[time-nuts] Re: Phase noise with a lock-in amplifier.
javier at nebulosa.org
Sun Apr 17 14:30:14 EDT 2005
I think that there is a problem with the mixing of the oscillator with a
delayed version of itself. Suppose that the noise affecting the
oscillator were exclusively a square wave lets say at 1KHz rate, lets
say modulating the oscillator at +/- 1Hz. At any given instant, and
suppossing that the cable length is far less that to produce a 1ms delay
(1ms delay would be around 200Km of RG-58/U, if I remember welll), the
frequencies entering the mixer would be exactly the same, producing a DC
- you would only have glitches at a 1KHz rate, corresponding to the
instants when the two frequencies differ.
This can be translated to that for measuring the phase noise mixing the
signal with a delayed version of itself, you must have a delay at least
comparable to the lowest noise frequency you want to determine, in order
to have truly uncorrelated phase noise at the two mixer inputs.
David Kirkby wrote:
> Bill Hawkins wrote:
>> Dr. David Kirkby wrote,
>> "I was not thinking of impedance matching at all. If the mixer is
>> 50Ohm input (as most are) and the oscillator has a 50 Ohm output, the
>> cable length would have no effect on this at all."
>> The cable length has no effect on frequency (if the output device is
>> buffered with an amplifier).
> But what if a 10MHz (for example) oscillator is not perfect? The
> output is no longer a pure sine wave at 10MHz, but could be considered
> as an infinite number of oscillators, all of different frequencies,
> with amplitudes and phases that are essentially random. Those
> oscillators closer to 10MHz would on average have higher amplitudes,
> but all oscillators will have some amplitude.
> So if the two inputs are fed to a mixer have different path lengths,
> the inputs to the mixers will *not* be identical.
>> As I understand it, a phase detector that has both inputs fed by the
>> same frequency is only capable of measuring systematic errors, such as
>> may be caused by differences in the FET response times. Of course, it
>> will also measure differences in cable length.
> But the inputs will not be of the same frequency normally, due to the
> phase noise on the oscillator.
>> There was a time when obscure discoveries could be rediscovered when
>> the right opportunity arose. These days, with information at the speed
>> of the Internet, the process of natural selection works much faster.
>> The phase noise question has been around longer than the Internet, so
>> it seems to be true that the best solutions have been found.
> Well, it never hurts to consider new ideas. That might be silly,
> fatally flawed in some way.
> I'm 99.9% sure the mixer, with two different cable lengths from a
> 10MHz oscillator will produce two ranges of frequencies. One will be
> close to DC, but will extend up a few hundred kHz or so. The other
> will be close to 20MHz, but extend a few hundred kHz either side.
> I'm pretty sure if you low-pass filter the output of that mixer, you
> will see the oscillator phase noise shifted from 10MHz down to near DC.
> I think if you put an audio spectrum analyser on the output of the
> mixer, after the low-pass filter, you would see the phase noise of the
> In fact, would you not hear the oscillator phase noise on a speaker if
> you amplified the low-pass filtered signal from the mixer? (I'm
> ignoring the fact the noise on the audio system might be lower than
> the oscillator phase noise).
> I'm less sure, but it it possible a lock-in might be able to see this
> phase noise, by setting the reference to the offset you want.
> Question, what would you see on the lock-in if you used its internal
> oscillator as as reference and added to this (op-amp configured as an
> adder) the output from the mixer? The op-amp would measure its own
> reference amplitude, but you would have introduced noise on it from
> the 10MHz reference, so the lock-in should see that noise too.
> Perhaps I'll sketch out what I mean and put a circuit diagram.
> The Standford dual-phase lock-in has functions to measure the noise on
> the signal.
More information about the time-nuts