Simple static Jitter buffer dimensioning

Hi all, how can I choose the size (in frames or packets, provided they are always of the same size) of a simple static jitter buffer implemented as a ring buffer that can overflow or underflow (so, no menagement of timestamps) if I know the variance of the interarrival times? The interarrival are very correlated because there are a lot of bursts ( a late packet and the subsequents all back to back). If the distribution was normal and the interarrivals independents, I could have said that taking three times the standard deviation I catch the 99% of them...

Thanks in advance

Breezer

Reply to
Breezer
Loading thread data ...

The fact that the delay on each packet isn't independent of the delay on previous and subsequent packets kills most of the assumptions you need in order to extrapolate to a Gaussian curve. At least on the small time-scale. I bet if you only did random sampling (say looked once every 10 seconds or once every 100 seconds then things would look more Gausian.

There are many systems where sampling at too high a rate give you too much high frequency crap (which can seriously change your observed curve shape). For a similar problem that ntp (the network time protocol) has to deal with, Google for "allan variance".

From a practical voip standpoint, I'm not sure what is gained by shooting for an arbitrarily chosen "benchmark" number like 99%. What you really need to optimize for user annoyance. You need to find the point in the delay / packet loss curve where the users total annoyance from highly delayed speech is balanced with the raspy sound from lost packets. Hopefully this won't be a common occupance but when it is, I'd hope the equipment has some reasonable *sounding* trade-off's that are set via real user tests.

-wolfgang

Reply to
Wolfgang S. Rupprecht

Cabling-Design.com Forums website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.