Unwanted modulation in crystal oscillator

[mauri]

hello,

a few years ago I built a Qrp kit transceiver, one with NE602, there are many
instances of those, I built the one by F6BQU, the 'Toucan', a cheap,
nice, little kit.
Last month, when trying to increase its output power a little bit,
I spotted a low frequency ripple in the transmitter output.
At first I thought there was a parasitic oscillation somewhere and
I put bypass caps, shunt resistors, etc,, to no avail.
Last week I traced back the cause at the very beginning of the
tx chain, in the 4MHz LO (which is also a NE602 - SA612AN, actually).
Measuring with a scope the LO I noticed what seemed to me a 800Hz
modulation over the 4MHz LO signal. 
I tried to replace the SA612 with another one, in case it was faulty,
but that didn't change anything. Then I removed the 4MHz crystal (HC49U) and
I put it in a different separated circuit, a simple test oscillator made by
one transistor (Colpitts, very simple), and I noticed the 800Hz modulation
was still there.  I replaced the crystal with a similar one (3.932 MHz) and I didn't see any spurious modulation, so I thought the 4MHz crystal was the culprit and I bought a
couple new (HC49S).
Last Monday I mounted the new crystals  and, to my dismay, I saw the
unwanted modulation again (this time is 300Hz with one of the new crystals and
100Hz with the other).
I attach scope pictures. I measured the tx output applying the scope probe right
upon the dummy load, and the LO on SA612 pin6.  As I'm a little bit paranoid,
I put capacitors between the scope probe and the circuit, on both the signal
and ground leads, to avoid loops, but that didn't make any difference.
I also attach the original diagram (the actual one is a little different as I have
modified it a bit to run on a different band, but the TX LO is basically the same). 
The transceiver is powered with batteries.
I looked for some insight in sacred texts as well as over the internet but I didn't
spot anything useful.
What could cause that ?  Can a simple crystal oscillator generate a modulated signal ?
Or is it just some scope artifact I'm not aware of ?

Thanks

Maurizio

[LA7SJA]

Check PIN1 on IC1 (78L05) with your scope, the VXO is an excellent source of FM modulation.

Johan-Fredrik

[nugglix]

Given that the modulation shows in a different circuit the TO should look for things common to both circuits.
Power supply, measurement (probe leads, ground leads), how the circuits are built and so on.

[mauri]

hi,

thanks for replying.

@ Johan-Fredrik : I checked pin1 on IC1 but didn't find any signal there, just DC

Today I have built a simple test oscillator running on batteries and I have tried 6 crystals with it.
That unwanted modulation appears with some crystal and not with others, which I find puzzlling.
The appearance of the unwanted modulation seems to be not dependent on the crystal frequency,
while it seems that branded crystals are less affected or not affected at all (crystal Q ?).
The frequency of the unwanted modulation is different for every crystal (even with the same frequency),
and is somewhat affected by the surrounding components (it changed a little when I replaced the
transistor with a different one ).

If it was just a scope artifact shouldn't it appears more or less on *all* crystals ?
At the same time, I doubt that most crystals around are faulty and nobody noticed that  ,
so something is going on that I don't understand.

Please let me know your take.

Thanks

Maurizio

[Tomorokoshi]

With the transceiver turned off, can you pick up any signals on the scope or spectrum analyzer in the 4 MHz range? Could the 400 Hz be the beat frequency due to interference between your transceiver and something else like a switching power supply?

Does the 400 Hz modulation change at all due to the orientation or position of the transceiver?

[David Hess]

If it was just a scope artifact shouldn't it appears more or less on *all* crystals ?

It looks like a digital storage oscilloscope artifact to me.  It could show up do to mixing between the sampling frequency and the oscillator frequency so it would vary with different frequencies.

To test this, I would put a trimmer into the oscillator and adjust its frequency slightly to see what effect that has on the oscilloscope's display.

[Ian.M]

Build a RF detector probe and see if that shows any signs of modulation on the oscillator.

[voltz]

I second the idea this is a scope artifact.
Here's why.
You have tried two completely different oscillator circuit designs, both show the problem - that pretty much rules out circuit design.
You only see the problem at multiples of 1Mhz (4Mhz in your case) and not at 3.932. Again this is suspicious.
You always see the modulated signal at multiples or 100Hz.. Like 300, 400 or 800 - again suspicious. Crystals simply dont do that.

Put that all together and its looking your scope is sampling at some multiple of 100Hz, this is manifesting itself as a beat note in effect which is looking like AM modulation, but in fact its not there. To test this as already suggested by David Hess, try a timmer cap across the crystal and adjust away. You should see is big change in the frequency of the 'am modulated signal'
Good luck with your QRP rig!

[mauri]

RF fellows,

thanks for your suggestions.

@ Tomorokoshi:
I have checked the spectrum and I can't see anything but the FM band.
I think that the probe cable acts as an antenna because I clearly
see in the spectrum a 88-108MHz rectangle.  I discovered that just after
buying the scope and since then I always use some filtering.
I don't have any switching power supply (I use batteries), so there
shouldn't be any RF source around.

@ Ian.M:
I have measured the oscillator output by means of a calibrated
RF probe, after applying a cap trimmer in series with the crystal.
I did that with the 4MHz HC49U crystal (the one that exhibits more
of that unwanted modulation) and the 6553kHz crystal (the one that
shows little or no modulation).
With the 4MHz HC49U crystal I got
- 1.168V DVM reading (which is about 1.3Vpp according to the
  probe manual) when the cap trimmer is fully closed (max.
  capacitance, 100pF),
- 1.163V (also about 1.3Vpp) with cap 3/4 closed
- 1.148V (1.25Vpp) with cap half closed
- 1.103V (1.2Vpp) with cap at 1/4
- 0.478V (0.6Vpp) with cap open (min. capacitance, it should be 5pF)

With the 6553kHz crystal I got
- 1.147V (1.25Vpp) with cap fully closed
- 1.121V (1.2Vpp) with cap half closed
- 0.894V (1 Vpp) with cap almost open (oscillation stops when it
  is fully open)

Such values seem pretty consistent with what I measured with the
scope and pretty linear. If I understand the purpose of this
test, an actual modulation should cause some spike in the readings,
correct ?

@David Hess and voltz:
as suggested, I have put a 5-100pF capacitive trimmer in series with
the crystal, and this is what I got.
The frequency of the unwanted modulation changes wildly when turning
the cap with the 4MHz crystals (both the HC49S and the HC49S), while
it basically doesn't change with the remaining crystals.
Unfortunately I haven't got any crystal which is exact multiple of
1MHz, other that one of 18MHz, so I tried that too and it also behaves
exactly like the 4MHz crystals in this respect.  In detail:
- the 4MHz HC49S crystal shows a 88Hz modulation when the cap is fully
  closed (it was 150Hz with no cap), 6.6Hz when it is half closed, and
  205Hz when the cap is almost open (oscillation stops if fully open)
- the 4MHz HC49U crystal shows a 463Hz modulation when the cap is
  closed (it was 600Hz with no cap), 290Hz when it is half closed, and
  2.2kHz when it is open
- the 18MHz crystal shows a 1.6kHz modulation when the cap is fully
  closed and 595Hz when it is half closed (oscillation stops turning
  the cap any further)

With the three crystal above, the scope behaves in a particular way:
when increasing the time base I see that sort of snake on the screen,
well, increasing the time base a little further makes a sinusoid
to appear on the screen which has the same amplitude as the fundamental,
it looks the same as the fundamental, but the scope cannot trigger it.
This behaviour is quite unclear to me and sounds very much like
an artifact. That doesn't happen with any of the remaining crystals.

- the 3932kHz crystal exhibits a small modulation with 40us period
  which doesn't change turning the cap
- the 3579kHz crystal exhibits a small modulation with 20us period
  which doesn't change much either turning the cap
- the 6144kHz crystal exhibits a small modulation with 200us period
  when the cap is closed and 140us when the cap is half closed, then
  disappears
- the 6553kHz crystal doesn't exhibit any measurable modulation, just
  the mains (50Hz) or harmonics of that appears, very low in amplitude
  (full data and pics are in the zipped attachment)

As I understand your point, this should prove that the unwanted
modulation I see on the scope screen is actually not there.
This raises a question though. When I noticed all that first in my
transceiver, it was caused by a 4MHz crystal in the LO that made
a 800Hz modulation to appear (that was the crystal I threw away
because I thought it was faulty ). What struck me was that I
found the very same modulation in the tx output measuring across
the dummy load, so I thought there was an actual parasitic modulation
that travelled through the entire tx chain up to the output (pictures
of that are in first post).
Now, if that 800Hz in LO was actually not there because it was just
"mixing between the sampling frequency and the oscillator frequency"
how do you explain the 800Hz in the tx output ?  Is it not there
either and also caused by the same artifact ?
If this is the case, that would also explain the discrepancy between
my wattmeter reading and the power computed based on the scope
measurement, which is always in excess.

And, most important, if there is indeed such artifact, how can I
workaround it ?  Is there a way to regain trust in my scope ?
You know, if I spot a 50Hz or 100Hz in the scope screen I easily
understand what the source is and dismiss it, but how am I supposed
to recognize the cause of a 833Hz ghost modulation which unfortunately
looks real to peasants like me ?

I'm looking forward to hearing your comments.

Thanks

Maurizio

[Ian.M]

@ Ian.M:
I have measured the oscillator output by means of a calibrated
RF probe, after applying a cap trimmer in series with the crystal.
I did that with the 4MHz HC49U crystal (the one that exhibits more
of that unwanted modulation) and the 6553kHz crystal (the one that
shows little or no modulation).
With the 4MHz HC49U crystal I got
- 1.168V DVM reading (which is about 1.3Vpp according to the
  probe manual) when the cap trimmer is fully closed (max.
  capacitance, 100pF),
- 1.163V (also about 1.3Vpp) with cap 3/4 closed
- 1.148V (1.25Vpp) with cap half closed
- 1.103V (1.2Vpp) with cap at 1/4
- 0.478V (0.6Vpp) with cap open (min. capacitance, it should be 5pF)

With the 6553kHz crystal I got
- 1.147V (1.25Vpp) with cap fully closed
- 1.121V (1.2Vpp) with cap half closed
- 0.894V (1 Vpp) with cap almost open (oscillation stops when it
  is fully open)

Such values seem pretty consistent with what I measured with the
scope and pretty linear. If I understand the purpose of this
test, an actual modulation should cause some spike in the readings,
correct ?

errrrrrr. . . .   You want to see if the crystal oscillator is amplitude modulated with a several hundred Hz signal.  Why would you use the RF probe on a DVM that cant respond fast enough to follow the modulation? 

USE A RF PROBE WITH YOUR SCOPE.  Put the 'raw' crystal output on one channel and the detected signal from the probe on the other.  If the detected signal follows the displayed envelope your oscillator is really modulated by *something*.  If there is no modulation, the detected signal will be a DC level, and then you know the observed envelope variation is a scope artefact.  As a final sanity check, connect the negative end of a 1.5V cell to your oscillator ground, and probe its positive end instead of the RF probe output.  If it doesn't give you a flat line you have serious scope problems or maybe a ground loop with a significant current flowing round it. 

[voltz]

Hi,
Your results from adding the trimmer cap and seeing a large change in the modulation is consistent with a scope artifact. And also why other crystals away from dead on 4Mhz frequencies appear not to show the problem. Also your observation of 'snake like' patterns while changing the scope timebase around 4mhz points the finger to artifacts too.

But what can be done?.. 

Well you could try an analog scope if you have one. That would show a true signal without any digital sampling.
Its the reason some of us keep both analog and digital scopes handy. Sometimes problems like this can happen.
There may be a menu option in your scope to change the sampling frequency but i cannot comment on that, its a maybe.

Another sanity check, try listening to the oscillator on a receiver. Tune it in for zero beat, observe if there is any modulation on the signal. If 400 or 800Hz is present, it would be clearly audible. The suggestion here is that there would be none, just a clear signal.

And yes as mentioned, you could try to demodulate the signal with a probe or make a simple diode detector and view that on the scope for any audio signals.

[Kalvin]

If you have a RTL SDR-dongle which can be tuned down to 4MHz, make a small loop antenna and place the loop close to the xtal oscillator. By zooming to a fundamental you should see whether there is any modulation or not. Does the modulation change when you probe the circuit with our oscilloscope? You can also use oscilloscope with a small loop antenna to pick up the oscillator's signal without touching the circuit itself. Is the modulation still present?

[mauri]

RF folks,

I have done my homework, and here is the result.

@Ian.M :
ok, it seems I have entirely missed your point
As I understand, we need a probe to discard the RF
and measure on the scope the AF-like signal which is
supposedly modulating the crystal oscillator.
My probe is a RF/AF and I'm not sure is suited for
this, so I just removed the clipping diode from
the oscillator and put it on the oscillator output
along with a 47nF cap, to act as a probe.
The 47nF cap should bypass the RF and still let
us measure something in the AF range, correct ?
Well, with this setup I can measure the oscillator
output  but I get nothing (in the scope) across
the 47nF cap, not even a DC.
On the other hand, I can see the 1.5V DC of the battery,
it looks flat with all timebases.


@voltz :
I have made the experiment with the receiver.
Now I remember that when I first saw that 800Hz ripple
in the tx output I did quite that, I grabbed my receiver
to see if I could hear the 800Hz tone in AM, and I heard
nothing, just the unmodulated carrier of my transceiver.
Unfortunatley, I trusted my scope more than my ears

I have tried the oscillator in three position of the cap
(three frequencies) and each time I zero-beated and heard
nothing different than what expected. The frequency of the
receiver and the one on the scope matched perfectly at zero
beat (zero beat is not quite perfect due to the rx SSB BW, but
I think I came close).  The scope showed a 1.3kHz modulation
with fully open cap, so it would have been quite audible.
I also tried in AM and heard nothing either, while a 1.3kHz
AM should be audible.


@Kalvin :
I haven't got a SDR dongle, so I attached a loop to the
second channel as suggested, and picked up a weak signal.
The loop signal behaves much like the osc. output, and
it *does* show the unwanted modulation exactly like the
first channel.  Is this result different than expected ?

[David Hess]

Here is another test you can do; hook your DSO to the output as usual, set the timebase to expand the waveform so you can see the individual RF cycles, and then set your DSO to infinite persistence or envelope mode to capture the variation if it exists.

[Ian.M]

Your RF probe circuit is no good.  You've got a capacitor in series with the diode with no other DC path so it cant rectify.
See http://www.qsl.net/g3oou/simplerfdetectors.html for probes that work.

[Kalvin]

@Kalvin :
I haven't got a SDR dongle, so I attached a loop to the second channel as suggested, and picked up a weak signal. The loop signal behaves much like the osc. output, and it *does* show the unwanted modulation exactly like the first channel.  Is this result different than expected ?

My thought was that the oscilloscope might induce or conduct some interference into the circuit which will the modulate the oscillator output. If you disconnect the oscilloscope from the circuit altogether and use only the loop antenna only to sniff the oscillator output do you still see the modulation? If you still see the modulation, I have no clue what is going on.

[Kalvin]

Looking at your oscilloscope time base, you are measuring MHz range signal with slow sample rate. Probably you are seeing aliasing due to slow sampling rate which shows as modulation due to timing difference between the oscilloscope and the oscillator?

[dmills]

Some digital scopes have a mode where they correctly interpolate sample values for screen display, as opposed to just joining the dots, try turning this on if you have it (It tends to make slow squareish waves look worse, but makes sines be actual sines on the display) ......

The issue is largely that the sample values lie ON the signal, but not necessarily on the PEAK values of the signal, so if your scope just joins the dots (Computationally cheap) you will see modulation like this as the scopes sample points slide past the signal peaks (It is particularly obvious when the signal frequency is **ALMOST** a rational fraction of the scopes sample rate). The cure is to place a resampler between the ADC and the screen display (in the digital domain) so that rather then joining the dots you are doing reconstruction, but that takes real processing power in the scope, so guess what gets dropped to hit a price point?

The other nasty digital scope trap is that they usually do not have anti alias filtering appropriate to the sample rate, especially when running at lower sample rates, you would think that always running the ADC flat out and doing the decimation in the digital domain would make this if not 'easy' (Nothing is at multi GHz rates) at least a reasonable thing to do.

You got to understand your instruments! And sometimes that old 465B really is the right tool for the job.

Regards, Dan.

[voltz]

@voltz :
I have made the experiment with the receiver.
Now I remember that when I first saw that 800Hz ripple
in the tx output I did quite that, I grabbed my receiver
to see if I could hear the 800Hz tone in AM, and I heard
nothing, just the unmodulated carrier of my transceiver.
Unfortunatley, I trusted my scope more than my ears

I have tried the oscillator in three position of the cap
(three frequencies) and each time I zero-beated and heard
nothing different than what expected. The frequency of the
receiver and the one on the scope matched perfectly at zero
beat (zero beat is not quite perfect due to the rx SSB BW, but
I think I came close).  The scope showed a 1.3kHz modulation
with fully open cap, so it would have been quite audible.
I also tried in AM and heard nothing either, while a 1.3kHz
AM should be audible.

Yup, this confirms it. There is no modulation taking place (as thought).. This is purely caused by scope artifact.

[David Hess]

The issue is largely that the sample values lie ON the signal, but not necessarily on the PEAK values of the signal, so if your scope just joins the dots (Computationally cheap) you will see modulation like this as the scopes sample points slide past the signal peaks (It is particularly obvious when the signal frequency is **ALMOST** a rational fraction of the scopes sample rate). The cure is to place a resampler between the ADC and the screen display (in the digital domain) so that rather then joining the dots you are doing reconstruction, but that takes real processing power in the scope, so guess what gets dropped to hit a price point?

This brings up a good point.  The sin(x)/x reconstruction filter should produce the correct result and it is not computationally difficult however the sin(x)/x reconstruction filter used in the Rigol DS1000Z series is known to be faulty producing a reconstruction which does *not* encompass the sample points.  Maybe that is what is causing the flawed display.

The other nasty digital scope trap is that they usually do not have anti alias filtering appropriate to the sample rate, especially when running at lower sample rates, you would think that always running the ADC flat out and doing the decimation in the digital domain would make this if not 'easy' (Nothing is at multi GHz rates) at least a reasonable thing to do.

The anti-aliasing filter if present should be a linear.  The waveform being displayed in this case shows AM modulation which no filter should be producing.

[mauri]

RF people,

here we are again with my homework


@David Hess :
I set the infinite persistence as suggested, pictures in attachments.
I didn't find an envelope mode in my scope though, so I have also tried
peak mode (also attached).


@Ian.M :
ok, I built the first RF probe described in the document you
kindly suggested.  I haven't got a Ge diode so I used an
ordinary Si one, after removing the clipping from the
oscillator its swing is large and we shouldn't have any
problem in detecting its signal.
I first hooked the RF probe to my receiver when listening
to a 1kHz beat whistle, to check if an AF-like signal gets
through it. The 1kHz signal does show up in the scope.
Then I hooked the RF probe to the oscillator, and the scope
did show a large DC, which seems to be quite flat with all
timebases (there is a small ripple in the DC, but after
magnifying it looks like the 4MHz signal itself, it's not
the AF-like thing we're looking for).


@Kalvin :

I have tried the loop alone as suggested, hovering over the
oscillator without touching it.
It is confirmed that the ghost modulation does appear in the
same way as when it was connected directly.


@dmills :
according to the scope manual there is a sin(x)/x interpolation,
which is On by default, but it is active only at 50nS timebase and
faster.  I have tried to set it on and off while the timebase
was 50nS and 20nS but I didn't notice any difference at all.

[Ian.M]

If you get a SOT23 dual series Schottky diode, you can make a very nice version of the second one.  Ideally use a RF Schottky diode, but an ordinary small signal one will work.  No PCB layout - scratch build it on a scrap of double sided copperclad with isolation gaps cut where you need them, insulate it with Kapton tape then cover it with copper foil soldered to the backside ground plane to form a screening can.

This won't be only time you need a sanity check on your scope!

[albert22]

This video may help with aliasing
https://www.tek.com/support/faqs/what-aliasing-and-how-do-i-detect-it-and-fix-it-my-oscilloscope

[David Hess]

@David Hess :
I set the infinite persistence as suggested, pictures in attachments.
I didn't find an envelope mode in my scope though, so I have also tried
peak mode (also attached).

Your photographs seem to show the normal peak-to-peak high frequency noise that infinite persistence, envelope, or peak detection will pick up.

So I think you have just been bitten by the DSO or perhaps Rigol DSO reconstruction bug; for whatever reason, the sample rate is insufficient to reconstruct the signal and you are instead seeing aliasing which is what it looks like.

[mauri]

RF commoners,

after reading your comments about aliasing I searched the
scope manual and it seems that the sampling rate decreases
as the timebase increases, while in a different mode called 'long-mem'
it decreases much less.
After doing some more tests, I found out that the snake-like
patterns you identified as aliasing starts appearing at 20us timebase
(which should correspond to 10MSa) while with such 'long-mem' on
it starts appearing at 1mS timebase (which should correspond to 20MSa).
So the artifact onset seems indeed due to the sampling rate.
Such long-mem mode is less affected, and it's not quite clear to me
while it is not the scope default mode, I suppose there is some drawback
around. As I understand based on your wise comments, with such cheap
instruments the choice is just as having the damn storage oscilloscope
to bite you on the left or right buttock.

I also found out that setting Peak Detect the artifact vanishes
entirely, at all timebases (irrelevant of long-mem or not).
Actually, the manual says "To Avoid signal aliasing, select Peak Detect
Acquisition".  That seems to be ideal for analyzing a RF with long
timebases, which is exactly what I was trying to do when I bumped
into the artifact, I was looking for the rise and decay time of my
tx output. But once again I still have to learn what the drawbacks
are with this mode, as dmills rightly pointed out I got to know
my instruments