[time-nuts] periodic jitter , frequency offset in phase lock LeCroy SDA 5000 vs Tektronix AFG 3252
stefan.brisken at fhr.fraunhofer.de
Wed Apr 7 07:35:16 EDT 2010
I did a little experiment. It comprised a Tektronix Arbitrary Function Generator
AFG 3253 and a digital oscilloscope LeCroy SDA 5000. The original intent was to
see the properties of the phase lock of the oscilloscope's 100 MHz external
I gave a 100 MHz output of the Arbitrary Function Generator onto the digital
a) I recorded data sets up to 10ms at 5GS/s (called "data")
In post processing I constructed a 100 MHz sinusoid digital
reference signal (called "ref"). I performed a hilbert transformation
on both. Then I calculated the phase difference "delta_phi" with respect
to the time vector:
delta_phi= angle(hilbert(ref)) - angle(hilbert((data))
It turned out that delta_phi in all measurements had roughly the same
slope(envelope) of 36000°/s so I concluded there must be a frequency
offset of 100 Hz. So far so good. I removed the slope and had a look at
the residual jitter, which seemed to be Gaussian distributed with a standard
deviation of 1.8°. But a closer look at delta_phi showed that there still was
a remarkable deterministic, periodic jitter left. Rather sinusoid than
sawtooth. With a frequency of 100 MHz.
A possible explanation for that would be that the sampling time of the ADC
of the oscilloscope depends slightly on the power of the received signal. If
this relation is linear, that would explain a sinusoidal phase shift.
Can anyone confirm or diffirm that theory from his or her experience?
My aim is to remove the deterministic jitter. But no fit in the time domain
The periodic jitter was also present in b).
b) Same as above, but I gave CH2 of the AFG on to the external 100 MHz reference
input oft the oscilloscope. I expected a constant phase difference
(=same frequency) in average, but it turned out that there is still 1-2 Hz
mean frequency offset left.
Is this the normal case? I had expected a non measurable frequency offset due
to the phase lock. Warm up effects can be ruled out.
To make confusion complete the header file of the oscilloscope data states (for
both a) and b) ) that the sampling rate was not exactly 5 GS/s but 66 S/s below.
Can I rather trust the header file or the preset 5 GS/s?
Thanks a lot.
More information about the time-nuts