RTP issues can be difficult to diagnose without a healthy amount of packet captures, etc.
Devices with Issues
Carriers known to be using this device: L3, WilTel, Bandwidth.com, Broadvox, Global Crossing, XO, Qwest and T-Systems (a division of Deutsche Telekom).
- 19:40, 9 December 2008 (PST)
- Known to support exactly
729 with rfc2833OR
711u with DTMF inband.
- If you're using FreeSWITCH then you'll be stuck using 711u (PCMU) with DTMF inband. ( also rfc2833 look below how )
- Dropped audio infrequently. Anthony discovered that they have a +500ms response time if you STOP streaming RTP data to the carrier.
Whatever hardware they use has a list of reasons to reset.
- If your installation of FreeSWITCH want to change timestamp base and send them mark bit, they reset with 2 seconds of silence.
- If you send 2833 on it's own timestamp base it resets.
If you are suffering from dropped audio, specifically regarding the first 2-3 seconds of a single audio stream then try the following:
<!-- This will continue a steady stream of RTP to the Sonus device and your dropped audio will nearly diminish --> <X-PRE-PROCESS cmd="set" data="send_silence_when_idle=400"/>
Take care when using this parameter, as it may suppress ringback tones by actually sending early media.
conf/sip_profiles/external.xml Make sure you are running a FreeSWITCH version newer than Dec. 9th, 2008
<!--This will rewrite all new timestamps instead of passing them from the other leg.--> <param name="rtp-rewrite-timestamps" value="true"/>
If you are having DTMF problems and Sonus is in your media path, you should make sure you are using the latest version of FreeSWITCH. As of FreeSWITCH revision 10744, FreeSWITCH auto-detects Sonus end-points and applies a hack to "fix" the issue. The hack is described as follows (from switch_types.h):
Sonus wrongly expects that, when sending a multi-packet 2833 DTMF event, the sender should increment the RTP timestamp in each packet when, in reality, the sender should send the same exact timestamp and increment the duration field in the 2833 payload. This allows a reconstruction of the duration if any of the packets are lost. final_duration - initial_timestamp = total_samples However, if the duration value exceeds the space allocated (16 bits), the sender should increment the timestamp one unit and reset the duration to 0. Always sending a duration of 0 with a new timestamp should be tolerated but is rarely intentional and is mistakenly done by many devices. The issue is that the Sonus expects everyone to do it this way instead of tolerating either way. Sonus will actually ignore every packet with the same timestamp before concluding if it's DTMF.
Note that the version of FreeSWITCH which has this patch auto-detects Sonus, so you don't need to configure anything.
If you are going through a provider who uses Sonus only as an SBC and not in your media path FreeSWITCH may incorrectly identify your call as going through Sonus and actually corrupt your media stream's DTMF. You can override FreeSWITCH's auto-detection of end-user agents with a flag in your sip_profiles/ profiles. Just add:
<param name="auto-rtp-bugs" value="clear"/>
In some cases, you may have the reverse situation where Sonus or Cisco equipment sits in your media path, but FreeSWITCH can't auto-identify it. In that case, try these settings:
<param name="auto-rtp-bugs" value="CISCO_SKIP_MARK_BIT_2833"/>
if you are having trouble with Cisco equipment. For Sonus, try:
<param name="auto-rtp-bugs" value="SONUS_SEND_INVALID_TIMESTAMP_2833"/>
If your Call to a Sonus still won't work
For those of you who have tried the auto-rtp-bugs and still cannot get DTMF to work for G711 negotiated calls, there appears to be an issue with the Sonus using Fax treatment which drops the out of band DTMF packets that come from FreeSWITCH (among other platform). Some more in depth information for Sonus users (at least for 5.0 and 5.1 users, I can't speak for those who may have the latest and greatest code revisions which may fix this) is below.
In the GSX Navigator (Sonus Insight), go to Sonus→Profiles→Packet Service Profiles→default. You may have some changes made to the Fax Treatment/Failure section. To resolve the DTMF you will have to change that to match the below examples.
An example of a base profile that will not work:
Audio Encodings: G.711 DTMF Modem Fax Pkt Size/ Law/ Relay/ Treatment/ Treatment/ Priority Codec Code Rate Send SID Remove Failure Failure 3 G711SS 20 LAWFROMCIR NONE G711FALLBACK G711FALLBACK 0BPS DISABLED DISABLED CONTINUE CONTINUE 4 G711 20 LAWFROMCIR NONE G711FALLBACK G711FALLBACK 0BPS DISABLED DISABLED CONTINUE CONTINUE
What you would need to change it to, to fix the DTMF:
Audio Encodings: G.711 DTMF Modem Fax Pkt Size/ Law/ Relay/ Treatment/ Treatment/ Priority Codec Code Rate Send SID Remove Failure Failure 3 G711SS 20 LAWFROMCIR NONE G711FALLBACK FAXRELAYORG711FA 0BPS DISABLED DISABLED CONTINUE CONTINUE 4 G711 20 LAWFROMCIR NONE G711FALLBACK FAXRELAYORG711FA 0BPS DISABLED DISABLED CONTINUE CONTINUE
The above profile will work if you are sending direct to the GSX using the dtg=TRUNKGROUP protocol. If you are routing your calls to the PSX, then you would make the appropriate changes to the Preferred Packet Service Profile that is attached to the trunk group that you are a) sending from or b) sending to.
It's a crazy world out there.
<!-- This will use the async timer (this one you can also set to none if you don't want asynchronous RTP traffic) --> <param name="rtp-timer-name" value="soft"/> <!-- RTP_BUG_CISCO_SKIP_MARK_BIT_2833 = (1 << 0), Some Cisco devices get mad when you send the mark bit on new 2833 because it makes them flush their jitterbuffer and the dtmf along with it. This flag will disable the sending of the mark bit on the first DTMF packet. RTP_BUG_SONUS_SEND_INVALID_TIMESTAMP_2833 = (1 << 1), Sonus wrongly expects that, when sending a multi-packet 2833 DTMF event, The sender should increment the RTP timestamp in each packet when, in reality, the sender should send the same exact timestamp and increment the duration field in the 2833 payload. This allows a reconstruction of the duration if any of the packets are lost. final_duration - initial_timestamp = total_samples However, if the duration value exceeds the space allocated (16 bits), The sender should increment the timestamp one unit and reset the duration to 0. Always sending a duration of 0 with a new timestamp should be tolerated but is rarely intentional and is mistakenly done by many devices. The issue is that the Sonus expects everyone to do it this way instead of tolerating either way. Sonus will actually ignore every packet with the same timestamp before concluding if it's DTMF. This flag will cause each packet to have a new timestamp. RTP_BUG_IGNORE_MARK_BIT = (1 << 2), A Huawei SBC has been discovered that sends the mark bit on every single RTP packet. Since this causes the RTP stack to flush it's buffers, it horribly messes up the timing on the channel. This flag will do nothing when an inbound packet contains the mark bit. RTP_BUG_SEND_LINEAR_TIMESTAMPS = (1 << 3), Our friends at Sonus get real mad when the timestamps are not in perfect sequence even during periods of silence. With this flag, we will only increment the timestamp when write packets even if they are eons apart. RTP_BUG_START_SEQ_AT_ZERO = (1 << 4), Our friends at Sonus also get real mad if the sequence number does not start at 0. Typically, we set this to a random starting value for your saftey. This is a security risk you take upon yourself when you enable this flag. RTP_BUG_NEVER_SEND_MARKER = (1 << 5), Our friends at Sonus are on a roll, They also get easily dumbfounded by marker bits. This flag will never send any. Sheesh.... RTP_BUG_IGNORE_DTMF_DURATION = (1 << 6), Guess Who? ... Yep, Sonus (and who know's who else) likes to interweave DTMF with the audio stream making it take 2X as long as it should and sending an incorrect duration making the DTMF very delayed. This flag will treat every dtmf as if it were 50ms and queue it on recipt of the leading packet rather than at the end. RTP_BUG_ACCEPT_ANY_PACKETS = (1 << 7), Oracle's Contact Center Anywhere (CCA) likes to use a single RTP socket to send all its outbound audio. This messes up our ability to auto adjust to NATTED RTP and causes us to ignore its audio packets. This flag will allow compatibility with this dying product. RTP_BUG_GEN_ONE_GEN_ALL = (1 << 8), Some RTP endpoints (and by some we mean *cough* _SONUS_!) do not like it when the timestamps jump forward or backwards in time. So say you are generating a file that says "please wait for me to complete your call, or generating ringback" Now you place and outbound call and you are bridging. Well, while you were playing the file, you were generating your own RTP timestamps. But, now that you have a remote RTP stream, you'd rather send those timestamps as-is in case they will be fed to a remote jitter buffer...... Ok, so this causes the audio to completely fade out despite the fact that we send the mark bit which should give them heads up its happening. Sigh, This flag will tell FreeSWITCH that if it ever generates even one RTP packet itself, to continue to generate all of them and ignore the actual timestamps in the frames. RTP_BUG_CHANGE_SSRC_ON_MARKER = (1 << 9), By default FS will change the SSRC when the marker is set and it detects a timestamp reset. If this setting is enabled it will NOT do this (old behaviour). RTP_BUG_FLUSH_JB_ON_DTMF = (1 << 10) Flush jitterbuffer when getting RFC2833 to reduce bleed through --> Examples usage in profile parameters: <param name="manual-rtp-bugs" value="SEND_LINEAR_TIMESTAMPS|START_SEQ_AT_ZERO|NEVER_SEND_MARKER|IGNORE_DTMF_DURATION|ACCEPT_ANY_PACKETS|GEN_ONE_GEN_ALL"/>
For more information on these and other Sonus issues:
- http://www.submityoursip.com/wiki/Sonus_NBS or http://web.archive.org/web/20100121175135/http://www.submityoursip.com/wiki/Sonus_NBS
VOP-Softclients don't like changes in the RTP timestamp, e.g. when a REFER to the VOP-client occurs and the RTP-stream is replaced by the stream from a different call with a different timestamp. This leads to one-way-audio for the callee, the VOP Softphone doesn't process the incoming RTP-pakets with the "new" timestamp and the callee will not hear the caller.
Setting rtp-rewrite-timestamps to True in the affected SIP-profile eventually solves the problem.