All testing at 4 Kelvin was done in a Precision Cryogenics dewar. I have two different methods of performing the test. Both methods employ a switch to select between the device under test (DUT) and a calibration source. The first method uses variable temperature 50 Ohm terminators for the power source. Each terminator has an attached heater and temperature sensor. The switch is set to the calibration terminator, which is heated to about 40 Kelvin. The output spectrum is repeatedly measured over a several minute period while the terminator cools down. The resulting data give a curve of output power versus input power at each frequency for many different input powers ranging from about 4.2 to 40 Kelvin. A line is fit to these points using linear regression, and the gain and noise of the test setup are computed from the slope and x intercepts. Then the DUT is switched in, and the process is repeated. The test set calibration noise and gain are subtracted out, giving the noise and gain of the bare amplifier.
This process works fairly well, but has the disadvantage of being slow. It takes several minutes to measure the noise a single time. It would be nice to be able to tune the DC bias while watching the output. Thus, the second test setup was devised. Instead of using terminators for the power source, large attenuators are used. Thus, where the original test dewar only had a single RF line coming out of the dewar, this new setup has three: the calibration input, the DUT input, and the output. If the two inputs are terminated, then this test setup can be used in exactly the same way as the first setup, except the experimenter must remember the 300 K of noise power generated by the room temperature terminators. However, the experimenter now also has access to the inputs, albeit through large attenuators. The network analyzer can be hooked up to measure the gain with high frequency resolution and large frequency bandwidth while the DC bias is adjusted to flatten the gain. If I had a lot of testing to do, I would also use a noise diode at the input of the test dewar to measure noise temperatures real-time. This would be fairly easy to set up, except perhaps for the modifications to the software on the amp test computer.