GB2382945A

GB2382945A - Scheduling data packets for transmission over a shared channel

Info

Publication number: GB2382945A
Application number: GB0128999A
Authority: GB
Inventors: Carlo Luschi; Louis Samuel
Original assignee: Lucent Technologies Inc
Current assignee: Nokia of America Corp
Priority date: 2001-12-04
Filing date: 2001-12-04
Publication date: 2003-06-11
Anticipated expiration: 2021-12-04
Also published as: GB2382945B; GB0128999D0

Abstract

A method of scheduling data packets and packet fragments for transmission from a first terminal to a second terminal over a channel shared with other terminals comprises scheduling at a packet data throughput rate that is determined by averaging the lengths of the packets and packet fragments for transmission then dividing the maximum supportable data rate by the average packet/packet fragment length determined.

Description

A METHOD OF SCHEDULING DATA PACKETS FOR TRANSMISSION OVER A SHARED CHANNEL, AND A TERMINAL OF A DATA PACKET TRANSMISSION NETWORK Technical Field The present invention relates to a method of scheduling data packets for transmission from a first terminal to a second terminal over a channel shared with other terminals. The present invention also relates to a terminal of a data packet transmission network, the terminal comprising a scheduler operative to schedule data packets for transmission over a channel shared with other terminals.

Background of the Invention High-speed packet transmission schemes are currently under development in the evolution of third generation (3G) telecommunication systems. Key factors that enable the high performance of these technologies include higher peak data rates via high order modulation such as 16 or 64 quadrature amplitude modulation, fast scheduling of the users within a common shared channel, and the use of multiple antenna techniques for transmission and reception. Features supporting fast scheduling are Hybrid-Automatic Repeat Request (H-ARQ) i. e. ARQ with Forward Error Correction (FEC) coding, and fast rate selection based on feedback of estimated link quality. Fast rate selection, combined with time domain scheduling on the shared channel, enables advantage to be taken of the short-term variations in the signal received by the mobile terminals, so that each user can be served on a constructing fading, i. e. each user is scheduled for transmission so as to minimise the chance of destructive interference.

In cellular communication systems, the quality of a signal received by a mobile user depends on distances from the serving base station and interfering base stations, path loss (i. e. attenuation), shadowing (signal reduction in the shadow of obstacles), and short-term multipath fading (i. e. scattering). In order to improve the system peak data rates and coverage reliability, link adaptation techniques are used to

modify the signal transmitted to and from a particular user to account for variation of the received signal quality. Two link adaptation techniques are Power Control and Adaptive Modulation and Coding (AMC). While the former allocates proportionally more transmitted power to disadvantaged users, with AMC the transmitted power is held constant, and the modulation and coding is changed to match the current channel conditions. In a system implementing AMC, users with favourable channel conditions, e. g. users close to the base station, are typically assigned higher order modulation with higher code rates, which results in higher data rates.

One of the fundamental requirements of high-speed packet networks over wireless channels is the capability of efficiently supporting guaranteed Quality of Service (QoS), in particular meeting the data rate and packet delay constraints of realtime applications like audio/video streaming.

The quality of service QoS of a data user can be defined in different ways. For real-time users, the delays of most of the data packets need to be kept below a certain threshold. A different notion of quality of service QoS is a requirement that the average throughput provided to a given user is not less than a predetermined value.

The need of efficiently utilize the wireless link capacity implies that a suitable scheduling algorithm should meet the above quality of service QoS requirements while optimizing the use of the scarce radio resources. For high-speed packet access, one way of obtaining efficient utilization of the radio link is to exploit the time variations of the shared wireless channel, giving some form of priority to users with better channel conditions. Along this line, the Third Generation Partnership Project (3GPP) High Speed Data Packet Access (HSDPA) scheme contemplates a method of scheduling based on a maximum signal to interference (C/1) ratio rule, where the channel is allocated to the packet flow with the highest supportable rate (the"maximum rate"rule). However, this approach is not effective in terms of maximum delay guarantee, especially for very low user mobility, which corresponds to nearly stationary channel conditions. Proportionally fair scheduling also has a drawback of poor delay performance. In fact, in this case a simple round-robin scheduling can provide better delay performance. On the other hand, a round-robin policy is not effective in terms of throughput.

A solution to the problem of supporting quality of service QoS while maximizing throughput is given by throughput-optimal scheduling schemes such as the

Modified Largest Weighted Delay First (M-LWDF) algorithm described in the paper: M. Andrews, K. Kumaran, K. Ramanan, A. Stolyar, P. Whiting, and R. Vijayakumar, "Providing Quality of Service over a Shared Wireless Link", IEEE Commun. Mag., February 2001. The basic principle of M-LWDF scheduling consists in serving at each time the queue (packet flow) for which the quantity yjrj rj is maximum, where 1 : j denotes the j-th queue head-of-the-line packet delay, rj represents the supportable rate (depending on the channel quality) with respect to thej-th queue, and yj is a positive constant that allows taking into account of different delay constraints. However, the actual capability of the scheduling method to provide the required quality of service QoS performance critically depends on the definition of the packet access mechanism, including the frame structure, H-ARQ mechanism, and control signalling.

Consider : ,(t) to be the data rate supported by the channel relative to the i-th data flow (i-th user), which depends on the channel quality and is assumed to be constant over one transmission time interval TTI of duration T ; /, to be the average data date supported by the channel relative to the i-th data flow (i-th user); Z-, to be the packet (fragmented IP packet) length at the HSPA scheduler for the i-th data flow (i-th user); #, (t) to be the actual supportable throughput for the i-th data flow (i-th user) over a transmission time interval TTI of duration T, taking into account the packet fragmentation; and 7 :, (t) to be the accumulated head-of-the-line packet delay for the i-th data flow (i-th user).

According to the above definitions, the Maximum Rate rule schedules the

data flow rate j with maximum efficiency A (t)

j = arg max C, (t), C, (t) =//, M, I

whilst the Proportionally Fair rule schedules the data flow j with maximum At, (ta

j =argmaxC, (t), C, (t) zu f. 1,

Neither of these two rules takes into account packet delay. On the other hand, the M-LWDF algorithm schedules the data flow j such that

j=argmaxC, (t), C, (f) =, (, f

for an arbitrary set of parameters y, > 0. It has been shown that a good choice of y, is given bey/, = a,/, where a, > 0, i =M, are suitable weights which characterize the desired quality of service QoS. The values of a, determine the trade-off between reducing the weighted delays and being proportionally fair when delays are small. The M-LWDF rule has been shown to be throughput-optimal, in the sense that it renders the queues at the scheduler as stable if any other rule could do so.

A similar approach is taken by a so-called Exponential rule in which given the sets of constants y, > 0 and a, > 0, = t...., M the scheduling operation is expressed as

j = arg maxC, (t), a, r, (t) - ar I /1 + ar where ar = (11 M) L, a, r, (t). For given values of r, and a,, this policy

tries to equalize the weighted delays a, r, (t) of all the queues when their differences are large. In fact, if one of the queues has a weighted delay larger than the others by more than order a/, then the exponential term of the above equation becomes very large, and overrides channel quality considerations. Conversely, for weighted delay differences less than order, the exponent is close to 1 and the policy approaches the proportionally fair rule.

The high-speed downlink packet access technologies discussed above are mainly concerned with best-effort Internet Protocol (IP) data traffic, which is transported by the Transport Control Protocol (TCP). To provide support for the

transmission of emerging streaming media traffic, it is necessary to meet the real-time requirements characteristic of these applications. Given the transmission delay requirements of audio and video transmission, the majority of streaming traffic over IP networks uses Real-Time Transport Protocol (RTP)/User Datagram Protocol (UDP) transport, which unlike TCP does not provide a flow or congestion control mechanism.

In these cases, a critical problem is given by the transmission delay associated to the presence of a wireless link, which is characterized by a time-varying capacity and variable delays due to link-level H-ARQ retransmissions. Current wireless systems have often to accept a degradation in terms of bit-error rate (for example, bit-error rates in the order of 2% for voice) in order to avoid the delay introduced by the H-ARQ.

Summary of the Invention The present invention provides a method of scheduling data packets for transmission from a first terminal to a second terminal over a channel shared with other terminals comprising scheduling at a packet data throughput rate determined by averaging the lengths of the packets and packet fragments for transmission then dividing the maximum supportable data rate by the average packet/packet fragment length determined.

Advantageously the preferred technique is based on considering the actual integer number of packets that can be transmitted over one transmission time interval TTI, for a given modulation and coding scheme and a given packet length, whilst taking into account the actual supportable data rate over the wireless link in the presence of variable packet lengths and packet fragmentation.

Preferably the packet data throughput rate for an i-th data flow at time t is the maximum supportable data rate minus a factor due to only whole packets being transmitted.

Preferably the supportable packet data throughput rate A, (t) for an i-th

data flow at time t is set to be u, (t)-6' (t) 77,, where p, (t) is the maximum supportable T data rate, T is transmission time interval, E, (t) is a rounding error due to only whole packets being transmitted, and 77, is the average packet/packet fragment length.

Preferably the scheduling is undertaken so as to maximise a cost function C, (t) that is

proportional to the packet data throughput rate 1,, (t) of the i-th data flow at time t.

At, (t) Alt (t) Preferably At, (t) is replaced by the quantity---'-----=-'-'-, in the cost 1+, -M] 1+, () function with k > 0, k'= k77,/T.

Preferably there are multiple data flows, the scheduling being such as to provide each data flow with a predetermined packet throughput rateThis provide an additional degree of freedom in meeting the quality of service QoS requirements while optimizing the use of the radio resources.

Preferably each data flow is for transmissions to a different mobile terminal of a network for wireless telecommunications.

Preferably the scheduling is for High Speed Data Packet Access

h I transmissions. High Speed Data Packet Access HSDPA allows time or code multiplexing of the different flows over a given Transmission Time Interval (TTI)..

Thus the proposed scheduling takes into account the possibility of time/code multiplexing over a predetermined time interval a number of users Q.

Preferably the simultaneous data flows for transmission are scheduled by selecting as the packet data throughput rate the average over all of the data flows of their optimised cost functions C,, (t).

Preferably the selected Q data flows y,....,/c are scheduled for simultaneous transmission such that

i Q ii,-.., jo arg max-Y C, (t), - Q fl=,, It -'1=1 C

where C, (t) is the cost function of the i-th data flow.

The present invention also provides a method of scheduling transmissions of data packets in both directions between a base station and a user equipment comprising using the method of scheduling for downlink packet transmissions in which the first terminal is a base station and the second terminal is a user terminal, and the method of scheduling for uplink packet transmissions in which the first terminal is a user terminal and the second terminal is a base station.

The present invention also provides a terminal of a data packet transmission network, the terminal comprising a scheduler operative to schedule data packets for transmission over a channel shared with other terminals, the scheduler being

operative to schedule at a packet transmission rate determined by averaging the lengths of the packets and packet fragments for transmission then dividing the maximum supportable data rate by the average packet/packet fragment length determined.

Brief Description of the Drawings A preferred embodiment of the present invention will now be described by way of example and with reference to the drawings, in which: Figure 1 is a diagrammatic illustration of a base station and one active mobile station (of many), Figure 2 is a diagrammatic illustration of High Speed Packet Access scheduling in a base station.

Detailed Description As shown in Figure 1, the preferred network 1 includes a base station 2 and many mobile stations UE communicating therewith (one UE being shown in Figure 1 for simplicity). The base station 2 and mobile UE each have a transmitter Tx and a receiver Rx, each transmitter Tx and receiver Rx having a respective antenna 4. On the downlink, a High Speed-Downlink Shared Channel (HS-DSCH) is used, and on the uplink, a High Speed-Uplink Shared Channel (HS-USCH) is used.

As shown in Figure 2, the base station 2 includes a High Speed Data Packet Access HSDPA scheduler 6 which takes into account channel quality information and quality of service (QoS) requirements of different user packet flows (8, user 1 to user M). The scheduler 6 is connected via a Hybrid-Automatic Repeat reQuest (H-ARQ) and Adaptive Modulation and Coding (AMC) unit 10 to the transmitter Tx.

The unit 10 schedules H-ARQ retransmissions, and controls modulation and coding so as to provide higher data rates to favoured users.

The HSPDA scheduler 6 schedules both uplink shared channel and dowlink shared channel transmissions, received signals at the base station 2 being processed by a further H-ARQ and AMC unit 12 to provide information to the scheduler 6.

Scheduling Techniques with Packet Fragmentation and Flexible Time/Code Multiplexing The effectiveness of the high-speed packet access scheduler 6 depends on its capability of taking into account the supportable rate (based on channel quality information) and quality of service QoS requirements of the different user packet flows.

However, the actual supportable rate over the wireless link is affected by the variable packet lengths associated with different flows, and/or by the presence of packet fragmentation. Packet fragmentation occurs where there is some remainder of the original packet left over from the segmentation process. The segmentation process is, of course, where a packet is fragmented into an integer number of blocks for transmission.

At any given instant any given data flow through a queuing system will have a packet length probability distribution associated with its packets. More importantly there is likely to be an average packet size associated with the flow through the queue. In a wireless environment for any given channel conditions this information is used to determine the most probable (average) packet transmission rate. This is effected by dividing the maximum supportable data rate by the expected (i. e average) packet/packet fragment length. More formally the above textual description can be stated as follows: Suppose that at a given time t the i-th queue data flow has a packet (fragmented IP packet) length distribution P, (L) = Pr {L, : ! L. Then, for a given channel condition, where the supportable data (transmission) rate is j. LI (t) of the i-th queue data flow at time t, the actual packet transmission rate is

where average packet length (or more specifically average packet fragment length) in the i-th queue data flow is 77, = E {L,}. However, within the transmission time interval TTI of (t, t + T) the number of packets actually transmitted is

1 ()-r-, (t) = (t)-T'J = , where (t) is the rounding 77, 1,

error of the i-th data flow at time t due to the fact that only whole packets are transmitted, the other parameters being as mentioned above. In other words, the actual number of packets transmitted corresponds to the integer value of the expected packet

transmission rate multiplied by the time interval over which the transmissions lasted.

The actual supportable throughput for the i-th data flow at time t is then

-1 Mz 171 T'T 77,-1 T T

i. e. the actual throughput rate is the maximum supportable data rate minus the rounding error due to the fact that only whole packets can be transmitted.

We can see that in order to efficiently use the channel an optimisation has to be performed to minimise the rounding error , (t). More formally : . In order to efficiently utilize the channel, one should therefore attempt to minimize the quantity #i, (t)/, (t). In short the proposed optimisation is that the effects of the rounding are incorporated into our chosen cost function. One way to do this is to use a scheduling algorithm that maximises a cost function C, (t) that is proportional to the supportable packet throughput rate A, (t) of the i-th data flow at time t instead of the supportable transmission rate JL, (t) of the i-th data flow at time t. Another way to do this is to replacez by the quantity

with k > 0, k'=k7J, IT. In the latter case, the cost function C, (t) is proportional to J. il (t) for small values of -, (t)///, (t), while if (t) is substantially smaller than a, (t), the supportable rate u, (t) is penalized by the scaling factor I/ [I + k'--, (t)].

Here we note that we have developed a cost function for a given data flow. However, there is seldom a single flow in existence at any one time across an air interface. In the following we apply the above to multiple flows.

The performance of the scheduling method can be affected by the time periodicity at which packets are scheduled. In Third generation partnership project 3GPP High Speed Data Packet Access HSDPA it is proposed that an update of the channel quality information is made every three 0.667 msec time slots, and it is therefore reasonable to assume a new scheduling operation by the scheduler 6 every 2 msec. However, it should be also taken into account that different user terminals UEs may have different transmission time interval TTI lengths. Assuming possible transmission time interval TTI lengths T of 1 or 3 time slots (respectively 0.667 or 2 msec), the scheduler 6 has therefore the flexibility of assigning the channel to a single UE with three-slot transmission time interval TTI, or to up to three different UEs with one-slot transmission time intervals TTI.

Additionally, High Speed Data Packet Access HSDPA allows codemultiplexing of different users over the common shared channel. This implies that, for each time interval T = 0.667 msec, the scheduler 6 can select for transmission a variable number of queues Q, ] < < < C. Correspondingly, assuming that the code resources

are uniformly distributed among the Q users, the data rate (i. e. bandwidth) assigned to c the i-th data flow will be given by a fraction Ai, (t)/0 of the supportable data rate /).

In other words High Speed Data Packet Access HSDPA will have up to a maximum of Qmnx Queues to service and each queue is equally likely to have code resources assigned to it, so that the data rate assigned to a given data flow is an equal share of the overall maximum data rate apportioned equally among the queues.

We observe that taking into account the flexibility of time and/or code multiplexing of the users in the scheduling scheme has the potential of providing an additional degree of freedom in meeting the quality of service QoS requirements while optimizing the use of the wireless channel. For this purpose, a simple strategy is to simultaneously schedule for transmission the selected Q data flows jo such that

where C, (1) denotes one of the scheduling cost functions defined in the previous section, possibly modified to take into account the packet lengths associated with the different flows. In other words the scheduler optimises simultaneous flows for transmission by selecting as the flow rate the average optimisation over all of the queues for the optimised cost function C, (t) outlined earlier.

The above technique can easily be extended to the case of a non-uniform allocation of the code resources among the Q users.

It will be seen that the preferred technique allows : optimization of the allocation of time/code space resources, efficient scheduling of real-time or best-effort traffic in the presence of variable packet lengths, scheduling of the different packet flows taking into account the respective packet lengths, and scheduling of a variable number of packet flows over a predetermined time interval by allocating the available time/code resources according to a given optimality criterion.

Claims

Claims:

1. A method of scheduling data packets for transmission from a first terminal to a second terminal over a channel shared with other terminals comprising scheduling at a packet data throughput rate'determined by averaging the lengths of the packets and packet fragments for transmission then dividing the maximum supportable data rate by the average packet/packet fragment length determined.

2. A method of scheduling data packets according to claim 1, in which the packet data throughput rate for an i-th data flow at time t is the maximum supportable data rate minus a factor due to only whole packets being transmitted.

3. A method of scheduling data packets according to claim 1 or claim 2, in which the packet data throughput rate #, (t) for an i-th data flow at time t is set to be

c (t) 77, /./, (t)-----, where /, (t) is the maximum supportable data rate, T is transmission time interval. c, (t) is a rounding error due to only whole packets being transmitted, and '7, is the average packet/packet fragment length.

4. A method of scheduling data packets according to claim 3, in which the scheduling is undertaken so as to maximise a cost function C, (t) that is proportional to the packet data throughput rate/1, (t) of the i-th data flow at time t.

5. A method of scheduling data packets according to claim 3, in which the scheduling is undertaken so as to maximise a cost function C, (t) that is proportional

to the quantity/-l, (t) - 11, (t), with k > 0, k'= kTJ,/T.

I I + k- [p, (t)-A, (t)] I + k'--, (t)

6. A method of scheduling data packets according to any preceding claim, in which there are multiple data flows, the scheduling being such as to provide each data flow with a predetermined packet throughput rate.

7. A method of scheduling data packets according to any preceding claim, in which each data flow is for transmissions to a different mobile terminal of a network for wireless telecommunications.

8. A method of scheduling data packets according to any preceding claim, in which the scheduling is for High Speed Data Packet Access transmissions.

9. A method of scheduling data packets according to any preceding claim, in which the simultaneous data flows for transmission are scheduled by selecting as the data rate the average over all of the data flows of their optimised cost functions C, (t)

10. A method of scheduling data packets according to any preceding claim, in which the selected Q data flows y,,..., j are scheduled for simultaneous transmission such that

.. 1 Q j Q =arg max-CJ (t), '''''" ; j=) - q= !

where C, (t) is the cost function of the i-th data flow.

11. A method of scheduling transmissions of data packets in both directions between a base station and a user equipment comprising using the method of scheduling according to any of claims 1 to 10 for downlink packet transmissions in which the first terminal is a base station and the second terminal is a user terminal, and the method of scheduling according to any of claims 1 to 10 for uplink packet transmissions in which the first terminal is a user terminal and the second terminal is a base station.

12. A terminal of a data packet transmission network, the terminal comprising a scheduler operative to schedule data packets and packet fragments for transmission over a channel shared with other terminals, the scheduler being operative to schedule data packets and packet fragments for transmission at a packet transmission rate determined by averaging the lengths of the packets and packet fragments for transmission then dividing the maximum supportable data rate by the average packet/packet fragment length determined.

12. A terminal of a data packet transmission network, the terminal comprising a scheduler operative to schedule data packets for transmission over a channel shared with other terminals, the scheduler being operative to schedule at a packet transmission rate determined by averaging the lengths of the packets and packet fragments for transmission then dividing the maximum supportable data rate by the average packet/packet fragment length determined.

Amendments to the claims hlave been filed as follows Claims : 1. A method of scheduling data packets and packet fragments for transmission from a first terminal to a second terminal over a channel shared with other terminals comprising scheduling data packets and packet fragments for transmission at a packet data throughput rate determined by averaging the lengths of the packets and packet fragments for transmission then dividing the maximum supportable data rate by the average packet/packet fragment length determined.

2. A method of scheduling data packets and packet fragments according to claim 1, in which the packet data throughput rate for an i-th data flow at time t is the maximum supportable data rate minus a factor due to only whole packets being transmitted.

3. A method of scheduling data packets and packet fragments according

to claim 1 or claim 2, in which the packet data throughput rate (t) for an i-th data flow

at time t is set to be, u, (t)----"-, where u, (t) is the maximum supportable data rate.

T is transmission time interval,, (t) is a rounding error due to only whole packets being transmitted, and 77, is the average packet/packet fragment length.

4. A method of scheduling data packets and packet fragments according to claim 3, in which the scheduling is undertaken so as to maximise a cost function C, (t) that is proportional to the packet data throughput rate A, (t) of the i-th data flow at time t.

5. A method of scheduling data packets and packet fragments according to claim 3, in which the scheduling is undertaken so as to maximise a cost function C,

(t) that is proportional to the quantity Jil (t) = Ji, (t), with k > 0, ! +, (t)-M] +, (t) k'= k7J,/T.

6. A method of scheduling data packets and packet fragments according to any preceding claim, in which there are multiple data flows, the scheduling being such as to provide each data flow with a predetermined packet throughput rate.

7. A method of scheduling data packets and packet fragments according to any preceding claim, in which each data flow is for transmissions to a different mobile terminal of a network for wireless telecommunications.

8. A method of scheduling data packets and packet fragments according to any preceding claim, in which the scheduling is for High Speed Data Packet Access transmissions.

9. A method of scheduling data packets according to any preceding claim, in which the simultaneous data flows for transmission are scheduled by selecting as the data rate the average over all of the data flows of their optimised cost functions Cla (t)

10. A method of scheduling data packets and packet fragments according to any preceding claim, in which the selected Q data flows yl,..., y are scheduled for simultaneous transmission such that

" 1 Q il,..., JQ == arg max- C (t), Q, e < .

where C, (t) is the cost function of the i-th data flow.

11. A method of scheduling transmissions of data packets and packet fragments in both directions between a base station and a user equipment comprising using the method of scheduling according to any of claims 1 to 10 for downlink packet transmissions in which the first terminal is a base station and the second terminal is a user terminal, and the method of scheduling according to any of claims I to 10 for uplink packet transmissions in which the first terminal is a user terminal and the second terminal is a base station.