Network Working Group | M. Petit-Huguenin |
Internet-Draft | Impedance Mismatch |
Updates: 3550 (if approved) | G. Zorn, Ed. |
Intended status: Standards Track | Network Zen |
Expires: May 27, 2014 | November 23, 2013 |
Support for Multiple Clock Rates in an RTP Session
draft-ietf-avtext-multiple-clock-rates-11
This document clarifies the RTP specification when different clock rates are used in an RTP session. It also provides guidance on how to interoperate with legacy RTP implementations that use multiple clock rates. It updates RFC 3550.
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at http://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."
This Internet-Draft will expire on May 27, 2014.
Copyright (c) 2013 IETF Trust and the persons identified as the document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
The clock rate is a parameter of the payload format as identified in RTP and RTCP by the payload type value. It is often defined as being the same as the sampling rate but that is not always the case (see, for example, the G722 and MPA audio codecs [RFC3551]).
An RTP sender can switch between different payloads during the lifetime of an RTP session and because clock rates are defined by payload format, it is possible that the clock rate will also vary during an RTP session. Schulzrinne, et al. [RFC3550] lists using multiple clock rates as one of the reasons to not use different payloads on the same Synchronization Source (SSRC). Unfortunately this advice has not always been followed and some RTP implementations change the payload in the same SSRC even if the different payloads use different clock rates.
This creates three problems:
Table 1 contains a non-exhaustive list of fields in RTCP packets that uses a clock rate as unit:
Field name | RTCP packet type | Reference |
---|---|---|
RTP timestamp | SR | [RFC3550] |
Interarrival jitter | RR | [RFC3550] |
min_jitter | XR Summary Block | [RFC3611] |
max_jitter | XR Summary Block | [RFC3611] |
mean_jitter | XR Summary Block | [RFC3611] |
dev_jitter | XR Summary Block | [RFC3611] |
Interarrival jitter | IJ | [RFC5450] |
RTP timestamp | SMPTETC | [RFC5484] |
Jitter | RSI Jitter Block | [RFC5760] |
Median jitter | RSI Stats Block | [RFC5760] |
This document first tries to list in Section 3 and subsections all of the algorithms known to be used in existing RTP implementations at the time of writing. These sections are not normative.
Section 4 and subsections then recommend a unique algorithm that modifies RFC 3550. These sections are normative.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119 [RFC2119]. In addition, this document uses the following terms:
The following sections describe the various ways legacy RTP implementations behave when multiple clock rates are used. Legacy RTP refers to RFC 3550 without the modifications introduced by this document.
One way of managing multiple clock rates is to use a different SSRC for each different clock rate, as in this case there is no ambiguity on the clock rate used by fields in the RTCP packets. This method also seems to be the original intent of RTP as can be deduced from points 2 and 3 of section 5.2 of RFC 3550.
On the other hand, changing the SSRC can be a problem for some implementations designed to work only with unicast IP addresses, where having multiple SSRCs is considered a corner case. Lip synchronization can also be a problem in the interval between the beginning of the new stream and the first RTCP SR packet.
The simplest way of managing multiple clock rates is to use the same SSRC for all the payload types regardless of the clock rates.
Unfortunately there is no clear definition on how the RTP timestamp should be calculated in this case. The following subsections present the algorithms used in the field.
This method of calculating the RTP timestamp ensures that the value increases monotonically. The formula used by this method is as follows:
timestamp = previous_timestamp + (current_capture_time - previous_capture_time) * current_clock_rate
The problem with this method is that the jitter calculation on the receiving side gives an invalid result during the transition between two clock rates, as shown in Table 2 (Appendix A). The capture and arrival time are in seconds, starting at the beginning of the capture of the first packet; clock rate is in Hz; the RTP timestamp does not include the random offset; the transit, jitter, and average jitter use the clock rate as unit.
Calculating the correct transit time on the receiving side can be done by using the following formulas:
The main problem with this method, in addition to the fact that the jitter calculation described in RFC 3550 cannot be used, is that is it dependent on the previous RTP packets, packets that can be reordered or lost in the network.
An alternate way of generating the RTP timestamps is to use the following formula:
timestamp = capture_time * clock_rate
With this formula, the jitter calculation is correct but the RTP timestamp values are no longer increasing monotonically as shown in Table 3 (Appendix A). RFC 3550 states that "[t]he sampling instant MUST be derived from a clock that increments monotonically[...]" but nowhere says that the RTP timestamp must increment monotonically.
The advantage with this method is that it works with the jitter calculation described in RFC 3550, as long as the correct clock rates are used. It seems that this is what most implementations are using (based on a survey done at Sipit26 and on a survey of open source implementations, see Appendix C).
The following subsections describe behavioral recommendations for RTP senders (with and without RTCP) and RTP receivers.
An RTP Sender with RTCP turned on MUST use a different SSRC for each different clock rate. An RTCP BYE MUST be sent and a new SSRC MUST be used if the clock rate switches back to a value already seen in the RTP stream.
To accelerate lip synchronization, the next compound RTCP packet sent by the RTP sender MUST contain multiple SR packets, the first one containing the mapping for the current clock rate and the subsequent SR packet(s) containing the mapping for the other clock rates seen during the last period.
The RTP extension defined in Perkins & Schierl [RFC6051] MAY be used to accelerate the synchronization.
start_offset += (capture_time - capture_start) * previous_clock_rate capture_start = capture_time
start_offset = random_initial_offset capture_start = capture_time
timestamp = (capture_time - capture_start) * clock_rate + start_offset
An RTP Sender with RTCP turned off (i.e. having set the RS and RR bandwidth modifiers [RFC3556] to 0) SHOULD use a different SSRC for each different clock rate but MAY use different clock rates on the same SSRC as long as the RTP timestamp is calculated as explained below:
Each time the clock rate changes, the start_offset and capture_start values are calculated with the following formulas:
Note that in all the formulas, capture_start is the first instant that the new timestamp rate is used. The output of the above method is exemplified in Table 4 (Appendix A).
An RTP Receiver MUST calculate the jitter using the following formula:
D(i,j) = (arrival_time_j * clock_rate_i - timestamp_j) - (arrival_time_i * clock_rate_i - timestamp_i)
An RTP Receiver MUST be able to handle a compound RTCP packet with multiple SR packets.
When the algorithm described in Section 4.1 is used the security considerations described in RFC 3550 apply.
The algorithm described in Section 4.2 is new and so its security properties were not considered in RFC 3550. Although the RTP timestamp is initialized with a random value like before, the timestamp value depends on the current and previous clock rates and this may or may not introduce a security vulnerability in the protocol.
This document requires no IANA actions.
Thanks to Colin Perkins, Ali C. Begen, Harald Alvestrand, Qin Wu, Jonathan Lennox, Barry Leiba, David Harrington, Stephen Farrell, Spencer Dawkins, Wassim Haddad and Magnus Westerlund for comments, suggestions and questions that helped to improve this document.
Thanks to Bo Burman (who provided the values in Table 4 of Appendix A).
Thanks to Robert Sparks and the attendees of SIPit 26 for the survey on multiple clock rates interoperability.
This document was written with the xml2rfc tool described in Rose [RFC2629].
[RFC2119] | Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997. |
[RFC3550] | Schulzrinne, H., Casner, S., Frederick, R. and V. Jacobson, "RTP: A Transport Protocol for Real-Time Applications", STD 64, RFC 3550, July 2003. |
The following tables illustrate the timestamp and jitter values produced when the various methods discussed in the text are used.
The values shown are purely exemplary, illustrative and non-normative.
Capt. time | Clock rate | RTP timestamp | Arrival time | Transit | Jitter | Average jitter |
---|---|---|---|---|---|---|
0 | 8000 | 0 | 0.1 | 800 | ||
0.02 | 8000 | 160 | 0.12 | 800 | 0 | 0 |
0.04 | 8000 | 320 | 0.14 | 800 | 0 | 0 |
0.06 | 8000 | 480 | 0.16 | 800 | 0 | 0 |
0.08 | 16000 | 800 | 0.18 | 2080 | 480 | 30 |
0.1 | 16000 | 1120 | 0.2 | 2080 | 0 | 28 |
0.12 | 16000 | 1440 | 0.22 | 2080 | 0 | 26 |
0.14 | 8000 | 1600 | 0.24 | 320 | 720 | 70 |
0.16 | 8000 | 1760 | 0.26 | 320 | 0 | 65 |
Capt. time | Clock rate | RTP timestamp | Arrival time | Transit | Jitter | Average jitter |
---|---|---|---|---|---|---|
0 | 8000 | 0 | 0.1 | 800 | ||
0.02 | 8000 | 160 | 0.12 | 800 | 0 | 0 |
0.04 | 8000 | 320 | 0.14 | 800 | 0 | 0 |
0.06 | 8000 | 480 | 0.16 | 800 | 0 | 0 |
0.08 | 16000 | 1280 | 0.18 | 1600 | 0 | 0 |
0.1 | 16000 | 1600 | 0.2 | 1600 | 0 | 0 |
0.12 | 16000 | 1920 | 0.22 | 1600 | 0 | 0 |
0.14 | 8000 | 1120 | 0.24 | 800 | 0 | 0 |
0.16 | 8000 | 1280 | 0.26 | 800 | 0 | 0 |
Capt. time | Clock rate | RTP timestamp | Arrival time | Transit | Jitter | Average jitter |
---|---|---|---|---|---|---|
0 | 8000 | 0 | 0.1 | 800 | ||
0.02 | 8000 | 160 | 0.12 | 800 | 0 | 0 |
0.04 | 8000 | 320 | 0.14 | 800 | 0 | 0 |
0.06 | 8000 | 480 | 0.16 | 800 | 0 | 0 |
0.08 | 16000 | 640 | 0.18 | 1600 | 0 | 0 |
0.1 | 16000 | 960 | 0.2 | 1600 | 0 | 0 |
0.12 | 16000 | 1280 | 0.22 | 1600 | 0 | 0 |
0.14 | 8000 | 1600 | 0.24 | 320 | 0 | 0 |
0.16 | 8000 | 1760 | 0.26 | 320 | 0 | 0 |
An alternate way of fixing the multiple clock rates issue was proposed by Wenger & Perkins [I-D.ietf-avt-variable-rate-audio]. This document proposed to define a unified clock rate, but the proposal was rejected at IETF 61.
This library uses the formula described in Section 3.2.2.
Note that this library uses gettimeofday(2) which is not guaranteed to increment monotonically, like when the clock is adjusted by NTP.
This library (which uses the oRTP library) uses the formula described in Section 3.2.2.
Note that in some environments this library uses gettimeofday(2) which is not guaranteed to increment monotonically.
This library uses the formula described in Section 3.2.2.
This library changes the SSRC each time the format changes, as described in Section 3.1.