&%PAGE&

-  &%PAGE&  -
Document AVC-206R
CCITT SGXV                                            	Document AVC-206R
Working Party XV/1                        			January 9, 1992
Experts Group for ATM Video Coding

SOURCE : CHAIRMAN
TITLE  : REPORT OF THE FIFTH MEETING OF THE EXPERTS GROUP FOR
 	    ATM VIDEO CODING IN SINGAPORE (January 6 - 9, 1992) 
Purpose: Report

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

			1. General
			2. Documentation
			3. Tape demonstration
			4. CCITT sole sessions
			5. Joint sessions with MPEG
			6. Next Meeting


1.	General

The fifth meeting of the Experts Group consisted of two parts; CCITT sole 
sessions and joint sessions with ISO/IEC JTC1/SC29/WG11 (MPEG).  The list of 
participants appears at the end of this report. 

The first part was held at Harbour View Dai-Ichi Hotel in Singapore on 6 
January 1992 with financial support of NTT.  The second part was held at 
Human Development Centre of Asia Matsushita Electric (S) Pte Ltd during 7-9 
January 1992 at the kind invitation of Asia Matsushita Electric (S) Pte Ltd.  
The Experts Group appreciated the support of these two organizations for 
having provided opportunities to meet in Singapore.

The Experts Group was advised that Belgian Coordination Member changed 
from Mr. Verbiest to Mr. Poncin.


2.	Documentation (TD2)

For this meeting, 28 AVC-numbered documents and 2 temporary documents were 
available as listed in Annex 1.  Furthermore 81 MPEG documents, including 21 
AVC-numbered ones, were considered during the joint sessions.


3.	Tape demonstration

Tape demonstrations related to AVC-202 (NTA), AVC-186 (Fujitsu) and AVC-184 
(RTT Belgium) were given for CCITT Experts Group members on 8 January.  
Other demonstrations were given during the VIDEO group joint sessions.


4.	CCITT sole sessions

4.1 Review of the Yokosuka meeting report (AVC-177R)

Chairman picked up some highlights from the report of the previous meeting 
in Yokosuka.  The meeting confirmed that the Experts Group concentrates on 
the joint work with MPEG for the development of H.26X video coding 
algorithm.

As a specific action toward the joint sessions, the meeting agreed to raise 
the desirability to test the coding algorithm against source signals that 
are converted from NTSC/PAL/SECAM signals to 4:2:2 digital component 
signals (see Section 4.1.1/AVC-177R).

4.2 Matters for consideration of the sole session only

	4.2.1 Picture format (AVC-203)

Mr. Bjoentegaard presented a new set of conversion filter coefficients for 
obtaining 576 lines x 720 pels x 59.94 Hz progressive SCIF from 625/50 or 
525/60 CCIR 601 signals with tape demonstration.  

Contributions are requested toward a decision on the format issue at the 
next Stockholm meeting (see Section 5.2.3 e for open items).

Mr. Brusewitz announced that his laboratory has facilities to display 
various picture formats, including progressive ones, which are different 
from those of the current television system.  Advance sending of test data 
tapes is requested to check appropriate operation of the display system. 

	4.2.2 VBR control (AVC-202)

Mr. Haskell presented AT&T's recent work on interaction between network 
restrictions and VBR video coding.  The meeting welcomed this stimulating 
input, awaiting responding contributions toward the next meeting to make 
progress on the following open issues;

     -	VBR advantage in obtainable picture quality
     -	Appropriate UPC for video (audiovisual) coding
     -	Required window or bucket size for average rate monitoring in the 
network

	4.2.3 Intellectual property (AVC-182)

Mr. May provided information on his patents concerning error concealment.  
The meeting appreciated this disclosure and discussed shortly what should 
be the policy of this Experts Group on intellectual property.  Chairman 
presented the following practice of the previous Specialists Group for 
H.261;

     -	Disclosure of filed patents: participating organizations agreed to 
disclose their patents as soon as being filed.  In case of patents 
being filed but not yet in the public domain, short description was 
given.

     -	Patent licensing policy statement: participating organizations agreed 
to submit their patent licensing policy which states non-
discriminatory granting on on a loyalty-free or reasonable-terms 
basis.

The meeting felt that disclosing filed patents which are not yet in the 
public domain may need careful consideration in each organization.  Another 
factor to take into account is that we are working together with other 
standardization groups this time.

This issue will be further discussed at the next meeting.

4.3 Review of contributions for consideration of the joint sessions

Due to the time constraint, the meeting could have only short presentation 
of each document.  Detailed discussion was left for the joint sessions which 
took place during January 7-9 with objectives to define TM0 and to identify 
action points toward TM1.  Topics and relevant documents were as follows;

	1) Architecture --- AVC-181

It was clarified that this flexible layering concept is independent of the 
flexible format concept.

	2) Pre-Processing to obtain 4:2:0 --- AVC-200

The meeting took note of a fact that 4:2:0 is friendly with with existing 
standards, but slight loss of quality due to 4:2:0 may not be accepted by 
quality-sensitive application areas such as secondary distribution.

	3) Prediction --- AVC-192,194,195

	4) Low delay mode or B-frames --- AVC-185,186,187,193,199

Effectiveness of B pictures depends on the test sequence.  It is thought 
that use of B pictures is most effective for those sequences where camera 
panning or zooming is contained, thus motion compensation works well. 

Mr. Bjoentegaard stressed that we need careful study as advantage of B-
pictures is related to various factors such as quantization, motion 
compensation.

	5) Compatibility/scalability --- AVC-181,183,188,189

AVC-183 clarified that for forward/backward compatibility the system should 
support multiplexing of two independent bitstreams; one represents a low 
resolution image according to the existing standard while the other 
represents an enhanced resolution image.

For AVC-188, there was a suggestion to plot "simulcast" cases at 4 and 9 
Mbit/s, not at 2.5 and 7.5 Mbit/s, in Figure 2

	6) Rate control --- AVC-191

	7) Future study items --- AVC-183,196,198

4.4 Position of the CCITT Experts Group to the joint sessions with MPEG

	4.4.1 Input papers from CCITT EG (AVC-179,190,196,197,205)

The meeting confirmed submission of two Experts Group documents AVC-179 
and AVC-196 which had originally been produced at the Yokosuka meeting and 
refined through correspondence afterward.

The meeting supported Japan's document providing a catalog of cell loss 
resilient techniques, but with addition of "layered coding" and "use of 
FEC/CRC cells".


AVC-197 was intended to be made complete through correspondence, but due 
to time limitation it was submitted for the Experts Group consideration as 
UK document.  After some discussion, the meeting concluded that this should 
be worked out during the week.  Mr. Parke coordinated collaborators and 
submitted AVC-205 (MPEG92/027) as an Experts Group document to the joint 
sessions. 

4.4.2 Particular efforts of the CCITT group (AVC-196)

The meeting confirmed that the following items are of our greatest concern 
and that we should make our efforts to reflect them in the joint work with 
MPEG;

     -	compatibility with H.261 (scalability),
     -	low end to end delay,
     -	cell loss resilience, and
     -	low bit rate operation.

4.4.3 Generic video coding standard and application oriented coding standard
      (AVC-204)

Since we are now fully involved in the joint development of "generic" 
standard with MPEG, the meeting started to discuss how to structure such a 
generic standard so that it includes necessary characteristics for 
communication applications in the ATM environment.  Mr. Haskell raised 
whether we have to sacrifice something to achieve the "end of 1992" goal.  
Mr. Princin pointed out that the video coding standard development be 
aligned with that of B-ISDN standards.

The meeting concluded that 

     -	we should avoid a situation that different requirements reach 
different standards, 

     -	this matter be considered continuously by ourselves and be also 
discussed during the joint sessions.

4.4.4 Representatives to the joint sessions

The meeting appointed the following representatives of the Experts Group 
for the joint sessions;

	Requirements sub-group			S. Okubo
	Video sub-group				J. Guichard
	Implementation study sub-group	D.G. Morrison


5.	Joint sessions with MPEG

5.1 Requirements sub-group (S. Okubo)

	5.1.1 General

REQUIREMENTS sub-group met on January 7, 8 and 9 to review whether the 
current video working program is conforming to the description of PPD 
(MPEG91/100 Issue 3, August 1991).  The main topic was how to structure the 
"generic" standard.  

The discussion concentrated on the video related activities this time since 
only three sub-groups met in Singapore; Requirements, Video and 
Implementation.  The meeting had a common view that similar works should be 
carried out on audio and system as well. 

	5.1.2 Process toward the generic standard

Starting from the PPD, we will reach the standard through Test Models 
iteratively as shown in Figure 1.  Each time Test Model is updated, it should 
be checked against the identified requirements.

At this meeting, we made an exercise of the first iteration as described in 
Sections 4-6 below.  

	5.1.3 Clarification of "generic standard"

Though the goal of "generic standard" for the second phase work and the 
work method using Test Model to reach this goal are generally accepted and 
supported, it is not yet clear how to structure such a generic standard in 
diverse interests of the participating members. 

There are three possibilities to structure a generic standard which is 
applicable to wide range of applications as illustrated in Figure 2.  

The first option (single standard solution) means that all the decoders 
conforming to the standard can understand all the bitstreams and 
reconstruct full pictures as far as the bitstream conforms to the standard.  
In the third option (toolkit solution), on the contrary, the standard 
provides all the necessary elements to cover wide range of applications.  
Application standards specify the elements to be used for their purposes. 
Decoders for a particular application may not understand the bitstreams of 
another application.  The second option (toolkit with maximum core solution) 
is intermediate between the first and the third solution.  There may be some 
standardized elements which are used only for a particular application. 

One benefit of using a generic standard is to achieve scale merit.  Another 
benefit of the generic standard is more importantly to allow cross 
application bitstream interchange.  Some application fields may require very 
cheap implementation dedicated to each particular application.  These 
factors are also indicated in Figure 2.

After some discussion, the meeting concluded that the objective of our work 
should be to achieve the first option as far as possible, but in practice the 
second option with maximum core may be the solution.  There is another 
notion "supercore" proposed in MPEG92/068 where all the decoders 
conforming to the standard can understand all the bitstreams but not 
necessarily reconstruct full pictures.  Further elaboration on the 
structure of generic standard is necessary.

	5.1.4 Review of the PPD

We extracted the requirements part from the PPD document and reviewed each 
item is appropriate or whether there are missing items.  The outcome is 
included in the left hand column of Annex 2.  During this work, the 
following items were identified as more appropriate for the system 
requirements;


    - handling of multiple programs
    - signal encryption and/or scrambling for authorized receivers etc.
    - mechanism to implement backward compatibility between systems with 
      different bit rates 
    - editing coded signals in such applications as home video 
    - error protection in the multiplexed level

It was felt that we need a neat document listing requirements with sufficient 
definition.  This should be worked out through correspondence toward the 
next meeting.

In response to the request of MPEG/VIDEO during the meeting, we provided 
"scalability requirements and definition" as in MPEG92/078 to help them to 
make their action plan.  

Another request of MPEG/VIDEO was to consider the necessity of 4:2:2 format 
instead of 4:2:0 as source coding input.  During the discussion of 
application profile, it was expressed that digital satellite broadcasting and 
secondary distribution of television signals are two demanding areas.

	5.1.5 Application profile

As a step to do the first iteration in Figure 1, we made an application 
profile as shown in Annex 2, which indicates relevant requirements for three 
typical applications in three typical fields; DSM, distribution and 
communication.  We need such profiles for other applications.  Contributions 
are requested. 

	5.1.6 Necessary items in the standard

We briefly reviewed what elements are required to implement each 
requirement listed in Annex 2 to identify missing techniques.  The outcome is 
contained in Annex 3.  Our conclusion was that the current video coding 
work program is sufficient for the current level of requirements analysis 
results except the following two items;

     - A work program is missing for "editing encoded bitstream without 
decoding."   It requires, however, more fundamental study.  

     - For "repetition of coding and decoding," we could not identify any 
explicit techniques to achieve this property.  For the moment, this should 
be tested at some stage of completion of the coding algorithm.  

	5.1.7 Test sequences

The following test materials should be used to check the robustness of the 
expected standard;

     - Stress materials (very difficult to encode)
     - NTSC/PAL/SECAM composite source
     - Progressive scan materials
     - Impaired sources (Note)
       
     Note: Sometimes coding may amplify defects included in the source.

	5.1.8 Action points toward the next meeting


Requirements sub-group agreed on the following recommendations for 
approval of the plenary; 

	1) Further elaboration is required on how to structure "generic 
standard."  Reference should be made to the outcome of the Singapore 
meeting and MPEG92/068.  Written contributions are solicited.

	2) A list of video requirements should be produced by extracting 
relevant part of the PPD document and by adding sufficient definition.  This 
work should be carried out through correspondence toward the Haifa 
meeting.  The list is to be used not only inside standardization groups but 
also for consultation with outside communities.

	3) Application profiles should be worked out for probable applications 
according to the examples obtained at the Singapore meeting.  Contributions 
are solicited.

	4) Necessary elements to implement each requirement should be tested 
and incorporated into TM and the final standard.  REQUIREMENTS group 
considers that the current video coding work program is sufficient for the 
current level of requirements analysis results except the item "editing 
encoded bitstream without decoding" which requires fundamental study.

	5) The following test materials should be used to check the robustness 
of the expected standard;

     - Stress materials (very difficult to encode)
     - NTSC/PAL/SECAM composite source
     - Progressive scan materials
     - Impaired sources (Note)
       
     Note: Sometimes coding may amplify defects included in the source.

5.2 Video sub-group (J. Guichard)

	5.2.1 General

The video group met three days with the following agenda;

    1)	Introduction
    2)	Allocation of contributions
    3)	Discussion of topics
    4)	Toward the Test Model
    5)	Proposal for experiments

The suggestion of the chairman was to have discussion on syntax, pre- and 
post-processing, quality, rate control, M-Quant, motion estimation in order 
to define experiments with two concerns; quality and requirements (or 
functionalities).

	5.2.2 Clustering of the documents (numbers are MPEG92/nn)

     -	B-frames vs delay 		03,04,10,15,16,17,28,29,37,58
     -	VBR					21
     -	TM0 proposal			28,32,39,41
     -	Compatibility/layering/		11,13,18,19,28,41,42,45,46,49,59
	scalability
     -	Motion estimation			06,28,33,42
     -	Cell loss				20,27,28
     -	Frame vs field prediction	03,22,23,24,25,28,41,45,46,60
     -	Subsampling of chrominance	28,30,42
     -	Proposal for core 		07,28,33,34,35,36,40,41
	experiments
     -	M-Quant/rate control		21,38,56,57
     -	Non transform coding		40,42,48
     -	Tape results			42,44,46,61
     -	Guidelines				26
     -	Subjective tests			61

	5.2.3 Discussion

The discussion of documents took place according to the following items;

     -	color subsampling
     -	frame structure: I,P,B; I,P,P'; I,P and frame/field issue
     -	scalability/compatibility
     -	rate control
     -	others

It has been noted that Japan has created an Ad Hoc Group to coordinate the 
work in that country (chair: Mr. Watanabe).  In Europe, the work is 
coordinated within the COST-VADIS project (chairs: Mr. Bjoentegaard and Mr. 
Nocture).  There is no such a framework structure in the U.S.A.

During the discussion, all the main points were raised; B-frames or no B-
frames, half pel requirements, field/frame prediction and 
scalability/compatibility were the most sensitive topics.  A view has been 
confirmed that the color subsampling issue (4:4:4/4:2:2/4:2:0) was for the 
requirement group.

In order to clarify the situation, several sub-groups were created;

     -	Frame/field syntax, mode decision, motion estimation 
     -	M-Quant and rate control
     -	Scalability/compatibility

Extensive discussion took place to identify the features of the Test Model, 
but no firm conclusions could be reached.

	5.2.4 Outcome of the video group

Points 4) and 5) of Section 5.2.1 above can be summarized as follows;

Toward the Test Model

No Test Model has been defined but a "simulation framework", which is based 
on;

     -	MPEG-1 CD plus SM3 for help if necessary 
     -	4:2:2 to 4:2:0 conversion according to MPEG92/028
     -	field/frame prediction according to MPEG92/080 and 079
     -	rate control and M-Quant according to MPEG92/077
     -	compatibility syntax extension according to MPEG92/076
     -	scalability syntax extension according to MPEG92/039
     -	Guideline for cell loss experiments according to MPEG92/027

Proposal for experiments

It has been explained that, due to the fact that there is no Test Model, 
only experiments can be carried out (the wording "core experiments" 
corresponds to experiments within a Test Model!). 

The following list of experiments has been agreed;

     -	Modified quantizer for non intra (as in MPEG92/028)
     -	Study impact of full search on subjective quality
     -	Half pel refinement on decoded pictures (as in MPEG92/028)
     -	Multi-field prediction vs adaptive field/frame 
	(as in MPEG92/046 and 079)
     -	Experiment on scalability in the frequency domain
     -	Field based coding option
     -	Compatibility: embedded bitstream vs simulcast
     -	I,B,P vs I,P vs I,P',P
     -	Cell loss resilience
     -	Adaptive VLC (as in MPEG92/028)
     -	Overlapped MC
     -	Alternative to DCT

Ad Hoc Groups

A document which formulates some recommendations has been discussed, and 
two Ad Hoc Groups have been created;

     -	"Preliminary Working Draft" group, chaired by Mr. Koster, in charge of 
gathering the relevant documents for the "simulation framework"

     -	"Experiments" group in charge of compatibility/scalability related 
experiments and quality related ones.  The former is chaired by Mr. 
Parke and Mr. Gonzales.  The latter is coordinated at the regional 
level by Mr. Nocture, Mr. Puri and Mr. Yonemitsu.

It has been noted that strong emphasis has been put by some US delegates 
and that Document MPEG92/078 (scalability requirements and definition) has 
been issued on this purpose by the Requirements Group. 

5.3 Implementation Study (D.G. Morrison)

	5.3.1 Kurihama proposal rankings

The work to evaluate encoder and decoder complexity was completed during 
the meeting as contained in Annex 4, which also includes analysis of 
influential factors. 

	5.3.2 Universal transform

The meeting considered MPEG92/048 (Hughes) proposing universal transform 
whose parameters are transmitted at the sequence layer, but did not reach 
any agreed views.

	5.3.3 Future work

The meeting considered the suggested "Hardware Reference Model" approach, 
but there was no support.  

	5.3.4 Action points toward the next meeting

Implementation Study sub-group agreed on the following recommendations for 
approval of the plenary; 

	1) MPEG confirm that the purpose of MPEG is to develop a standard 
which will find widespread use through a combination of several 
characteristics which include low implementation cost.  In particular it 
remains unclear that B frames are required for reasonable picture quality 
in low end systems and the implementation group recommends that further 
study on restricting the Test Model to forward prediction coding schemes be 
carried out. 

	2) The Implementation Studies Group has provided results which should 
be of value for the MPEG second work item.  We request a response from MPEG 
Video detailing how the Implementation Studies Group work is to be 
incorporated into the algorithm design.  The Implementation Studies Group 
remains open to suggestion as to how the content and format of its results 
may be improved as work continues.

	3) Means be sought to improve the coupling of implementation studies 
with other MPEG activities.  We consider that modifications could be made to 
the meeting schedules to allow more interaction between us and other MPEG 
groups.


6.	Next Meeting

	March 18-20 in Stockholm (sole sessions)
	March 23-27 in Haifa (joint sessions with MPEG)


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

Annexes

  Annex 1	Documents for the fifth meeting of the Experts Group
  Annex 2	Examples of Application Profile
  Annex 3	Possible techniques to implement each requirement
  Annex 4	Kurihama proposal rankings


END


Participants of the fifth meeting of
Experts Group for ATM Video Coding 
(6-9 January 1992, Singapore)

FRG		Mr. F. May			Daimler-Benz		CM
		Mr. G. Zedler		DBP Telecom		CM

Australia	Mr. H.G. Lim			Monash University
		Mr. J. Princin		Telecom Australia		(CM)
		Mr. G. Smith		AUSSAT

Belgium	Mr. O. Poncin		RTT Belgium		CM

Canada     	(advised that no one can attend) 

USA		Mr. B.G. Haskell		AT&T Bell Labs
		Ms. A. Wong			Bellcore			(CM)

France 	Mr. G. Eude			CNET				CM
		Mr. J. Guichard		CNET

Italy		Mr. Gandini			CSELT			(CM)

Japan  	Mr. T. Fukuhara		Mitsubishi
		Mr. K. Hibi			Sharp
		Mr. Y. Katayama		GCT
		Mr. S. Okubo		NTT				Chairman
		Mr. K. Sakai		Fujitsu 
		Mr. T. Tanaka		NTT				CM
		Mr. M. Tsujikado		Oki
		Mr. H. Ueno			Toshiba
		Mr. T  Yukitake		Matsushita 

Norway	Mr. G. Bjoentegaard 	Norwegian Telecom		(CM)

Netherlands	Mr. A. Koster		PTT Research       	(CM)

UK      	Mr. I. Parke		BT
		Mr. D.G. Morrison		BT				CM

Sweden	Mr. H. Brusewitz		STA				CM
		Ms. C. Verreth		STA

Korea		Mr. J-G. Choi		ETRI				Observer
		Mr. J-Y. Nam		ETRI				Observer

							  CM: Coordinating Member
							(CM): Substitute for CM