1/f noise measurement with lock-in amplifier

I am just wondering how to use lock-in amplifier to measure the 1/f noise spectrum of a device or an amplifier (there may be other ways but I am interested in how to apply lock-in principles in this case). Some people did something similar but they didn't show how exactly they were doing it http://www.radionet-eu.org/fp7wiki/l...h_ew_artal.pdf

A first thought would be directly feeding the noise into the lock-in amplifier. But I am not sure if it is going to work because the phase of noise is random and it is hard to lock the phase. Can anyone explain how the measurement is done?

I assume it will be used as small band receiver with programmable bandwidth and center frequency in unlocked state. The output will be an AC voltage in this case, you have to meaure it's power respectively Vrms.

The reference describes that the mm-wave cooled LNAs, while offering extremely low noise temperatures at mm wave bands, have quite poor 1/f low-frequency noise.
The same problem was solved in 1946 by Dicke who modulated (interrupted) the antenna output noise to separate its level from receiver noise, and used a lock-in amplifier (or synchronous demodulator) to obtain the amplified difference between the antenna output noise temperature and the reference noise temperature.

All this is even more important in projects like Planck where the modulation of antenna output noise is generated by antenna rotating in space. The small variations of antenna output as it scans the sky are relatively slow, therefore 1/f noise of the receiver is quite significant obstacle.

The authors describe how they tried to quantify the component noise. One of the methods utilized the lock-in amplifier which synchronously separates component noise from the reference input while the frequency was set to ~550 points over the band 1 mHz to 100 kHz.

Another test was made with the signal analyzer which I assume uses FFT.

Both methods confirm the high 1/f noise mainly in LNAs and resulted in designing the radiometer system as correlation-type that can in part reduce the 1/f noise effect.

I believe that was the case