DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0062] Referring now to FIG. 2 , shown is a priority-based state scheduler provided by an embodiment of the invention. This might for example be implemented within a base station transceiver such as BTS 10 of FIG. 1 .

[0063] This state scheduler is generally indicated at 20 , and has a set of inputs for each user/mobile station to be state scheduled and ultimately defines an active mode user pool 26 which is used by a scheduler 32 . In the illustrated example, the input 22 for “user 1 ” is shown to include channel conditions, packet transmission deadline, minimum data rate, drop rate, buffer occupancy, buffer empty time, delivered data rate per user. These are very specific input parameters which are included for the purpose of illustration only. Additional parameters may also be in employed and/or completely different parameters. A set of inputs is shown for each user in the system. In the illustrated example, it is assumed that there are N users. The input parameters may be actively passed to the state scheduler 20 , or alternatively and/or in combination, the state scheduler may simply access stored data for each user. Typically, the state scheduler would be implemented as a combination of software and processing hardware. For example, this could be implemented in an ASIC, a general purpose processor with software, an FPGA, DSP, etc.

[0064] The state scheduler 20 processes the data received in respect of each of the users and performs a priority ranking of the users. The actual priority ranking is discussed in further detail below. The users that are prioritized at the top, for example the first Na users, are then assigned the most resource intensive state, for example in the illustrated example the previously described “active state”. Similarly, the next highest prioritized group of users, for example users prioritized from rank Na+1 to Na+Nc, would be assigned the next most resource intensive MAC state, for example previously described “control-hold” state. The remaining users, namely those ranked from Na+Nc+1 through to user N are assigned the least resource intensive state, for example the previously described “dormant” state. FIG. 2 shows the priority ranking produced by the state scheduler generally indicated by 24 , and shows the active mode user pool 26 , a control-hold user pool 28 , and a dormant user pool 30 . The base station with MAC layer protocol includes messaging which is designed to inform users of changes in their state. For example, a user transitioning from control-hold to active mode would be informed of this change. Such messaging is conventional and will not be described in detail here.

[0065] Referring again to FIG. 2 , after state scheduling has been performed by the state scheduler, and the priority ranking and state updating completed, the active mode user pool 26 alone is used by the packet scheduler 32 . Packet scheduling can proceed as in conventional systems without the need for any change. However, as discussed previously, the packet scheduler typically does perform a prioritization algorithm of its own in determining which user is to be selected for packet transmission. In a preferred embodiment of the invention, the state scheduler 20 employs a priority algorithm which is similar to in as many respects as possible the prioritization algorithm employed by the packet scheduler 32 .

[0066] It is noted that the rate of update of states implemented by the state scheduler 20 would typically be substantially slower than the rate of prioritization and packet scheduling by the packet scheduler 32 . For example, where packet scheduling might be performed by the packet scheduler every 1.25 ms, it may only be necessary to update the states of each of the users every 100 scheduling periods, i.e. every 0.125 s. For this reason, it is not possible for the state scheduler 20 to perform identically to the packet scheduler 32 . The state scheduler 20 necessarily has to operate on different information than the packet scheduler 32 . In a preferred embodiment, where the packet scheduler 32 may perform prioritization on the basis of real-time or short-term data, for a given parameter, for example real-time C/I, the state scheduler employs long-term statistics, for example filtered C/I. The state scheduler 20 can use identical equations but employs filtered or long-term values instead of real-time or short-term values for example.

[0067] The embodiment shown in FIG. 2 assumes three MAC layer states. FIG. 3 is a flowchart of the state scheduling in its most simple implementation, namely when there are only two MAC layer states. The method begins at step 3 - 1 with the ranking of all power on users. Next, at step 3 - 2 , subsets of power-on users are identified according to rank for the number of different states. At step 3 - 3 , layer 2 or layer 3 state transition messages are generated and transmitted such that the required state transitions are actually implemented in the mobile stations. Finally, at step 3 - 4 , packet scheduling is performed for active users until the next update of the states.

[0068] As an example of a state scheduler priority algorithm, the following equation may be employed:

P _i =f ( r _{selected — i} , T _i , {overscore (r)} _i *, D _i , E _i , d _r,i )

[0069] where r _{selected — i} is the selected data rate for user i based on C/I, T _i is the average data rate for user i, {overscore (r)} _i * is the minimum required data rate for user i, D _i is the minimum packet deadline for user i, E _i is the data buffer empty time for user i, d _r,i is the drop rate for user i, and P _i is the calculated priority for user i.

[0070] If some parameters are not available for users in a certain state, a constant value is used then. For example, the channel condition for a dormant user is not available, and a constant value, e.g., 0 dB is used.

[0071] As discussed above, certain parameters may not be available for one or more of the states and as such constant values may be assigned. Alternatively, different equations may be employed for users in different states.

[0072] FIGS. 4 and 5 show example performance evaluation results. In FIG. 4 , the outage rate is shown for a 2-state system with mixed traffic without video. In this example the traffic mix is 27.1% HTTP, 62.6% WAP, and 10.3% FTP. Performance results are shown at curve 40 for timer-based state transitions, and at curve 42 for priority-based state transitions. The horizontal axis is the number of users. It can be seen that for the same outage rate, the priority-based state transition scheduling can support a larger number of users. For example for an outage rate of 20%, approximately 80 users can be supported in the priority-based scheduling scheme, where only about 55 can be supported in the timer-based system.

[0073] Table 1 below shows further results for this same performance evaluation example. Table 1 shows the sector throughput, system capacity, and the respective gains in these two parameters when switching from the timer-based state transition to the priority-based state transition. The system capacity for this example consists of the numbers of users supported with an outage rate of 2%. 1

[0074] FIG. 5 shows similar results for the 3-state example. Curve 44 shows the outage rate for the timer-based state transition system, and curve 46 results for the timer-based approach. Again, it can be seen that for a given outage rate, a larger number of users can be supported using the timer-based approach. Once again, Table 2 is provided below showing further performance results. 2

DETAILED EXAMPLE OF THE STATE SCHEDULER IMPLEMENTATION

[0075] A detailed example of a state scheduler implementation will now be presented starting with an assumed packet scheduler design.

[0076] Packet Scheduler:

[0077] This example assumes that a proportional fairness packet scheduler with β set to 1.

[0078] The packet scheduler calculates a selected data rate r _{selected — i} for each user assuming all users' buffers are full based on the reported C/I from each user in active state and in control-hold state. The relationship between C/I and the selected data rate is typically not a linear one and may be implemented using a table look-up.

[0079] The packet scheduler calculates an average throughput T _i for each user with full data buffer (buffer full indicator B_f _i =1) based on:

T _i ( k )=α· T _i ( k− 1); if user i was not selected at ( k− 1)th scheduling instant. Eq(2-1)

T _i ( k )=α· T _i ( k− 1)+(1−α)· N _i ( k− 1); if the user was scheduled N _i ( k− 1) bits at ( k− 1)th scheduling instant. Eq. (2-2)

[0080] where α=1− ^{1.25·10 −3} / _1.5

[0081] The packet scheduler calculates the packet priority 1 $\begin{array}{cc}\mathrm{by}P_i=\frac{r_{\mathrm{selected\_i}}}{{\left(T_i\right)}^{\beta }}& \mathrm{Eq}.\left(3\right)\end{array}$

[0082] The operation of the packet scheduler for this example is summarized in FIG. 6 . The input generally indicated 60 by to the scheduler 62 consists of the C/I readings for the users, and B_f indicators for all the users. The selected data rate is determined for each user, for example based on predefined tables. The average throughput is calculated as above, and the priority is also calculated above and output, as indicated at 64 . These calculations might for example take place every 1.25 ms. These priorities 64 are then low-pass filtered by the state scheduler with low-pass filter 66 to generate the state priorities, P _{state — i} for active users. The packet scheduler may also calculate priorities P _i for Control-Hold users and filter these to generate P _{state — i} . Only active users are ranked for packet scheduling. It is not important where the various priorities are generated.

[0083] State Scheduler

[0084] As indicated above, the state scheduler (or packet scheduler) filters the packet priority to acquire the state priority. The state scheduler also generates state priorities for dormant users, preferably using the same mechanism as was used for active users, although not all the information may be available. C/I is not typically available for a dormant user for example and might be assumed to be some constant value, for example 0 dB.

[0085] The operation of the state scheduler will be described with reference to FIG. 7 . The inputs to the state scheduler 70 consist of the state priorities, P _{state — i} 68 and buffer empty time periods E _{t — i} , and the output consists of an active user list, control-hold user list and dormant user list.

[0086] At the state update instant:

[0087] a) The state scheduler ranks all full buffer users based on their state priority—this includes all users in active state, control-hold state and dormant state with full buffer;

[0088] b) For all empty buffer users in active state and in control-hold state, the state priority is modified by P _{state — i} =P _{state — i} /E _{t — i} , where E _{t — i} is the buffer empty time period of user i. The state scheduler ranks those users based on the modified state priority;

[0089] c) Dormant users with empty buffer are not ranked;

[0090] d) The two ranks are then combined with the users with full buffer having higher priority.

[0091] In some embodiments, the ranking thus determined is used as is to generate the Active, Control-hold and Dormant user groups, simply by taking the top Na users for Active, the next Nc users for control-hold, and the remaining users for Dormant.

[0092] In some embodiments, one or more constraints are imposed on which state transitions are allowed to occur, and these constraints are used to adjust the rankings prior to generating the user groups.

[0093] In some embodiments, the rule that a dormant user with full buffer is not allowed to be moved to control-hold state is followed. For example, a previously dormant user with full buffer may be moved out of the ranking altogether if its rank places it in the control-hold group.

[0094] The state scheduling process is performed less frequently than the packet scheduling, and might be carried out every M×1.25 ms for example, where M is an integer >1. In one embodiment, M is 100.

[0095] Ranking Strategy 1

[0096] In this ranking strategy, a full buffer user in active state or in control-hold state is allowed to be moved to the dormant state. This strategy will be described with reference to FIG. 8 .

[0097] Generally indicated at 86 is the ranking of users based solely according to state priority. This includes a top most ranked set of Na users, the next highest rank set of Nc users and the remaining users. Each users previous state is indicated by an appropriate shading. Generally indicated at 80 are the users that were previously in the active state or control-hold state and have a full data buffer. Generally indicated at 82 are users with an empty data buffer. Finally, generally indicated at 84 are users previously in the dormant state that have a full buffer. It can be seen in this example that the top Na users include one dormant user with a full buffer, and the next Nc users include one dormant user with a full buffer.

[0098] With this ranking strategy, dormant users are not allowed to move from dormant state into control-hold state, but are allowed to move into active state. As such, any dormant users that are initially ranked to put them in the control-hold state are removed from the ranking. This is indicated symbolically by arrow 87 which shows a dormant user with full buffer being moved from the group of Nc users down to the bottom group. Generally indicated at 88 is the ranking after this re-organization has taken place. The user that was formerly in the group Nc users that was previously in dormant state has been moved down to the bottom, and one of the users which formerly fell outside of the group of users ranked in the top Na+Nc has now been moved into the control-hold user ranking range.

[0099] Ranking Strategy 2

[0100] In this ranking strategy, a full buffer user in active state or in control-hold state is not allowed to be moved to dormant. This strategy will be described with reference to FIG. 9 .

[0101] In this example, the ranking based on state priority is generally indicated by 90 . The users in active or control-hold state with full data buffer are indicated at 92 , users in dormant state with full data buffer are indicated at 94 , and users with an empty data buffer are indicated at 96 .

[0102] It can be seen that after the initial ranking, the top Na users include users in the active or control-hold state with full buffer, and include two users in dormant state with full buffer. The next Nc highest ranked users include active or control-hold users with full buffer, and include a user in the dormant state with full data buffer. The users ranked below Na+Nc include two users previously in the active state with full data buffer, indicated at user _— 3, user _— 2 respectively, and include one user, user _— 1, in the dormant state with full data buffer.

[0103] In this ranking strategy, as before users which would be moved from dormant to control-hold state are removed off the ranking. Thus the user in dormant state ranked in the Nc users is moved out of that group off the ranking, or equivalently to the bottom. This leaves one space for an active user that was not previously in the top Na+Nc users to move into the group of Nc. However, after this takes place there is still one user, user _— 2, which is ranked outside the top Na+Nc, and this ranking strategy does not allow this. As such, the lowest ranked dormant user which remains in the top Na is moved off the ranking to the bottom and the remaining users are bumped up accordingly. This provides room in the group Nc for the remaining active user, user _— 2. The arrangement of rankings after such an arrangement is indicated generally at 98 .

[0104] Priority Plus Threshold Algorithm

[0105] In order to decrease the possibility of moving an active user or control-hold user to dormant state during a packet call (e.g., a HTTP page download), the buffer empty time threshold can be used. It means that a user with an empty buffer is not moved to dormant if its buffer empty time is not larger than the pre-defined threshold. Such a user can still be viewed as a virtual full buffer user. Therefore the ranking strategy 2 can be employed.

[0106] As indicated previously, messaging is employed to instruct mobile stations to move between the various states. While a user is in active or control-hold state, a channel has been set up, and a base station can communicate directly with the mobile station to instruct an appropriate state transition to occur. On the other hand, when a mobile station is in the dormant state, there is no such channel available, and a broadcast type channel suggests a paging channel would typically be employed. Also, in a multi-base station environment, it is preferred to keep all base stations that are communicating with the mobile station up to date on the state of the mobile station.

[0107] The State scheduler is preferably implemented in the BTS. However the BSC should be updated with all user's state. The indication of state transitions to a mobile by a BTS is via over-the-air messages and the communication between BTSs and BSC is through internal signaling. A dormant user associated with a BTS is a user who is currently monitoring the paging/broadcast channel of this BTS. This is known by receiving a resource request from this dormant user or receiving the user's response to a page by this BTS.

[0108] A symbol example of how messaging might take place is shown in FIG. 10 where it is assumed that there is a base station controller 100 , a first base station BTS 1 102 , and a second base station BTS 2 104 communicating with a mobile station MS 106 . This scenario begins with BTS 1 102 instructing mobile station 106 to move into the active state with message 108 . BTS 1 102 also informs the base station controller 100 which in turn informs any base stations including BTS 2 104 of the state change. Next BTS 1 102 instructs the mobile station 106 to transition to the control-hold state as indicated at message 110 . This message is also conveyed through to the base station controller 100 and BTS 2 104 as before. Next, BTS 1 102 moves the mobile station 106 back into the active state as indicated by message 112 . Once again the BSC 100 and BTS 2 104 are updated. Next, it is assumed that a handoff takes place between BTS 1 102 and BTS 2 104 and as such the next message comes from BTS 2 104 instructing the mobile station 106 to move into the control-hold state. The ranking of a user might change when it moves from a first base station such as BTS 1 102 to a second base station such as BTS 2 104 and this may result in such a state transition. Now, BTS 2 104 sends messages to update the BSC 100 which in turn updates BTS 1 102 . The next step in the scenario has BTS 2 104 sending a message to the mobile station 106 generally indicated at 116 instructing the user to go into the dormant state. Again, BSC 100 and the BTS 2 104 are updated. Now the user is in dormant state and there is no active channel from either of BTS 1 or BTS 2 to send state transition messages. Thus, when it then becomes time to move the user back into the dormant state, the BSC 100 initiates the messaging by instructing both BTS 1 102 and BTS 2 104 to generate paging messages to the user instructing it to move into the active state. After a delay, the user does go back into the active state. It should be understood that this is a very specific example provided solely for the purpose of illustration, and that any sequence of state transitions is typically possible subject to any constraints which might be imposed upon the system. For example, the state transition from dormant to control-hold is not allowed in some scenarios. There may be more than two base stations involved at a given time or there may be only one involved at a given time in which case there is no updating of other base stations. Furthermore, some systems may not require a separate base station controller.

[0109] Numerous modifications and variations of the present invention are possible in light of the above teachings. It is therefore to be understood that within the scope of the appended claims, the invention may be practised otherwise than as specifically described herein.