1)  What is the aperture of your horn? 

2)  How much gain do you have after the two LNA4ALL preamps?

3)  What is the loss of your BPF for 1420?

4)  What is you detection bandwidth?

5)  What is the loss of your coaxial cable(s)?

The reason for all the questions is based on our experience with a 14-foot off-boresite fed dish reflector as an antenna, a full-wave loop antenna element placed 1/4-wavelength

in front of a splash plate.  That is fed to (all Minicircuits) a 0.5 dB NF 23 dB gain preamp.  That feeds a 1 GHz HPF to eliminate (mostly, but not completely) a near-by cell tower

installation (grrrr....).  The HPF feeds another preamp of the same ilk.  Then some 70-feet of low loss rigid coaxial cable into the 'warm room' where we have installed

roughly another 40 dB of gain.  Inside, we have also installed 3 dB attenuators between active stages for impedance stabilization and gain vs. SWR stabilization. 

That finally feeds an AirSpy SDR.  The system operates under SDR#.  Our noise temperature is nominally 100 to 120 K, established (somewhat painstakingly) by several different and independent

methods (not all 'calibrated' noise sources exhibit consistency in the third decimal place!!).  Installation of the 1 GHz filter to get rid of a very dirty cell tower

 installation is responsible for half the noise temperature......  As you are aware, the H1 emission can be up to 500 to 750 kHz wide.  You are detecting only a small

amount of the power of the H1 emission with the SDR#-limited available max bandwidth.

I'm very confused by your bandwidth comments here.  I don't know much about SDR#, but if it shows a couple of MHz around the notional
  line frequency of 1420.40575MHz, and allows deep integration, then you've covered essentially ALL of the hydrogen line doppler range within
  the galaxy.

The next reasonable "target" for an amateur setup is the emissions from our sister galaxy, Andromeda, which are up-shifted to around 1422.5MHz
  or so due to blue shift (Andromeda is hurtling towards us).  Very long integration times would be required, due to having to deal with
  inter-galactic, rather than inter-stellar, distances.

Here at CCERA, we operate a 21cm spectrometer, using AirSpy receivers on a 1.8m dish.   Our final bandwidth is about 2MHz, which allows us
  to "see" features at +/- 170km/sec from the rest frequency.   Our software is Gnu Radio based, rather than based on AstroSpy or SDR#.

I'll likely do a write-up sometime soon.