Video Streaming Data Scheduling

autsys.aalto.fi

Video Streaming Data Scheduling

VIDEO STREAMING DATA

SCHEDULING

Control Engineering Laboratory

Karina Gribanova

09/09/2003 1


Introduction

The purpose of the discussion:

to familiarize the audience with the video streaming data

transmission and scheduling

09/09/2003 2


Topics of Discussion

Video streaming

Video transmission system

Video coding

Video source modelling

Video streaming data scheduling goals

09/09/2003 3


Video Streaming (1)

Video streaming is a technique that enables simultaneous

delivery and playback of the video. Conceptually it can be

thought to consist of the following steps [1]:

partition and compress video into packets

start delivery of the packets

begin delivery and playback at the receiver while the video is still

being delivered

Video streaming provides number of benefits, including low

delay before viewing starts and low storage requirements

since only a small portion of the video is stored at the client at

any point of time.

09/09/2003 4


Video Streaming (2)

The video streaming communications can be:

point-to-point

multicast

broadcast

The transmitted video can be:

Captured and encoded for real-time communication (interactive

videophone, video conferencing, life broadcast, sporting events)

Pre-encoded and stored for later viewing (video-on-demand

(VoD))

09/09/2003 5


Video Transmission System (1)

Basic video transmission system:

Original

video

Source

Encoder

Channel

Encoder

Channel

Channel

Decoder

Source

Decoder

Reconstructed

video

Transmitter

Receiver

Error

Resilience

Error

Concealment

The encoded data bit-stream is segmented into fixed or

variable length packets, modulated and sent over the

network/channel. In the wireless transmission environment,

path loss, shadowing and fast fading, co-channel, intersymbol

and multi-user interference, and power control errors

corrupt the transmitted data.

09/09/2003 6


Video Transmission System (2)

Recently the concept of Intelligent Multimode, Multimedia

Transceivers (IMMTs) has become popular in the context of

wireless systems [2].

Video

Encoder

Packet

Assembly

Mapper

n

n-class

FEC

Encoder

Mapper

&

TDMA MPX

Modulator

(Adaptive

modulation)

Channel

Feedback information

Video

Decoder

Packet

Disassembly

Mapper

n

n-class

FEC

Decoder

Mapper

&

TDMA DEM

Demodulator

09/09/2003 7


Video Transmission System (3)

The advantage of burst-by-burst IMMTs is that in hostile

wireless propagation environments their system is able to

configure itself to achieve the highest possible throughput

mode while maintaining the required transmission integrity.

09/09/2003 8


Video Transmission System (4)

cdma 2000 1xEV-DO downlink channel:

Time division multiplexed (TDM) transmission using a 1.25 MHz

carrier

1.67 ms

(Estimated (Estimated data rate) data rate) (Requested data rate) (TX (TX at requested at data data rate) rate)

Pilot-DRC

Pilot-DRC

Pilot-DRC

09/09/2003 9


Video Transmission System (5)

cdma 2000 1xEV-DO downlink channel:

Adaptive modulation: depending on the reported SIR value

the modulation is chosen to be QPSK, 8PSK or 16QAM,

providing the data rate from 38.4 kbps to 2.4576 Mbps

Forward error correction (FEC)

Automatic repeat request (ARQ)

09/09/2003 10


Video source codec (1)

The video source codec takes advantage of the inherent

video data’s feature to contain statistical redundancy:

Spatial redundancy due to the correlation of neighbouring

samples (pixels) within an image or video frame. The spatial

redundancy is reduced applying transforms and entropy coding.

Temporal redundancy due to the correlation of neighbouring

regions in successive video frames. The temporal redundancy

is reduced by the prediction of future frames based on motion

vectors.

Subjective redundancy, i.e. data to which human visual

system (HVS) is not sensitive. It was discovered that HVS is

more sensitive to low-frequency components of the image.

09/09/2003 11


Video source codec (2)

Transform Quantize Reorder

Entropy

encode

ENCODER

Motion-compensated

Prediction

DECODER

Current

frame

Subtract

+

Im age

Encoder

Encoded

frame

Im age

Decoder

+

Add

Decoded

frame

-

+

Create

prediction

Motion

estimation

Motion

vectors

Create

prediction

Previous

frame(s)

Im age

decoder

Previous

frame(s)

Inverse

transform Rescale Reorder

Entropy

decode

09/09/2003 12


Video source codec (3)

Video coding standards are developed to ensure

interoperability or communication between encoders and

decoders made by different people and different companies

Currently there are two families of video compression

standards:

the standards developed by ITU such as H.261, H.263, H.26L

the standards developed by the ISO Moving Pictures Expert

Group (MPEG) such as MPEG-1, MPEG-2, MPEG-4, MPEG-7,

MPEG-21

H.26L is being finalized by the Joint Video Team (JVT) from

both ITU and ISO MPEG.

09/09/2003 13


Video source codec (4)

There are three basic types of coded frames:

Intra-coded frames, or I-frames, where the frames are coded

independently of all other frames

Predicted frames, or P-frames, where the frame is coded

based on a previously coded frame

Bi-directionally predicted frames, or B-frames, where the frame

is coded using both previous and future frames

MPEG encodes frames in a deterministic pattern, for

example “IBBPBBPBB”

09/09/2003 14


CBR vs. VBR

In Constant Bit Rate (CBR) video coding the output of the

encoder is constant, i.e. the encoded frames are assigned a

pre-defined amount of bits and according to this the coder

adjusts its output

In Variable Bit Rate (VBR) video coding the output of the

encoder is variable due to the time-varying complexity of a

video sequence

Coding video to achieve a constant visual quality requires a

variable bit rate, thus coding for a constant bit rate would

produce a time-varying quality

09/09/2003 15


Video Traffic Modelling (1)

Most papers about video modeling focus on the VBR video

traffic

The VBR video traffic has a non-Gaussian marginal

distribution, high variance and complex correlation

properties. It is shown to exhibit self-similar, long-range

dependent (LRD) characteristics.

Of particular interest in video traffic modelling are the frame

size distribution and the traffic correlation.

09/09/2003 16


Video Traffic Modelling (2)

The modeling approaches for VBR video traffic can

roughly be divided into several main classes [4]:

Histogram-based models

Markov chain models

Auto-regressive processes (AR)

Self-similar or fractal models

09/09/2003 17


Video Traffic Modelling (3)

Video frame size sequence generated using auto-regressive

(AR) processes model [11]

15 x 104 Synthetic video sequence

10

Frame size

5

0

0 200 400 600 800 1000 1200 1400 1600

Frame nr.

09/09/2003 18


Playout buffer (1)

Playout buffer provides a number of important advantages [1]:

Delay jitter reduction

Error recovery through retransmissions

Error resilience through interleaving

Smoothing throughput fluctuation

09/09/2003 19


Playout buffer (2)

Time

Loss

Playout

Playout

Delay

Buffering

Packet

Reception

Delay

Packet

Transmission

Packet Number

09/09/2003 20


Video streaming challenges (1)

Each frame must be delivered before and decoded by its

playback time, therefore the sequence of frames has an

associated sequence of deliver/decode/display deadlines. If

the interval between displayed frames is denoted as ∆, then:

Frame N must be delivered and decoded by time T N

Frame N+1 must be delivered and decoded by time T N + ∆

Frame N+2 must be delivered and decoded by time T N +2 ∆

Etc.

The data that arrives after its decoding and display deadline

is too late to be displayed. Certain data may still be useful

even if it arrives after its display time, for example, if

subsequent data depends on the “late” data.

09/09/2003 21


Video streaming challenges (2)

In the wireless environment the transmitted video data

encounters unknown and dynamic:

Bandwidth

Delay jitter

Bit error rate (BER)

The goal to overcome bandwidth problem is to estimate the

available bandwidth and than match the transmitted video bit

rate to the available bandwidth

Delay jitter can be reducted including the playout buffer – but

the frame deliver/decode/display deadlines still remain

Error control can be introduced through forward error

correction (FEC), retransmissions, error concealment and

error-resilient video coding.

09/09/2003 22


Delay requirements [1]

For interactive applications < 150 ms

For non-interactive applications much looser, can be 5-10

seconds

09/09/2003 23


Proportionally Fair Rule [13] (1)

Q

pf

⎛DRCi[ n]


[ n] = argmax⎜ ⎟

i

⎝ Ri[ n]


where

⎛ 1⎞

1

Ri[ n] = ⎜1 − ⎟Ri[ n− 1] + DRCi[ n] δ[ Qpf[ n] −1]

⎝ tc

⎠ tc

Q pf denotes the selected user, DRC i [n] is the requested data

rate of the ith user at the nth slot, d [.] is a dirac-delta

function, R i [n] is the average throughput of the ith user during

the moving window size of about t c .

09/09/2003 24


Proportionally Fair Rule (2)

The bad channel user means a user with low DRC compared

with its average throughput. Even if the user is far from the

base station so that its propagation loss is large, it can be still

served as often as the nearer user, since each user’s priority

takes into account its average throughput.

Designed for non-real time services, but can be included into

the delay-sensitive (DS) data transmitting user decision

estimation in combination with decision based upon delays.

09/09/2003 25


Modified largest weighted

delay first [13] (M-LWDF)

Q

aiDRCi

( n)

arg max Ci

( n)

i DRC

mlwd

=

Q mlwd denotes the selected user, C i [n] is either the number of

packets or the delay of head-of-line (HOL) packet in the ith

user buffer at the nth slot, a i > 0, i=1,...,N are suitable weights

characterizing desired QoS.

The rule is throughput optimal, i.e. it renders the queues at

the base-station stable if any other rule can do so.

The rule tries to balance the weighted delays of packets and

to utilize the channel in a good manner.

i

09/09/2003 26


The EXP rule [15] (1)

Q

xp

⎛ ⎛aC i i[ n] − aC⎞DRCi[ n]


[ n] = arg max exp

i ⎜



1 aC DRC ⎟

⎝ ⎝ + ⎠ i ⎠

aC

where

N

1

= ∑ aiCi[ n]

N i = 1

Q xp denotes the selected user, C i [n] is either the number of

packets or the delay of head-of-line (HOL) packet in the ith

user buffer at the nth slot, a i > 0, i=1,...,N are suitable

weights characterizing desired QoS.

09/09/2003 27


The EXP rule (2)

For “reasonable” values of a i the policy tries to equalize the

weighted delays a i C i [n] of all the queues when their

differences are large. Depending on the weighted delay

differences the EXP rule gracefully adapts from a

proportionally fair to the one that balances delays.

The rule is throughput optimal, i.e. it renders the queues at

the base station stable if any other rule can do so.

09/09/2003 28


Video streaming data

scheduling goals

To support multiple users with good quality of service (QoS) for all

users, i.e. with packet delays not exceeding given thresholds with

high probability

To avoid the receiver playout buffer underflow/overflow

To take advantage of the channel variations (due to fast fading) by

giving some preference to a user whose channel is currently good,

thus achieving the best possible throughput

To assure fairness among the users

To meet the bit error rate (BER) requirements

09/09/2003 29


Future Work

In the first phase, to build a simulator for a scheduler which

would take transmitting user decisions according to the

packet delays in the transmitters’ buffers (=the data level in

the receiver play-out buffer) and the channel conditions,

using as data the video frame sizes generated by the autoregressive

(AR) video traffic model.

In the next phase, to include the source rate control (i.e. to

modify the video traffic model) and other possible

improvements.

09/09/2003 30


References

1. John G. Apostolopoulos, Wain-tian Tan, Susie J. Wee, “Video streaming: concepts, algorithms and systems”, HP Laboratories, 2002

2. L. Hanzo, C.H. Wong, P. Cherriman, “Channel-adaptive wideband wireless video telephony”, Signal Processing Magazine, IEEE , vol. 17, pp. 10-30, July 2000

3. I. Richardson, "Video Codec Design: Developing Image and Video Compression Systems", John Wiley & Sons, May 2002

4. O. Rose, “Simple and efficient models for variable bit rate MPEG video traffic”, Technical Report 120, Institute of Computer Science, University of Wirzburg,

July 1995.

5. G. Chiruvolu, T.K. Das, R. Sankar, N. Ranganathan, "A scene-based generalized Markov chain model for VBR video traffic", ICC 98. Conference Record.1998

IEEE International Conference on, vol. 1, pp. 7-11, Jun. 1998

6. P. Abry, D. Veitch, “Wavelet Analysis of Long Range Dependent Traffic”, IEEE Transactions on Information Theory, vol. 44, No. 1, pp. 2-15, 1998

7. N. Ansari, Liu Hai, Y.Q. Shi, Zhao Hong, "On modeling MPEG video traffics", Broadcasting, IEEE Transactions on, vol. 48, pp. 337-347 Dec. 2002

8. Li Xue Ming, Men Ai Dong, Yuan Bao Zong,"Traffic model for MPEG compatible video service", ICSP '98. 1998 Fourth International Conference on, vol. 1, pp.

831 -836, 12-16 Oct. 1998

9. D.P. Heyman, T.V. Lakshman, "Source Models for VBR Broadcast-Video Traffic", IEEE/ACM Transactions on Networking, vol 4, pp. 40-48, Feb. 1996

10. V.S. Frost, B. Melamed, "Traffic modelling for telecommunications networks", IEEE Communications Magazine, March 1994

11. Derong Liu, Sara, E.I., Wei Sun, "Nested auto-regressive processes for MPEG-encoded video traffic modelling", IEEE Trans. Circuits Syst. Video Technol., vol.

11, pp. 169 -183, Feb. 2001

12. Marwan Krunz, Satish K. Tripathi, "On the characterization of VBR MPEG streams”, SIGMETRIC’97, Cambridge, MA, pp. 192-202, Jun. 1997

13. S. Shakkottai and A.L. Stolyar, “Scheduling Algorithms for a Mixture of Real-Time and Non-Real-Time Data in HDR”, Proceedings of the 17th International

Teletraffic Congress - ITC-17, Salvador da Bahia, Brazil, pp. 793-804, 24-28 September, 2001

14. Rhee Jong-Hun, J.M. Holtzman, Kim Dong-Ku, “Scheduling of real/non-real time services: adaptive EXP/PF algorithm”, Vehicular Technology Conference,

2003. VTC 2003-Spring. The 57th IEEE Semiannual , vol. 1, pp. 462-466, April 22-25, 2003

15. Jong Hun Rhee, Dong Ku Kim “Scheduling of Real/Non-real Time Services in an AMC/TDM System: EXP/PF Algorithm”, CDMA International Conference

2002, 506-513

09/09/2003 31

More magazines by this user
Similar magazines