--
StevenWhite - 29 Oct 2007
Here's my[Roger Norrod] promised input on the spectral baseline stability requirements:
A handy GBT sensitivity calculator is at:
http://www.gb.nrao.edu/GBT/setups/senscalc.html
This will calculate the expected noise RMS across a position-switched baseline from the GBT Spectrometer, depending on the channel bandwidth, the number of sampling levels, and the smoothing used. It is convenient to consider RMS/Tsys which is measured on a receiver by taking two integrated spectra with a stable input load, calculating (Spec 1 -Spec 2)/Spec2 channel-by-channel, and then the standard deviation across the bandwidth (or a portion thereof). (Caution: there seems to be a bug in the calculator in that when Tsys=1K is input, the program instead uses a default Tsys as shown at the bottom of the results. It will take, say, 10K correctly; then you divide the displayed RMS by 10.)
Two representative results are:
9-level, 50MHz, 8192 chs, Hanning, 3min On&Off -> dT/Tsys = 0.985e-3
3-level, 800MHz, 1024 chs, Hanning, 3min On&Off -> dT/Tsys = 0.104e-3
One wants this level of stability in the product of Total Power(f) = dTrx(f)*dGain(f) over a reasonable position switched On/Off period (~6 mins is the minimum desirable). I can tell you that this ain't easy.
It is useful to keep sensitive transmission paths short so that drifts in the path length cause ripple periods greater than the spectrometer bandwidth (at least in principal easier to fit out with simple polynominals). Since the ripple period is equal to C/(2*L) (C is the velocity of propagation in the line), setting L < C/(4*BW) keeps less than half a ripple period across the bandwidth. This works out to 9.4 cm electrical length for BW = 800MHz.