1. doc.: IEEE 802.11-09/0496-00-00aa Submission Apr 2009 Graham Smith, DSP GroupSlide 1 Considerations for Statistical Multiplexing Support in OBSS Proposal - QLoad Date: 2009, April 29 Authors:

2. doc.: IEEE 802.11-09/0496-00-00aa Submission Apr 2009 Graham Smith, DSP GroupSlide 2 Abstract This presentation first looks at the statistics of video streams and then how fields in the QLoad Element, proposed in OBSS solution “OSQAP”, could be added in order to support statistical multiplexing of the video loads. This presentation recommends a new version of the QLoad Element

3. doc.: IEEE 802.11-09/0496-00-00aa Submission Apr 2009 Graham Smith, DSP GroupSlide 3 Introduction The original OBSS Proposal “OSQAP” suggested a new QLoad Element for the sharing of overlapping QAPs – 08/457r4, 08/1260r1, 09/230r0 This QLoad element included fields for: Overlap QLoad Self QLoad Total The QLoad Total represents the aggregate of “QLoad self” from all the QAPs in the OBSS graph. The use of simple addition of the QLoad Totals by overlapping QAPs was suggested and basically using total Peak Load as basis for Sharing. Ed Reuss (Plantronics) and Brian Hart (Cisco) suggested that the QLoad should support statistical multiplexing so to be more efficient. In this presentation, this is investigated.

4. doc.: IEEE 802.11-09/0496-00-00aa Submission Apr 2009 Graham Smith, DSP GroupSlide 4 Video Throughputs Samples of throughputs of three actual individual video clips is shown below. Video 1Video 2Video 3 MAX Mbps11.410.08.6 MIN Mbps3.38.13.6 MEAN Mbps7.99.25.8 Video 1 Video 2 Video 3

5. doc.: IEEE 802.11-09/0496-00-00aa Submission Apr 2009 Graham Smith, DSP GroupSlide 5 Video Throughputs The TOTAL throughput of all three videos, “composite video”, is shown below MAX Mbps27.6 MIN Mbps16.6 MEAN Mbps22.8 Composite Stream for all 3 Videos

6. doc.: IEEE 802.11-09/0496-00-00aa Submission Apr 2009 Graham Smith, DSP GroupSlide 6 Simple Addition of the three does not result in the Composite MAX Mbps30.0 MIN Mbps15.0 MEAN Mbps22.8 Addition of statistics for all 3 Videos MAX Mbps27.6 MIN Mbps16.6 MEAN Mbps22.8 Composite Stream for all 3 Videos Video 2 is relatively constant, so based upon Videos 1 and 3, we get: Based upon MAX Mbps, then simple addition produces 8.7% Over allocated MAX Mbps17.7 MIN Mbps7.6 MEAN Mbps13.7 Composite Stream for Videos 1 and 2 MAX Mbps20.0 MIN Mbps6.9 MEAN Mbps13.7 Addition of statistics for Videos 1 and 2 Based upon MAX Mbps, then simple addition produces 13% Over allocated

7. doc.: IEEE 802.11-09/0496-00-00aa Submission Apr 2009 Graham Smith, DSP GroupSlide 7 VIDEO STATISTICS Video 1Video 2Video 3Composite MEAN7.929.165.7622.84 MAX11.4010.018.5527.65 MIN3.318.143.5716.62 STDEV1.840.371.412.22 Video 1Video 2Video 3Composite MEAN7.929.165.7622.84 +2σ11.599.918.5727.27 -2σ4.258.412.9418.41 Statistics for the Video streams, including “standard deviation”, are: Note that MAX and MIN can be estimated as MEAN +/- 2 STDEV

8. doc.: IEEE 802.11-09/0496-00-00aa Submission Apr 2009 Graham Smith, DSP GroupSlide 8 Video Statistics HISTOGRAMS and NORMAL DISTRIBUTIONS

9. doc.: IEEE 802.11-09/0496-00-00aa Submission Apr 2009 Graham Smith, DSP GroupSlide 9 Video Statistics If each video stream can be represented by a Normal Distribution, then the sum of the streams is also a Normal Distribution Note: Summation of Normal Distributions: Meanµ = Σµ i Stddevσ = sqrt(Σσ i 2 ) Very good correlation between Actual composite and Sum of three videos

10. doc.: IEEE 802.11-09/0496-00-00aa Submission Apr 2009 Graham Smith, DSP GroupSlide 10 Video Statistics So far we can conclude the following: Based upon the three sample videos: Individual Video stream statistics can be reasonably modeled by a Normal Distribution Composite video can be modeled by a Normal Distribution Summation of the individual normal distributions for each video stream produces distribution that is close to the actual composite video normal distribution Max and Min can be estimated as –MAX = Mean + (2 x Standard Deviation) –MIN = Mean – (2 x Standard Deviation) HENCE: We now know how to sum the individual streams

11. doc.: IEEE 802.11-09/0496-00-00aa Submission Apr 2009 Graham Smith, DSP GroupSlide 11 Mean, Max and Min We have shown accurate representation of the composite video by summation of the Normal Distributions for each stream: Sum of Normal Distributions: Meanµ = Σµ i and Stddevσ = sqrt(Σσ i 2 ) AlsoMAX = Mean + 2σ andMIN = Mean - 2σ Hence, we estimate the total MEAN and STDEV from the individual streams: MEAN µ = Σ MEAN i STDEV σ = 0.25 sqrt {Σ (MAXi – MINi) 2 } (see note) Using resulting µ and σ, we can calculate total MAX and MIN MAX Mbps27.65 MIN Mbps16.62 MEAN Mbps22.84 Actual Composite Stream 3 Videos MAX Mbps27.68 MIN Mbps17.99. MEAN Mbps22.84 Estimated Composite Stream 3 Videos Very good!! NOTE: Calculating STDEV from just MAX or just MIN does not give accurate result MAX calculated based on square root of MAXi 2 produces MAX tot = 31.55Mbps

12. doc.: IEEE 802.11-09/0496-00-00aa Submission Apr 2009 Graham Smith, DSP GroupSlide 12 CDF The CDF then shows the probabilities of transmitting at a certain data rate. MAX 90% NOTE: 90% = 1.3sigma 80% = 0.83sigma

13. doc.: IEEE 802.11-09/0496-00-00aa Submission Apr 2009 Graham Smith, DSP GroupSlide 13 SUMMARY SO FAR 1.If TSPECs include MAX, MIN and MEAN information, the HC can calculate the MAX, MIN and MEAN for the composite requirement for that AP 2.If this information included in QLoad, overlapping QAPs can calculate the total MAX, MIN and MEAN for total traffic Could add MAX and MIN to QLoad but ALTERNATIVELY and BETTER Just include MEAN and STDEV in QLoad –QAP calculates STDEV from the MAX, MIN and MEAN given in the TSPECs Given this information, we know how to calculate the total requirement. Is it practical?

14. doc.: IEEE 802.11-09/0496-00-00aa Submission Apr 2009 Graham Smith, DSP GroupSlide 14 TSPECs and MAX, MIN, MEAN What if TSPEC does not include MAX, MIN and MEAN? e.g. Admission Control TSPEC only mandates MEAN In case of voice or CBR traffic: MEAN=MAX=MIN In case of Audio/Video: Unknown and variable (VBR traffic) OPTIONS Assume ‘standard’ STDEVs for Audio and Video –1.84, 1.41 and 0.87 were values for videos used in this presentation –Could look at many samples, audio and video, and determine “standard” values for Video and Audio (related to codec?) Assume MEAN=MAX=MIN –If STA did not generate full information, it does so at its own peril. IF only MAX and MEAN provided, then STDEV can still be calculated

15. doc.: IEEE 802.11-09/0496-00-00aa Submission Apr 2009 Graham Smith, DSP GroupSlide 15 QLoad Simple addition of the QLoad traffic is now not used, therefore: Proposal QLoad Element is amended to include the MEAN and STDEV for the total traffic for that QAP: Note that each QAP must calculated the Self MEAN and STDEV using: –MEANµ = Σ MEANi –STDEVσ = 0.25 sqrt{ Σ (MAXi – MINi) 2 } Note: The original “rules” for simply adding the QLoads no longer apply

16. doc.: IEEE 802.11-09/0496-00-00aa Submission Apr 2009 Graham Smith, DSP GroupSlide 16 Proposed Extended QLoad Element SEE LATER

17. doc.: IEEE 802.11-09/0496-00-00aa Submission Apr 2009 Graham Smith, DSP GroupSlide 17 QLoad Element Fields Overlap Number of APs that are sharing this channel and are overlapping QLoad MEAN and STDEV The mean and standard deviation of the total traffic presented to the QAP by TSPECs from STAs associated to that QAP QAP ID First octet = random number (0 to 255) Second octet = octet 6 of MAC Address Once selected, QAP retains this ID Chosen so that it is still possible to know which specific QAP this is QAPs need recognize their own QLoad Visible Bit If the QAP that corresponds to ID, MEAN and STDEV values is directly visible to the QAP Self, then this is set to 1 Visibility bit set to 1 for Self Channel Priority Used only if QAP is operating with HCCA, indicates HCCA Supervisor QAP Priority Streams Number of streams on ACs 2 and 3 per QAP. Used so that the contention overhead can be estimated.

18. doc.: IEEE 802.11-09/0496-00-00aa Submission Apr 2009 Graham Smith, DSP GroupSlide 18 Example

19. doc.: IEEE 802.11-09/0496-00-00aa Submission Apr 2009 Graham Smith, DSP GroupSlide 19 Benefits of changes to QLoad Element Each QAP in the OBSS Graph now knows the following information: OBSS size –The sum of all the QAP IDs in its QLoad Element How many hidden QAPs in the OBSS Graph –The sum of all the Visibility Bits = 0 The individual QLoads of each QAP in the OBSS Graph The QLoads of those QAPs that are directly overlapping (visible) and therefore contend for the same air time –Important as EDCA efficiency reduces as traffic increases on same Access Category HOW ABOUT SHARING RULES?

20. doc.: IEEE 802.11-09/0496-00-00aa Submission Apr 2009 Graham Smith, DSP GroupSlide 20 Sharing Given the MEAN and STDEV values for each QAP in the OBSS Graph, every QAP can now calculate: Total Peak traffic Total Mean Traffic Total Traffic Allocation based upon: 1.MAX traffic = µ tot + 2 σ tot 2.90% Traffic = µ tot + 1.3 σ tot 3.80% Traffic = µ tot + 0.83σ tot 4.Other? Total Traffic Allocation limit is also affected by –EDCA Overhead Contention overhead reduces the total traffic bandwidth. Important as number of streams increases –HCCA Allocation Limit Need to allow bandwidth for non-QoS traffic, say only 90% of total bandwidth should be reserved

21. doc.: IEEE 802.11-09/0496-00-00aa Submission Apr 2009 Graham Smith, DSP GroupSlide 21 EDCA Overhead – Capacity drops with # streams Maximum throughput on (shared) channel decreases as number of video streams increases As number of video streams increases, the contention also increases. In order to keep latency low the capacity of the Channel is decreased. 1 stream @ 33Mbps 2 Streams @ 14Mbps = 28Mbps total 5 Streams @ 4.5Mbps = 22.5Mbps total HENCE: Total Allocation MUST take account of the number of streams Note: This is also for Admission Control on each QAP Limits to ensure low loss: NOTE: Above graph is for independent streams. Downlink streams from QAP may be better due to queuing at the AP

22. doc.: IEEE 802.11-09/0496-00-00aa Submission Apr 2009 Graham Smith, DSP GroupSlide 22 Basic Sharing Requirments ? 1.Ability to calculate the Total Traffic Requirement of all the sharing QAPs Calculate Total Traffic 100%, 90%, 80%, Other? 2.Ability to adjust for EDCA contention Note QLoads of QAPs that are “Visible” and the number of streams To enable this, add # of Streams per QAP to the QLoad Element 3.Adjust for HCCA limit Do not allow allocation over 90% to allow for other traffic

23. doc.: IEEE 802.11-09/0496-00-00aa Submission Apr 2009 Graham Smith, DSP GroupSlide 23 Extended QLoad Element Added # of EDCA streams

24. doc.: IEEE 802.11-09/0496-00-00aa Submission Apr 2009 Graham Smith, DSP GroupSlide 24 Sharing Conclusion Conclusion is that in order for the OBSS scheme to provide reliable service on a regular basis: Basic or Recommended Sharing Rules are required Especially true for HCCA –HCCA needs to have a common Sharing procedure so that QAP with CHP = 1 has a known timing scheme for allocation of time within the fixed slot for TXOPs A separate Presentation on Sharing will be prepared

25. doc.: IEEE 802.11-09/0496-00-00aa Submission Apr 2009 Graham Smith, DSP GroupSlide 25 Conclusions New QLoad Element has significant advantages –Need to decide if extended version is preferred Information enables QAPs to make better decisions on individual allocations Recommendations: 1.Use extended version of QLoad Prepare “Sharing” presentation 2.Revise main proposal accordingly