Efficient Transmission of H.264 Video over WLANs

Transcription

1 Efficient Transmission of H.264 Video over WLANs Yaser P. Fallah March 2007 UBC 1

2 Multimedia Communications Multimedia applications are becoming increasingly popular Video on mobile devices (cell phones, ipod, ) Video streaming (IPTV, Home HDTV streaming) Video Conferencing and Telephony Wireless access networks are being deployed throughout the world at a rapid pace Wireless Local Area Networks ( or WiFi) Cellular 3G networks Video Transmission over Wireless LANs UBC

3 Video over IEEE e WLAN QoS- Internet Video Streams may incur significant packet loss, delay and delay jitter in Wireless Environments. UBC

4 WLAN Access Network Problems What is different about wireless networks? Delay, Jitter, Packet Loss and Data Corruption are considerably higher in wireless networks than in wired networks WLAN problems may come from two sources Physical Layer (PHY): fading, noise and interference may corrupt packets. MAC Layer: collision in MAC layer may not be resolved on time, so packets are dropped or delayed. UBC

5 Enhancements, through cross-layer solutions Enhancements may be achieved in several layers, either separately or in a cross layer fashion Application Layer ( H.264 Video ) Error Resiliency (Flexible Macroblock Ordering) Scalable Video Coding Data Partitioning Slicing Video frames App. Layer: H.264 Video RTP/IP MAC Layer Packet size adjustments (fragmentation) to reduce packet loss Differentiated Service Provisioning (EDCA) Guaranteed Service Provisioning (HCCA) PHY Layer Unequal Error Protection MAC: e PHY: b,g,a UBC

6 Agenda Overview of H.264 Slice Coding Flexible Macroblock Ordering Data Partitioning Solutions Frame size control MAC QoS Provisioning UBC

7 H.264, Slice Coding When packet sizes exceed the maximum allowable amount for reliable transmission, the frames may be sliced into smaller decodable data units the alternative is to let the network fragment the packet. Frame Data NALU slice1 NALU slice2 NALU slice3 NALU slice4 Slices Self contained, all information for decoding slice available within slice. Each slice encapsulated in a NAL (Network Abstraction Layer) Unit. Advantages: losing a slice only affects part of a picture, not an entire frame Disadvantages: More overhead, less compression efficiency UBC

8 H.264 Error Resiliency - Flexible Macroblock Ordering (FMO) The H.264 provides various error resiliency schemes, including Flexible Macroblock Ordering (FMO) H.264 allows macroblocks within pictures to use a variety of slice mapping patterns.these include: Interleaving Dispersed Foreground Box-out Raster Scan Wipe Explicit user defined UBC

9 H.264 Data Partitioning The Compressed Frame is partitioned into : Partition A, contains the most important information such as MB types, quantization parameters, and motion vectors. Partition B (intra partition), contains intra coded block pattern (CBP) and intra coefficients. Partition C (inter partition), contains only inter CBPs and inter coefficients. Partitions may be treated differently by the delivery layer Unequal Error Protection in PHY and MAC Prioritized or Guaranteed Services in MAC UBC

10 Enhancing H.264 Video Transmission over WLANs Frame size control Optimal packet size is found in the MAC/PHY layers to achieve lowest Packet Loss Ratio (PLR) Video slicing is done in the application layer, instead of fragmentation in MAC, to achieve optimal packet size. QoS provisioning H.264 Baseline profile H.264 Extended profile (with Data Partitioning) UBC

11 Frame Size Control: Effect of Packet Size on Packet Loss Ratio Using large packets: reduce RTP/IP/UDP, and more importantly PHY overhead reduce contention and collision (MAC operation overhead) Higher PLR in PHY layer: PLR = 1 (1-BER) pkt_len ~ BER * pkt_len Using small packets: Lower PLR in PHY layer Higher Packet loss in MAC due to congestion and increased PHY overhead There MAY exist an optimum value for packet length Packet Loss Ratio MAC and PHY (BER 10E-4) MAC only (BER = 10E-5) PHY only (BER 10E-4, Retx 5) Packet Size (Bytes) UBC

12 Packet Size Adjustment. Average PSNR Fragment or Slice Size (Bytes) Slice Size: 2300B 700B 300B UBC

13 Packet Size Adjustment: Video Slicing vs. MAC Fragmentation Assuming an acceptable or optimal packet size, we prefer video slicing to MAC fragmentation Packet Error Rate: fragments or slices of size K, picture of size L, n fragments, X transmission attempts, e: BER. UBC

14 Packet Size Adjustment: Video Slicing vs. MAC Fragmentation Adjusting MAC packet size to optimal value can be done in MAC layer ( fragmentation) Application layer (video slice coding) MAC Fragmentation MAC packet Video Frame (NAL packet) MAC packet MAC packet MAC packet MAC packet Video Slicing Video Slice Video Slice Video Slice Video Slice Video Slice MAC packet MAC packet MAC packet MAC packet MAC packet UBC

15 Packet Size Adjustment: MAC vs. App. PSNR NoSlice 340B Frag Slice 300B Frame # Average PSNR MAC Fragmentation Video Slicing Slice or Fragment Size Application Layer frag. (Video Slicing) UBC MAC Fragmentation

16 MAC QoS Solutions QoS in WLAN e Standard defines two access methods EDCA (Enhanced Distributed Channel Access) Prioritized contention access: higher success probability for video, voice, Aggregate QoS HCCA (HCF Controlled Channel Access) Scheduled access user defined scheduler Per-session QoS UBC

17 DCF/EDCA Operation Prioritized Access provided through traffic specific parameters: 1. Initial Deferment, AIFS 2. Backoff window size 3. Allowed Burst size (TXOP) Immediate access when medium is idle >= AIFS[TC]+SlotTime AIFS[TC] +SlotTime Busy medium AIFS[TC] DIFS PIFS SIFS Contention Window from [1,1+CWmin[TC]] Backoff Window Next Frame SlotTime Defer Access Select Slot and decrement backoff as long as medium stays idle UBC

18 HCCA Operation Access Point can use PIFS waiting time and generate a contention free duration (CAP) at almost any time QAP (downstream) PIFS SIFS SIFS PIFS SIFS Poll Data Ack Data Ack QSTAs (upstream) EDCA Polled TXOP EDCA TXOP TXOP obtained by QAP Controlled Access Phase (CAP) EDCA UBC

19 QoS Support using e EDCA Based only provides aggregate and prioritized QoS, poor delay and jitter performance does not perform well at heavy loads, very sensitive to MAC parameters HCCA Based Scheduled access to the medium is possible through HCCA, thus service guarantee is feasible The scheduling scheme is user-defined in HCCA Challenges of Scheduling in e CSMA/CA: channel shared between Uplink/downlink at all times Mixed contention based & contention free (controlled) access Multi-rate operation of PHY: Each packet may be sent at a different PHY rate Delay (sec) UBC

20 CAPS: Controlled Access Phase Scheduling ACCESS POINT UPSTREAM, Requests from stations DOWNSTREAM packets Actual Packets Virtual Packet Generator (VPG) classifier Virtual Packets Time Stamping HCCA Queues 8 EDCA queues Select access mode Inner Scheduler EDCA Contention Access Indicates whether the next service round is a contention access using EDCA or an HCCA CAP generation Virtual Packets VAP classifier Actual Packets Generate Poll messages Station sends packet, AP sends Ack AP sends packet, Station sends Ack Scheduler/Shaper UBC

21 MAC QoS Based Solutions Techniques for H.264 Video Transmission over WLANs TECHNIQUE Supported H.264 Profiles Supporte d e WLAN Network-Awareness In The Multimedia Source Media-Awareness In The WLAN 1. Single Stream served by EDCA baseline, extended (all profiles) EDCA Limited: Tagging all frames with type of service Serving tagged video in priority levels (AC2) 2. Single Stream served by CAPS baseline, extended (all profiles) EDCA & HCCA with CAPS Limited: Tagging all frames with traffic stream ID Requires video pattern information, serving tagged video in guaranteed access traffic session 3. Partitioned Video served by EDCA extended EDCA Tagging different partitions for different priority levels. Serving packet of each partition in a different priority level (A: AC2, B & C: AC1) 4. Partitioned Video served by CAPS extended EDCA & HCCA with CAPS Tagging different partitions for different traffic streams. Requires video pattern information, serving partition A packets in a separate guaranteed access traffic session from partitions B & C 5. Partitioned Video, Partially Aggregated, served by CAPS extended EDCA & HCCA with CAPS Tagging different partitions for different traffic streams, aggregating partitions B and C (or A & B). Requires video pattern information, serving partition A packets in a separate guaranteed access traffic session from the aggregated packets of partitions B & C UBC

22 MAC QoS Based Solutions: CAPS vs. EDCA Technique 1 vs. Technique 2: EDCA vs. CAPS packet delay (sec) UBC

23 CAPS vs. EDCA, Home video scenario One uplink or downlink video Background Traffic (all classes): Heavy downlink load (10Mbps), light uplink load (<200Kbps) Frame # packet delay (sec) UBC

24 Video Quality: CAPS vs. EDCA CAPS all cases EDCA 100ms EDCA 100ms with 1E-5 BER EDCA 250ms UBC

25 MAC QoS Based Solutions Techniques for H.264 Video Transmission over WLANs TECHNIQUE Supported H.264 Profiles Supporte d e WLAN Network-Awareness In The Multimedia Source Media-Awareness In The WLAN 1. Single Stream served by EDCA baseline, extended (all profiles) EDCA Limited: Tagging all frames with type of service Serving tagged video in priority levels (AC2) 2. Single Stream served by CAPS baseline, extended (all profiles) EDCA & HCCA with CAPS Limited: Tagging all frames with traffic stream ID Requires video pattern information, serving tagged video in guaranteed access traffic session 3. Partitioned Video served by EDCA extended EDCA Tagging different partitions for different priority levels. Serving packet of each partition in a different priority level (A: AC2, B & C: AC1) 4. Partitioned Video served by CAPS extended EDCA & HCCA with CAPS Tagging different partitions for different traffic streams. Requires video pattern information, serving partition A packets in a separate guaranteed access traffic session from partitions B & C 5. Partitioned Video, Partially Aggregated, served by CAPS extended EDCA & HCCA with CAPS Tagging different partitions for different traffic streams, aggregating partitions B and C (or A & B). Requires video pattern information, serving partition A packets in a separate guaranteed access traffic session from the aggregated packets of partitions B & C UBC

26 QoS Based Solutions - Data Partitioning Loss Ratio Prt B & C EDCA Prt A EDCA Prt B&C CAPS-Aggreg Prt A CAPS-Aggreg CAPS single stream EDCA single stream Num of Video Sources Packet loss ratio for different partitions, using different schemes UBC

27 Conclusion Adjustment of packet sizes is necessary to minimize the effect of packet loss and congestion Delegating the packet size adjustment to the application layer results in increased video quality EDCA may be used : for delay and jitter tolerant video applications (streaming) when capacity is not a concern CAPS based guaranteed service provisioning can provide the required QoS for real time video When Data Partitioning (DP) is available CAPS with DP provides better quality than EDCA with DP Aggregation of partitions B & C or A & B is necessary in WLANs UBC

28 Thank You UBC