Empirical Formula for EDCA Bandwidth Factor March 2010 doc.: IEEE 802.11-08/0304r0 March 2010 Empirical Formula for EDCA Bandwidth Factor Date: 2010, March 12 Authors: Graham Smith, DSP Group Graham Smith, DSP Group

March 2010 doc.: IEEE 802.11-08/0304r0 March 2010 Abstract The Medium Time returned for an EDCA Admission Control TSPEC includes the packet length, SIFS and ACK. It does not include the overhead required for medium access. This presentation: Investigates the medium access overhead with respect to the number of VO and VI streams Investigates equal streams, mixed streams, equal and mixed data rates, PHY rates, aggregations Derives ‘simple’ empirical formulas and rules to estimate EDCA Bandwidth Factor Graham Smith, DSP Group Graham Smith, DSP Group

March 2010 Medium Time Medium Time = Surplus Bandwidth Allowance * pps * medium time per frame exchange where: pps = ceiling( (Mean Data Rate / 8) / Nominal MSDU Size ); medium time per frame exchange = duration (Nominal MSDU Size, Minimum PHY Rate) + SIFS + ACK duration; duration() is the PLME-TXTIME primitive defined in the standard that returns the duration of a packet based on its payload size and the PHY data rate employed Note that it represents the time that the packet is on the air, and does not include the time between packets, the Medium Access Time The Medium Access Time includes SIFS, AIFSN and the Contention Window value Graham Smith, DSP Group

March 2010 Medium Access Time The Medium Access Time includes SIFS, AIFSN and the Contention Window value As the number of streams increases, each stream is held up in the contention window while another stream is transmitting. The total time required for EDCA streams is therefore greater than the Medium Time and it varies with the number of streams. This total required time compared to the Medium Time is termed the EDCA Overhead Factor. Graham Smith, DSP Group

Investigation into EDCA Overhead Factor March 2010 Investigation into EDCA Overhead Factor A simulation program was written to measure the throughput and delays for multiple streams. The following can be set for each stream: EDCA parameters Packet Size Data rate, Mbps PHY Rate, Mbps The outputs are: Packets in, Packets out Actual data rate Maximum and Average Packet Delay Graham Smith, DSP Group

March 2010 Method For a set number of streams, the data rate was varied and the average packet delay noted. For the values to be acceptable, or a PASS, the criteria used was: Average Delay < 2 times the SI of the packet Where SI = reciprocal of Packets per Second = 1/pps And pps = data rate/packet length Graham Smith, DSP Group

Simulation Example March 2010 10 Video Streams @ 54Mbps Graham Smith, DSP Group

March 2010 Methodology For a set number of streams, the data rate was varied until the average delay for every stream was acceptable The total Medium time was then noted, in µsecs Medium Time Factor = 1sec/Medium Time (overhead) For VI the packet size was set to 1470Bytes The VO packet size was set to 160Bytes and the data rate to 64 kbps 11a/g assumed to be the same (6us extension on OFDM to make SIFS 16us)) 11n 2SS with Aggregations of 0, 8k, 16k, 32k and 64k Graham Smith, DSP Group

Example Results for equal streams March 2010 Example Results for equal streams 36 Mbps VI Streams Mbps/stream Medium Time O/H Factor 1 23.65 853628 1.17 2 24 12 866261 1.15 3 23.19 7.73 837025 1.19 4 22.68 5.67 818617 1.22 5 22.5 4.5 812120 1.23 6 22.32 3.72 805623 1.24 7 22.19 3.17 800931 1.25 8 22.08 2.76 796960 9 22.05 2.45 795878 1.26 10 22 2.2 794073 Results taken for 54, 36, 24, 18, 12 and 6Mbps Graham Smith, DSP Group

Results for Equal VI streams March 2010 Results for Equal VI streams Graham Smith, DSP Group

EDCA Overhead Formula 11a/g AC_VI March 2010 EDCA Overhead Formula 11a/g AC_VI The results were examined and the following formula were derived for estimating the EDCA Overhead Factor, F for 11a/g For PHY Rate R, and number of equal streams S 1 stream F = 1.01 + 0.004R 2 streams F = 1.04 + 0.003R 3 to 10 streams F = 1.1 + 0.007S + .0025R >10 streams, S = 10 Graham Smith, DSP Group

Comparison of Results vs. Formula AC_VI March 2010 Comparison of Results vs. Formula AC_VI Formula result is within 1% of simulated Graham Smith, DSP Group

AC_VO EDCA Overhead Factor using F formulas, Equal Streams 11a/g March 2010 AC_VO EDCA Overhead Factor using F formulas, Equal Streams 11a/g Streams 54 36 24 18 12 6 1 1.23 1.15 1.11 1.08 1.06 1.03 2 1.20 1.09 3 1.26 1.21 1.18 1.17 1.14 4 1.22 1.19 1.16 5 1.27 1.28 7 1.24 8 1.29 1.25 9 1.30 10 1.31 Graham Smith, DSP Group

AC_VO Simulations March 2010 Example with 8 bi-directional voice streams plus one video stream. Video stream data rate is increased until delay limit is reached. AP has 8 x Mbps Graham Smith, DSP Group

March 2010 VO Results Simulations were carried out using bi-directional voice packets of 160 Bytes at 64kbps (20ms SI) The results were examined and the following formula were derived for estimating the EDCA Overhead Factor, F for 11a/g For PHY Rate R, and number of streams S 1 to 10 stream F = 1.175 + 0.007S + 0.0025R >10 streams, S = 10 Graham Smith, DSP Group

Comparisons of Simulations vs Formula AC_VO March 2010 Comparisons of Simulations vs Formula AC_VO NOTE: Number of streams is number of Bi-Directional streams Formula provides results for all PHY Rates, including 18Mbps Graham Smith, DSP Group

AC_VO EDCA Overhead Factor using F formula, 11a/g March 2010 AC_VO EDCA Overhead Factor using F formula, 11a/g Bi- Streams 54 36 24 18 12 6 1 1.32 1.27 1.24 1.23 1.21 1.20 2 1.28 1.25 1.22 3 1.33 1.29 1.26 4 1.34 5 1.35 1.30 1.31 7 1.36 8 1.37 9 10 1.38 Graham Smith, DSP Group

VO VI Mix Measurements taken with 1, 2, 3 and 4 video streams March 2010 VO VI Mix Measurements taken with 1, 2, 3 and 4 video streams Overhead factor does not change with number of video streams RULE: Just count the AC_VO streams Graham Smith, DSP Group

Mixed AC_VI Data Rates, 11a/g Same PHY Rate March 2010 Mixed AC_VI Data Rates, 11a/g Same PHY Rate For a fixed PHY Rate, with fixed number of streams, select one data rate, vary other for Delay Limit E.g. 24Mbps, 2 streams 10 and 7.25Mbps 12 and 5.1Mbps 14 and 2.6Mbps 15 and 1.2Mbps Repeated this for more streams and other PHY Rates. By examining results, established following “Rule” S equivalent, Sequ = 0.5 x Total Medium Time of all streams Medium Time of Stream with least Medium Time Max Sequ = 10 Enter Sequ into the formula to obtain F for fixed R Formulas for 1 and 2 streams is modified for a ‘fraction’ For 1< Sequ<2 F = 0.97 + 0.04S + .004R For 2< Sequ<3 F = 0.97 + 0.04S + .003R Graham Smith, DSP Group

March 2010 Mixed PHY Rates, 11a/g Simulated several mixes of PHY Rates and data rates Investigation resulted in the following Rules, similar to Mixed Data Rates Sequ = 0.5 x Total Medium Time of all streams Medium Time of ‘lowest stream’ R = PHY Rate of the ‘lowest stream’ (‘lowest stream’ is stream with shortest Medium Time) Hence, use Sequ and Rlowest in the F formula Graham Smith, DSP Group

Summary 11a/g EDCA Overhead Factor F March 2010 Summary 11a/g EDCA Overhead Factor F AC_VI Basic Formula 1 stream F = 1.01 + 0.004R For 1< Sequ<2 = 0.97 + 0.04S + .004R 2 streams F = 1.04 + 0.003R For 2< Sequ<3 = 0.97 + 0.04S + .003R 3 to 10 streams F = 1.1 + 0.007S + .0025R >10 streams, S = 10 For Mixed Data Rates and PHY Rates Sequ = 0.5 x Total Medium Time of all streams Medium Time of stream with least Medium Time R = PHY Rate of the stream with least Medium Time AC_VO Basic Formula 1 to 10 stream F = 1.175 + 0.007S + 0.0025R For Mixed VO and VI Just count the AC_VO streams For mixed PHY Rates R = Highest PHY Rate of the VO streams Graham Smith, DSP Group

EDCA Overhead Factor for 11n March 2010 Break for Questions on 11a/g, then tackle EDCA Overhead Factor for 11n This is a BIG subject!! Graham Smith, DSP Group

EDCA Overhead Factor 11n The 11n networks investigated were: March 2010 EDCA Overhead Factor 11n The 11n networks investigated were: 2SS No Aggregation A-MPDUs for 64k, 32k, 16k, 8k In case of aggregation, Medium Time was calculated as the ‘on-air time’ including BA Implicit BA was assumed Note: As far as I can tell, there is no set method for the STA to indicate it is using A-MPDUs and whether 64, 32, 16 or 8k and this makes it difficult for an AP to calculate the correct Medium Time. Proposals are being considered within Wi-Fi Alliance but these will not be discussed here. In the following slides, it is assumed that the Medium Time is correct. Graham Smith, DSP Group

11n Same procedure as for 11a/g, March 2010 11n Same procedure as for 11a/g, For No Aggregation and 64k, 32k, 16k and 8k Aggregation Equal VI streams for different PHY rates Calculate F formulas VO bi-directional streams with VI stream(s) Calculate F formula Investigate Mixed Aggregation Graham Smith, DSP Group

11n No Aggregation VI - Results March 2010 11n No Aggregation VI - Results For PHY Rate R, and number of equal streams S 1 stream F = 1.05 + 0.0026R 2 streams F = 1.076 + 0.002R 3 to 10 streams F = 1.12 + 0.0073S + .0015R >10 streams, S = 10 Graham Smith, DSP Group

March 2010 AC_VI EDCA Overhead Factor using F formulas, Equal Streams 11n No Aggregation Streams 130 117 104 78 52 39 26 13 1 1.39 1.35 1.32 1.25 1.19 1.15 1.12 1.08 2 1.34 1.31 1.28 1.23 1.18 1.13 1.10 3 1.30 1.26 1.22 1.20 1.16 4 1.27 1.21 1.17 5 1.33 6 1.36 1.24 7 1.37 1.29 8 9 1.38 10 Graham Smith, DSP Group

11n No Aggregation VO - Results March 2010 11n No Aggregation VO - Results For PHY Rate R, and number of streams S 1 to 10 stream F = 1.21 + 0.004S + 0.0016R >10 streams, S = 10 Graham Smith, DSP Group

AC_V0 EDCA Overhead Factor using F formula, 11n No Aggregation March 2010 AC_V0 EDCA Overhead Factor using F formula, 11n No Aggregation Bi-Dir Streams 130 117 104 78 52 39 26 13 1 1.42 1.40 1.38 1.34 1.30 1.28 1.26 1.23 2 1.43 1.41 1.24 3 1.39 1.35 1.31 4 1.29 1.27 1.25 5 1.44 6 1.36 1.32 7 1.45 8 1.37 1.33 9 10 1.46 Graham Smith, DSP Group

11n Aggregation Equal Streams, same PHY Rate March 2010 11n Aggregation Equal Streams, same PHY Rate Individual formulas calculated for each Aggregation Then a ‘combined formula’ derived 1 stream F = 1 + 0.0075R/A For mixed streams = 0.94 + .06S + 0.0075R/A (1<Sequ<2) 2 streams F = 1.05 + 0.0044R/A For mixed streams = 0.94 + .06S + .0044R/A (2<Sequ<3) 3 to 10 streams F = 1.1 + 0.007S + .0041R/A >10 streams, S = 10 Where R = PHY Rate S = number of streams A = 8, 16, 32 or 64 (corresponding to 64k, 32k, 16k, 8k) Graham Smith, DSP Group

11n Aggregation F formula Results March 2010 11n Aggregation F formula Results Packet Duration > 10ms Packet Duration > 10ms Graham Smith, DSP Group

11n Aggregation F formula Results March 2010 11n Aggregation F formula Results Graham Smith, DSP Group

11n Aggregation Results by Formula March 2010 11n Aggregation Results by Formula Left to Right In each A Block 130 -> 13Mbps 64k 32k 16k 8k 130M 13M Graham Smith, DSP Group

Simulation Example March 2010 O/H Factor Two 8k Streams Eight 64k Streams Graham Smith, DSP Group

Mixed Aggregation March 2010 Same PHY Rate, same data rate per stream, mixed aggregation, example Note: actual readings (not formula) Compare: Two 8k streams is better throughput than one 8k and one 64k! Also, note that the O/H Factor is constant for the mix AND the value is higher than 8k O/H, even for 9 and 10 streams. Repeating for two and three 8k streams, similar O/H Factor results Graham Smith, DSP Group

More “Strange” results with Aggregation March 2010 More “Strange” results with Aggregation 1.27 Factor as per previous slide Note that the total throughput is often better with lower Aggregation!! AND The Overhead Factor is worse For higher aggregation EXCEPT when 8k portion is low Graham Smith, DSP Group

8k + 8k Example March 2010 Sequ = 0.5 x MT Total/MTmin Factor is equivalent to For 1 - 2 Sequ F = 0.94 + .06Sequ + 0.0075R/A (accounts for fractions) For 2 - 3 Sequ F = 0.94 + .06Sequ + .0044R/A (accounts for fractions) For 3 - 10 Sequ F = 1.1 + 0.007S + .0041R/A THIS IS SAME RULE FOR Sequ AS PER 11a/g Graham Smith, DSP Group

16k + 8k Example March 2010 Rule goes wrong here And gets worse for 32k and 64k Need to account for aggregation somehow Graham Smith, DSP Group

64k + 8k March 2010 78Mbps One 64k and one 8k Stream Data Rate, Mbps FACTOR 64k 8k Tot 56.2 5 61.2 1.15 49.3 10 59.3 1.18 43.5 15 58.5 35 20 55 1.24 28 25 53 1.27 22 30 52 1.28 13.6 48.6 1.34 8.3 40 48.3 4.3 45 1.30 Dominated by 64k but Factor is high for 64k Equal streams, F = 1.27 as per slide 34 Much worse than 10+ at 8k (1.21) Why is this worse? When equal, 8 x as many 8k packets as 64k packets, but 64k packets are 8x longer than 8k packets Graham Smith, DSP Group

EDCA Overhead Factor and Aggregation March 2010 EDCA Overhead Factor and Aggregation When equal streams and fixed PHY Rate, the formulas are straightforward Mixing data rates with same aggregation, the “rule’ derived for 11a/g holds Mixing aggregation, unable to derive a ‘rule’ based upon the “equal stream’ formulas Therefore, new formulas derived Graham Smith, DSP Group

Mixed Aggregation and Data Rate March 2010 Mixed Aggregation and Data Rate Looking (real hard) at the results, the following empirical formula results: Specify a new parameter, r r = MT of lowest aggregation stream (total Medium Time) Then, for mixed aggregation streams: F = 1.12 + [0.005 + 0.028 (Amax/Amin)] r Graham Smith, DSP Group

Results March 2010 Mixed Aggregation formula Equal stream, Sequ formula See over Graham Smith, DSP Group

11n Aggregation Formula March 2010 r = MTmin/MTtot Note: Could use this formula for same aggregation, using r = MTmin/MTtot Not so accurate, but is ‘safe’ here Graham Smith, DSP Group

March 2010 11n Aggregation Formula r = Total MT of lowest aggregation streams (total Medium Time) Then, for mixed aggregation streams, at all PHY Rates: F = 1.12 + [0.005 + 0.028 (Amax/Amin)] r Graham Smith, DSP Group

Examples using Aggregation Formula March 2010 Examples using Aggregation Formula Reasonable results, Graham Smith, DSP Group

SUMMARY 11n VO Bi-directional streams (no aggregation) March 2010 SUMMARY 11n VO Bi-directional streams (no aggregation) 1 to 10 stream F = 1.21 + 0.004S + 0.0016R (>10 streams, S = 10) Equal VI streams, same aggregation For 1 - 2 Streams F = 0.94 + .06S + 0.0075R/A For 2 - 3 Streams F = 0.94 + .06S + .0044R/A For 3 - 10 Streams F = 1.1 + 0.007S + .0041R/A (>10 streams, S = 10) Where R = PHY Rate S = number of streams A = 8, 16, 32 or 64 (corresponding to 64k, 32k, 16k, 8k) Mixed streams, same Aggregation S = Sequ = 0.5 x MT Total/MTmin Mixed Streams, mixed Aggregation F = 1.12 + [0.005 + 0.028 (Amax/Amin)] r Where r = Total MT of Amin aggregation streams Total Medium Time Graham Smith, DSP Group

March 2010 Next Steps These empirical formulas could be used by the AP to estimate the required bandwidth for EDCA Admission control, as the AP will have all the information required For OBSS, to estimate the combined Overhead Factor, each AP would need to know the details of individual streams, and this is probably too much data Next step is to use these results to look at the minimum amount of data required in the OBSS QLoad, so as to make a reasonable estimate This could be as simple as always assume F = 1.30 At the moment the QLoad reports, number of streams and total medium times Could be based upon each AP reporting its Overhead Factor and then combine them Graham Smith, DSP Group