Hi, francx:
According to my understanding,when people deal with experiment data by pre-processing the data with a scale of log, they make nonlinear data linear to facilitate analysis. Regarding the BER-sensitivity theoretical background, kindly search the maxiam website **broken link removed** yourself for BER etc. You should be able to find the invaluable information in the application note section.
Well, when you talk about "test time vs. bit ratio" do you mean the sensitivity test also? Usually, we do not go for "gating by time" as each LOP level will give you different error bits. You may like to try "gating by error bits" for certain confidence level since you keep counting the error until the error bits meets your criteria.
Hope this helps.
Best regards,