission of bursts of packets, and comprising means for receiving position signals from the terminals with at least some of the packets, indicative of those packets' positions in a burst. The scheduler may comprise means for identifying, from the position signals received from the terminals, which terminals have transmitted part of a burst but have further packets of that burst awaiting transmission, wherein the allocation means is arranged to allocate capacity to those terminals to allow transmission of further packets of the burst. It may also comprise means for setting a flag to a first indication in respect of a terminal when the first packet of a burst is received from that terminal, and resetting the flag to a second indication when the last packet in the burst is received.

The scheduler may include means for identifying, from the position signals transmitted by the terminals, which terminals have transmitted part of a burst, but

still have further packets of that burst awaiting transmission, the allocation means having means to preferentially allocate capacity to such terminals to allow transmission of further packets of the bursts. The scheduler may have means for detecting an initial position signal indicating the length of a burst, and means for allocating capacity in a plurality of frames for the transmission of the said burst. It may be operable, in response to a single request from a terminal for capacity to transmit a burst, to continuously allocate capacity to the burst in each of a plurality of frames until the complete burst has been transmitted The scheduler may also have for identifying, from the position signals transmitted by the terminals, whether any of the terminals are not part way through transmission of a burst, and allocation means for allocating capacity to allow such terminals to request transmission of new bursts. The allocation means is preferably arranged to vary the proportions of the capacity allocated to terminals not part way through a burst, and the capacity allocated to terminals already part way through a burst, according to the current number of terminals in each of those conditions..

An embodiment of the invention will now be described, with reference to the drawings, in which Figure 1 shows a frame format used in the conventional system Figure 2 shows a frame format suitable for use in the invention Figure 3 shows a request algorithm for use in the conventional prior art arrangement already discussed Figure 4 shows a request algorithm for use in the system according to the invention Figures 5 and 6 show the results of comparative tests between a system according to the invention and a conventional system Figure 7 shows schematically the elements co-operating to perform the invention.

Referring firstly to Figure 7, each satellite ground station 70 is associated with one or more terminals 71 which operate to transmit packet data to the satellite 72, which relays them to other ground stations (not shown). The allocation of slots in the packet frame is controlled by a sequencer 73 located in the satellite 72.

The allocation scheme according to the invention has a frame format as shown in Figure 2, which is similar to the conventional arrangement shown in Figure

1 in that it consists of a series of data transmission slots D, F interleaved with DAMA request slots R. The request algorithm is given in figure 3. As in the conventional system, ground terminals 70 transmit requests for bandwidth with their uplink packet transmissions, using the adjacent request slots R as and when required.

As has been stated, in the prior art arrangement the ground terminals 71 each request a number of slots D based on the instantaneous ground terminal total queue size (irrespective of whether they form part of one burst or several), and the number of outstanding slot requests. With the request strategy according to the present invention, as shown in Figure 4, requests take the form of bandwidth signaling on a burst-by-burst basis. Terminals 71 are flagged in the scheduler 73 as existing in one of two states, ON or OFF. When a packet in a burst is transmitted on the uplink (step 41) the terminal 71 determines whether the packet is the first in a burst, the last in the burst, or an intermediate packet (step 42).

In this embodiment the terminal 71 labels the first and/or last packets in order to indicate their position in the burst to the scheduler 73. The first packet of a burst may carry a label not only indicating to the sheduler 73 that it is the first packet, but also indicating the length of the burst. In this case the scheduler 73 need only count the packets to identify the last one in the burst, and thus can re-set the terminal status without a further specific signal being transmitted. However, this arrangement is only possible if the lengtrh of the burst is already known to the terminal, so it is not possible to use this technique if unless the entire burst is already in the queue at the terminal 71. Alternatvely, the last packet of each burst may be labelled, allowing the scheduler to then recognise the next packet to arive from the same terminal as being the first packet of the next burst.

When the scheduler 73 identifies a packet as the first of a burst, it changes the setting for the originating terminal to the ON state (step 43), and sets its status flag to ON (step 44). When the last packet of each burst is transmitted, the scheduler 73 changes the status of the relevant terminal 71 back to the OFF state (step 45) and returns its status flag to OFF (step 46).

The scheduler 73 in the satellite 72 maintains two lists : one containing the terminals that have signaled ON and one containing the terminals that have signaled OFF. Each frame consists of a variable number of free assigned slots determined by: Numberof freeslots = LN] STOT FREEMAX NOFF < NTOT NOT Lrcr.

- STOT NOFF-NTOT<BR> Number of dema7ld slots = STOT-Number of free slots Where, NOFF Number of terminals in the OFF state NTOT = Total number of terminals STOT = Total number of slots in the frame FREEMAx = Maximum proportion of free slots Thus, provided that at least one terminal is in the"ON"state, a proportion of the slots (at least 1-FREEMAX) are demand-assigned. The value of FREEMAX is set such that when demand is low (but non-zero) demand-assigned slots are not delayed by an excessive number of free-assigned slots. This proportion increases as the number of terminals in the"ON"state increases. Similarly, provided that at least one terminal is in the"OFF"state, the number of free slots is non-zero.

The scheduler then allocates bandwidth to each terminal on a frame-by-frame basis, with the demand-assigned and free-assigned slots allocated to the ON and OFF terminals respectively. Unlike the process of Figure 3, there is no need to request capacity based on queue size. The number of available demand-assigned slots D and free-assigned slots F changes dynamically to suit the instantaneous requirements, by changing the position of the boundary B shown in Figure 2. When all the nodes are OFF, the entire frame is free-assigned to minimise the signalling delay for terminals following the start of a burst.

If the labelling of the packets includes an indication of burst length, the scheduler 73 can be arranged to distribute the demand-assigned slots between the bursts currently in progress in accordance with burst size. This may be done in several different ways, for example in order to weight the allocation in favour of the larger bursts so that subsequent bursts from the same terminal are not delayed unduly, or to give prefererence to any bursts which can be completed in the current frame, thereby releasing capacity for other bursts from the same or other terminals 71.

It should be noted that it is burst size, and not queue size, which determines the allocation of capacity in these embodiments. The size of a queue is constantly changing, requiring much more frequent signaling from the terminal 71 to the scheduler 73 than is the case when burst size is the determining factor. However, the size of an individual burst is fixed. With conventional DAMA schemes for data, traffic requests have to be made for transmission of each packet in the queue, resulting in delays whilst such requests are processed. In a satellite system such delays can be significant because of the distance of the satellite from the terminals. In the present invention, delays are minimised because once a terminal has signaled ON, slots are continually provided without the need for repeated requests. in effect, the connectionless packet bursts are treated like a individual connection-oriented (circuit- switched) applications by providing a continual supply of capacity to a terminal for as long as necessary to transmit an individual burst. The minimum delay for each slots is reduced to one satellite hop instead of two, since no further requests for capacity have to be made whilst the status is set at ON.

The invention provides terminals with access rights to the demand assigned bandwidth, which is shared equally between them. The maximum channel throughput is dependent on the number of terminals currently being supported with no hard limit on the number of terminals that can be supported, simply a gradual reduction in the bandwidth available to each terminal.

The conventional system and a scheme according to the present invention have been simulated for a star based satellite network consisting of a number of terminals communicating with a hub station via a GEO satellite with on-board scheduler. The results are shown in Figures 5 and 6 Figure 5 shows the distribution of end-to-end delay values of packets with the the conventional (CFDAMA) scheme at 70% channel load and the scheme according to the invention (BRMA-FD) at various channel loads. These results show that a large majority of packets are received within a very narrow range of end-to-end delay times with the scheme according to the invention, indicative that the bandwidth is successfully targeted to terminals that require it. As the proportion of demand assigned slots is increased, more bandwidth is targeted to terminals within bursts resulting in a larger percentage of packets experiencing low end-to-end delay values.

The maximum end-to-end delay increases, however, as terminals have to wait longer

before they can signal ON at the start of a burst. With a maximum proportion of 30% free assigned slots, 64% of packets are transmitted with an end-to-end delay of less than 1.2 satellite hops with a maximum end-to-end delay of 2.7 satellite hops. The distribution of end-to-end delay values is more evenly spread with the prior art arrangement, ranging from 1 to 2.5 satellite hops.

Figure 6 shows the cumulative distribution function of the end-to-end delay difference between consecutive packets within bursts. It can be seen that at the same channel loading (70%), the difference in end-to-end delay values is extremely low for the scheme according to the invention, with 80% of consecutive packets experiencing less than 0. 01s delay variation compared with only 70% experiencing less than 0.04s delay variation with the conventional scheme.