Presentation is loading. Please wait.

Presentation is loading. Please wait.

Improvements to for Video Applications

Similar presentations


Presentation on theme: "Improvements to for Video Applications"— Presentation transcript:

1 Improvements to 802.11 for Video Applications
Month Year doc.: IEEE yy/xxxxr0 November, 2007 Improvements to for Video Applications Date: Authors: Graham Smith, DSP Group John Doe, Some Company

2 Month Year doc.: IEEE yy/xxxxr0 November, 2007 Abstract The possible impairments that jeopardize the user experience when using video over are investigated The Video Problem The RF Problem QoS OBSS DLS 11n Methods or options that could improve the experience are introduced and discussed Recommendations are made for future amendments that might be useful Graham Smith, DSP Group John Doe, Some Company

3 Objectives The objectives are
Month Year doc.: IEEE yy/xxxxr0 November, 2007 Objectives The objectives are To investigate the possible impairments that jeopardize the user experience when using video over Look into methods or options that could improve the experience Future amendments that might be useful and their impact Graham Smith, DSP Group John Doe, Some Company

4 CONTENTS November, 2007 The Video Problem The RF Problem QoS OBSS DLS
Recommendations Graham Smith, DSP Group

5 Video Problem November, 2007
What problems have existed that have caused user dissatisfaction? When considering video streaming, the required data rates are in the range 1 to 23Mbps. Codec SDTV DVD HDTV MPEG2 1-5Mbps 8Mbps 15-23Mbps MPEG4 1-2Mbps 2-3Mbps 8-12Mbps Practical range of a/g Video products, using of MPEG 4/H.264 Graham Smith, DSP Group

6 Video – Effect of Lost Packet
November, 2007 Video – Effect of Lost Packet Effect of Packet Loss on Video A video MSDU contains 7 x 188 video octets, or 10,528 ‘video’ bits. One packet is lost, then 10,528 bits are lost. Examples: TV (PAL) 720 x 480 = 345,600 pixels, Frame rate of 30 fps TV = M Pixels/sec SDTV at 8Mbps, then 1 bit = /8 = pixels Hence, if one packet is lost 10528 bits = pixels = 13644/720 = 19 lines LOST HD 1920 x 1080 = 2,073,600 pixels HD = 62.2M Pixels/sec HDTV at 20Mbps, 1 lost packet is a minimum of 19 lines lost Losing a minimum of 19 lines will definitely cause an observable video error!! Packet Loss possible because of limit on # of re-tries. Graham Smith, DSP Group

7 November, 2007 Video Errors Packet Loss 0.1% gives 44 Errors per minute Graham Smith, DSP Group

8 FEC Coding for Video Data Higher Layer Solution - Basic Methodology
November, 2007 FEC Coding for Video Data Higher Layer Solution - Basic Methodology To allow video coding, MAC changes required Might allow low packet loss over PHY c.f. zero Graham Smith, DSP Group

9 Header November, 2007 Headers (LLC, IP, UDP, RTP and MAC) = 640 bits
Graham Smith, DSP Group

10 FEC vs 3 x Re-try Debatable if FEC is better November, 2007
1 error/hour @ PER 0.26% 1 error/hour @ PER 0.75% 3 x Retry error is complete packet (10500 bits) R-S error is a number of bit errors Header error rate is insignificant. Debatable if FEC is better Graham Smith, DSP Group

11 R-S Concatenated Coding after Encryption on TX
November, 2007 R-S Concatenated Coding after Encryption on TX Combination of FEC plus Re-Tries produces very robust scheme Huge improvement – 1 error per PER 29% Graham Smith, DSP Group

12 FEC Coding at Application Layer
November, 2007 FEC Coding at Application Layer Requires Checksum on Transport and Network layer Headers plus MAC Header Encryption could/will increase errors Retry Mechanism on Header checksum only Indication that Stream is using FEC Results show that no real improvement over standard method of re-tries on Packets Graham Smith, DSP Group

13 Requires changes to MAC and probably Baseband
November, 2007 FEC at MAC Layer Requires changes to MAC and probably Baseband Could be done in Software Potentially very powerful Graham Smith, DSP Group

14 November, 2007 Video To support the use of extra FEC coding, however, does require changes to MAC FEC coding after encryption and before Convolution coding Huge robustness, relaxes need for zero Packet loss at PHY layer For purely Layer 2 considerations the need for zero or very low packet loss and controlled latency requires: A good PHY with good diversity to overcome dynamic multi-path The inclusion of Wi-Fi MAC necessary buffers and processor speed, i.e. a Wi-Fi device that is designed for the application[1] And very important: The use of an effective QoS scheme so as to guarantee the required latency and throughput [1] We do not refer here to the video buffers used by the codec. We refer to the need for the MAC to have sufficient buffers to handle the packets as they are presented or passed to the upper layers. Graham Smith, DSP Group

15 RF Problem November, 2007 Not a MAC Problem, but interesting to note.
Will skip quickly through this Graham Smith, DSP Group

16 Need for Good PHY - the RF Problem
November, 2007 Need for Good PHY - the RF Problem Multiple signals arrive at receiver with differing amplitudes and phases Here shown are the “first order” paths, “single bounce” Graham Smith, DSP Group

17 Signal Strength – 12 x 12 room
November, 2007 Signal Strength – 12 x 12 room 2437MHz 10 -15dB variations over very short distances. Caused by multi-path Peak to trough in 1½ inches Peak to trough in 2/3 inch Graham Smith, DSP Group

18 November, 2007 Indoor Testing Formula used to predict the mean signal strength at any point in the building/house The average error was +1.57dB and the standard deviation of the error was 4.45dB Graham Smith, DSP Group

19 Indoor Testing November, 2007 The STA was swung through a small arc of
11 The STA was swung through a small arc of about 2 feet at a rate of about 2 feet per second Note that this is the total signal strength. Even selected antenna shows variations of >10dB, no better than each antenna, but the selected signal strength is higher. Note the change in signal strength over very short distances. This can only predict the mean data rate, it does not show the effects of multi-path fading. Graham Smith, DSP Group

20 Selective Fading of sub-carriers
November, 2007 Selective Fading of sub-carriers Received signal is effected by: Diversity Delay spread (up to 50ns for a home) Position Frequency Room size For example, frequency and OFDM Sub Carriers Both antennas can be faded Combined diversity, on sub carrier basis (MRC) performs better And these are just for simple open room with no furniture or other reflecting surfaces! For zero or very low Packet Error Rate, good Diversity required Part Explanation of why poor Video in past is that almost all Wi-Fi Cards used switched antenna diversity 802.11n Devices do have good diversity Graham Smith, DSP Group

21 Video needs QoS November, 2007
Problem is that is designed as a Contention Based Scheme Graham Smith, DSP Group

22 EDCA is a priority based QoS and four Access Categories (AC) are used:
November, 2007 Introduction - EDCA EDCA is a priority based QoS and four Access Categories (AC) are used: AC_VO the highest priority. Also referred to as ACI 11 (Access Category Index). AC_VI (ACI 01) is the second highest priority. AC_BE (ACI 00) is the third highest priority. AC_BK (ACI 01) is the lowest priority. EDCA priorities are realized by setting the parameters of the contention functions, CWmin, CWmax, and AIFSN. Graham Smith, DSP Group

23 HCCA is a scheduled access QoS
November, 2007 Introduction - HCCA HCCA is a scheduled access QoS It is always the STA that sends the TSPEC. Hence, the STA must be aware of the details of the traffic it wants to send or receive. Certain recommended and acceptable TSPECS are included in Specification [3]. The calculation of the bandwidth requirements by the HC is standard and repeatable across different vendors. STAs must be confident that the HC allocations are correct. The HC must not allocate all the bandwidth; it must leave a proportion, say 10 to 15% unallocated to allow other traffic in the network. Graham Smith, DSP Group

24 Introduction - EDCA Admission Control
November, 2007 Introduction - EDCA Admission Control Adds bandwidth control to an EDCA network STA sends a TSPEC that is a subset of the one used in HCCA As per HCCA, in Admission Control it is always the STA that sends the TSPEC. Hence, the STA must be aware of the details of the traffic it wants to send or receive. There are no recommended or ‘allowable’ TSPECs (except those used in Wi-Fi WMM-SA Test Plan). There is no requirement for standardizing the calculation for bandwidth allocation in Admission Control. It is possible, and hoped, that the calculation of Medium Time is standardized but the policy for allocation is not. Note that an HCCA compliant AP must also be compliant with EDCA and Admission Control (is this true for e?) Graham Smith, DSP Group

25 We will compare EDCA and HCCA for video applications.
November, 2007 Throughput HCCA uses contention free periods and hence the channel is more efficient when using HCCA than EDCA or DCF. This by itself, however, is not enough to state that HCCA is the preferred approach We will compare EDCA and HCCA for video applications. Graham Smith, DSP Group

26 That is the purpose of QoS!
November, 2007 QoS For the purposes of this paper, we will describe QoS using an example: A video stream starts – it looks great – it remains looking great until it finishes. That is the purpose of QoS! The video experience must stay great despite changes in network load and operational variations Addition of further traffic Operational variations such as movement or range Graham Smith, DSP Group

27 EDCA Video Performance
November, 2007 EDCA Video Performance Consider the “recommended” settings (WFA): CWmin CWmax AIFSN OFDM DSSS AC_VO 3 7 15 2 AC_VI 31 AC_BE 1023 Fundamental in this are the SIFS and Slot times, these are: SIFS OFDM 10µs DSSS 16µs SlotTime Short 9µs Long 20µs The total maximum back-off time is: SIFS + (AIFSN +CWmax) x SlotTime Unfortunately, no tolerance is provided on SIFS or SlotTime values by IEEE or by Wi-Fi Alliance. ALSO Undoubtedly different settings would benefit different applications Graham Smith, DSP Group

28 EDCA Video Performance
November, 2007 EDCA Video Performance Figure 2 – Four 6Mbps streams EDCA The STAs from Vendor A had much more bandwidth than did the STAs from Vendors B and C. Not one STA achieved 6Mbps A pretty damning result for a “QoS” system NOTE: All of the STAs were Wi-Fi Certified for WMM. Graham Smith, DSP Group

29 EDCA Video Performance
November, 2007 EDCA Video Performance Simulated Four 6Mbps AC_VI Streams, 36Mbps, with ±1µs tolerance on SlotTimes For Slot time = 9us CWmin = 7 = 63us CWmax =15 =135us For a Slot time = 8us CWmin = 56us ~ 6 slots CWmax =120us ~ 13 slots For a Slot time = 10us CWmin = 70us ~ 8 slots CWmax =150us ~ 17 slots 1µs variation on SlotTime results in a 20% variation in throughput. Without any set tolerances on SlotTime, it is very difficult to predetermine the bandwidth allocations when more than one stream is admitted on the channel on the same Access Category Graham Smith, DSP Group

30 EDCA Video Performance
November, 2007 EDCA Video Performance Figure 5 – Video Test Network A MPEG2, SD video clip, 4-7Mbps streamed from STA2 to STA1, via the AP, using AC_VI. 6Mbps Chariot Video Stream was sent from STA3 to the AP also using AC_VI. In addition STA4 was transmitting TCP data (Chariot Script FLSNDLG). The result was that the video clip played flawlessly on STA1. Graham Smith, DSP Group

31 EDCA Video Performance
November, 2007 EDCA Video Performance Then STAs 2 and 3 reversed roles. STA3 now sent the video clip and STA2 the Chariot Video Stream, as shown in Figure 6 Figure 6 - Video Test Network B The result was that the video clip failed totally with constant stuttering and freezes. Demonstrates the problems not only of EDCA but also of Admission Control. If the result is unknown, then the Admission Control policy must either err on the conservative side, but how conservative? Graham Smith, DSP Group

32 HCCA Video Performance
November, 2007 HCCA Video Performance For the tests, A and B, as per figures 5 and 6, the video clip worked flawlessly when QSTAs with HCCA were used. Figure 7 – Four 6Mbps Streams, HCCA, EDCA, 36Mbps Actual Chariot Streams Not simulated Note that the three HCCA streams all achieve 6Mbps, and the EDCA stream is just below. Compare to Figure 2 Graham Smith, DSP Group

33 Throughput - Conclusions
November, 2007 Throughput - Conclusions EDCA relies on every individual STA to control the priorities and access to the medium. This allows much room for variation, at best; or cheating, at worst. If EDCA cannot be relied upon to give ‘equal’ or predictable throughput to STAs on the same Access Category, then it is difficult, if not impossible, for Admission Control to allocate bandwidth accurately or efficiently. In HCCA the timing and access to the medium is completely controlled by the AP (HC). This allows little or no room for variations or cheating The absence of limits to the tolerances on SIFS and SlotTime is a major flaw for EDCA Predictable throughput obtained with HCCA Graham Smith, DSP Group

34 OBSS November, 2007 This subject is VERY IMPORTANT for Video.
This part of presentation also goes into details of proposed solutions which Are also detailed in an OBSS presentation for 11v. Graham Smith, DSP Group

35 OBSS – Estimation of Size of Problem
November, 2007 OBSS – Estimation of Size of Problem Floor Plan of Apartments Each Apartment 26 x 40 feet, about 1000 square feet Imagine similar floors above and below this one. Indoor propagation loss formula used: Lp = – log F + 40 log d + WAF (p) + FAF (q) F in MHz, d in feet At shorter distances, the Free Space formula dominates, Lp =– log F + 20 log d + WAF (p) + FAF (q) The predicted propagation loss is the higher of the two. Each wall (WAF) and floor (FAF) between apartments is assumed to be 10dB penetration loss (fireproof). Ceiling height is assumed to be 10 feet. Graham Smith, DSP Group

36 Received Signal Strengths
November, 2007 Received Signal Strengths 1 Inside same apartment 2 Next door (one each side) x2 3 Two away (one each side) x2 4 Three away (one each side) x2 5 Opposite 6 Opposite, across one (one each side) x2 7 Opposite, across two (one each side) x2 8 Opposite, across three (one each side) x2 9 Directly up and down x2 10 Up or down, neighbor (one each side) x4 11 Up or down, two away (one each side) x4 12 Up or down, three away (one each side) x4 13 Opposite, up and down x2 14 Opposite, up and down, two across x4 15 Opposite, up and down, one across x4 16 Opposite, up and down, three across x4 17 Two floors directly up and down x2 30dB power control Graham Smith, DSP Group

37 Number of OBSS – DFS and TPC
November, 2007 Number of OBSS – DFS and TPC Table 1 – Theoretical OBSS for Apartments sq. ft. Frequency Band Number of Interfering Networks Interfering Networks per 20MHz Channel Interfering Networks per 40MHz Channel 2.4GHz 31 10 5GHz 27 0-1 3 Ideal DFS reduces problem significantly! 5GHz for Home! Received signal strength within each apartment is high, better than -40dBm. Theoretically, therefore, the power could be reduced by 30dB with no deterioration in the throughput. Solves OBSS! Table 2 – Theoretical OBSS with 30dB Power Reduction Frequency Band Number of Interfering Networks with 30dB power reduction Interfering Networks per 20MHz Channel with 30dB power reduction Interfering Networks per 40MHz Channel with 30dB power reduction 2.4GHz 8 3 5GHz 4 Graham Smith, DSP Group

38 Effects of OBSS - 1 November, 2007 # Network A OBSS Network B Effect
Result 1 Legacy Traffic simply competes Reduced bandwidth in each network No lost packets Not recommended for streaming 2 EDCA Higher priority traffic in Network A will drive down traffic in Network B AC_VO and AC_VI traffic dominates. Could be OK for streaming traffic but no admission policy Network A “wins” 3 Traffic competes on a priority basis. Networks compete on an ‘equal’ basis No real protection for streaming traffic in either network Graham Smith, DSP Group

39 presently certified and it breaks down in OBSS!
November, 2007 Effects of OBSS - 2 # Network A OBSS Network B Effect Result 4 Admission Control Legacy Higher priority traffic in Network A will drive down traffic in Network B AC_VO and AC_VI traffic dominates. Could be OK for streaming traffic Network B bandwidth can be drastically reduced 5 EDCA Traffic competes on a priority basis. Admission Control in Network cannot control traffic in Network B No protection for admitted traffic in Network A 6 Admission Control Traffic competes on a priority basis. Admission Control in either Network cannot control traffic in other Network No protection for admitted traffic in either Network These cases are cause for concern, Admission Control is the highest QoS presently certified and it breaks down in OBSS! Graham Smith, DSP Group

40 Effects of OBSS - 3 November, 2007 7 HCCA Legacy
Scheduled TXOPs in Network A also apply CFP to Network B. Full protection for scheduled traffic in Network A Network B bandwidth reduced 8 EDCA 9 Admission Control Scheduled TXOPs in Network A also apply CFP to Network B Admitted traffic Network B is lower priority than scheduled traffic in Network A Both Networks using TSPECS 10 Each HCCA AP will admit streams and allocate time to them BUT each AP and STA will obey the TXOP allocation of the other. No guarantee that each Network can allocate time when it needs to. , Reduced protection for scheduled traffic in either network. Graham Smith, DSP Group

41 Conclusions: OBSS – EDCA on EDCA
November, 2007 OBSS – EDCA on EDCA Table clearly shows that OBSS is a problem for when it is intended to be used for applications that require QoS. EDCA/WMM does not address the problem at all. EDCA Admission Control only solves the bandwidth allocation problem within its own network and does not address OBSS. HCCA does overcome OBSS problems in all but the case where two HCCA networks overlap. Conclusions: EDCA is not providing QoS in OBSS situation and any higher bandwidth streaming application is not protected If we wish to solve OBSS problem in Wi-Fi then the use of HCCA would seem to be mandatory and we need to look into solving the OBSS situation for two HCCA networks Graham Smith, DSP Group

42 Investigation carried out that shows how:
November, 2007 Solving OBSS One clear recommendation would be to initiate mandatory methods for DCF and TPC If so, it could be assumed that the OBSS situation could be eliminated or limited to a maximum of two QAPs Investigation carried out that shows how: Two HCCA networks could co-operate HCCA and Admission Control QAPs could co-operate Two Admission Control QAPs co-operate Note: Still not protected against WMM OBSS Graham Smith, DSP Group

43 A separate presentation on OBSS and solution made for 11v
November, 2007 OBSS A separate presentation on OBSS and solution made for 11v Graham Smith, DSP Group

44 Channel Selection NOW November, 2007
A possible basic method for QoS networks to co-operate is: QAP selects a Channel Listen for another Beacon on that channel If clear, then allow STAs to associate If Beacon indicates AP is not a QAP, note this but continue to seek a clear channel. If after cycling through the channels, no clear channel is found then Select a channel that is shared with a non-QoS AP. Allow STAs to associate Issue a standard Beacon Request/Report (802.11k) to the STAs. If no other QAP is reported, then the QAP may choose that channel If another QAP is reported (on a different ESS), try to find another channel. No idea of QoS Load, No method of Sharing Bandwidth (but should be used NOW) Graham Smith, DSP Group

45 November, 2007 DLS Problem Graham Smith, DSP Group

46 DLS Hidden STAs November, 2007
Condition A, STA 2 can receive the traffic being sent from STA3 to 4, but STA 1 cannot. STA2 will not acknowledge STA1 traffic, DLS between STAs 3 and 4 effectively blocks the other link. Condition B, neither link can hear the other so each thinks it has the entire bandwidth Transfer traffic at the highest possible rate. AP is saturated and other stations in the network are blocked. Potential problem for Video Graham Smith, DSP Group

47 HCCA overcomes DLS hidden STA problem
November, 2007 HCCA overcomes DLS hidden STA problem Now consider the same situations when HCCA is used. STAs 1 and 3 both request their required bandwidth of the QAP. The QAP will now schedule the TXOP, as required to each direct link. Everything is under control and the QAP is managing the network efficiently. HCCA will cause DLS to operate correctly Graham Smith, DSP Group

48 November, 2007 802.11n Graham Smith, DSP Group

49 A-MPDUs will occupy channel for long durations
November, 2007 802.11n Concerns for Video A-MPDUs will occupy channel for long durations Very efficient when used within TXOP 8Mbps, 16ms SI, 16000B per SI 23Mbps, 16ms SI, 46000B per SI Not restricted to streaming or QoS requirements If used on data at 65000B, over 5ms at 100Mbps Will cause QoS latency problems Graham Smith, DSP Group

50 November, 2007 A-MPDU, Multiple MAC Protocol Data Units (MPDU) are sent in one PHY Service Data Unit (PSDU). This means that a Block ACK is necessary. 8191 to a maximum of 65535Bytes are possible. Maximum length of MPDU within an A-MPDU is 4095Bytes Very efficient when used within TXOP 8Mbps, 16ms SI, 16000B per SI 23Mbps, 16ms SI, 46000B per SI 8Mbps, 10ms SI, 10000B per SI 23Mbps, 16ms SI, 28750B per SI Graham Smith, DSP Group

51 A-MPDUs can occupy channel for “long” time
November, 2007 Effect of A-MPDU A-MPDUs can occupy channel for “long” time 100Mbps Can be used for data, not only streaming. Hence, will prevent efficient timing and latency Scary point is what does all this do with QoS? Which wants to deliver at bounded latency (even with EDCA) Need to think of 11n with QoS hat on. Just relying on bandwidth overkill is not a QoS approach Possible solution is to set limit on TXOP for AC_BE and AC_BK. Is this enough? Good news is A-MPDUs works real good within the HCCA TXOP Graham Smith, DSP Group

52 MUST use the OBSS proposals to:
November, 2007 40MHz OBSS 40MHz Channels Easy/intuitive to see how two 40MHz overlapping networks will be less efficient than separate, independent 20MHz channels. MUST use the OBSS proposals to: Try to find clear channel If not clear, look for 20MHz channel MUST introduce procedures for preventing or controlling OBSS and usage of 40MHz channels OBSS Channel selection method proposed can be applied Graham Smith, DSP Group

53 November, 2007 Recommendations Graham Smith, DSP Group

54 Work with 802.11v or VTS TG for inclusion of:
November, 2007 Recommendations Work with v or VTS TG for inclusion of: “Q LOAD Element” for HCCA and EDCA Admission Control QAPs “OBSS” Beacon Request Report Fixed 10ms Slot time for HCCA Use of bit 7 in QoS CF Poll to indicate start of Slot Time Allow use of concatenated coding at Application Layer Allow removal of FCR error filter Allow use of encryption on individual stream basis Addition of recommended practices for OBSS and concatenated coding Investigate 11n and QoS. Use of A-MPDUs within QoS. Possibly restrict outside QoS Introduce strict usage for 40MHz channels with OBSS procedures Graham Smith, DSP Group

55 November, 2007 Thank You Graham Smith, DSP Group

56 November, 2007 Back up - hidden AP Graham Smith, DSP Group

57 November, 2007 Hidden QAPs If QAP stays after Beacon Report, set CHP to 0 and sends OBSS Beacon Request QAP B now knows of QAP A and its Q Load QAP ‘A’ and QAP ‘B’ calculate their maximum allocated bandwidth, based upon their Q Loads and the SPNB method. QAP A and QAP B must now harmonize their Scheduled Allocations Graham Smith, DSP Group

58 Harmonizing SI – Direct Method
November, 2007 Harmonizing SI – Direct Method Direct Method (as per non-hidden QAPs) Could be possible using a common STA BUT The QSTA may be in power save mode If the first TXOP has been granted then the QSTA is prevented from transmitting, so sending the timer onto the other QAP is not possible The only legitimate transmission from a STA to an AP outside its network, is the Probe Request It is not advisable, or even allowed to change a scheduled time by too much. Graham Smith, DSP Group

59 Harmonizing SI – Indirect Method
November, 2007 Harmonizing SI – Indirect Method QAP A CHP = 0; QAP B CHP = 1 QAP A determines that a scheduled stream to a particular QSTA is blocked and suspects that it is due to scheduling from the QAP B. In this case, QAP A shifts its TSF timer, at DTIM, in the positive direction by 5% of the slot time, i.e. 500us. Similarly, QAP B determines that a scheduled stream to a QSTA is blocked and suspects that it is due to scheduling from the QAP A. In this case, QAP B shifts its TSF timer, at DTIM, in the negative direction by 5% of the slot time, i.e. 500us. Graham Smith, DSP Group


Download ppt "Improvements to for Video Applications"

Similar presentations


Ads by Google