PFM5 Spectrometer Microphonics Analysis

OBSID 300122DD, 2007-02-22 (PFM5)

OBSID 300122DE, 2007-02-22 (PFM5)

OBSID 300122DF, 2007-02-22 (PFM5)

OBSID 300122E0, 2007-02-22 (PFM5)

OBSID 300122E1, 2007-02-22 (PFM5)

OBSID 300122E2, 2007-02-22 (PFM5)

OBSID 300122E3, 2007-02-22 (PFM5)

OBSID 300122E4, 2007-02-22 (PFM5)

OBSID 300122E5, 2007-02-22 (PFM5)

OBSID 300122E6, 2007-02-22 (PFM5)

Summary

There is insufficient bias frequency coverage at the low end.
The observed noise spikes are for the most part not stable/repeatable.
The SL array has significantly lower microphonic response than the SS array.
In SS channels, there seems to be a consistently clean region around 210 Hz. Above 260 Hz is bad. Much of the range less than 200 Hz is usually OK.

1. Data

I used the noise spectra from Bernhard posted at the bottom of this page. In particular, I used the powerspect1_all.pow.fits files in subdirectories like this one. I am not sure what these power spectra are, but they might be a combination of the 4 bias amplitude settings for each frequency. During these tests, the lab log notes "microphonics excitation started with random noise filtered to 10Hz to 1 KHz".

For the bias frequencies, I used the values from the PFM5 data log. They are: 75.41, 100.68, 126.01, 151.41, 175.96, 201.35, 227.11, 253.65, 279.02, and 305.18 Hz.

2. Observational Results

An observed "IF" (intermediate frequency) spectrum has response to features in the audio frequency spectrum below and above the bias frequency (lower and upper sidebands). In the first set of plots, each noise spectrum for each bias frequency has been plotted twice: one curve as if all of the noise comes from the lower sideband (f = f_bias - IF), and the other as if all of the noise comes from the upper sideband (f = f_bias + IF).

Here is a directory with all of observed noise spectra.

Some comments:

The spectra are slightly smoothed (~0.2 Hz FWHM gaussian).
About each bias frequency, there is a mirrored pair of spectra, one red (lower sideband) and one green (upper sideband). Due to the 1/f noise in many of the spectra, there are often false "features" around the bias frequencies, even though I omitted the IF < 1 Hz parts of the spectra.
Note the gap in coverage for one sideband between the lowest two frequencies. A bias frequency between 75 and 100 Hz is needed. It wouldn't hurt to have another one between 100 and 125 Hz, too.
A stable microphonic feature, fixed in audio frequency, should appear in these plots as overlapping lower sideband/upper sideband spikes (plus two false "image" spikes). However, stable microphonic features appear to be uncommon in these data.

3. Finding the Microphonic Spikes

I tried a decomposition technique to try to reconstruct the true audio frequency spectrum (without the false image spikes), but it failed miserably, probably due to the unstable spikes and 1/f noise. As an alternative, I simply multiplied the lower sideband spectra by the upper sideband spectra, which should show the repeatable microphonic features as the largest spikes.

Here is the directory with those plots.

Here is the median over all SS channels.

And here is the median over all SL channels. The noise in the typical SL channel is much less than that in the typical SS channel.

CDD, 2007 Sep 10