Non-linear Response (FB)

I was comparing the different SNR methods and noticed that the correlation plots look funny (non-linear relation) and the funny channels (20, 48, 120 ...) show a time-variable response/gain compared to e.g. channel 1. The correlated noise removal assumes that the relative response does not vary in time during the scan, thus this makes it tricky do remove the skynoise (in one step). Does anyone have an idea wher this non-linear gain comes from and how to deal with it?

indir('/mnt/hgfs/VMshare/Temp/')
proj('T-79.F-0002-2007')
read(str(12468)) 
execfile('Test/cabling.py')
be=data.BolometerArray.BEGain
data.Data *= array((VtoJy * be),'f')
updateRCP('master-laboca-may07.rcp')
data.zeroStart()
flat()
flagC(resistor)
flagC(cross)
flagC(sealed_may07)
flagC(bad_may07)
data.flagSpeed (below=30.)
data.flagSpeed(above=500.)
data.flagAccel(above=800.)
medianBase(subscan=0)
flagC([93,257])   
data.correlatedNoiseFFCF(chanRef=1,plot=1,skynoise=0)     
signal(20)
signal(1,overplot=1,ci=2)
signal(48,overplot=1,ci=3)
data.correlatedNoiseFFCF(chanList=[1,20,19],chanRef=1,plot=1,skynoise=0)