1. Impact of Radio Resource Allocation Policies on the TD-CDMA System Performance JSAC, Vol. 19, No. 10, October 2001

2. Main Research Axes  Assessment of how important the impact of various parameters and policies could be on the overall TD- CDMA system performance. –User behavior: mobility –Service constraints: source activity, delay constraint –Radio algorithms: power control, handover, joint detection –Network topology: inter-site distance –Radio resource management –Loading metric: number of codes vs. interference (per time slot) –Neighboring list size

3. Simulation Scenario

4.  It is assumed that all the BSs are synchronized.  The propagation model takes into account both propagation along street segments and above rooftops.  Hard handover is supported.  Simulated services are speech service in both circuit and packet switched mode, and 64kb/s circuit data service.  Only downlink case is presented.

5. Simulation Scenario  System performance measure –capacity (Erlangs) at a certain level of satisfied users (e.g. 98%). –capacity at a target value for the frame error rate (FER).  Quality of Service Criteria –Blocking criterion –Call quality criterion –Dropping criterion

6. Impact of Environmental Properties  A. Circuit vs. ON/OFF Sources

7. Impact of Environmental Properties  FER Behavior in light traffic load –the FER is even worse for packet switching than for circuit due to the activity of ON/OFF sources.  FER Behavior in medium-heavy traffic load –The gap between the two operation modes increases dramatically in favor of packet switching.

8. Impact of Environmental Properties  B. Impact of Manhattan Structure

9. Impact of Environmental Properties  The bloc size affects the capacity. –A slight improvement in terms of capacity is expected with smaller bloc size. This is mainly due to shorter distances between BSs, which increases the number of accepted users connected to farther BSs. –A higher FER is introduced with smaller bloc size, for users being connected to farther BSs have stronger impact on the global interference level and contribute more to the FER statistic. The impact becomes more important at heavy traffic load since users connected to farther BSs suffer from combined effect of loading (more new calls accepted) and interference.

10. Impact of User Behavior  C. Impact of Mobility

11. Impact of User Behavior  Mobility has serious impact on FER statistic and satisfied users statistic, both for circuit and ON/OFF sources. –There is less FER loss, with respect to the static scenario, in packet switching mode than in circuit mode. It is why a more complex satisfaction users’ criterion (considers also blocking statistic for new access requests) is more adapted for circuit switching while an FER criterion is more suitable for packet switching. –For circuit connections there was more limitation due to satisfied user criteria than to FER, while for ON/OFF sources FER imposes a strong capacity limitation. It is why a more complex satisfaction users’ criterion (considers also blocking statistic for new access requests) is more adapted for circuit switching while an FER criterion is more suitable for packet switching.

12. Impact of Service Constraints  D. Influence of User Dropping Threshold  Which action (dropping /keeping alive) should be taken for users falling in bad radio conditions.  Which finite value of the “dropping threshold” is desirable? Which value approaches the “no dropping” boundary?

13. Impact of Service Constraints  E. Impact of Handover Margin  At a similar performance, it is better to operate at higher margin values to reduce the signaling.  Optimal handover margin has to be evaluated for each particular network and propagation models since it is highly dependent on the particular scenario.

14. Impact of Service Constraints  F. Impact of Initial Power Value  Basic algorithms require a constant initial value, this they are not sensitive to the current loading and individual radio conditions.

15. Impact of Service Constraints  1) FER statistic: –When starting at the highest allowed power level, it takes time for the MS to converge. Once the convergence is reached, other new calls are likely to be generated, entering always at the maximum allowed power. Consequently C/I value might be no longer good, thus the power control loops will try to ask for more power, and all frames will be lost until the C/I condition is again satisfied. –When starting at the lowest allowed power level, the applied initial power is very likely to be less than the required value for an acceptable quality. The power control loop will request more power to reach the satisfactory C/I value. During the convergence steps, the erroneous frames will be lost.

16. Impact of Service Constraints  G. Joint Detection Imperfections  Joint detection receivers are multiuser detectors which combat both ISI and MAI by exploiting the knowledge about the spreading sequences and channel impulse responses.

17. Impact of RRM Policies  H. Evaluation of Some Resource Allocation Schemes

18. Impact of RRM Policies

19.  Channel allocation strategies: –Neighboring list codes loading (NLCL) the number of codes in use per time slotThe load is expressed as the number of codes in use per time slot. any BSA call is admitted to any BS in the current neighboring list, provided the C/I target is reached and there are enough resources available. –Best available codes loading (BACL) the number of codes in use per time slotThe load is expressed as the number of codes in use per time slot. best BSA call can only be connected to the best BS among those in the list. –Neighboring list interference loading (NLIL) the interference level measured in each time slotThe load is expressed as the interference level measured in each time slot. Time slots are ordered from the lowest to the highest interference level. The TS of the lowest interference level is selected.

20. Impact of RRM Policies  a) Circuit-switched mode This is because resources available are better distributed among BSs. –At light traffic load, a large neighboring list of potential servers and a strategy based on the interference level measured by the user exhibits great advantage. This is because resources available are better distributed among BSs. This is due to the fact that power budget is less affected since connections to farther BSs are not allowed. –At heavy traffic load, a strategy based on a smaller neighboring list is suitable. This is due to the fact that power budget is less affected since connections to farther BSs are not allowed. (the operators’ job) –The neighboring list size can be tailored based on the desired tradeoff between carried traffic and individual radio link quality (the operators’ job).

21. Impact of RRM Policies  Implementation trend: a large neighboring list at light traffic load while a small list at medium-heavy one (the switching point is operator-defined) –The access to farther BSs is excluded at medium-heavy traffic load. –Advantages Better link quality. Less measurements. Reduce the time need to scan the neighboring list at heavy load. Reduced signaling. Better battery life. Simplified power control and DCA operations. less carried traffic at heavy traffic load –Main inconvenience is less carried traffic at heavy traffic load.

22. Impact of RRM Policies  b) ON/OFF sources –Due to the ON-OFF properties, the FER performance is worse, and an explosion of the number of users having bad quality calls is resulted. And hence there’s no similar interest to have a reduced neighboring list at heavy traffic load (but the advantages in signaling, battery life, DCA and PC simplification hold true). –An aggressive CAC is absolutely necessary in packet switching mode to support a desirable FER at a reasonable traffic load and to avoid severe call dropping. –In packet switching mode there’s no similar interest to have a strong reduction of neighboring list size at heavy traffic load: the link quality is not really much better.

23. Impact of Mobility on High-Rate Sources  I. 64kb/s LCD Service (circuit mode)

24. Conclusion  The system capacity depends highly on the selected satisfaction criteria. –More complex criteria mixing new call blocking and individual radio link quality are more appropriate for circuit switching while a simple FER criterion is more suitable for packet switching. “bad quality call”  Packet switching is much more sensitive to the “bad quality call” criterion than to other criteria.  Packet switching generally outperforms circuit switching. –However, at light traffic load the FER is worse, for the activity of ON-OFF sources induces more disturbance into the system.  Shorter intersite distance introduces higher capacity. –The number of available servers is increased. –The FER is worsened, however.

25. Conclusion  Mobility has strong impact on the system capacity.  Fewer users are dropped with a higher dropping threshold, but more calls will end up as “bad quality calls”.  An adaptive initial power control to C/I at the channel assignment phase gives the best capacity performance.  In circuit mode, there’s a considerable increase in capacity if the neighboring list size is adaptive. It is not the same case with packet mode, however.  The most exigent criterion is the FER and it is the real satisfaction indicator for packet switching operation.