&%PAGE&

-  &%page&  -
Annex 4 to Doc. AVC-65R
Annex 4 to Doc. AVC-65R


COMPARISON OF VBR/CBR CODING
FRAMEWORK FOR FURTHER STUDY
                               
1 - INTRODUCTION

During the meeting, the relevant contributions (TD No.1, ��4.4) were scanned 
for all issues related to VBR/CBR coding and its comparison.  These issues 
are listed in this document in a more or less ordered structure.  Some 
issues were analysed into more detail, other issues not.  Some suggestions 
are given to help compare VBR/CBR comparison results with minimal 
ambiguity.

The document also suggests issues for further study, an order of priority 
could be added.  Due to the limited time budget the document is not 
complete.  Refinements are necessary, some issues may be missing.  Probably 
the document can be used as a living document, that can be updated at 
future meetings.

2 - PRESENTATION OF VBR/CBR COMPARISON RESULTS

Up to now, multiple contributions were received comparing CBR vs VBR.  
Comparison of these results are difficult because:

    1)	A lot of variables are changed at the same time (e.g. different bit 
rate, different picture quality, ...).

    2)	Different input sequences are used.  Therefore it is suggested, in 
order to be able to compare the results, and to come to a consensus, 
that only one variable is changed at a time, or at least as less 
variables as possible.

It is also felt beneficial to define a video sequence that can be used to  
compare statistical results (simulations, measurements).

	2.1. Variables

Following variables were identified, influencing the VBR/CBR comparison 
results:

    1)	Buffer size (delay)
    2)	Picture quality (e.g. subjective, SNR)
    3)	Bit rate (related to network model)
    4)	Codec complexity (related to coder efficiency, ...).

These variables are dependent, there are less than 4 degrees of freedom.  
Other variables are searched for.  It is suggested that for further study, 3 
variables are frozen, in order to determine the relationship between the 
SMG and the 4-th variable.  (e.g. to compare a CBR and VBR codec with 
respect to bit rate, where both coders have the same buffer size, subjective  
picture quality and coding algorithm, in order to determine the relationship 
between bit rate and SMG).

In the next figure, these variables are mapped into Fig. AVC-38/1.



























	2.2. Reference sequence(s)

It is not the intention to create a sequence for study of coding algorithms 
(typical critical sequence), as a lot already exist.  The idea is to provide  
a relatively long sequence for statistical evaluations.  First of all, the 
requirements for such a sequence should be defined:

    -	length
    -	burstiness (+ definition how to measure it)
    -	interactive services (highest priority)
    -	listening and active speakers
    -	pointing to graphics; panning; switching to document, between persons;
     	...
    -	noisy camera, good lighting, ...
    -	input auto-focus, auto-iris, ...
    -	single sequence or multiple, number?
    -	suitability for VBR coding
    -	format
    -	format for physical distribution (e.g. D1).

It is proposed to use the CIF format as it is felt that the format will have  
little impact on VBR/CBR comparison results and as it solves the 30 Hz/25 Hz 
conversion problem.

For the generation of the sequence following procedure is proposed:

    1)	Agreement on contents
    2)	Shooting of sequence on D1, 50 or 60 Hz
    3)	Conversion to CIF (HW, SW)
    4)	Conversion back to D1, 50 & 60 Hz
    5)	Distribution

Volunteers are searched to take up one (or more) of these tasks.

3 - STUDY OF ISSUES IMPACTING ON SMG RESULTS

Following issues were identified that have an impact on SMG results.  It is 
proposed to invite contributions on these topics.

	3.1. Use of priority bit and layered coding.  Impact on SMG, and on the 
choice of VBR or CBR.

	3.2. Will the B-ISDN network be able of offering different QOS's, what 
is the possible impact on SMG's?

	3.3. Relation between SMG and CLR

           -	impact on network model
           -	impact on picture quality
           -	blocking probability vs average cell loss
           -	impact of CLR behaviour (evenly spread or bursty) on experience 
by individual source, SMG and service degradation

	3.4. Correlation.  Different sources of correlation were identified:

           -	source periodicity
		 * frame basis
		 * MPEG intrafield 1:15
           -	multimedia?
           -	video sequence contents?

	3.5. Control of VBR.  Multiple contributions emphasized the need for a 
controlled behaviour of a VBR coder.  Following study issues are identified:

	    -	averaging windows
	    	 * size
	    	 * jumping vs sliding (or triggered jumping)
	    	 * network & coder desires (requirements, where to meet?)
	    -	network constraints, UPC, ...
	    -	control strategies within the encoder: credit building, ...

	3.6. Network model
        
	    -	comparison of model with measured data (simulation, HW 
experiments)
	    -	use of single source model
	       * generate synthetic bit rate files
	       * use for statistical multiplexing experiments
	    -	refinement of the model
	    -	...

4 - STUDY OF OTHER ISSUES, RELATED TO VBR/CBR COMPARISON

Following issues are not directly related to SMG, but may impact the 
selection of VBR/CBR.

	4.1. Clock recovery

	    -	does VBR requires dedicated clock recovery schemes between 
encoder and decoder?  What is the impact on complexity and 
performance?

	    -	do layered coding schemes require special clock recovery 
schemes?  Impact on complexity and performance?

	4.2. Can any VBR coder act as a CBR coder?

	    -	relation with buffer length
	    -	symmetry?
	    -	CBR as a special case of VBR:
		 * go directly for VBR, CBR as fall back solution?
	    -	...


END