Thursday, May 13, 2010

What is Optimum Value of T200?

What is Optimum Value of T200?
we like to know the optimum value for T200 on LAPDm. All vendors have different default values (Siemens 145 ms, Nokia 220 ms, Satellite Abis 400 ms etc.) for this timer - but which value is the best to reduce SDCCH drops and to keep the retransmissions at an acceptable level ?

Example: SDCCH/8

During a 51er multiframe the SDCCH/8 occupies four consecutive TDMA frames (four bursts are sent). Than the MS / BTS has to wait for the next 51er multiframe (i.e. 235 ms) before the next Layer 2 frame could be sent. 145 ms / 220 ms are shorter than the 51er multiframe (235 ms) so in case of an missing acknowledgement this is always a T200 expiry. The SDCCH drop will occur if T200 expired N200+1 times. If the T200 is increased (for example to 500 ms) we have two 51er multiframe to get the acknowledgement and the SDCCH drops are reduced.
The Qos Stats show a clear & strong corelation between the KPI and the parameter T200. That's amazing !

Regarding the Ack from the BTS, I'm not sure (and I'm tired to look in the 3GPP specs :) ). Take the subchannel "0" from the SDCCH ts.

in DL : the BTS sends SDCCH/0 on burst 0, 1, 2, 3 and the SACCH/0 on burst 32, 33, 34, 35

in UL : th MS sends SDCCH/0 on burst 15, 16, 17, 18
and the SACCH/0 on 47, 48, 49, 50

And I ***believe*** that the Lapdm acknowledgments can be sent on either the SACCH or the SDCCH, since both of them are sent over the same LapDm link. I'm not sure at all about this though, but it sounds logical.

The value of the SDCCH Drop due to Radio failures (in ALU) is usually around 1% in a fairly good network. The SDCCH Drop due to Radio Failures is a counter that encompasses both the Radio Link Timeout and the "T200*N200+1 times" failures.

I am not able to test your changes because I am not working on a live network (i am a gsm trainer, living in a world of theory...)

T200 = 220ms for sDCCH SAPI0
= 450ms for SDCCH SAPI3
= 900ms for SACCH associated to SDCCH

No comments:

Post a Comment