&%PAGE&

-  &%page&  -
Annex 5 to Doc. AVC-65R
Annex 5 to Doc. AVC-65R

LIAISON STATEMENT TO SGXVIII 



CCITT							Temporary Document        (XVIII/8)
STUDY GROUP XVIII
Geneva, 11-28 June 1991

Questions: 3,4/XV; 2,13,22/XVIII

SOURCE : EXPERTS GROUP FOR ATM VIDEO CODING IN SGXV
TITLE  : LIAISON STATEMENT TO SGXVIII 
Purpose: Action

----------------

1.	Introduction

The second meeting of the CCITT SGXV Experts Group for ATM Video Coding 
was held in Paris, 23-31 May 1991, to progress studies on video coding for 
services on the B-ISDN.  

Many network related issues impact upon the work of the Experts Group, as 
reflected by the list of questions put to CCITT SGXVIII after the first 
meeting in The Hague in November 1990.  Many of those questions cannot be 
answered in detail at this stage, but responses from SGXVIII to some of the 
questions were welcomed.  Some additional issues were raised at the second 
meeting, and these are detailed below in Section 2.  

The Experts Group also considered the Integrated Video Services (IVS) 
Baseline Document initiated by SGXVIII in Matsuyama in Nov./Dec. 1990.  The 
SGXV Experts Group welcomes this initiative taken by SGXVIII under its 
coordinating role, and wishes to advise that it intends to offer substantial 
contributions during the evolution of the document.  As an initial step, 
some text is offered in Section 3 below.

2.	Network Related Questions

The SGXV Experts Group would welcome a response from SGXVIII on the 
following issues to assist in the progress of video coding and video system 
architecture developments.

CLR for services with differing rate behavior

Will VBR and CBR services be subject to the same CLR (in both low and high 
priority classes)?

Differential delay between virtual channels

The Experts Group recognises the advantages offered by the multiplexing of 
multimedia connections on the basis of virtual channels (VCs).  However, it 
will be necessary to limit any differential delay between VCs.  Is there any 
technique to limit the difference of delay for multiple VCs?  Will the 
network offer a mechanism whereby a request that multiple VCs should be 
supported over the same transmission path can be satisfied?

AAL Type 2

The Experts Group is considering the functionality that it may require from 
the AAL for video services support.  Will the function of AAL Type 2 be 
determined only from the standpoint of video coding?  The required 
functionality may not necessarily be uniform across the range of services, 
applications and coding methods.

CLP bit

The Experts Group wishes to clarify whether the CLP bit could be changed 
by the network after it has been set to "high priority" by the user.  If this 
change could occur, it would be of serious concern to the Experts Group.

3.	IVS Baseline Document

The SGXV Experts Group welcomes the creation, and recognises the value, of 
the IVS Baseline Document as a means of coordinating video coding studies 
for the B-ISDN.  Continued support of the document, and contributions to it, 
will be provided.

3.1 Video Service Interworking

The CCITT SGXV Experts Group for ATM Video Coding proposes that the text 
of the IVS Baseline Document of Dec. 1990 (Matsuyama) be modified as follows.

	The text contained in Annex 4 (Video Service Interworking) 
should be deleted, and replaced with the following:

Annex 4.  Video Service Interworking

Integration of video services is recognised as a key objective for 
ATM Video Coding.  It is an agreed target for the video coding system 
under study by the SGXV Experts Group.  Several options exist for 
interworking between services:
	
Negotiation Approach:
     
	At the commencement of a connection, terminals negotiate a set of 
parameters with which both can cope.  A set of standards of 
increasing quality would be defined and a basic capability assumed 
for all terminals.

Simulcast Approach:

	Transmitting terminals contain multiple encoders, operating at a 
variety of resolutions and quality levels so that broad 
interconnectivity can be achieved.  Receiving terminals could be 
simple devices able to receive one of the bit streams, or could 
contain multiple decoders allowing a selection.

Layered Signal Approach:

	A hierarchical representation of the video signal is defined.  
Coders transmit a baseband signal which provides a basic quality 
service.  Incremental signals, which can be used along with the 
baseband to recover a high quality signal, are also transmitted.  
Receiving terminals utilise the baseband and an appropriate number 
of incremental signals to recover the video signal to the quality 
which they are capable of displaying.  Transmitting terminals 
provide the number of signals which is commensurate with their 
input signal quality.  Note that "embedded bitstream" and "syntactic 
extension" techniques are also versions of layered coding (see TG 
CMTT/3 liaison statement to SGXVIII dated 17 April 1991 for the 
terminology).  

A range of issues needs to be considered in comparing these different 
approaches, including complexity, coding rate penalties and 
performance.  Negotiation would seem inappropriate for multipoint and 
distribution services, whereas simulcast seems inappropriate for 
storage applications (e.g. store and forward video mail).  Layered 
coding seems suited to the widest application range.  "Flexible 
layering" in which any number of layers can be used in any particular 
application, appears to provide broad interworking capability with few 
restrictions, and is currently one of the options being studied.  

It is recognised that to provide easy interworking or conversions 
between services, and to use common display components on a terminal 
device intended to access multiple video services, the definition of a 
family of picture formats would be beneficial.  Picture formats 
represent an important area that will influence video coding and it is 
being studied actively in the SGXV Experts Group.

- END -

3.2 Network Issues

The CCITT SGXV Experts Group for ATM Video Coding proposes that the text 
of the IVS Baseline Document of Dec. 1990 (Matsuyama) be modified as follows.

	The text contained in Annex 2 (Network Aspects) should note the 
needs of the Video Coding Experts to be advised of certain parameters 
having important impact on the coding:

Annex 2. Network Aspects

Add a Section with title "Network Parameters Impacting on Video 
Coding Definition" as follows:

A number of parameters and operational procedures concerning the B-
ISDN network will have significant impact on the definition of 
appropriate coding schemes for the support of video services.  The 
areas requiring definition are listed below:

Cell loss ratio

This is an important determinant of the quality of service achievable 
for a video application.  It determines the means, and even necessity, 
for providing cell loss protection for different services.  It is 
recognised that there is a degree of flexibility in this figure, since 
the network operators have some flexibility to dimension the network 
to provide certain cell loss ratios if they are considered essential 
for some video services, while the codec design can also be changed to 
accommodate different figures.  Progress needs to be made, though, 
perhaps by considering the impact of a range of cell loss ratios on 
both network and codec.  The cell loss ratios for both priority levels 
need to be defined.  The SGXV Experts Group believes that guaranteed 
overall cell loss ratios, for both priority levels, will be essential to 
satisfy video quality of service requirements.  Guaranteed 
performance, at least within certain time intervals, will also be 
required.

If the cell loss ratio is sufficiently small, no cell loss protection 
may be necessary.  For example, a high quality videoconference 
connection operating at 10 Mbit/s would suffer only one cell loss 
every 10 hours with a CLR of 10-9.  This may be acceptable even if the 
cell loss caused visible degradation.  

Studies are required to determine the quality of service parameters 
available to the user, and to relate these to cell loss ratio.

Cell loss burst behavior

It is understood that cell losses may occur in bursts.  This impacts on 
the means of cell loss protection; the use of forward error correction 
may be too expensive and delay may be excessive for conversational 
services if multiple consecutive lost cells must be detected and 
corrected.  Cell loss burst behavior may be modeled by the Gilbert 
model (a two-state Markov model requiring four transition 
probabilities, with one state representing no cell loss and the other 
constant cell loss).

Open questions remaining are:

-	How will the cell loss burst behavior depend upon the service 
rate?

-	Will the burst behavior of high priority cells differ from that of 
low priority cells and, if so, how?

-	How can we estimate the average interval time, T, in which no cell 
loss occurs?  If T>>1/(bitrate x CLR), the requirement for CLR might 
be relieved.

Use of CLP bit

The CLP bit is seen as a useful mechanism to provide protection 
against cell loss by controlling that information which might be lost.  
It is crucial that, after a cell is labeled "high priority" by a terminal 
device, this is not changed by the network.

Open questions:

-	Will there be separate negotiations for the two priority levels?

-	Will the usage monitoring structure encourage use of both high and 
low priority cells?
 
-	What options are available in selecting the quality of service?

Usage parameters

The rate statistics required of a video encoder have a significant 
impact on its performance (in terms of picture quality and delay).  For 
circuit switched networks, the target was straightforward; minimise the 
rate and keep it constant.  For the B-ISDN (with the possible 
advantages of variable rate over constant rate operation), entirely 
different rate control strategies may be appropriate, and these could 
have a significant impact on codec performance.  At this stage, the 
only clear decision is that peak rate will be an important parameter 
that is monitored.

In our group the term "window" means the policing time for the 
average bit rate.  The following methods are considered for policing 
in the network:

	Jumping window: 

		There is no time interval between two successive windows.

	Moving window (sliding window):
 
		The window is sliding at a time step smaller than the 
window size.

	Stepping window:
 
		There is a time interval between two successive windows, 
which always start at a valid cell.

	Leaky bucket: 

		Cells are put into a buffer and taken from the buffer at an 
average bit rate.  If the buffer overflows, cells are discarded.

If a codec does not know when the network measuring window starts, it 
should control the bit rate by sliding window (the most severe 
method).  Is there any way in which the starting time of the network 
measuring window can be known?

Open questions:

-	What parameters will be used for policing and admission 
control?

-	What policing mechanism will be used?

-	What averaging intervals can be used to measure mean, peak, etc.?  
Longer intervals (significantly greater than a video frame period 
which is typically 33-40 ms) are preferred for video services.

-	When the network capacity is very large, the bit rate requirements 
of a single user will be relatively small.  In this situation it seems 
there will be very little difference in the required network 
resources for low and high priority cell loss classes.  Will the high 
priority cell loss class continue to exist in the future?

Multimedia connections

Multiplexing of multiple media has been carried out within the 
terminal device for circuit switched networks.  The B-ISDN already 
offers the flexibility to use cell-based or virtual channel based 
multiplexing instead.  An important factor in the choice between 
terminal-based or cell-based multiplexing is whether there will be a 
penalty caused by the use of an ensemble of virtual channels instead 
of one composite one, although the overall rate characteristic, for 
example, would be the same.  Most importantly, would the two options 
have the same transmission costs?

Some multimedia connections (most obviously associated audio, stereo 
in particular, and video channels) require synchronism.  A concern 
arises, therefore, if the differential delay between virtual channels 
became noticeable in some service applications.  This is unlikely to be 
a problem unless the cumulative differential delay exceeds some tens 
of milliseconds from end to end. 

Open questions:

-	How will multimedia services be handled in the B-ISDN? 

-	What signalling methods are being proposed?

-	What kind of multimedia multiplexing method is preferred from the 
standpoint of network resource management?

Bit error rates

Cell payloads will be subject to a small probability of transmission 
error on the B-ISDN.  The statistics of such errors will determine the 
need for, and type of, error correction mechanism and the overhead 
necessary to achieve this.  It could also influence approaches to, and 
efficiency of, video coding and choice of codeword assignment scheme.  
Estimates of the likely bit error rates are required by those working 
on video coding schemes for the B-ISDN.

For interworking between video codecs on N-ISDN and B-ISDN networks, 
the B-ISDN bit error rate must be no greater than that for the N-ISDN.  
It should also be noted that the H.261 coding scheme for N-ISDN 
provides bit error correction, so this would not be a necessary 
function of the AAL in this case.  

SGXVIII should work in close collaboration with the video coding 
experts to define any capability within the AAL concerning bit error 
detection or correction.

Cell delay and jitter

The fixed component of end-to-end network delay contributes to the 
total service end-to-end delay and therefore is a determining factor 
in the overall quality of service.  Estimates of the limits of B-ISDN 
delay are required to quantify such performance and determine its 
impact on video encoders and decoders. 

The variation in delay, or jitter determines the size of receiver 
buffers necessary for its removal, and therefore again influences 
total end-to-end delay.  The expected statistics of cell delay jitter 
need to be known to determine the impact on the video coding system 
and overall quality of service.

- END -

4.	Conclusion

This document has raised some important network related questions and 
offered text for the evolving IVS Baseline Document, as part of continuing 
collaboration between the SGXV Video Coding Experts Group and SGXVIII on 
the development of video services for the B-ISDN.


END