Hirose and Izumi,
Related logs: 18452, 18442, 18437, 17659
Since PLL X exhibited the mystery peak again today, we went to the floor and looked at various signals both at RF and audio frequencies.
Currently, we are analyzing the data, but we have a feeling that they will not nail down the cause. Please be patient until we come back with the analysis results.
By the way, the plots below are the ones from the summary page today, highlighting that we had a few hours where the peak was hight.
The plot below shows two spectra of the PLL X feedback signals when the mystery peak was visible and when it wasn't.
The red trace is the one with the mystery peak visible at around 250 Hz, labeled as "loud time." The blue was the same signal but measured when the mystery peak was not visible, labeled as "quiescent time". The raw data are attached as text files to this log entry. There are two data files as follows.
Note that the spectra are stitched together from multiple measurements with different bandwidth settings. Therefore, the variance of the power spectra don't look consistent across the frequency band. The measurements were all done by a Moku-lab configured for
The last one I would like to show here is the beatnote signals. The plot below shows the comparison of the beatnote when it was in loud and quiescent states.
The legend and color scheme are identical to the plot for the feedback signals. As expected, the data for the loud time show an excess at around 60 kHz.
However, something is wrong below 40 kHz because the noise level for the quiescent time is greater than the one for the loud time. This is counterintuitive given the feedback spectra show at the top. Now, I have to confess that I made a mistake which I believe is properly corrected. The quiescent data was taken with a different physical unit setup i.e., dBm rather than dBm/Hz. I thought I corrected the data back to dBm/Hz by normalizing the data by the resolution bandwidth. But, I might have made another mistake there. Otherwise, this feature may be a hint for resolving the mystery.
Anyhow, I plan to repeat the same measurement to make sure that the blue trace is really greater than the red (or vice versa), hopefully tomorrow.
Sorry for confusions. In the original post at 18466, I wrote something like that there could be some relations between the beatnote's morphology and the mystery peak. However, repeating the same measurements in this morning, I quickly learned that my theory was wrong. The morphology of the beatnote spectrum is (almost) independent of whether the mystery peak is present or not.
The plot below is a fresh measurement telling an essence.
In the measurement, the mystery peak at around 250 Hz with a moderate height was present all the time. I varied the overall gain by changing the IN1 gain in the common mode servo for ALS PLL X. The result is very obvious --- peaks and structures above offset frequencies of 50 kHz are consequences of the loop instability. Also, the noise suppression effect is visible nearby the beatnote frequency as a function of the overall loop gain. Note that the IN1 gain has been nominally set to -20 dB.
I suspect that the IN1 gain setting was not properly set during the measurements yesterday. In fact, I remember me changing the IN1 gain multiple times, but don't remember when.
