Tabletop frontend experiments

[slosar]

Nov 14, 2016: Same signal fed into two ports (AS)

A 1 GHz signal split via 180 hybrid with 50Ohm terminator at one port, fed into two inputs running at 1.25GS. Extracting full resolution (1/0.107 Hz) spectra at 1GHz+-5kHz. Ampltiudes normalised at beginning of the run. Delta if the offset freq.

Start at 1855: 2000: 2100: 2200:

Briefly stop to restart in screen. Note a new transient 2206: 0000: 0100: 0200: 0300: 0400: 0500: 0600: 0700: 0800: 0900: Pause, restart at 11.10, note transient again.

[slosar]

Nov 14, 2016: One part of the signal goes through "front-end"

After 180 deg hybrid, one goes straight into CH2, the other goes through 9dB attenuation and is then coupled at -30dB to a 50Ohm terminator line that is then amplified by ~+15x2 and fed into CH1. Picture of the setup:

[slosar]

Nov 28, 2016: Better analysis of this data

The first set (tone into both inputs) is analyzed here: https://github.com/bmxdemo/bmxproject/blob/master/ipynb/tonetone.ipynb

The data notebook should be self-explanatory, but here are some findings:

• chan1 and chan2 follow each other very well modulo gain fluctuations
• the cross-correlation coefficient is 1 within 1e-8
• the phase drifts a bit, by 0.005 radians. At 1GHz this corresponds to a path length of around 2mm, but time difference is miniscule
• frequency varies by around 2kHz
• power spectrum of gain fluctuations show clear 1/f: they are perfectly correlated above knee frequency, some 0.9 below.

The second setup is analyzed here: https://github.com/bmxdemo/bmxproject/blob/master/ipynb/amptone.ipynb

The data notebook should be self-explanatory, but here are some findings:

• chan1 is going through amplifiers, chan2 straight
• chan1 and chan2 are now pretty much decoupled
• cross-correlation coefficient 1 within 1e-6 (rather than 1e-8)
• beautiful 1/f noise spanning 3 decades in both channels, but considerably more pronounced for the amplifier leg, as expected.
• Knee frequency around 0.1Hz so one can really average the tone to get good SNR
• At higher frequencies, power in chan2 ampltiude fluctuations 3% higher, but power spectra very very flat
• cross power spectrum correlated at 0.97 above knee frequency, decorrelates below, but only to some 0.65 at 1e-3 Hz, so majority of 1/f still coming from digitizer.

[slosar]

Dec 5, 2016: tone and broadband, one hour (AS)

Took one hour of data the contains both broadband and tone

The mean broadband (with 1GHz bin, cancelled out, otherwise it is a spike with 100x power):

We clearly see some structue in power, perhaps due to reflection and missing attenuator between amplifiers.

I took log of this power and calculate eigenvectors and eigenvalues in this space. The first eigenvalue is ten times bigger than the second one. However, the covariance matrix is still diagonal, while the first eigenvalue is 30 times bigger than if all variance was distributed nominally. Below are first three eigenvalues:

We see that the first three have some structure, the rest seems noise. The peak at 1GHz in the first one is simply due to 1GHz bin being blinded.

Next I decomposed each 100ms record in this space and plotted the time line:

We see that the second and third eigenvalues correspond to some (regular?) RFI events. The first one really looks more like 1/f ish behaviour. Note that they are vertically offset for easier viewing.

Finally, I plot what tone and what first components are doing as a function of time, smoother by 40s moving average. This looks a bit like I expected, but I did expect somewhat closer matching on shorter timescales.

The next step is to do the same for the full dataset.