[bojan]

Malcolm,
I'm afraid the bandwidth can't be reduced in our case - the received signal is also noise-like and broadband, so the only way to increase S/N is reducing the losses in front of LNA.. that means larger dish, optimized illumination of the feeder, best possible feeder (I am trying with this helix design, however I am not entirely sure how good it is, so this will have to be checked as well at some stage.. and whenever we are dealing with polarized signals, we have that 3dB loss), impedance matching to LNA and shortest possible cabling in between).
And a lots of signal averarging/integration (appropriate software should be providing this..)
The biggest problem here is experimenting - it takes a long time between changing something and seeing the results of the change, so this work requires a lot of experience and confidence in what has been done already..