spaper04_slides - Video Capacity of WLANs...

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: Video Capacity of WLANs With a Mul7user Perceptual Quality Constraint  ­ Jing Hu, Sayantan Choudhury, and Jerry D. Gibson, Fellow, IEEE  ­NeelKamal (49112218)  ­Rajkumar (85494936) ABSTRACT •  WLANs became part of the n/w infrastructure and it is cri7cal to understand the following –  Performance provided to the end user –  Capacity of these WLANs in terms of the number of supported flows (calls). •  This paper is about video user capacity of wireless subject to a mul7 ­user perceptual quality constraint. •  AVC/H.264 coded streams over an IEEE 802.11a WLAN is used as an example in this paper which uses packe7zed video over n/w. INTRODUCTION •  Numerous studies have been performed on WLANs, especially on cross ­layer designs for video over WLANs, which includes – Designing the network matched to special characteris7cs of video. –  compressing and transpor7ng video adap7vely in OSI stack. – Op7mizing by solving Cross layer design problems over mul7ple OSI layers. Mo7va7on •  The video capacity has received only a lidle aden7on even though it is a fundamental limit in video communica7on. •  Videos can be compressed to any desired bit rate, resul7ng in different reconstructed quality and video capacity should always be accompanied by quality constraint. •  Measuring the video quality is very difficult and we have MSE or PSNR as the standards which rates the video distor7on as poorly to perceptual video quality. •  Objec7ve percep7on video quality measures that are based on lower processing of HVS (human vision systems). •  Even though quality of mul7ple users is defined, it is not clear how to calculate the capacity. •  The lowest rate required for a certain video quality depends on the specific video codec that is employed. Cont.. •  Each applica7on has its own encoder according to its specific compression requirements. •  Rate Control (RC) and Rate Distor7on Op7miza7on (RDO) techniques have been included to op7mize the video codec's to compress videos with minimum distor7on at a certain bit rate. •  RC selects the quan7za7on step sizes and RDO selects the intra/inter predic7on modes, mo7on vectors and other coding parameters. •  RC/RDO problem itself is a subject of intensive study. Contribu7ons •  The simula7on of AVC/H.264 coded video over 802.11a WLAN with mul7path fading. •  In this simula7on it is observed that even the avg PSNR over all transmided frames of a video with packet losses is reasonably high, PSNRs vary significalntly across the video frames •  Video quality varies drama7cally across different transmissions over the channel. •  In order to capture these distor7ons, a new sta7s7cal indicator PSNRr,f which is defined as PSNR achieved by f% of the frames in each one of the r% of the transmissions. •  This quan7ty has the poten7al to capture the performance loss due to damaged frames in a par7cular video sequence (f%), as well as to indicate the probability of a user experiencing a specified quality over the channel (r%). WLAN Simula7on Setup •  AVC/H.264 coded video over 802.11a WLANs is simulated which provides a huge selec7on of coding schemes and values for their parameters. •  90 frames each from a group of videos are coded using combina7ons of group of picture sizes (GOPS)(10,15,30,45 frames). •  Quan7za7on parameter(QP)(26 for fine, 30 for coarse quan7za7on) •  Payload sizes (PS) (small ­100 bytes and large ­1100 bytes. Cont.. •  QP: QP dominates the quan7za7on error and has major effect on coded video rate. •  GOP: GOP determines the inter ­frame refresh frequency and plays major role when there is a packet loss. •  PS: PS is the parameter that is carried forward from the source to the PHY layer. AVC/H.264 encoder Parameters inves7gated in video over WLAN Cumula7ve distribu7ve func7on(cdf) PSNRs of each frame of silent.cif Packet Loss and Video Quality •  From the figure it is clear that even for the same video, coded using the same parameters for the same average channel SNR, the quality of the delivered video in terms of PSNR varies significantly across different channel realiza7ons. •  It can be concluded that neither avg PER or avg PSNR is a suitable indicator of the quality a video user experiences and therefore these quan77es should not serve as the basis for developing or evalua7ng video communica7ons schemes for WLANs. Coded video Data Rate •  In order to study the number of video users that can be supported by any network, we need to study the coded/compressed video data rate. •  Now in this sec7on we study the compressed intra (I) frame and inter (P) frame sizes of a video. •  When all the parameters are fixed, the coded I ­frame size of a video mostly depends on the complexity in a scene, and the coded P ­frame size depends on the mo7on across the frames. •  In the figure, we plot the coded frame size of each 90 frames of 3 different videos (silent.cif,paris.cif and stefan.cif). All using QP = 22, PS = 100, and GOPS =10. Cont.. •  We observe the following: •  I ­frames consume a much higher bit rate than the P ­frame. •  Although the I ­frame and P ­frame sizes do vary throughout the 90 frames of each video, they stay close to a certain level for each video. •  Among the 3 videos –  Stefan.cif is sports video (busiest scene) –  Silent.cif is a videoconference (simplest scene) –  Paris.cif is a news streaming video (intermediate level) 3 videos compressed at QP=22, PS=100, GOPS = 10 Defini7on of PSNR(r,f) • PSNR varies significantly • A sta7s7cal video quality measure PSNR (r,f) which is PSNR achieved by f% of the frames in each one of the r% of the realiza7ons • r captures the reliability of channel over many users and is a number between 0 ­100 according to desired consistency of user experience Observa7ons in video quality Assessment •  The frames of poor quality in a video sequence dominate viewing experience of the video •  If poor quality frames are very small then quality drop due to these is not perceivable by human eye •  When PSNR’s higher than a threshold increasing it further does not increase perceptual quality Experiment •  In order to conform the observa7ons an experiment using s7mulus ­comparison methods is conducted •  Two videos sequence’s played side by side •  On len perfect quality video is played and on right video that is compressed and reconstructed with possible packet loss and concealment is played •  3 human subjects are asked to pick a number represen7ng the perceptual quality of processed video compared to original perfect video Cont… •  50 video pairs with 20% appearing twice for tes7ng the consistency •  18 of them silent.cif,16 of them are paris.cif,16 videos of stefan.cif •  In the experiment it is observed that PSNR(f) with f=90% correlates to the opinion score the best •  This experiment shows that PSNR(r,f) can serve as an effec7ve video quality measure and f should be set around 90% for medium video frame rates Mean Opinion Score (MOS) •  Linear Mapping from PSNR to MOS achieved by r % of transmissions is •  A new mul7user perceptual video quality indicator PSNR(r,f)/MOS is defined •  PSNR (r,f) focuses on the distribu7on of the video quality across the video frames and channel uses, while MOS(r) provides guidance on the perceptual quality across different users Video Capacity of WLAN with DCF •  Two different mechanisms : Conten7on based Distributed Coordina7on Func7on ( DCF) and the polling based Point Coordina7on Func7on (PCF) •  DCF achieves automa7c medium sharing between compa7ble sta7ons by implemen7ng Carrier ­Sense Mul7ple Access with Collision Avoidance (CSMA/CA). Assump7ons for video capacity calcula7on •  All video users are communica7ng same type of videos •  H.264/AVC is the default video codec used by all the users and all the parameters are chosen same •  Collisions are not considered in capacity calcula7on •  Always assume that frames are received without errors Reason for calcula7ng max no of users supported •  it provides the network operators an idea of how many users can be supported in the network in an iden7cal traffic category •  if there is a mix of video users, the capacity number can be approximated by an interpola7on of the capacity values for each traffic category present Video Capacity with no extra buffer at Receiver •  Worst Case : When all users have I ­frames to be transmided at the same 7me, the capacity, C(m) is given by •  SI/PS is the compressed I ­frame size,10^3/FR is transmission deadline and T(PS,DR) is the transmission 7me of a packet with certain payload PS Cont.. •  Best Case : all users happen to be coordinated in I ­ frame refreshing, The video capacity with C(m) video users is given by •  The numerator in this formula calculates the total transmission 7me allowed for a group of pictures, T (PS,DR) is the transmission 7me of a packet with certain payload PS, (Si+Sp) * (GOPS ­1)/PS is the number of packets contained in a compressed group of pictures of a single video user Video Capacity with extra buffer at Receiver •  The playout buffer is only for the frames that have more bits than the other frames, such as I frames, to take more 7me to transmit, and then the extra 7me taken by these frames is compensated by the following less intensive frames, such as P frames •  For the video capacity to achieve its upper bound, the buffer length needs to be Cont.. •  The video capacity with extra buffer (length of ms), fluctuates between the lower bound C (m,b) and the upper bound C(m) is given by •  Summariza7on of the results Conclusion •  A new mul7user perceptual video quality indicator PSNR (f,r) / MOS(r) is proposed to capture the distribu7on of the distor7on across the video frames and channel uses (realiza7ons or number of users). •  WLAN operated under the Distributed Coordina7on Func7on (DCF) are formulated for the case when there is buffering and no buffering at the receiver •  Combining the mul7user perceptual quality indicator and the video capacity calcula7on, a methodology for video over WLAN communica7on system design and evalua7on is proposed Ques7ons ...
View Full Document

Ask a homework question - tutors are online