# How can you tell if a system is oscillating?

Discussion in 'Electronic Design' started by Dr. David Kirkby, Sep 15, 2004.

1. ### Dr. David KirkbyGuest

I have an electronic system consisting of 3 main parts, and whilst I
don't think it is, there is a possibility the sytem is oscillating, as
clearly a signal can be seen on an oscilloscope or spectrum analyser,
with no input whatever.

The system consists of:

1) A special photodetector called an avalanche photodiode (APD), which
detects light. The actual detector device has a gain of 30, and there
are a couple of other stages of amplfifiction on the PCB.

2) The output of this detector module is fed into an RF amplifer (500
MHz bandwidth)

3) The output of this amplifer is fed in to a band-pass filter, with a
centre frequency of 70 MHz, and a bandwidth of +/- 15 MHz. (i.e. 3 dB
points of 55 and 85 MHz).

4) There are further stages of amplification, and variable
attenuation. There are over 100 dB of RF amplification.

If I look at the output of the system on an oscilloscope, there is a
very noisy sine wave, with a period of about 15 ns (about 70 MHz). I
don't have the unit in front of me, but assuming the peak-to-peak
amplitude of the sine wave is 1 V, then there is about 300 mV p-pk of
noise, so the sine wave looks very noisy indeed.

If I look at the sytem on a spectrum analyser, I see a 'signal' that
is some 30 MHz wide, centred on 70 MHz.

There are two possibilities here.

1) The detector device generates wide-band noise (which I know it
does), which gets filtered, so no is narrow band. Then this is
amplified, so the output consists of noise passed through a filter,
which then takes on the shape of the filter.

This is what I think is happening, but there is another possibility
too.

2) The system, with a very high gain (over 100 dB) is acting as an
oscillator. Since the gain is much higher between 5 and 85 MHz due to
the filter, it will osciallate somewhere between there, where the loop
gain is > 1. Whether or not this would result in a clean oscillation,
rather than something that is noisy on a spectrum analyser, I don't
know. I suspect it would, which makes me think it is just the
amplified noise I am seeing, and no osciallations.

If I put into the optical detector a light source which is RF
modulated, the output RF modulation can be seen on the spectrum
analyser if the bandwidth scanned is small (20 kHz or less), but can't
be seen on the oscilloscope at all. It is probably burried beneth the
noise.

But how can I determine for sure if the system is oscillating or not?
I don't think it is, but the question arrises how does one prove this
one way or the other?

Dr. David Kirkby BSc MSc PhD CEng MIEE

2. ### Ken TaylorGuest

How well shielded is the RF? 30MHz is around what I'd expect for a satellite
transponder. Could you be picking up an external source? If you narrow down
your spec an on this 30MHz can you see anything that looks like a signal of
some sort?

Ken

3. ### Tim WescottGuest

I wouldn't count on oscillation being clean, although it usually is.

If it's oscillating there will be some critical gain where it will
suddenly drop out of oscillation. If it's amplifying noise reducing the
amplification will drop the output noise proportional to the decrease in
gain.

If you can reduce gain by removing stages or varying some component and
see the effect suddenly pop into and out of existence then it's
oscillating, otherwise it's noise.

4. ### Chuck OlsonGuest

Usually if it no longer obeys ohm's law, it's oscillating.

5. ### Rene TschaggelarGuest

You lost already. Gains of 40dB on one pcb are a limit.

You have to seperate the stages with faraday boxes around.
Such that the loop is not closed.

Rene

6. ### Mike MonettGuest

(Dr. David Kirkby)
wrote

[...]
Sounds like you are looking at wideband noise with a center frequency
of 70MHz. If the system were oscillating, there is nothing to limit
the gain and you would see a square wave at the output instead of a
sine wave. So the sine wave is a good sign

When you add a small modulated signal and can see it on the spectrum
analyzer, it means the amplifier is linear. If it were oscillating, it
would be nonlinear and you would no longer be able to see this small
signal in the output.

I use a dental tool that has a short J-shaped piece of stainless steel
with a sharp point mounted in a plastic holder. This has ~1pF stray
capacity and acts as a small antenna. I use it to probe sensitive
inputs to see if they break into oscillation when I touch them, or if
a parasitic oscillation stops.

With the large gain in your system, I would expect touching an input
with a probe would cause the amplifier to break into oscillation, and
you would see a square wave at the output. A small rf signal at the
input would no longer be visible on a spectrum analyzer.

So my conclusion is the amplifier is operating as it should, and you
are to be congratulated on achieving such a high gain at these
frequencies without oscillation. It would be a good idea to check the
gain and phase vs frequency with a network analyzer to verify there is
no unwanted peaking in the response. But you have to be very careful
how you apply a signal to the input.

Most single-shield coax cable starts leaking around -60 to -80dB, so
the coax cables to the network analyzer can introduce unwanted
coupling from the input to the output. You can check for this by
grasping the cables at different points and see if the response
changes. Try double-shielded coax if there is any coupling.

Mike Monett

7. ### Tim ShoppaGuest

I think you're just seeing amplified noise, too. You've got so much
gain that I'd be surprised if you weren't getting limiting in there too,
and when the oscillator limits you're likely to see ringing at a much
higher frequency (>>70 MHz).

The "70 MHz wave" you see on your scope probably only extends a couple
of cycles before disappearing into the noise, right? That's excactly
what you'd expect from noise fed into a filter with a center of 70MHz
and a Q of 4 or so.

If you see pip(s) on the spectrum analyzer output, that could be
oscillation. Complex active systems don't necessarily oscillate at
only one mode, and especially when there's a lot of noise you'll be
driven from one mode to another, but I'd expect if there were any relevant
modes you'd see it as a pip on the noise bump. If the modes are so
spread out and mushed around that they aren't pips, I'd contend that
they are indistinguishable from noise so it doesn't matter. (Indeed,
thermal noise is physically a whole bunch of nearly uncorrelated oscillators).

Tim.

8. ### Tim ShoppaGuest

Wow, better not tell that to the guys who've been making IF strips
with 80 dB of gain for the past half a century

Tim.

9. ### Rich GriseGuest

I agree with Tim Wescott about the gain threshold - that's very easy to
test, with a variable attenuator. Probe various parts of the circuit,
to see what the signal looks like, because
1. how noisy is the APD, and what's its output?
2. 100 dB gain is a _lot_
3. If there is some stray positive feedback, but not enough to oscillate,
you'd have even more gain, so just before it oscillates it will be
amplifying the system noise by as much as it can. But the spectrum
you've described sounds just like well-filtered noise, albeit it
does indicate some nonlinearity in the circuit, because it seems to
be modulating, and not just adding, but that could be more an artifact
of the filter's response.

What happens if you replace the sensor with a short? (or if there's
supposed to be an offset or bias current, you should simulate that,
but as quietly as possible.)

Good Luck!
Rich

10. ### Dr. David KirkbyGuest

The system is seeing 70 MHz, not 30, which is to be expected as the
filter is centred there.

The very front end device (a commerical unit) is not particulary well
shileded it must be said - it will be when boxed up. All amplifers are
well shielded, and for the most part, semi-rigid coax with a solid
outer is used connecting the various stages together. Where that is
not possible due to mechnanical reasons, good quality coax is used.

I have never seen anything that looks like a signal, unless I either
put the light into it, or wave a piece of coax nearby connecting to a
tracking generator. Clearly with a local source of transmitter
(tracking generator for the spectrum analyser), the device does detect
this.

11. ### Dr. David KirkbyGuest

Yes, that was my feeling too.
Good point. I have reduced the gain by 32 dB with a PIN diode
attenuator, and notice the output drop by the same amount, indicating
it is linear.

It all points to a stable system.

12. ### Dr. David KirkbyGuest

These are not on one PCB.
That is done. Each gain stage is a shielded enclosure, with good
qualtiy coax (usually semi-rigid with a solid outer) connecting the
stages.

13. ### John WoodgateGuest

I read in sci.electronics.design that Tim Shoppa <[email protected]
edge.com> wrote (in <>)
about 'How can you tell if a system is oscillating?', on Wed, 15 Sep
2004:
Before WW2, 110 kHz IF amplifiers in well-designed BC receivers achieved
approaching 80 dB, most of it in one stage!

14. ### Dr. David KirkbyGuest

But for a square wave one needs harmonics. And the gain at the second
harmonic (140 MHz) is << 100 dB, due to the attemiatom of the filter.
Seeing a signal does not reall prove it is linear, but the output is
proportional to the input, so I think that proves it is linear.
But there is nowhere really for the signal to leak out back to the
input, due to the use of good coax, and fully shielded boxes. So
whilst touching the input will no doubt cause it to pick up radio
signals, I would not expect it to osciallate, due to the shielding.

I don't have a network analyser. Could perhaps borrow one, but don't
I tend to use semi-rigid, or double-shielded. I avoid cheap coax.

Dr. David Kirkby

15. ### Dr. David KirkbyGuest

Limiting is a problem, if I run with
a) Little attenuation.
b) The noisy front end.

The final output stage is rated at +10 dBm (1 dB compression). I only
need +4 dBm, but was getting a lot less power (around -20 dBm) at the
fundamental frequency, when an RF modulated light source is put in to
the detector module. The RF modulated light source produces a signal
of around -90 dBm at the exact frequency of the modulation.

With the -90 dBm at 70 MHz from the noisy detector, the system has a
gain of only 70 dB, so produces -20 dBm. With the same -90 dBm from a
clean signal generator, the gain is 100 dB, giving an output of +10
dBm. So the noisy detector module is limiting the gain, there is no
doubt.

In fact, if I increase the increase the attenuation (reduce gain), the
output will increase. This I am convinced is the output stages
saturating. But if a signal generator is put into the system, it works
exactly as expected, in a linear fashion.

I think what is happening is that the power in the noise is
significant over the bandwidth of the filter, so the final amplifer
saturates - using a larger amplifer, rated at +20 dBm, gives more
apparent gain and more output.

I have a variable (0 to 31.5 dB) attenuator in the system. If the
attenuation is set too low, harmonics are indeed seen, as the system
saturates.

I intend substituing a narrower (crystal, rather than LC) filter, to
reduce the noise power, so the amplifer does not have to amplify all
the unwanted rubbish. The noise power should be proportional to
bandwidth.
Not sure what you mean there?
No, only harmonics are seen if overdriven.

Everyone is convincing me the system is not oscillating, but I am just
filtering noise.

16. ### Fred BartoliGuest

"Dr. David Kirkby" <> a
écrit dans le message de
You should be able to calculate the expected output noise. Is it near that
value ?

Also, you mention possible saturation :
I've built an IF amp, at a much lower freq though (100kHz), that have 145dB
gain. With 2nV/root(Hz) input referred noise and about 7Mhz (uncut) BW this
is 40µVpp and I "only" need 110dB to saturate the stages to +/-10V, leaving
no room for the signal. You have 4 times that BW and you'll probably have to
cut your BW, or gain to restore dynamique range.

I was also prepared to write that at 70MHz, 50R, 100dB you only need 0.5fF
(femtoF) to close the loop at unity gain. You seem to have done it right,
but 0.5fF is pretty low.

17. ### Rene TschaggelarGuest

At what frequency ?

I found broadband amplifiers to be tricky. If -80dB of the
output leaks to the input you're there. At 0 to 500MHz,
the phase condition is easily met at a frequency.

Rene

18. ### Rene TschaggelarGuest

And the shielded enclosures are bolted to a solid aluminum ?
Any reflections on the coax appear as shield currents and
propagate from there.

Rene

19. ### Mike MonettGuest

(Dr. David Kirkby) wrote
Ah, so the bandpass filter is after the 500MHz amplifier. You
described this clearly at the beginning.

In this case, if there were coupling from the 500MHz amp to the
input of the detector, the system could oscillate at any frequency
where the phase shift was an exact multiple of 360 degrees and the
loop gain was greater than unity.

The oscillation would be highly unlikely to be centered on the 70MHz
bandpass filter. In fact, it could be outside the filter bandpass
and you would not see any signal at the output!

If the system oscillated at some frequency within the bandpass of
the filter, you would see a sine wave, but it would be free of noise
since the output amplifier would be limiting.

So the noisy sine wave centered on 70MHz is a clear indication the
system is operating properly and not oscillating. Which is the
conclusion you have already reached. But we need to explain the
reason for the noisy sine wave.

I searched for examples of narrowband gaussian noise, but had no
luck. So I will try a hand-waving thought experiment.

Suppose you take a perfect 1KHz sine wave and display it on a scope,
which is triggered at the positive-going zero crossing. You are now
viewing a standard sine wave. If you go 1 millisecond downstream,
you see a second positive-going zero crossing with no jitter.

Now a real signal is actually a collection of noise signals in a
narrow bandwith. So lets open the bandwidth a little and see what
happens.

If we increase the bandwidth of the noise to 10Hz, you will start to
see some fuzz in the second zero crossing. At a bandwidth of 100Hz,
the entire sine wave will show considerable noise, nut not quite as
bad as what you observed at 70MHz.

However, if we increase the noise bandwidth of the signal to 428Hz,
it should look very much like what you observe at 70MHz. (from your
center frequency and bandwidth: (30/70)*1e3=428)

The first zero crossing is a tight point due to the trigger
threshold, but the signal at the second zero crossing is full of
noise. The sine wave deteoriates rapidly beyond this due to the
noise, and at some point is no longer recognizable.

Does this help explain how the input noise in your system can look
like a sine wave after the bandpass filter?
Mike