&%PAGE&
-  &%PAGE&  -
Annex 1 to Doc. AVC-355R
Annex 1 to AVC-355R

Documents for the Tarrytwon Meeting 
(Sep. 28 - Oct. 1, 1992)


Normal Documents

[ ] indicates MPEG92/???.

AVC-317R         Report of the seventh meeting in New Jersey and Rio de
                 Janeiro - July 1992, Part 1 (Chairman)
AVC-318R         Report of the seventh meeting in New Jersey and Rio de
                 Janeiro - July 1992, Part 2 (Chairman)

These two documents report the outcome of the previous meeting in New Jersey 
and Rio de Janeiro.

AVC-319  to be considered in Ipswich
AVC-320  to be considered in Ipswich
AVC-321  to be considered in Ipswich  
AVC-322  to be considered in Ipswich

AVC-323 [N0245]  Test Model 2 (Test Model Editing Committee)
AVC-326 [413]    TM2 Erratum (Test Model Editing Committee)

These two documents give specifications for the Test Model algorithm and all 
agreed experiments.

AVC-324 [229rev] Information on requirements for MPEG-2 video 
                 (Requirements Group)
AVC-325 [230rev] Guide for the video work, Rio revision 
                 (Requirements Group)

These two documents are the outcome obtained in Rio de Janeiro regarding video 
requirements.

AVC-327 [431]    Results of low delay core experiment (Japan)

Low delay core experiments are carried out with TM1, mainly adopting FAMC as 
prediction mode and Frame M=1 as image structure.  And the first picture is 
treated as an intra picture.  Through these experiments, it has been confirmed 
that the techniques of forced intra slice and picture skipping are useful in 
terms of delay time.

AVC-328 [469]    Scene change handling without picture skipping in low 
                 delay mode (Japan)

In order to reduce buffer delay, forced intra slices is a very efficient 
technique.  When scene change takes place, however, temporal or spatial 
picture quality of the first picture in the new scene is degraded.  Picture 
quality degradation of the coding 'without picture skipping' is considered.  
The simulation based on TM1 shows that bit allocation for a I-picture 
influences the picture quality of the I-picture and following P-pictures.  
However, the influence fades out in about 30ms (10frames).  In the worst 
cases, visual degradation is detected.  It is recommended in the low delay 
coding system without picture skipping that target bits for I-picture should 
be larger than twice of target bits for P-pictures.

AVC-329 [455]    Intra slice/column and leaky prediction (Japan)

Intra Slice and Intra Column methods are compared, and only a minor difference 
was found.  Leaky prediction method is also studied, and SNR degradation curve 
for various leaky factor values is obtained using TM2.  From the view point of 
SNR, LF=1-1/16 is concluded as the minimum for sufficient picture quality.

AVC-330 [470]    Buffering for low delay mode (Japan)

This contribution analyzes the buffering problem for the low delay mode (M=1 
frame coding and M=2 field coding) where occasional picture dropping takes 
place to cope with scene changes or other forced update situations.  An 
extension of the existing VBV specifications is proposed to keep the steady 
state low delay operation by momentarily sacrificing temporal reproduction of 
the coded pictures.  It is also proposed that the headers of dropped pictures 
are sent with a flag indicating picture dropping and that Temporal Reference 
(TR) is numbered according to the source picture.

AVC-331 [471]    Coding efficiency of leaky prediction (Japan)

Coding efficiency of the leaky prediction as defined in Core experiment No.6 
of TM2 has been measured by an open loop experiment with fixed step sizes.  
Data for the information generation and macroblock type distribution are 
provided against the leak factor.  Coding loss due to leaky prediction is 
competitive with the use of periodic intra pictures.  Coded pictures are also 
observed.  It is concluded that if we can find a means to remove artifacts 
peculiar to leaky prediction, such as busy or blocky backgrounds, it will 
contribute to making the coding algorithm simple and robust, hence generic.

AVC-332 [472]    Cell-loss compensation scheme (Japan)

A cell-loss compensation scheme, which consists of leaky prediction, 
structured packing and concealment at the decoder, was introduced at the  
previous meetings.  In this paper, the effects of each element, especially 
leaky prediction, are evaluated in detail through computer simulation for TM2 
IP mode and low delay mode.  The results indicate that three elements work 
cooperatively and are effective for cell loss.  It is proposed that the 
standard should provide both the spatial and temporal domain compensation 
schemes.  It is also pointed out that the 8 bit arithmetic accuracy for the 
leaky prediction can produce residual error of +/- (2**n-1) at maximum for 
LF=1-1/2**n.

AVC-333 [473]    Experiments on cell loss resilience (Japan)

This document discusses two techniques which should be considered to realize 
cell loss resilience, i.e.
   1) Spatial localization of lost macro block
   2) Temporal localization of degraded image
For temporal localization, intra picture, intra slice and leaky prediction 
have been compared.  Leaky prediction gives the most favorable impression for 
the recovery of degraded image, however we have to improve the image quality 
at the stationary state.

AVC-334 [458]    Simulation results of compatibility core experiment 
                 (Japan)

In this document, prediction from the base layer mode and prediction for the 
prediction error mode were compared again within the TM2 framework.  The same 
tendency as that of the previous document (MPEG 92/257) was obtained.  As a 
conclusion, this document describes that the prediction from the base layer 
with a switchable compatible type for each field is better than the prediction 
for the prediction error as a candidate for the compatible mode from the 
coding efficiency point of view.

AVC-335 [474]    Proposal of TM2 rate control modification for low delay 
                 mode (Japan)

The current TM2 rate control for the low delay mode makes the number of bits 
per picture decreases as time passes by, thus SNR is degraded toward the end 
of sequence.  This document provides two alternative solutions for this 
problem based on experiments.

AVC-336 [475]    Picture header modification for source clock recovery 
                 (Japan)

Modification of picture header to include a new field for video source clock 
recovery is proposed.  Source clock frequency measured with a network clock is 
transmitted using this field.

AVC-337 [432]    Simulation results on prediction and DCT mode coupling for
                 S-FAMC (Japan)

This document addresses the S-FAMC prediction and DCT mode coupling for inter 
macro block, which is related to core experiment NO.8 on prediction mode.  S-
FAMC with frame DCT and with field/frame DCT have been compared and it is 
concluded that S-FAMC needs the field/frame DCT for rapid motion sequences.

AVC-338 [433]    Simulation results on S-FAMC and Dual' (Japan)

This document addresses Simplified FAMC (S-FAMC) and Dual' prediction, which 
are core experiment NO.1 and No.3 on prediction modes.  S-FAMC defined in TM2 
has no interpolation mode for B-picture.  However this exclusion of 
interpolation suffers from loss of coding efficiency.  This document proposes 
the introduction of frame base averaged macro block as interpolation mode of 
S-FAMC.  This introduction makes the coding efficiency high as almost same as 
original FAMC with keeping memory band width same as field or frame 
prediction.  The field/dual' simulation at 4Mbit/s is also carried out and it 
is concluded that dual' improves the coding efficiency slightly from 
field/frame prediction but less than S-FAMC.

AVC-339 [456]    Simulation results on prediction core experiment (Dual-
                 prime, SFAMC) in TM2 (Toshiba)

Performance of the dual prime prediction was evaluated comparing with that of 
SFAMC.  Simulations results shows that both DUAL' and SFAMC similarly improve 
coding efficiency if they are combined with Frame and or Field.  As a 
conclusion, the dual prime prediction mode is proposed for its high prediction 
efficiency and its hardware implementation simplicity.

AVC-340 [454]    Results of low delay core experiments on TM2 - Among
                 prediction modes (KDD)

Several prediction schemes are simulated on the low delay mode with M=1, N=15, 
concluding that;
   1) SFAMC with Frame or Field MC gives about 1dB gain in FG, however the 
gain is limited to about 0.2dB in other sequences.
   2) SVMC is effective for both the slow and fast movement and is the best 
among tested prediction modes in most of all sequences even if its motion 
calculation load is about 60% of that of Frame/Field MC.
   3) DUAL' with Frame and/or Field MC gives better performance than SFAMC 
with Frame or Field MC in most of all sequences, but less than SVMC especially 
in BC and CL sequences. 

AVC-341 [430]    Results on the comparison of error prediction versus
                 reconstructed signal prediction (PTT Research)

Based on the experimental results showing that there is a small difference 
(o.o to 0.2 dB) in favor of the prediction from the base layer, it is 
concluded that the selection should be made according to complexity or 
functionality.

AVC-342   -      Short term PSTN videotelephone standardisation 
                 (UK, France, Italy, Denmark, FRG, Belgium, The
                  Netherlands, Sweden)

Several suggestions are mode toward developing standards, which complement the 
existing ISDN ones, in a rather short period of time (draft proposal by summer 
1993).
   
AVC-343 [488]    Results of Core Experiments I.4 - Scalable Side 
                 Information (Aus. UVC consortium)

This document discusses results obtained within the framework of Core 
Experiment "Scalable Side Information" whose purpose is to investigate the 
possibility of reducing the side information on the lowest layer scale_2 in 
scalable coding schemes.  Four coding architectures were experimented in 
comparison to a single layer TM2 implementation.  Results show that a target 
bit rate of 750 kbit/s at the lowest layer is realizable using scalable side 
information and appropriate scaling of quantization between the layers.

AVC-344 [489]    A Frequency Pyramid Architecture with Improved Coding
                 Efficiency (Aus. UVC consortium)

A new coding structure is introduced to improve the coding efficiency and to 
solve the drift problem in frequency scalability schemes.  This structure 
makes use of the prediction from the lower scales by using "path_Y".  
Experimental results show that this structure improves the coding efficiency 
of the scalable coder.  The following is proposed;
   - this encoder/decoder structure be investigated as a follow-up 
     experiment for Core Experiment I.2 and I.3,
   - different bit rate distributions be investigated,
   - an 8 bit word be transmitted in the sequence header to indicate the 
     structure of the decoder which is necessary in decoding the bitstream.
 
AVC-345 [490]    Considerations on ATM cell loss experiments 
                 (Aus. UVC consortium) 

The comparison of the cell loss resilient properties of a spatial layered 
coded to those of a single layered codec requires careful consideration if a 
fair comparison is to be made.  This document raises important issues and 
provides Table 1, listing parameters and their values, for spatial layered 
codec cell loss resilience experiment.

AVC-346 [491]    Results of Core Experiment I.1 - Interlace-in-Interlace
                 extraction (Aus. UVC consortium)

This documents reports on the results of Core Experiment for the interlace-in-
interlace extraction.  Comparisons are made with a simple upper 4x4 extraction 
technique both in terms of PSNR and bit rate in each layer.  In addition, the 
effect of using simple extraction on images coded in field mode and adaptive 
field/frame mode are presented.  It is concluded that the Method A extraction 
technique provides the highest quality reconstructed scale-4 image sequence.  
It is pointed out that any method of extraction results in a very high rate 
but only fair quality for the scale-4 service, suggesting the use of a coder 
and decoder with more than one coding loop.

AVC-347 [492]    Result of the compatibility experiment 1c : H.261 SIF 
                 based prediction of Prediction Error vs Prediction of 
                 Input in Frame structure pictures (CNET-FRANCE TELECOM) 

The base layer is coded by H.261 (RM8).  The results show that the compatible 
coding using the prediction of input outperforms the simulcast.

AVC-348 [461]    Clarification of Appendix H in TM2 (T. Yukitake)

Handling the first picture for "intra picture" case and skipped pictures is 
proposed to clarify the low delay mode experiment. 

AVC-349 [493]    Leaky prediction: Eliminating the limit cycle (AT&T)

The limit cycle problem of the leaky prediction which is caused by finite 
precision of the digital filter is described.  The following two solutions to 
eliminate the limit cycle are provided;

    - error spectrum shaping which generates the auxiliary signal using the 
      n least significant bits from the previous truncated signal, 
    - using a pseudo-random auxiliary signal

Based on experimental results showing faster convergence, the second solution 
is recommended.
 
AVC-350 [494]    Leaky prediction: Experimental results (AT&T)

Leaky prediction was experimented using the second solution described in AVC-
349.  For five sequences, values of the leak factor which give equivalent 
quality compared to the case using cyclic I pictures are obtained as 7/8 - 
31/32.  Then, cell losses are applied to conclude that the leaky prediction 
gives less visually objectionable pictures.

AVC-351 [484]    Progress Report for Ad-hoc Group on ATM, Packet Loss
                 and General Error Resilience (Chair, MPEG Ad-hoc
                 Group on ATM, Packet Loss and General Error)  

An overview is given for the current status of the work, listing various 
approaches and action points .

AVC-352 [495]    TM2 compatibility experiments (BT)

A two-layered coding scheme is experimented where coded SIF base layer 
pictures are upsampled and used as a prediction for the CCIR601 layer.  A new 
prediction using the average of upsampled SIF and locally decoded CCIR601 is 
also experimented.  It is concluded that the two layer coding scheme can 
improve the SNR picture quality over the simulcast approach, and that the new 
compatibility prediction mode gives subjectively better pictures and warrants 
further study.

AVC-353 [503]    Core Experiment I.8 results (BELGACOM, UCL)

The aim of Core Experiment I.8 is to compare the efficiency of the
frequency scanning method versus the block scanning one.  This document
collects the first results obtained.  It is concluded that;
   - at this stage of experimentation, both techniques present similar 
     efficiency,
   - the higher the bit rate, the higher the efficiency of frequency 
     scanning technique,
   - further studies are required to adapt frequency scanning technique to 
     the context of TM2.

AVC-354 [498]    Simulation results of the basic TM2 (CNET)

This document presents TM2 simulation results for four sequences with 
statistical data and processed pictures.


Temporary Documents

TD-1  Agenda for the eighth meeting in Tarrytown (Chairman)
TD-2  Available documents (Chairman)
TD-3  Review of the Experts Group work plan (Chairman)
TD-4  Draft status report - Issue 3 (Chairman) 
TD-5  Coding structure vs delay in terms of SNR (T. Yukitake)


END