i’ve seen such issues in ALU as well, cells with low “erlang” on sdcch are having congestion.
There is a possibility that all LU are coming exactly at the same time (let’s say 100 LU Request every 10s)
With only 5 SDCCH, the cell can answer 40 LU Req simultaneously. The 60 others are rejected due to congestion.
In terms of erlang, those microscopic peaks of traffic are invisible.
Of course, it’s difficult to confirm..
One other assumption : the erlang are not properly counted. the erlang due to LU from 3G MS are not triggering the qos counters used for erlang computation.