[time-nuts] Using Timelab with Frequency Data (Long)

[Mark Goldberg]

As I have posted before, I am trying to use an SDR to make peak frequency
measurements at intervals using Spectrum Lab software to control the SDR
and record the data.I don't have the full info about what algorithms are
used to capture the peak frequency and was trying to look into various
methods to estimate the peak frequency between FFT bins. The software also
has a Phase measurement function that is described as:

Multiply the incoming signal with a local oscillator, here: a numerical
controlled oscillator with two outputs (90°)

Low-pass filter and decimate the mixed signal until the required (low)
bandwidth is reached

If the signal is MSK (minimum shift keying, as most VLF transmitters),
square the decimated signal, and examine the 'peaks' in the spectrum at
f_center +/- bitrate.
The same principle is also used in the continuous sampling rate calibrator .

Calculate the amplitude and the phase angle of the decimated signal (for
MSK: Angle of the recovered carrier signal).

I am not using MSK. I then use this function they provide to get the
frequency at each sample time:

pamN.freq returns the precisely measured frequency in Hz . Ideally the same
as the configured center frequency (can be used as an additional "health
check").

I don't know any more than that about the algorithms, but ideally it
provides a measured frequency of the peak of the signal of interest. So, I
should have a file with the measured frequency, sampled at the sample
interval. I have checked this statically with specific input frequencies
generated by an HP 8642A signal generator and it appears to provide a good
estimate of the frequency. The algorithm wraps every Hz, so I use the
poorer estimation of the peak to decide which Hz offset the frequency
actually is. This seems to work well, at least at 0.1 Hz intervals a few Hz
around the center frequency of the SDR. The measured frequency always is
within 0.2Hz of what was set, including all variations.

As a check, I connected both the reference clock of the SDR and the
measured signal to the same reference source, a Trueposition GPSDO. Ideally
this would provide a straight line on the ADEV plot that represents the
inherent noise of the measurement system.

So I have taken the inputs which represent the measured frequency at 0.6816
second intervals and imported them into Timelab as follows:

File | Import ASCII Phase or Frequency Data
Sampling Interval 0.6816 Sec
Input Frequency 10e6 Hz
Bin Density 29
Bin Threshold 4
Trace History 1
Numeric Field  <chosen column> X 1.0
Data Format Decimal
Comment Prefix USF(
# if Channels 1
Frequency (Hz) Selected

I have then taken readings at a slower rate, every 5.453 seconds and again
checked that they match every 0.1 Hz around the SDR center frequency.

I then connected both inputs to the same source again to reproduce the
measurement of the internal noise, this time with frequency measurements
every 5.453 seconds and imported as above with the following change:

Sampling Interval 5.453 Sec

Otherwise the connections and measurements are the same, just using a
different sampling interval. I had Timelab calculate the ADEV of both. I
would have expected them to be similar with no data below each sampling
interval. The result is two curves of basically the same shape, but the
curve of the higher sampling interval is offset vertically by a factor of
about 4.5. Since these are two different runs a day apart, some variation
is expected, but not a constant offset. This is repeatable for other
measurements, the results from the larger sample rate are always offset
vertically above the lower sample rate.

I have looked at http://www.leapsecond.com/pages/adev-avg/ and I can see
that the data has a small downward turn at the low end of the fast sample
rate, which is likely due to some averaging, but the two graphs are very
similarly shaped otherwise. I do not see an example there of such an offset.

The Screen capture from Timelab and the raw data are at:

https://drive.google.com/drive/folders/1jkORB_wujXOw-rknBt2pFCqLSvp91In6?usp=sharing

There are three columns in each file. Column 1 is from the peak frequency
algorithm and shows an offset and other anomalies. Columns 2 and 3 are from
the phase measurement algorithm frequency measurement function. All data is
in Hz at the respective sampling rates.

Can anyone tell me what I may be doing wrong?

Unfortunately, I can't get a TIC at this time, so I am trying to work with
what I have.

Regards,

Mark