BACKGROUND OF THE INVENTION Within the International Telecommunications Union (ITU), several different air in- terfaces are defined for third generation mobile communication systems, based on either Code Division Multiple Access (CDMA) or Time Division Multiple Access (TDMA) technology. Wideband CDMA (WCDMA) is the main third generation air interface and will be deployed in Europe and Asia, including Japan and Korea, in the same frequency band, around 2GHz.

The WCDMA air interface has been defined to provide, in the first phase, data rates up to 2 Mbps in the 3GPP (third generation partnership project) Release 99 and Release 4 specifications. In the Release 5 specification, peak data rates up to 10 Mbps are possible with a High Speed Downlink Packet Access (HSDPA) fea- ture to thereby support packet-based multimedia services. In HSDPA, the intelli- gence of the base station device or Node-B (which is the third generation equiva- lent to the former base station) is increased for handling of retransmissions and scheduling functions, Thereby, L1 retransmission between the user equipment (UE, which is the third generation equivalent to the former mobile station) and Node-B are introduced and the round trip delay of these are lower than'normal' RLC retransmission round trip delays. However there will still be (a little fraction of) RLC retransmission also for HSDPA.. This makes retransmission combining feasi- ble in the mobile device due to reduced memory requirements. In general, all HSDPA users share the channel in both time and code domains. Adaptive modu- lation and coding is used to support multiple rate transmissions for different types of multimedia services.

With HSDPA, the Node-B is responsible for scheduling different users on the High Speed Downlink Shared Channel (HS-DSCH) and as such, the user data to be scheduled must to some extent be buffered in advance in the Node-B. The user data is sent from the RNC to the Node-B by means of a mapping operation bet-

ween a packet data unit of the Radio Link Control protocol layer, i. e. RLC-PDU, and a packet data unit of the Medium Access Control protocol layer, i. e. MAC-d- PDU, which is transmitted over the lub or lur interface, i. e. the interfaces between the RNC and the Node-B and between different RNCs, respectively. When the data has been sent to the Node-B, the RNC is unaware of its status until it re- ceives a response or acknowledgment. In particular, the sender side polls the re- ceiver which on its turn answers with a status report, telling the status of the out- standing RLC blocks, after some significant time delay. The RLC layer response algorithm can be controlled using the so-called polling bit, but some minimum de- lay is inherent to the network.

To provide the highest system capacity gain, it is assumed in 3GPP standardiza- tion that the so-called proportional fast scheduling strategies will be employed.

This scheduling method chooses that user for scheduling, which has the best rela- tive channel quality at a given time, thereby optimizing cell throughput while pro- viding the users with approximately the same access probability to the channel.

The basic concept for this type of scheduling is"multi-user diversity" ; e. g. the abil- ity to choose the best user from a large set of users. Hence, for this method to work optimally, it is important that the scheduler can pick the most optimum user at any given time instance which means that an excess amount of data, compared to what is actually scheduled on the channel, has to reside in the Node-B MAC-hs buffer.

However, the provision of such large buffers leads to the following problems: At some point, RLC PDUs, which are located in the Node-B may be retransmit- ted by an RLC retransmission from the RNC. Reason for this is that in the status report it is indicated that there are some RLC blocks missing. It should be noted that the RNC does not know whether the data is still waiting to be scheduled in the Node-B, whether it was sent but the UE failed to detect it, or whether the uplink response or acknowledgement was lost during transmission from the UE. If there are always provided e. g. 3-4 times as much data as the average capacity stored in the Node-B in order to facilitate sufficient multi-user diversity, quite some data sent over the lub/lur interface has to be retransmit- ted, which leads to an excessive and highly undesirable load on this link. It is however noted that the RLC layer will only retransmit those blocks missing, not all blocks. It is thus a selective retransmission algorithm.

Non-real time (NRT) services are often running over services which use the Transport Control Protocol (TCP). TCP has its own congestion avoidance algo- rithm, including features like fast retransmissions, timeouts, etc. TCP timeouts works as follows : Every time a TCP segment is sent, a timer is started. This timer is compared to the RTO (retransmission timeout), which equals the fil- tered estimated TCP RTT (round trip time) plus 4 times the filtered estimated variance of the TCP RTT. If the timer is exceeding the RTO value, congestion actions are started, as TCP assumes the segment to be lost or delayed due to congestion. These congestion actions consist among others of retransmitting all segments from the segment in question and starting the slow start algo- rithm. TCP slow starts are undesirable since they lead to low utilization of the transmission medium. From the above it can be understood that sudden very large delays for some blocks may lead to TCP timeouts. The delay caused by a RLC retransmission is considerable, since the retransmission is put at the back of the queue in the Node-B. Since RLC retransmissions only happen very sel- dom, the RTO algorithm cannot adjust to them. Thus, an RLC retransmission creates a sudden delay increase which may be interpreted by TCP as congestion, leading to undesirable congestion actions.

The RLC transmission is based on an RLC window. The working of this can be explained with an example, which is shown in Fig. 7. The amount of out- standing data can never be more than the size of the window. As soon as the latest block (most to the right) is acknowledged the window is shifted, so that new data can come in. In the example of Fig. 7, the last oldest block (hatched with left tilt) in a window of an RLC 30,32 is negatively acknowledged. There- fore, this block is retransmitted. All horizontally hatched blocks have positively been acknowledged, whereas the hatched blocks with right tilt are waiting for acknowledgement. They can be in the buffer of a Node-B 20,22, in the air in- terface or their acknowledgement can be on its way. The block with the re- transmission will without the ideas presented in this report be put at the end of the queue in the Node-B 20,22, which will mean that the time spend before a positive acknowledgement is received will increase. Before this block is ac- knowledged, the RLC 30,32 cannot send any new blocks since this block is the last block in the RLC window. This is called stalling and will lead to a lower throughput for the user.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a scheduling scheme for access networks, by means of which retransmission delay can be decreased even with large buffer storage at the Node-B.

This object is achieved by a method of scheduling data packets in a network de- vice of an access network, the method comprising the steps of: providing to said network device an information indicating a retransmission status of a data packet; and allocating a scheduling priority to said data packet based on said retransmis- sion status information.

Furthermore, the above object is achieved by a system for scheduling data pack- ets in an access network, said system comprising: at least one network controller device for providing an information indicating a retransmission status of a data packet; and at least one network device for allocating a scheduling priority to said data packet based on said retransmission status information.

Accordingly, retransmitted data packets can be prioritized during scheduling to thereby reduce the retransmission delay and increase the throughput as seen by the application or user. Large buffer storages can thus be allowed for at the base station devices, while the risk of RLC stalling and TCP retransmission is mini- mized.

The information may be provided as a flag information in a payload portion or in a header portion of a data frame comprising the data packet. This flag information can be allocated to the data packet. It may indicate whether the data packet or the data frame, respectively, is retransmitted or not. Thereby, a straightforward solu- tion requiring minor modifications at the network controller devices can be pro- vided. Any spare bit in the header or payload portion can be used to convey the flag information set by a dedicated setting operation at the network controller de- vice.

As an alternative, the information may be provided in an out-band signaling.

The data packet may be a MAC-d PDU, and the data frame may be an HS-DSCH data frame. The provided information may be set based on a corresponding infor- mation received from the RLC protocol layer.

Furthermore, the allocation step may comprise the step of placing the data packet at a predetermined position in a buffer queue of the base station device. Specifi- cally, the data packet may be placed to the front position of the buffer queue if the provided information indicates a retransmission. The allocation step may comprise the step of adjusting a reordering of the buffer queue if the information indicates a retransmitted data packet.

Alternatively, the allocation step may comprise the step of placing the data packet based on the provided information either to a retransmission buffer or to a first- transmission buffer. Then, data packets stored in the first-transmission buffer are scheduled when the retransmission buffer is empty. Hence, the network device may be arranged to schedule data packets of the first-transmission buffer after all data packets of the retransmission buffer have been schedule.

BRIEF DESCRIPTION OF THE DRAWINGS In the following, the present invention will be described in greater detail based on preferred embodiments with reference to the accompanying drawings, in which: Fig. 1 shows a schematic diagram of a network architecture in which the present invention can be implemented; Fig. 2 shows a schematic diagram of a frame structure according to a first em- bodiment; Fig. 3 shows a schematic diagram of a frame structure according to a second em- bodiment; Fig. 4 shows a schematic diagram of a signaling procedure for forwarding a re- transmission information according to the first and second embodiments;

Fig. 5 shows a schematic block diagram of a scheduling function according to a first example ; Fig. 6 shows a schematic block diagram of a scheduling function according to a second example; and Fig. 7 shows a schematic block diagram of a conventional window-based RLC transmission with a stalling situation.

DESCRIPTION OF THE PREFERRED EMBODIMENTS The embodiments of the present invention will now be described on the basis of a third generation WCDMA radio access network architecture as shown in Fig. 1.

Fig. 1 shows a terminal device or UE 10 connected via an air interface to a first Node-B 20 and/or a second Node-B 22. The first and second Node-Bs 20,22 are connected via respective lub interfaces to first and second radio network control- lers (RNCs) 30,32 which are connected to each other via a lur interface. The Node-Bs 20,22 are logical nodes responsible for radio transmission and reception in one or more cells to/from the UE 10 and terminate the lub interface towards the respective RNCs 30,32. The RNCs 30,32 are in charge of controlling use and integrity of radio resources within the radio access network. Furthermore, the RNCs 30,32 provide connections to a third generation core network 40, e. g. a UMTS (Universal Mobile Telecommunications System) network for both circuit- switched traffic via a lu-CS interface and packet-switched traffic via a lu-PS inter- face. The existence of an open standardized lur interface is essential for proper network operation, including soft handover support in a multi-vendor environment.

According to the embodiments of the present invention, a mechanism is provided to put data packets of RLC retransmissions in front of a buffer queue of a schedul- ing function provided in the Node-Bs 20,22 and related to the relevant traffic stream. To achieve this, a retransmission flag RF is introduced indicating a re- transmission of the concerned data packet or data frame. Of course, other indica- tion methods based on in-band and/or out-band signaling, i. e. within the con- cerned user channel and/or via a separate control channel, can be used as well.

The Node-Bs 20,22 can then adjust a reordering of the queue of data packets, if a data packet with a set retransmission flag arrives. An alternative is that the Node-B uses two queues per traffic stream. One can be used for first transmissions and

the other for RLC retransmissions. Once a user is to be scheduled, data is first read out of the retransmission buffer. When this one is empty, the original data can be scheduled.

These measures allow for large buffer storage in the Node-Bs 20,22 and therefore fast scheduling while simultaneously minimizing the RLC retransmission time..

This provides the advantage that the risks of TCP timeouts and RLC stalling are reduced, while L2 buffer requirements are loosened, since retransmission ac- knowledgments arrive faster due to the reduced RTT.

At the RNCs 30,32, MAC layer processing is arranged to set the retransmission flag RF for a retransmitted data packet or data frame.

Fig. 2 shows a first embodiment of a structure of an HS-DSCH data frame, where each MAC-d PDU has an allocated RLC retransmission flag RF. As can be gath- ered from Fig. 2, retransmission flags RLC are provided for each MAC-d PDU in the payload portion of the data frame. Some of the spare bits available in the data frame can be used as RLC retransmission flags. The first embodiment relates to the case that each MAC-d PDU can be marked as an RLC retransmission or not.

The RLC retransmission flag is set by the MAC-d protocol layer according to the retransmission status of the concerned MAC-d PDU, which the MAC-d protocol layer derive or obtain from the RLC layer. In the present embodiments, one bit is used per MAC-d PDU, which contains one RLC PDU. The location of the bit of the retransmission flag RF can be just in front of the MAC-d PDU. However, it can also be located at some other place. In Fig. 2, spare bit No. 4, of the spare bits in front of each MAC-d PDU is used as the retransmission flag RF. For example, if the retransmission flag RF is set to"1", the respective MAC-d PDU is a retransmitted data packet, and if it is set to"0", the respective MAC-d PDU is transmitted for the first time. Of course, the opposite setting logic may be used as well.

Fig. 3 shows a second embodiment of the structure of the HS-DSCH data frame, where one HS-DSCH data frame has only one RLC retransmission flag RF allo- cated to the whole data frame, i. e. to all MAC-d PDUs in the payload portion. In this case, there is only one RLC retransmission flag provided in the header portion of the data frame. This means that all the MAC-d PDUs in the payload portion of the HS-DSCH data frame are retransmission packets in case the RLC retransmis- sion flag is set.

Fig. 4 shows an example of a signaling and processing flow in case RLC retrans- mission flag (s) RF is/are included in the HS-DSCH data frame. This signaling flow covers both of the above first and second embodiments.

In the example of Fig. 4, one of the RNCs 30,32 of Fig. 1, e. g. the lower RNC 32, is a Drift RNC (D-RNC), and the other, e. g. the upper RNC 30, is a serving RNC (S-RNC), as regarded from the viewpoint of the UE 10. The S-RNC 30 is in charge of the RRC (Radio Resource Control) connection between the UE 10 and the radio access network. Thus, the S-RNC 30 terminates the lu connection. The D-RNC 32 supports the S-RNC 30 with radio resources when the connection between the radio access network and the UE 10 needs to use cell (s) controlled by the D-RNC 32, e. g, via the Node-B 22. According to Fig. 4, the S-RNC 30 generates a HS- DSCH data frame with a retransmission flag setting as RLC retransmission indica- tor and forwards the data frame to the D-RNC 32 using an HS-DSCH frame proto- col (HS-DSCH FP) signaling (step 1). Then, the D-RNC 32 forwards the HS-DSCH data frame to the Node-B 22 again using the HS-DSCH FP signaling (step 2). At the Node-B 22, a packet scheduling functionality or packet scheduler considers the retransmission flag setting during its scheduling operation (step 3). If the UE 10 is located within the coverage area of the cells controlled by the S-RNC 30, the HS-DSCH data frame can be forwarded directly to the UE 10.

Once the Node-B 22 has received the flag information RF, it will utilize it during the packet scheduling. Possible ways of prioritizing RLC retransmissions in the Node- Bs 20,22 are described in the following based on Figs. 5 and 6.

The manner in which queued data packets are selected for transmission on the link is known as link scheduling discipline. Figs. 5 and 6 show first and second ex- amples of a queuing model abstraction, e. g. for a Fist-in-First-out (FIFO) link scheduling principle. Packets arriving at the link output queue are queued for transmission if the link is currently busy transmitting another packet. The conven- tional scheduling function selects packets for link transmission in the same order in which they arrived at the link queue.

Fig. 5 shows a scheduling function according to the first example which can be used in connection with both first and second embodiments. The Node-Bs 20,22 comprise a checking functionality or unit 210 for checking the retransmission flag setting of a received data frame or data packet.. Based on the detected flag set- ting, a scheduler functionality 220 is controlled so as to allocated a scheduling pri-

ority to the concerned data packet or data frame, e. g. , by adjusting the queue or- der of data packets in a buffer memory 230 to place retransmitted data packets at a front position of the queue. Thereby, the queue of data packets in the buffer memory 230 is reordered to prioritize retransmitted data packets. In particular, re- transmitted data packets may be placed in front of the queue.

Fig. 6 shows a scheduling function according to the second example which also can be used in connection with both first and second embodiments. Here, two buffer memories 230,240 are provided, i. e. a retransmission buffer memory 240 for storing retransmitted data packets and a first-transmission buffer memory 230 for storing data packets transmitted for the first time. Furthermore, a controllable switching functionality 250 is provided which can be controlled by the scheduler 220 in response to the checking result of the checking unit 210 so as to switch or place the received data packet to one of the buffer memories 230,240 based on the setting of the allocated retransmission flag RF. The switching unit may be implemented as hardware function, e. g. a demultiplexer, a semiconductor switch or a logical gate, or a software function controlling a memory read/write operation.

The scheduler functionality 220 is adapted to schedule the data packets stored in the retransmission buffer memory 240 at first. Then, a selection functionality 260 can be provided which selects the first-transmission buffer 230 as soon as the re- transmission buffer memory 240 is empty, so that the non-retransmitted data pack- ets are then scheduled by the scheduler 220. As an alternative, the selection of the selection functionalty 260 may be provided by the scheduler functionality 220, such that the selection functionality 260 can be dispensed with.

The scheduling functionalities according to the above first and second examples may be implemented in the Node-Bs 20,22 as concrete hardware functions or as software routines controlling a processing unit with an allocated read/write or ran- dom access memory.

It is noted that the present invention is not restricted to the above preferred em- bodiments but can be used in any network device having a scheduling functional- ity. Moreover, any kind of signaling can be used for forwarding the retransmission information to the scheduling network element. The preferred embodiments may thus vary within the scope of the attached claims.