Re: [codec] #28: Layered bit-stream

["codec issue tracker" <[e-mail]>]

#28: Layered bit-stream
------------------------------------+---------------------------------------
 Reporter:  hoene@…                 |       Owner:     
     Type:  defect                  |      Status:  new
 Priority:  minor                   |   Milestone:     
Component:  requirements            |     Version:     
 Severity:  Active WG Document      |    Keywords:     
------------------------------------+---------------------------------------

Comment(by hoene@…):

 [Vladimir]:
 We'd like to express SPIRIT's point of view on the topic. It is as
 follows:
 1/ we support layered coding in our IPMR-codec. Such approach is used by
 ITU-T Recs G.711.1 and G.718, for example.
 2/ we think that VoIP and Videoconferencing systems are users of such
 codecs.
 3/ we consider that it is not needed to drop scalability and layered
 coding.

 [Christian]:
 As far as I understand, layered coding helps if multiple streams having
 the sample content but different rates must be generated.
 For example, if a conferencing system stream the same audio stream to N
 users but each users has a different bandwidth. Just encode all layers and
 drop the higher layers for the low bandwidth users. This approach is easy
 and efficient and reduces the encoding complexity.

 The arguments against are simple.
 a) First, this use case is a local optimization only. Thus, the must not
 be standardized.
 b) Second, instead of layered coding one can use other ways of tweaking
 the implementation performance. For example, if you calculate a 512 FFT do
 get two 256 FFTs for free. I bet there are thousand other shortcuts which
 I am not aware of.

 Thus, I have the opinion that layered coding is not worth the extra
 bandwidth of 20 or more percentage. It is just good locally but not needed
 for interoperability.

 [Dmitry]:
 From application point of view, the layered stream structure allows server
 manipulate channel bandwidth individually for each user with zero
 performance overhead. Obviously, conferencing is the most important use-
 case.
 […]
     Shall layered coding be supported? - we think "yes", because ... (see
 my first sentence)
     Who needs it?                      - answered
     Can we drop this requirement?      - only if we have real good reasons
 for it. Do we have them?

 [Christian]:
 Because […] an application programmer want[s] to have an easy life, the
 codec designer shall develop a more complicated codec? In addition,
 everybody should suffer from a higher bit rate? No, that is not fair.
 > Obviously, conferencing is the most important use-case.
 No, end-to-end connections are more frequent than conference calls.

 > "local optimization" of what?
 I mean that the layered coding is only used within one computer. It is not
 important in-between computers. And, it is only a performance optimization
 that makes the conference gateway faster.

 [Stephen]:
 A couple of observations on this
 (a) I agree with Christian that layered codecs usually need higher
 bitrates to achieve the same quality. Of course we do want good quality
 over our desired bitrate range, and it is likely to be more difficult to
 achieve that with a layered codec.

 (b) I agree with Dimity that layered codecs reduce the complexity of VOIP
 gateways and perhaps conference bridges.  Also, with layered codec designs
 you get wire-speed management of channel bandwidth, so there can be a
 delay benefit as well as a complexity reduction.

 (c) Arguing about the relative priority of multipoint conferences vs point
 to point calls is pointless, because they are clearly both MUSTS.

 I am not sure if Christian is arguing that layered codecs SHALL NOT be
 considered, or if the requirements allow but are not biased towards
 layered proposals.  If might be useful to clarify this point.

 [Christian]:
 Let me explain my arguments a bit more by discussion point b.

 (b) I agree with Dimity that layered codecs reduce the complexity of VOIP
 gateways and perhaps conference bridges.
 I see that there is an important need to reduce the computational
 complexity at gateways and conference bridges. Layered coding is an easy
 solution that helps to reduce complexity. However, I think that there are
 many other ways to achieve similar complexity reduction WITHOUT requiring
 layered coding. One example is a special encoder that encodes multiple
 streams at the same time.

 Also, with layered codec designs you get wire-speed management of channel
 bandwidth, so there can be a delay benefit as well as a complexity
 reduction.

 Also, a fast management of channel bandwidth can be achieved by
 controlling the encoder (via a feedback channel not locally)

 Thus, I would vote for a NEED NOT or even MUST NOT because the costs of
 layered encoding are imposed to all users of the codec but the benefit is
 only to the gateways.

 [Dimitry]:
 Definitely, there are number of applications that can (more or less
 efficiently) utilize layered stream structure. At the same time,
 scalability makes codec more complex and cause bitrate penalty due to a
 simple fact that entropy(a+b)<=entropy(a)+entropy(b).
 The only question is: what penalty is acceptable and what is not? Based on
 IP-MR codec experience I consider the price for scalability as not too
 high.

 > I mean that the layered coding is only used within one computer.
 > It is not important in-between computers. And, it is only a
 > performance optimization that make the conference gateway faster.
 This is correct. No other benefits except performance saving.

 [Dimitry]:
 […]
 But the thing is that this is not a point we should stand on. Let’s stay
 on a point of problems and solutions.
 The problem is: gateways have to recompress data to fit user bandwidth and
 it would be very desired to decrease recompression cost:
                     Does this problem live only in my mind? - any
 considerations are welcome.
 Possible solution :
                     The layered stream structure allows zero cost
 recompression (but increases bitrate/quality ratio).
 All other solutions are also welcome.

-- 
Ticket URL: <http://trac.tools.ietf.org/wg/codec/trac/ticket/28#comment:1>
codec <http://tools.ietf.org/codec/>