Re: [AVTCORE] Benjamin Kaduk's Discuss on draft-ietf-payload-tsvcis-03: (with DISCUSS and COMMENT)

["Roni Even (A)" <[e-mail]>]

Hi,
About the reference to TSVCIS.
The RTP payload is about how to encapsulate the payload in an RTP packet. The objective is to define how an RTP stack can insert the tsvcis frames and  extract the tsvcis frames from the RTP packet. Typically it is not required to understand the payload structure in order to be able to perform the encapsulation.
This is why the reference to the payload is Informational and we did not require to have it publically available.  If there is a need to understand the payload itself for the encapsulating than we need more information in the RTP payload specification and a publically available normative reference. I think this is not the case here

Roni Even 

AVTCore co-chair (ex Payload)

-----Original Message-----
From: [e-mail] [mailto:[e-mail]] 
Sent: Saturday, October 05, 2019 12:18 AM
To: 'Benjamin Kaduk'; 'The IESG'
Cc: [e-mail]; 'Ali Begen'; [e-mail]; [e-mail]; 'Victor Demjanenko, Ph.D.'; 'Dave Satterlee (Vocal)'
Subject: RE: Benjamin Kaduk's Discuss on draft-ietf-payload-tsvcis-03: (with DISCUSS and COMMENT)

Everyone,

Thanks for the comments.  I think I mis-understood the ambiguity with respect to to changing rates within a RTP packet.  That was not plan.  An RTP packet must have MELP speech frames of the same rate.  What is possible is that the amount of augmented TSVCIS speech data may vary from one speech frame to the next.  This allows for a dynamic VDR as suggested by the NRL paper.  So an RTP packet may have varying TSVCIS data but must always have MELPe 2400 data.

Again backwards parsing is necessary but the timestamp uniformly increments 22.5msec per combined MELP/TSVCIS speech frame.

The NRL is a good public reference on the VDR aspects.  The actual TSVCIS spec we had was FOUO so we could not replicate its detail.  (I believe a later spec is public or at least partially public.  I am trying to get this.)  The opaque data is pretty obvious with the TSVCIS spec in hand.

We will address the issues/concerns raised next week.  Other business had priority.

Thank you and enjoy the weekend.

Regards,

Victor & Dave

-----Original Message-----
From: Benjamin Kaduk via Datatracker <[e-mail]> 
Sent: Wednesday, October 2, 2019 10:40 PM
To: The IESG <[e-mail]>
Cc: [e-mail]; Ali Begen <[e-mail]>; [e-mail]; [e-mail]; [e-mail]
Subject: Benjamin Kaduk's Discuss on draft-ietf-payload-tsvcis-03: (with DISCUSS and COMMENT)

Benjamin Kaduk has entered the following ballot position for
draft-ietf-payload-tsvcis-03: Discuss

When responding, please keep the subject line intact and reply to all email addresses included in the To and CC lines. (Feel free to cut this introductory paragraph, however.)


Please refer to https://www.ietf.org/iesg/statement/discuss-criteria.html
for more information about IESG DISCUSS and COMMENT positions.


The document, along with other ballot positions, can be found here:
https://datatracker.ietf.org/doc/draft-ietf-payload-tsvcis/



----------------------------------------------------------------------
DISCUSS:
----------------------------------------------------------------------

I support Magnus' point about the time-ordering of adjacent frames in a packet.

Additionally, I am not sure that there's quite enough here to be interoperably implementable.  Specifically, we seem to be lacking a description of how an encoder or decoder knows which TSVCIS parameters, and in what order, to byte-pack or unpack, respectively.  One might surmise that there is a canonical listing in [TSVCIS], but this document does not say that, and furthermore [TSVCIS] is only listed as an informative reference.  (I couldn't get my hands on my copy, at least on short notice.)  If we limited ourselves to treating the TSVCIS parameters as an entirely opaque blob (codec, convey these N octets to the peer with the appropriate one- or two-byte trailer for payload type identification and framing), that would be interoperably implementable, since the black-box bits are up to some other codec to interpret.

In a similar vein, we mention but do not completely specify the potential for using CODB as an end-to-end framing bit, in Section 3.1 (see Comment), which is not interoperably implementable without further details.


----------------------------------------------------------------------
COMMENT:
----------------------------------------------------------------------

Where is [TSVCIS] available?

Is [NRLVDR] the same as
https://apps.dtic.mil/dtic/tr/fulltext/u2/a588068.pdf ?  A URL in the references would be helpful.

Is additional TSVCIS data only present after 2400bps MELPe and the first thing to get dropped under bandwidth pressure?  The abstract and introduction imply this by calling out MELPe 2400 bps speech parameters explicitly, but Section 3 says that TSVCIS augments standard 600, 1200, and 2400 bps MELP frames.

It's helpful that Section 3.3 gives some general guidance for decoding this payload type ("[t]he way to determine the number of TSVCIS/MELPe frames is to identify each frame type and length"), but I think some generic considerations would be very helpful to the reader much earlier, along the lines of "MELPe and TSVCIS data payloads are decoded from the end, using the CODA and CODB (and, if necessary, CODC and others) bits to determine the type of payload.  For MELPe payloads the type also indicates the payload length, whereas for TSVCIS data an additional length field is present, in one of two possible formats.  A TSVCIS coder frame consists of a MELPe data payload followed by zero or one TSVCIS data payload; after the TSVCIS payload's presence/length is determined, then the preceding MELPe payload can be determined and decoded.  Per Section 3.3, multiple TSVCIS frames can be present in a single RTP packet."  This (or something like it) would also serve to clarify the role of the COD* bits, which is otherwise only implicitly introduced.

Section 1.1

RFC 2736 is BCP 36 (but it's updated by RFC 8088 which is for some reason an Informational document and not part of BCP 36?!).

Section 2

   In addition to the augmented speech data, the TSVCIS specification
   identifies which speech coder and framing bits are to be encrypted,
   and how they are protected by forward error correction (FEC)
   techniques (using block codes).  At the RTP transport layer, only the
   speech coder related bits need to be considered and are conveyed in
   unencrypted form.  In most IP-based network deployments, standard

Am I reading this correctly that this text is just summarizing what's in the TSVCIS spec in terms of what needs to be in unencrypted form, so the "only the speech coder related bits[...]" is not new information from this document?  I'm not sure I agree with the conclusion, regardless -- won't the (MELPe) speech coder bits be enough to convey the semantic content of the audio stream, something that one might desire to keep confidential?

   link encryption methods (SRTP, VPNs, FIPS 140 link encryptors or Type
   1 Ethernet encryptors) would be used to secure the RTP speech
   contents.  Further, it is desirable to support the highest voice
   quality between endpoints which is only possible without the overhead
   of FEC.

I think I'm missing a step in how this conclusion was reached.

   TSVCIS will be characterized.  Depending on the bandwidth available
   (and FEC requirements), a varying number of TSVCIS specific speech
   coder parameters need to be transported.  These are first byte-packed
   and then conveyed from encoder to decoder.

Per the Discuss point, how do I know which parameters need to be transported, and in what order?

   Byte packing of TSVCIS speech data into packed parameters is
   processed as per the following example:

      Three-bit field: bits A, B, and C (A is MSB, C is LSB)
      Five-bit field: bits D, E, F, G, and H (D is MSB, H is LSB)

           MSB                                              LSB
            0      1      2      3      4      5      6      7
        +------+------+------+------+------+------+------+------+
        |   H  |   G  |   F  |   E  |   D  |   C  |   B  |   A  |
        +------+------+------+------+------+------+------+------+

   This packing method places the three-bit field "first" in the lowest
   bits followed by the next five-bit field.  Parameters may be split
   between octets with the most significant bits in the earlier octet.
   Any unfilled bits in the last octet MUST be filled with zero.

I agree with Adam that this is very unclear.  A is the MSB of the three-bit field but the LSB of the octet overall?
We probably need an example of splitting a parameter across octets as well, to get the bit ordering right.

Section 3.1

   It should be noted that CODB for both the 2400 and 600 bps modes MAY
   deviate from the values in Table 1 when bit 55 is used as an end-to-
   end framing bit.  Frame decoding would remain distinct as CODA being

Where is the use of CODB as an end-to-end framing bit defined?  If we're going to provide neither a complete description of how to do it nor a reference to a better description, we probably shouldn't mention it at all.

Section 3.2

   RTP packet.  The packed parameters are counted in octets (TC).  In
   the preferred placement, shown in Figure 6, a single trailing octet
   SHALL be appended to include a two-bit rate code, CODA and CODB,

I'd consider saying something about this being the preferred format
("placement") due to its shorter length than the alternative, and say that it "SHOULD be used for TSVCIS payloads with TC less than or equal to 77 octetes".

Section 3.3

When a longer packetization interval is used, is that indicated by signaling or RTP timestamps or otherwise?

   TSVCIS coder frames in a single RTP packet MAY be of different coder
   bitrates.  With the exception for the variable length TSVCIS
   parameter frames, the coder rate bits in the trailing byte identify
   the contents and length as per Table 1.

Maybe also note that the penultimate octet gives the length there?

   Information describing the number of frames contained in an RTP
   packet is not transmitted as part of the RTP payload.  The way to
   determine the number of TSVCIS/MELPe frames is to identify each frame
   type and length thereby counting the total number of octets within
   the RTP packet.

terminology nit: if a frame is the combination of MELPe and TSVCIS payload data units then there are two layres of decoding to get a length for the frame, since we have to get the TSVCIS length and then the MELPe length.

Section 4.2

   Parameter "ptime" cannot be used for the purpose of specifying the

nit: missing article ("The parameter")

   will be impossible to distinguish which mode is about to be used
   (e.g., when ptime=68, it would be impossible to distinguish if the
   packet is carrying one frame of 67.5 ms or three frames of 22.5 ms).

So how is the operating mode determined, then?
(I think this is the same question I asked above)

Section 4.4

   For example, if offerer bitrates are "2400,600" and answer bitrates
   are "600,2400", the initial bitrate is 600.  If other bitrates are
   provided by the answerer, any common bitrate between the offer and
   answer MAY be used at any time in the future.  Activation of these
   other common bitrates is beyond the scope of this document.

It seems important to specify whether this requires a new O/A exchange or can be done "spontaneously" by just encoding different frame types.
(It seems like the latter is possible, on first glance, and this is implied by Section 3.3's discussion of mixing them in a single packet.)

Section 5

Please expand PLC at first use (not second).

Section 6

I don't understand the PLC usage.  Is the idea that a receiver, on seeing an SSRC gap, constructs fictitious PLC frames to "fill the gap"
and passes the resulting stream to the decoder?

Section 8

   and important considerations in [RFC7201].  Applications SHOULD use
   one or more appropriate strong security mechanisms.  The rest of this
   section discusses the security-impacting properties of the payload
   format itself.

I thought we described TSVCIS itself (much earlier in the document) as requiring encryption for some data; wouldn't that translate to a "MUST"
here and not a "SHOULD"?