Adaptive media playback (AMP) techniques dynamically adjust playback speed based on buffer conditions and network performance to reduce interruptions. Early work in 1996 introduced buffer-based smoothing to vary playback when buffers were full or empty. Later approaches in 2001 considered channel conditions, slowing playback during losses and speeding up during recoveries. Additional techniques estimated frame properties to selectively play frames most important for the viewer experience. Recent research focuses on optimizing buffer sizes and frame selection algorithms while accounting for packet-level transmission characteristics of high-definition video over specific network types like 802.11 wireless networks.
2. Background
Proposed concept falls under the general category of
“Adaptive Media Playback or (AMP)”
l
2l
AMP:
l ×s
/f
3. Historical Perspective
Adaptive Media Playout
(AMP) Content Adaptive
Optimal frame
VDoP [Laoutaris] selection
Laoutaris and
Yet another AMP
buffer design
algorithm
Video Smoothing First HD video
used for
evaluation
Motion based AMP
4. Video Smoother -1996
Playout buffer (size N)
Video frames playout of frames
Threshold
(TH)
Playout rate Time
controller
If number of frames in playout buffer exceeds
TH, maximum playout rate (m) is employed.
Otherwise smoother uses proportionally reduced rates to
eliminate playout pauses due to empty buffer.
[1] M. C. Yuang, S. T. Liang, Y. G. Chen, and C. L. Shen, “Dynamic video playout smoothing method for
multimedia applications,” in IEEE International Conference on Communications, 1996, vol. 3, pp. 1365–1369.
5. Video Smoother
Metrics as TH increases (l: mean frame arrival rate)
p0: probability of empty buffer pL: frame loss probability
8. Adaptive Media Playout (AMP) - 2001
Adaptive Media Playout: the adjustment of the playout
speed of the media packets depending on the condition
of the channel and the current client buffer fullness.
Video playout based on channel conditions:
Bad conditions: slow down playout, virtual increase in buffer
Good conditions: following recovery of bad
conditions, playout is done faster than normal
[2] E. Steinbach, N. Farber, and B. Girod, “Adaptive playout for low latency video streaming,” in Image
Processing, 2001. Proceedings. 2001 International Conference on, 2001, vol. 1, pp. 962–965.
9. AMP
Assumptions: Audio+Video frame fit into one packet
Lost packets:
retransmission requests are
sent from client to server
(application layer)
10. AMP
Packet burst error and the arrival of retransmissions
If a packet burst loss exceeds the maximum playout time,
we get buffer underflow: freeze video
11. AMP
Two main criteria for evaluation
Probability of buffer underflow
Average value of max burst length:
Average end-to-end delay introduced by adaptive playout
12. AMP
Metric: Mean Time Between Buffer Underflow (MTBBU)
s: slowdown factor
(s≥1)
f: speed-up factor
(f≤1)
13. Variance of Distortion of Playout
(VDoP) - 2001
New metric to gage interruptions in video playback
Extension of 1996 work done by Yuang which suffered
from an undesirable fast forward effect
[3] N. Laoutaris and I. Stavrakakis, “Adaptive playout strategies for packet video receivers with finite buffer
capacity,” in Communications, 2001. ICC 2001. IEEE International Conference on, 2001, vol. 3, pp. 969–973.
15. Recent
Take into consideration motion characteristics of frames
Choosing specific frames (frame selection)
Most work revolves around finding the optimum buffer
size, threshold and corresponding frame rate adjustment
in the context of AMP
16. Many of the works consider the network to be a cloud
(possibly internet, LAN, etc.)
We consider specific home network, ad hoc, 802.11 based
Most deal with frames as a whole, and don’t get into details
of packetization
We are relying on frames split across packets
Assumptions are made on frame arrival rates
In our scenario a more accurate estimation of frame arrival
rates can be considered
HD video is not considered
Our work is centralized around HD video
Notes de l'éditeur
Selection of TH is critical. Overestimation: playout rate declines giving bad playout performance. Underestimation: higher probability of having an empty buffer = playout discontinuityOptimal TH trade off rise in probability of empty buffer against increase in playout rateQueue analysis is performed and the frame arrival processes is modeled by a Poisson process. This is to simulate random networking delays, etc.Authors solve for optimal probabilities of empty buffer and frame loss probability, which is the metric for this study. Seems to be either fast or slow!!!!
Stanford University
When target buffer size has been reached (similar to TH in previous work) playout begins
For B_target from 50 to 200 packets in increments of 10. Higher MTBBU is good! Channel model is a two-state Markov (good or bad) . Packet loss probability is 0.05No video is used…. Everything is based on packet assumptions
Again Poisson arrival process of frames (they admit the poor choice of modelling the frame arrival rate)Makes use of variance of discontinuity (VoD)
MPR: Mean Playout RateSmall TH allows buffer overflow to take place thus losing frames