Re: [AVTCORE] RFC3550: RTP Jitter value calculation

[Nataraja Hosahalli <[e-mail]>]

Hi Kevin,

Yes you are right. I ran both the formulae and also your improved formula for 500 times with constant d value, the value rr->jitter is converging near to average delay d after couple of iterations (~100 for d=2 for example).
However your new formula below is resulting to more accurate average.
rr->jitter = (s->jitter + 8) >> 4;
Older formula is almost always lagging by nearly 1 ms (0.5 ms floating value) with the average delay.

May be your new formula can be updated in RFC.

Thank you.

Regards,
Nataraj

From: Kevin Gross [mailto:[e-mail]]
Sent: Wednesday, 26 March, 2014 7:22 PM
To: Nataraja Hosahalli
Cc: [e-mail]
Subject: Re: [AVTCORE] RFC3550: RTP Jitter value calculation

The formula looks basically correct. You need to look at the settling value for rr->jitter, not s->jitter (which is a state variable). In your example, s->jitter will settle at 32 and the correct value (2) will be reported for rr->jitter.


I think the accuracy of the reporting could be improved by changing the second line to:

        rr->jitter = (s->jitter + 8) >> 4;





Kevin Gross - AVA Networks

On Wed, Mar 26, 2014 at 6:55 AM, Nataraja Hosahalli <[e-mail]<mailto:[e-mail]>> wrote:
Dear avtcore members,

I have a query on the RTP RFC 3550 which looks be an issue in RFC to me, please help by provide your expert advice.

RFC 3556 specifies 2 formulae for calculating RTP jitter in section 3.8. Couples of the RTCP RFC implementations are following these formulae.

          Section 3.8/ RFC 3550

      int transit = arrival - r->ts;

      int d = transit - s->transit;

      s->transit = transit;

      if (d < 0) d = -d;


One calculation option proposed,

        s->jitter += (1./16.) * ((double)d - s->jitter);

        rr->jitter = (u_int32) s->jitter;


other calculation option proposed
        s->jitter += d - ((s->jitter + 8) >> 4);

        rr->jitter = s->jitter >> 4;

RFC 3550 says both formulae are same. But if I look with sample numbers I see that the second formulae may not be appropriate.
The result of the formula "s->jitter += d - ((s->jitter + 8) >> 4)" LHS always keep increasing infinitely and could not be a reflecting an average value of jitter.

For example, in an hypothetical case if d = 2 each time, Jitter will be 2, 4, 6, 8, ... ever increasing, with the second formula.
Whereas with the first formulae each time it will be corrected towards d by fraction of 1/16. That is If d=2, rr->jitter will  be near to 2 in long run.

I request your opinion on this.

Thanks in advance.

Regards,
Nataraj

[1 (Custom) (2)]<http://www.dspg.com/>
[dsp_mal signature-Tali_180613-14]<mailto:[e-mail]>[dsp_mal signature-Tali_180613-15]<http://www.dspg.com/>[dsp_mal signature-Tali_180613-16]<https://www.facebook.com/#!/DSPGroup2013>[dsp_mal signature-Tali_180613-17]<http://www.linkedin.com/company/6894?trk=tyah>

Nataraja Hosahalli

Phone: [phone]<tel:%2B91%2080%204024%208408>
Mobile: [phone]<tel:%2B91%2099860%2025262>








______________________________________________________________________
DSP Group, Inc. automatically scans all emails and attachments using MessageLabs Email Security System.
_____________________________________________________________________

_______________________________________________
Audio/Video Transport Core Maintenance
[e-mail]<mailto:[e-mail]>
https://www.ietf.org/mailman/listinfo/avt


______________________________________________________________________
DSP Group, Inc. automatically scans all emails and attachments using MessageLabs Email Security System.
_____________________________________________________________________

______________________________________________________________________
DSP Group, Inc. automatically scans all emails and attachments using MessageLabs Email Security System.
_____________________________________________________________________