&%PAGE&

-  &%PAGE&  -
Document AVC-107R
CCITT SGXV                                            	Document AVC-107R
Working Party XV/1                        			August 23, 1991
Experts Group for ATM Video Coding

SOURCE : CHAIRMAN
TITLE  : REPORT OF THE SECOND MEETING OF THE EXPERTS GROUP FOR
 	    ATM VIDEO CODING IN SANTA CLARA (August 14-23, 1991) 
	    - PART II 
Purpose: Report

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

PART II - JOINT SESSIONS

				1. Requirement Sub-group
				2. Test Sub-group
				3. Video Sub-group
				4. System Sub-group
				5. Implementation Sub-group

1.	Requirement Sub-group (by Sakae Okubo)

1.1 General

REQUIREMENTS/TEST subgroups met in cascade on August 20-22 under 
chairmanship of Mr. S. Okubo and Mr. T. Hidaka.  Before starting the sessions, 
time and input document allocations were decided between the two sub-
groups.  These meeting sessions were jointly held by MPEG and CCITT EG.

On August 20, the two groups had a joint meeting of one hour with VIDEO 
subgroup to discuss MPEG-2 matters.  In the evening, a close pre-view for 
demonstration tapes was held at Apple with a purpose of allowing critical 
viewers to observe processed pictures at appropriate viewing distance. 

The meeting confirmed the following targets for the REQUIREMENTS/TEST 
meetings; 

   1)	Identification of additional general requirements to MPEG-2
   2)	Finalization of the practical matters for the Kurihama Tests
   3)	Finalization of the testing methods for the Kurihama Tests
   4)	Finalization for the video related part of the PPD document

1.2 Time schedule

August 20 (Tue)  9:00-13:00  REQ - general requirements
                14:00-15:00  VIDEO/REQ/TST - MPEG2 matters
                15:30-17:30  REQ - general requirements

August 21 (Wed) 11:00-13:00  TST - testing and analysis methods
                14:00-16:00  TST - testing and analysis methods 
                16:00-17:00  RQ/TST - reflection to PPD

August 22 (Thu)  9:00-13:00  REQ/TST - meeting report

In the joint session among VIDEO, REQUIREMENTS and TESTS, possible 
methodologies for convergence after the Kurihama Tests were discussed.  The 
details are contained in the report of VIDEO subgroup.  According to the 
discussion, the following request is added to Section 6.2.1 4) a.:

	"Good documentation containing the following materials is requested 
to make convergence easier after the Kurihama tests (see Section 5.3 
for objectives of the subjective tests)"

1.3 Documentation

REQ/TST: MPEG/100 and its revision (CHAIR-REQ)

REQ    :  094(CHAIR), 099(CMTT), 108(HUGHES), 115(MITSUBISHI), 139(CCITT),
         140(CCITT), 144(CCITT), 151(CCITT), 155(MIT), 158(VADIS), 160(INTEL),
         163(PRISM)

TEST   : 093(CHAIR-TST), 105(CMTT), 112(BNR), 116(JVC), 136(TCE), 168(Hughes)

1.4 Agreements of REQUIREMENTS subgroup

1.4.1 Review of the draft PPD revision (MPEG91/100 and its revision - Issue 2 
      updated of July 29)

The revisions added to MPEG91/100 (AVC-70, parts indicated with bar in the 
left hand margin) were approved with the following amendments;

   1)	p.2, Section 1.1: include the statement on the tope of page as part of 
the body. 
   2)	p.4, Section 3.3: add "assuming CCIR-601 input" at the end of sentence  
above the table showing two categories;
   3)	p.5, Section 3.4 1), CCIR601: change "240" to "243 (Note)" and add 
"Note - It should be noted that CCIR Rec. 601 defines active numbers 
per field as 243 for 525/60 systems.
   4)	p.5, Section 3.4 1), Progressive scan format: change "25/30" to 
"25/30/50/60".
   5)	p.10, Section 4.1: add "for existing television standards" after "-2 
picture frames ... + 3 picture frames (see CCIR Rec. nnn)".  
   6)	p.13, Section 6.2.1 3): add Section 3.2 3) of MPEG91/094 between g. and 
h.
  
"h. The greatest degree of compatibility would be achieved by a core 
MPEG1 decoder operating on a bit stream in the range of 1.0-1.5 
Mbit/s." 

1.4.2 Review of the input documents to this meeting

	1.4.2.1 MPEG1/099(CMTT) Preliminary functional requirements for 
	        secondary distribution of digital TV and HDTV 

It was agreed to incorporate the following parts into the PPD document;

   1)	Section 4/099 into Section 3.3 of PPD, after the table of categories.
   2)	Section 5/099 into Section 3.4 5) of PPD, at the end.

During the discussion, the following comments and questions were given:

    -	Quality objectives: We are considering whether future MPEG standards 
should guarantee performance in terms of picture quality.  It is appreciated 
if CMTT/2 could provide information on how the picture quality of video 
codecs conforming to Rec. 723 have been measured or to be measured and how 
the results are represented.
    -	Features: We believe some features such as random access lead us to 
common technical requirements to the video coding if we consider such 
features as channel hopping required for the secondary distribution.

    -	Compatibility: It is appreciated if we could receive specific guidelines 
based on the "20-30 year long life of the standard" requirement so that we 
can interpret it into design objectives of video coding which we intend to 
standardize in the time frame of 1992-1993.

The meeting agreed to contain these comments and questions in a liaison 
statement to CMTT/2 SRG.

	1.4.2.2 MPEG91/108 (Hughes) Scalability  

It was agreed to add a note in Section 3.4 1) the second last item as follows;

	Broadcast television and scalable window system (Note) are desirable.
                                                       ^^^^^^
	Note - A scalable video format is defined as one where the parameters 
of decoding are independent of those of encoding.  A bitstream is 
scalable when some coded bits can be disregarded and a usable image 
still results.  Scalability facilitates decoding the images at different 
rates and resolution scales through the design of the bitstream or 
data representation itself (see MPEG91/108).

	1.4.2.3 MPEG91/115 (Mitsubishi) A program to generate moving Campbell 
                                        chart

Information was provided on a program which generates the moving Campbell 
chart for checking particular elements of the coding algorithm.  Those who 
are interested in can get contact with Mr. Nishida for further details.

	1.4.2.4 MPEG91/139 (CCITT EG) Additional submission materials for the
                                      "Kurihama Tests"

It was agreed to adopt the proposed additions to the PPD document 
encouraging consideration of cell loss resilience;

   1)	Add to Section 6.2.1 3) i.

	Any claims for additional features (such as cell loss and random bit 
                                          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
	error resilience) should be supported by demonstrations.
	^^^^^^^^^^^^^^^^^

   2)	Add note to Section 6.2.1 4) c. the last hyphenated item

    -	any other functionalities if claimed (Note)
                                            ^^^^^^
	Note - To demonstrate cell loss resilience, it is recommended to 
simulate 1E-3 cell loss ratio, e.g. by replacing 384 coded bits every 
0.1 sec with '0', decoding the resultant bit stream and reconstructing 
pictures.  Exact method of simulation should be described.

	1.4.2.5 MPEG91/140 (CCITT EG) H.26X requirements

A list of updated requirements to the projected ATM video coding standard 
H.26X was presented.  Items in Section 7 of MPEG91/140 are reflected in 
Section 3.4 7) b. as follows;

   1)	Add "high/low priority cell utilization" to the the list of B-ISDN 
transport flexibility in the 4th hyphenated item.
   2)	Add at the end of list "The video coding must provide rate control for 
conforming to the usage parameter control of the network".
   3)	Add "(e.g. for low complexity)" to the 2nd last paragraph of Section 
3.4 9).

	1.4.2.6 MPEG91/144 (CCITT EG) Remote video surveillance as a B-ISDN 
	                                service

It was agreed to add this item to the list of applications in the PPD 
document as follows;

   1)	Add "RVS Remote Video Surveillance" between "NDB" and "SSM" in 
Section 3.1.
   2)	Add "RVS" after "ENG/SNG" with the following parameters;
	Start from an arbitrary point: yes
	Symmetry in allowed complexity: C << D

	1.4.2.7 MPEG91/151 (EBU) Basic audio quality requirements for digital 
	                          audio bit-rate reduction systems for broadcast
	                          emission and primary distribution

The meeting took note of the "indistinguishable from the compact disc 
quality" requirement and the precise definition of being distinguishable.  
How to reflect this content into the PPD document awaits AUDIO group 
discussion.

	1.4.2.8 MPEG91/155 (MIT) Operations on the bitstream

A necessity of "manipulation on the coded bitstream" and some applications 
were presented.  The necessity was also stressed by CCITT EG so that the 
continuous presence multipoint system can be realized.  Study progress in 
this topic is expected. 

Having heard this presentation, the meeting agreed to make the following 
amendments in the PPD document;

   1)	Add "(see MPEG91/155)" at the end of Section 3.4 14) 
   2)	Add the following items to the end of Section 6.2.2 2)
	"- partial decoding and recoding for rescaling"
	"- on-line transmission bit rate selection on demand

	1.4.2.9 MPEG91/158 (VADIS) EUREKA 625 (VADIS) press release 1

Requirements related activities in a European project VADIS was presented 
for information.

	1.4.2.10 MPEG91/160 (Intel) Proposal for parallel decode requirement 
	                             for MPEG-2

It was presented that potential reduction of hardware may be obtained by 
using parallel structure implementations.  

	1.4.2.11 MPEG91/163 (PRISM) Some hooks for error recovery

Necessity of some hooks for error recovery was presented for information.  
It was pointed out that other means such as use of INTRA may give a 
solution.  How to implement the error recovery is for further study. 

1.4.3 Clarifications found necessary during the meeting

	1.4.3.1 Small difference in number of coded pictures

The meeting agreed to allow small difference in number of coded picture 
elements such as 704 pels/line instead of 720 pels/line, and 240 lines/field 
instead of 243 lines/field on the assumption that this small difference does 
not affect the coding performance.  The algorithm proposers should include 
these parameters in the description.

According to this agreement, the second line of Section 6.2.1 3) f. should 
now read 

"Some coding algorithms may subsample or crop the input signal ..."
                                      ^^^^^^^ 
	1.4.3.2 Random access specification for the test (Section 6.2.1 3) c. and
	        d./PPD)

The following modifications were agreed to clarify the current specification:

   1)	Replace the current Item c. including Note with; "Maximum interval 
between two entry points for random access must be less than about 2/5 
second (equivalent to 10 frames at 25 Hz, 12 frames at 30 Hz)."

   2)	Add at the end of Item d. "(including a single frame (namely two fields
in a set) random access)".

	1.4.3.3 Paper listing for the coded bitstream file

The following item is included at the end of in Section 6.2.1 4) a. of the PPD 
document for completeness of the required documentation for proposal 
submission.

	"- Paper listing for each sequence which indicated the corresponding 
coded bit stream file in a format "ls -l" output (see Section 7 of this 
document"

	1.4.3.4 Obsolete items

Delete Section 3.4 of the PPD document.

	1.4.3.5 Questionnaire for hardware complexity estimation

The following outcome of IMPLEMENTATION subgroup is included in the PPD 
document.

   1)	Add the following at the end of Section 8;

"It is a very desirable objective that that it be possible to implement fully 
automatic encoders.  The extent to which any non-automatic adjustment of 
coding parameters has been used to generate the pictures submitted to the 
Kurihama tests must be declared.  The nature of these must be described 
such that an assessment can be made of the feasibility of eventual 
automation in real encoders."
   2)	Add the following list at the end of Section 6.2.1 4) a.

  " -	number and sizes of all picture buffers, including necessary display 
        buffers
    -	size of coded date buffer
    -	for each module

	    -	size and width of memory (on/off case) 
    -	memory bandwidth (on/off case)
    -	number and width of additions per second
    -	number and width of multiplications per second
    -	table sizes (specify fixed or downloaded)
    -	number of table lookups per second (specify fixed or 
downloaded)
	    -	a functional description including data, control, and address 
generation function and flow. Similarly, if any, to well known 
modules such as DCT, VLC decoder should be pointed out.  Any 
"tricks" that can simplify implementations e.g. symmetry in 
tables should be stated.

    -	global

	    -	Any implementations on coder and decoder when altering modes, 
parameters etc. to suit applications.
	    -	Any claims with supporting justification that the algorithm is 
amenable to simple.cheap implementation.
	    -	Block diagrams etc. and explanation of algorithm operation 
(likely to be similar to, if not same as, information required 
by VIDEO group) "

	1.4.3.6 Test sequences for the Kurihama tests

Replace Table 2 with that in the report of TEST subgroup.   

1.4.4 Outstanding items for the Kurihama tests

	1.4.4.1 Schedule

Two alternatives were considered for the Kurihama meeting including the 
subjective test; November 14-22 (14-19 for subjective test, 18-22 for WG11) or 
November 18-26 (18-21 for subjective test, 18-26 for MPEG1).  Some members 
expressed preference for the first alternative, while some others for the 
second alternative.  Most of the members kept silence.  These feelings were 
conveyed to Convenor for facilitating his decision.

	1.4.4.2 Testing bit rates and sequences

After having reviewed the tape demonstrations, the meeting agreed to keep 
the testing bit rates as they are according to the agreement at the Paris 
meeting (see Section 3.3 of MPEG91/094).

Testing sequences were agreed as in the report of TEST subgroup based on 
the number of pre-registered coding schemes and the capacity of the hosting 
laboratory.

	1.4.4.3 Proposal document submission

Every proposer is requested to submit the proposal document according to 
the MPEG document management agreement (see Recommendation nn of the WG11 
Santa Clara meeting).

It is noted that the hosting laboratory, JVC, does not provide copying 
service for these algorithm proposal documents.  

	1.4.4.4 Deadline for D-1 demonstration tape

Each proposer is requested to bring his/her D-1 tape directly to the 
meeting for demonstration of those features other than normal play back 
(see Section 6.2.1 4) c.).

1.4.5 Modifications made during the MPEG Plenary on August 23

   1)	Section 3.4 7) c. 1st line: change "optimized" to "adequate"

   2)	Add the following as new Section 3.5
 
	3.5 Implementation aspects

	Algorithm developers seek solutions which allow flexibility in the 
	choice of implementation architectures, thus giving manufacturers the 
	opportunity to design equipment for the widest range of applications.

   3)	Section 4.1: change "-2 picture frames" to "-20 ms" and "+3 picture 
frames" to "+40 ms", and reference be made to CCIR Rec. 772.

   4)	Section 6.2.1 4) a.: add the following at the end

     -	It is a very desirable objective that it be possible to implement
  	fully automatic encoders.  The extent to which any non-automatic
  	adjustment of coding parameters has been used to generate the
  	pictures submitted to the Kurihama tests must be declared.  The
  	nature of these must be described such that an assessment can 
  	be made of the feasibility of eventual automation in real 
  	encoders.

1.5 Improvement of document handling

Related to the discussion in Section 4.4.3, a proposal was raised to improve 
the document handling in meetings that each document or at least its 
abstract be distributed in advance of the meeting among participants.  This 
allows participants to review the contribution before the meeting starts, 
thus increases the productivity of the meeting.   The meeting supported this 
proposal, but felt practical difficulties at the same time.  Convenor should 
be consulted if we could establish such a mechanism. 

1.6 Recommendations of REQUIREMENTS subgroup

   1)	Video coding algorithms be submitted according to the revised PPD 
document made available at the WG11 plenary of the Santa Clara meeting. 

   2)	Identification of functional requirements for the audio and system be 
initiated at the next meeting.

   3)	Evaluation for functionalities of the proposed algorithms be initiated
at the next meeting.

2.	Test Sub-group (by Sakae Okubo)

2.1 General

MPEG/TEST meeting took place in two sessions under chairmanship of Mr. T. 
Hidaka; 14:00-17:00 of August 21 and 12:00-13:00 of August 22.

2.2 Finalization of the boundary conditions for the kurihama test

2.2.1 Test sequences

Since the number of intended proposals is so large as 39, the number of test 
sequences have to be reduced as follows;

	----------------------------------
	   Sequence		    4M   9M
	----------------------------------
	Flower Garden            X    X
Popple                        X
Table Tennis             X    X
Mobile and Calendar      X    X
Football                 X
----------------------------------
No. of Test Sequences    4    4
----------------------------------


Note of Chairman - At the plenary of WG11 on August 23, Mr. Koster pointed 
out that the 625 line Exabyte version of "Football" test sequence contain 
some errors.  After the meeting, Mr. Morris checked the D-1 version but 
found the same errors.  Due to the short time to the D-1 tape submission 
deadline, it was decided to drop "Football" from the Kurihama tests.  

2.2.2 Assessors for the subjective tests

    1)	Two group of assessors were agreed for the Kurihama tests;

    2)	50 assessors with normal vision of color are required; 25 assessors are 
per group.

    3) Algorithm proposers have the first priority to send assessors for the 
test;

    -	one assessor per proposal 
    -	two to three assessors per joint proposal 

    4)	Other MPEG/CCITT/CMTT members can take part in the subjective tests 
on the second priority basis.

During the meeting, an application sheet for assessors was circulated, 
obtaining about 40 registraters.

2.2.3 Test conditions

Mr. Mead, Hughes, proposed the use of 2H and 4H viewing distances for the 
Kurihama tests to get better distinction among proposals in MPEG91/168.  
Some supported this proposal, while others stressed the use of well 
established CCIR Rec.500-3 method. 

The meeting concluded to maintain the official test method as it is, but at 
the same time agreed that JVC would provide additional test sessions beside 
the official ones to experiment the shorter viewing distances.  Volunteers 
are welcome to joint this experiment.

2.2.4 Final schedule for the Kurihama test

    -	4 to 5 sessions per day
    -	4 days

2.3 Discussion on next steps in MPEG decision process; ranking criteria

Mr. Lemay, BNR, proposed to use the variance analysis method in MPEG91/112 
for the Kurihama test results, which identifies a group of algorithms not 
significantly different from the best ranked algorithm.

The meeting agreed to use this method in addition to the current method for 
better interpretation of the results.

2.4 Recommendations of the TEST sub-group

    1)	D-1 tapes shall be submitted to JVC-Kurihama by 18 October 1991.

    2)	In addition to formal subjective testing in accordance with CCIR Rec. 
500-3, an experimental subjective test environment with viewing 
distances of 2H and 4H will be provided at the Kurihama meeting.


3.	Video Sub-group (by Jacques Guichard)

The video group met four times during the week.  It also had a common 
meeting with the audio group and another with the requirements and test 
groups.

-	The first meeting mainly concerned the review of the MPEG1 CD 
(Committee Draft).  This document is very close to being finalised since very 
little modification was reported.

-	The second meeting was held together with the requirements and test 
groups, and concerned:

     		1 - Methodology for convergence
     		2 - Technical inputs to MPEG2

Point 1 was discussed with the three groups.  The main issue was the 
procedure for the Kurihama tests.  The outcome of the Kurihama tests should 
be followed by a clustering phase to reduce the 39 proposals to n, and then 
to a second phase to go from n to a single TM (Test Model).  According to 
the MPEG1 experience, the importance of a verification group during the 
tests has been emphasized as well as the concept of a core experiment model 
which was very helpful.

Point 2 was discussed within the video group only.  It was limited to a brief 
presentation of the following documents:

    *	H. Sandgrind (Norway Telecom) CCITT AVC-104
    *	J. Guichard (CNET France Telecom) CCITT AVC-103
    *	K. McCann (National Transcom Ltd.) MPEG91/157, 158, 159 (some work 
which is currently done in VADIS) 
    *	I. Park (British Telecom) CCITT AVC-100
    *	A. Koster (RNL) CCITT AVC-94
    *	D. Anastassiou (Columbia University) MPEG91/131 (some consideration 
regarding the advantages of coding pictures on a field basis versus a 
frame basis)

-	The third session was devoted to tape demonstrations of coded picture 
concerning the current work for MPEG/H.26X.  Only European members 
disclosed information regarding their coding schemes.

-	The last session was devoted to patent issues.


4.	System Sub-group (by Barry G. Haskell)

The  WG11 (MPEG) Systems Committee met in Santa Clara August 19-23, 1991.  
During this period a number of improvements and simplifications were made 
to the multiplexing specifications, culminating in Revision 8/22/91 of the 
Committee Draft (CD).

Recall that at the Paris meeting a System Target Decoder (STD) was defined 
for demultiplexing data packets into individual buffers, one for each 
elementary stream.  Decoder buffer sizes are assumed known to the 
encode/multiplex system, and it is the responsibility of that system to not 
overflow or underflow the decoder buffers of the STD.  Demux System Time 
Clock (STC) recovery is the responsibility of the designer.

In Santa Clara, the following agreements were arrived at:

   1)	A System Header Packet was agreed to that contains information about 
the ISO11172 (MPEG) Stream, including maximum data rate, maximum number of 
audio and video elementary streams, whether or not the data rate is fixed, 
and the maximum sizes of STD buffers for each constituent elementary 
stream.  The first packet of the ISO11172 stream must be a System Header 
Packet.

   2)	Two private elementary stream types were defined.  Private1 uses the 
normal packet header protocol, including optional stuffing, buffer size 
indication and time stamps.  Private2 is only required to provide 
packet_start_code and packet_length.

   3)	STD buffer size specification now has two scales.  Scale 0, required    
for audio, is in multiples of 128 bytes.  Scale 1, required for video, is in 
multiples of 1024 bytes.  Other elementary streams can use either scale.

   4)	A tolerance of 50 ppm was agreed to for the demultiplexer System Time 
Clock (STC).

   5)	A maximum time of 0.7 seconds between packs was agreed to.

   6)	A start was made on defining a Constrained System Parameter Stream    
(CSPS).  In video, this concept had been referred to as "core", but the term 
seems to have fallen from favor.  So far, we have agreed only on a maximum 
packet rate within a pack of 300 per second.

A little progress was made toward understanding how variable data rates 
might work.  However, a common vocabulary has not yet been established, and 
the situation is far from being completely understood.

For a constant (known) data rate, the System CD wording is fairly 
consistent.  Bytes arrive at the STD at times known (fairly accurately) to 
the multiplexer and demultiplexer.  If the DSM is master, the demux STC tries 
to lock itself to the SCRs as they are received.  If the DSM is not master, 
the DSM (plus DSM buffer, if needed) must follow the master clock, in order 
to produce a (fairly) constant data rate.

For variable data rates, there are a number of scenarios.  If the STD byte 
arrival times are known to the multiplexer, which then sets SCR values 
accordingly, there is not much problem, since the demux STC can still lock 
to the received SCRs.

If the multiplexer does not know the STD byte arrival times, it has to work 
with hypothetical values.  For ATM networks, the byte transmission times may 
be sufficient if cell arrival jitter is small.  In that case, a receiver buffer 
would try to output bytes to the STD so that demux STC and received SCR 
values match.  Demux master clock and STC would use the fullness of this 
receiver buffer for tracking purposes.

For bursty networks or DSMs, the multiplexer might simply use a 
hypothetical constant data rate for setting SCRs.  A large receiver buffer 
would then be needed to smooth the data rate, so that Demux STC and 
received SCRs could be made to match.

The System CD does not yet distinguish between hypothetical and actual STD 
byte arrival times.

Further study and refinement of these and other aspects of multiplexing 
continue.


5.	Implementation Sub-group (by Geoff Morrison)

The Implementation Studies Group met on the 21st and 22nd of August 1991.  
Discussions centered on three areas.

1)      Verification of CD11172 by a decoder implementation operating in real 
time.  A separate report on this has been prepared.

2)      Document MPEG91/160 proposed that the list of requirements for MPEG-
2 include the use of parallel structures for the video algorithm decoding 
operations which map on to parallel hardware implementations.  Although the 
meeting recognised that parallel processing can be useful especially at 
higher coded data and computation rates, there were also thought to be 
disadvantages such that parallel structured algorithms should not be 
especially favoured.  For example, a parallel implementation generally 
requires a larger silicon area and hence higher cost.

It was agreed that flexibility of implementation approaches was important 
and algorithm designers should endeavor to allow this.  A recommendation 
was drafted to this effect.

3)      The major part of the meeting was spent on finalising the procedure 
to be used for addressing the complexity of algorithms to be proposed for 
MPEG-2 and evaluated in November 1991.  The basic principle of rank 
ordering had been established at the May 1991 meeting.  Document MPEG91/113 
proposed a methodology which was favourably reviewed.  In particular, the 
use of standard deviation to flag inconsistencies between reviewers was 
agreed.

A suggestion at the previous meeting had been to provide a list of 
parameters etc for proposers to provide information about.  This was 
prepared for inclusion in the updated Proposal Package Description.

The meeting was of the opinion that encoders should also be assessed.  The 
reasoning was that the cost and complexity of encoder will be important in 
some applications and it would therefore be unwise for MPEG to proceed 
after the November subjective picture quality assessment exercise without 
relevant information.  It was agreed to rank order the encoder and decoder 
complexity separately and that less precision in the encoder assessment 
would be acceptable.

It was pointed out that manual "tuning" of coding parameters for each test 
sequence, individual pictures within sequences and even individual blocks 
within pictures was possible under MPEG-2 proposal rules.  While the 
Implementation Group did not wish to limit the coded picture quality 
presented by proposers there were obvious concerns about the practicality 
of achieving such results under real conditions.  Accordingly a 
recommendation was drafted requiring proposers to state the extent and 
nature of such hand crafting so that an assessment can be made of the 
possibility and cost of its eventual automation.

The reason for performing the complexity evaluation is to bring the 
implementation cost considerations into the MPEG-2 decision making process.  
However costs also depend on market size and hence general applicability of 
the chosen algorithm.  A recommendation was prepared stating that all the 
requirements listed in the PPD be considered along with the picture quality 
and complexity results when selecting paths for further study.

The expected number of proposals for assessment at the November meeting 
raised several questions of logistics.  It was agreed that the first 2 or 3 
days of the November meeting be spent on a series of 30 minute interview 
sessions with proposers.  These would consist of a 15 minute presentation 
followed by 15 minutes of a question and answer session.  To aid the 
assessors in obtaining an understanding of the proposals by the first few 
days of the November meeting it was agreed to request submission of all 
complexity assessment documentation from proposers by 1 November 1991 
directly to all assessors (expected to number 12 to 15).  The remaining 
timetable for the complexity evaluation exercise would be as follows:

    -	By day 5 of November meeting - assessors make initial         
assessments and consult to check that the procedure is satisfactory, 
but these results not published.

    -	By 17 Dec 91 - Assessors finish individual rankings.

    -	On the first day of next meeting after November (ie 8 Jan 92 in 
Singapore) - Assessors meet to complete the evaluation process and 
issue the results.


END