On Sun, Apr 6, 2008 at 9:21 PM, Jeremy King <jerry.ipe.thomas at gmail.com> wrote: > > > Hi Benny, Hi Jeremy, (since you Cc-ed your second mail to the list, I'm taking the freedom to Cc the list on my reply too. I apologize if that's not what you wanted). > I am facing a problem in RTP receive side of PJ-MEDIA. I apologise for the > length of this mail. > > My test set-up is as shown below from a media perspective. > > <---[LAN]--->| |<---[INTERNET]--->| |<---[SERVICE_PROVIDER]---> > | NAT | | | > | + | | RTP | > [PJSUA]----| F/W |------------------| |----[MEDIA SERVER] > | + | |PROXY| > | G/W | | | > | | | | > > When a UA makes a call to a particular service number, the media server > ends > up playing out a series of announcements to the UA. G.711 u-law codec is > used > between the media server and PJ. The media server sends out RTP packets > at a > more or less regular 20 ms interval. However, in the RTP packets the media > server sends to the UA, the timestamp and SSRC fields are both '0'. > Nevertheless, the sequence of arrival of RTP packets as well as the > sequence > numbers in those RTP packets seem correct. If the timestamp in RTP packets are always zero, then basically it's broken. > However, the announcements, when played out at PJSUA, sounds very > "choppy". > The playout sounds as if there is massive packet loss. Yep, it's quite expected. > I am using the latest SVN check-out of PJ without any modifications and > have > set the PJ_CONFIG_MAXIMUM_SPEED group of compile-time options. > > Since a Wireshark capture showed that all the RTP packets are arriving > correctly at the PC where PJSUA is running, I tried investigating further > focusing on how the incoming frame is inserted into the jitter buffer. > > On seeing that in "on_rx_rtp" in "pjmedia\src\pjmedia\stream.c", just > before > "pjmedia_jbuf_put_frame", the frame sequence number is calculated based on > the RTP timestamp, I tried to manually generate a timestamp from the RTP > sequence number inside "on_rx_rtp". > > That is when I found out that the incoming RTP packets are coming > completely > out of order i.e. there is a sequence difference of around 250 packets, > which > I assume the jitter buffer will never be able to handle. I think probably you missed converting the numbers to host byte order, hence the 256 difference. All the values in RTP header (including in pjmedia_rtp_hdr) are in network byte order. Cheers Benny > I removed the timestamp generation code and retained the following RTP > packet > header log line. > > PJ_LOG(3,(stream->port.info.name.ptr, > "RTP Pkt #%05hu: PT->[%05hu] TS->[%010lu] SSRC->[%010lu]", > hdr->seq, hdr->pt, hdr->ts, hdr->ssrc)); > > I am attaching the following with this mail: > > 1. Wireshark Packet Capture -> pktCap_20080406235400_TS4.cap > 2. PJSUA Generated Log File -> pjsua_log_20080406235400_TS4.txt > 3. Recording of PJSUA Playback -> audioCap_20080406235400_TS4.wav > 4. PCM extracted from the RTP -> audioRef_20080406235400_TS4.wav > > The build environment is WinXP SP2 with Feb 2003 edition of platform SDK, > Oct > 2004 edition of DXSDK and VC6 IDE. > > I hope I haven't missed out something that was right in front of me and > that > I am not doing something wrong! > > Thanks for taking the time to read this. I would appreciate any help on > this > issue. > > - Jerry >
