Re: [xrblock] Transport of I.D-xrblock-rtcp-xr-qoe style quality datain IPFIX

[Michael Krüger <[e-mail]>]

Hi, Qin,

Am 24.07.12 05:20, schrieb Qin Wu:
> Hi,Michael:
> Thank for your clarification.
> One thing I am not clear to your idea  is how to calculate MoS value or Rating value based on stream data from multiple sources?
> Currently in QoE XR draft, we support MoS calculation per SSRC or MoS calculation per audio channel?
> It seems you claim you can calculate MoS values based on audio data from multi-channels? 
> Is that MoS value is the weighted mean of MoS value of each channel? If in that case, what the weight for each 
> channel MoS calculation?
> Maybe this is just implemenation specific issue, however I am interested to listen to that.
> 
Thank you for your interest.
Of course we have not reinvented MOS. Our approach is related to ITU-T
P.564. The audio test samples used in that standard have a minimum
length of 8-seconds. We therefore assume that one must evaluate at least
this period of time, to determine a reliable MOS value. What we try to
achieve with monitoring live traffic, is to determine for every
individual flow of a VoIP call MOS values for every 10-second segment.
This provides us with many MOS values over the duration of an individual
call. Most network impairments are of transient nature, so for every
call the result will be a distribution of MOS scores falling into one of
the 5 classes. Depending on the number of seconds in each of the
classes, it is possible to assess how many seconds of a call have been
with which quality on the MOS scale 1-5. If this data is available for a
single flow, call, etc. the seconds of the 5 classes can be aggregated
over all flows/calls performed between VoIP operators, operators and
customers, or simply SIP-trunks. This will e.g. allow to come up with
new SLA agreements where the agreement could possibly be that 99% of all
VoIP minutes must be in the highest MOS class. The accuracy of this is
pretty high, since one determines MOS values for only a few seconds of a
call, bringing every impairment to attention via a lower MOS value for
the segment in which the impairment occurred. No averaging over a call
is done, so the distribution of MOS values like this, truly reflects the
monitored quality.

Does this make sense to you as well?

Regards,
Michael

> Regards!
> -Qin
>