Forums

PM demodulation

Started by kl31n November 17, 2005
For my final project at university I'm working on a special measurement
hardware.

I've a tone at a certain frequency. I want to analyze the phase noise
which is present in a bandwidth of 1 MHz around the tone.

For doing that, stated that I'm using an FPGA with an ADC and given
that the frequency of the tone is known, I demodulate the tone, getting
I and Q and then I take the atan of their fraction, having like this
the phase.

The method works fine and the results are as expected, but I'm under
the impression that could exist some other and more elegant method to
achieve even better results, like some sort of extimators and in
telling this I've in mind ESA for example.

In my case I cannot make the step FM demodulation to PM demodulation
because indeed I'm not retrieving a signal, but noise in a wide band
and so such a step would reduce and pollute the bandwith over which I'm
carrying me analisys out(the integrator would work as a lowpass filter
and also the removal of the mean would require a highpass filter).

Could anybody point me to any estimation method I could use(being able
to estimate the phase would also put me in the condition to use
bandpass sampling which doesn't upset me at all)? or even to any paper
that deals with such a problem? Also any insight on the problem itself
would be greatly appreciated.

Thanks in advance,

loom

"kl31n" <vishuddi@gmail.com> wrote in message 
news:1132236351.964146.134830@o13g2000cwo.googlegroups.com...
> For my final project at university I'm working on a special measurement > hardware. > > I've a tone at a certain frequency. I want to analyze the phase noise > which is present in a bandwidth of 1 MHz around the tone. > > For doing that, stated that I'm using an FPGA with an ADC and given > that the frequency of the tone is known, I demodulate the tone, getting > I and Q and then I take the atan of their fraction, having like this > the phase. > > The method works fine and the results are as expected, but I'm under > the impression that could exist some other and more elegant method to > achieve even better results, like some sort of extimators and in > telling this I've in mind ESA for example. > > In my case I cannot make the step FM demodulation to PM demodulation > because indeed I'm not retrieving a signal, but noise in a wide band > and so such a step would reduce and pollute the bandwith over which I'm > carrying me analisys out(the integrator would work as a lowpass filter > and also the removal of the mean would require a highpass filter). > > Could anybody point me to any estimation method I could use(being able > to estimate the phase would also put me in the condition to use > bandpass sampling which doesn't upset me at all)? or even to any paper > that deals with such a problem? Also any insight on the problem itself > would be greatly appreciated. >
Hi Loom - I'm not sure what your problem is , are you seeing more AM than you expected ? Are you seeing higher discrete phase noise spikes than expected ? Your mention of trying to do something with the energy separation algorithms of Morango et al on an input that has very little AM or FM by design is confusing me - can you give some typical numbers for your phase noise measurement set-up ( 60 dBc at 500 Hz offset type of thing)?. It looks, from your description, as though you have implemented a straight forward measurement set-up , it works well and you have some spare time to see what else you can do that might be useful - is this right? I'm hoping someone else here has a clearer idea of what you are trying to do because, just assuming that you are still interested in different methods of measuring phase noise on fairly clean carriers with no intentional modulation and are not really interested in trying to extract features that might indicate what is causing the phase noise, I can only ask whether you have looked at Allan Variance (despite http://horology.jpl.nasa.gov/papers/specambi.pdf#search='allan%20variance') or looked at physically beating a delayed version of the carrier with itself or at methods which might imporve the sensitivity of your present equipment (helping to automatically extract slowly changing differences in gain and DC offset on your I and Q channels for example) like not trying to mix down quite to 0 frequency on average. Best of Luck - Mike
I apologize for my quick presentation of the problem, I didn't want to
overload the occasional reader with a in depth exposition of what I'm
really doing because otherwise the lengths itself of the explanation
would have scared him/her away.

What I'm actually doing is a little bit different than just seeking an
estimation of the phase noise.

I've a precise RF oscillator outputting a tone which is in frequency
around 100MHz. This tone is beaten with a coherent signal to lower the
actual frequency(this is something I would like to get rid of in the
near future) and adapt it to the possibilities of my ADC and FPGA. The
"quality" I've on this tone is very high(-60dBc at 500Hz is pretty much
what I got) and the spurious freedom of my tone is ruled(ok, the
harmonics are there, but -50dBc under the first and around -70dBc more
the second) mostly by my ADC(-93dBc).

This tone, whose frequency is fixed and accurately known, is decimated
in its samplerate(the final resolution in time I need is 1us and I
provide around 0.3us in the end), beaten inside my FPGA with a
generated sine and cosine of the same frequency(yes, the clock of the
FPGA is coherent with the input signal itself) and filtered to bring
the bandwitdh of 1 MHz around the tone to baseband.
Here the I convert in hardware everything to floating point, that
division and atan to attain the phase.

In that bandwidth of 1MHz I'm not just looking at the noise itself, but
I'm searching for deterministic behaviours in the phase of the tone.
This means that I don't actually care about the measurement of the
white noise(which is what rules and as big as expected), but that I'm
averaging what I measure to discover the presence of coloured noise and
of deterministic signals which, if present, have to be somehow
corrected in the source.

In this context my approach already works and outputs what was
expected, but I've the impression there must be something more elegant
than the brute force my design represents. I haven't been able to find
a direct way to extract the phase of a tone once it's frequency is
known and this is actually what my question is about...is there any
estimators/operator I could use to strip the phase out of the tone
directly? The reference to ESA was purely as an example of the sort of
tool I could need, I know in this situation it's far from being a good
choice.

Thank you very much,

loom

"kl31n" <vishuddi@gmail.com> wrote in message 
news:1132324235.228765.44640@g14g2000cwa.googlegroups.com...
>I apologize for my quick presentation of the problem, I didn't want to > overload the occasional reader with a in depth exposition of what I'm > really doing because otherwise the lengths itself of the explanation > would have scared him/her away. > > What I'm actually doing is a little bit different than just seeking an > estimation of the phase noise. > > I've a precise RF oscillator outputting a tone which is in frequency > around 100MHz. This tone is beaten with a coherent signal to lower the > actual frequency(this is something I would like to get rid of in the > near future) and adapt it to the possibilities of my ADC and FPGA. The > "quality" I've on this tone is very high(-60dBc at 500Hz is pretty much > what I got) and the spurious freedom of my tone is ruled(ok, the > harmonics are there, but -50dBc under the first and around -70dBc more > the second) mostly by my ADC(-93dBc). > > This tone, whose frequency is fixed and accurately known, is decimated > in its samplerate(the final resolution in time I need is 1us and I > provide around 0.3us in the end), beaten inside my FPGA with a > generated sine and cosine of the same frequency(yes, the clock of the > FPGA is coherent with the input signal itself) and filtered to bring > the bandwitdh of 1 MHz around the tone to baseband. > Here the I convert in hardware everything to floating point, that > division and atan to attain the phase. > > In that bandwidth of 1MHz I'm not just looking at the noise itself, but > I'm searching for deterministic behaviours in the phase of the tone. > This means that I don't actually care about the measurement of the > white noise(which is what rules and as big as expected), but that I'm > averaging what I measure to discover the presence of coloured noise and > of deterministic signals which, if present, have to be somehow > corrected in the source. > > In this context my approach already works and outputs what was > expected, but I've the impression there must be something more elegant > than the brute force my design represents. I haven't been able to find > a direct way to extract the phase of a tone once it's frequency is > known and this is actually what my question is about...is there any > estimators/operator I could use to strip the phase out of the tone > directly? The reference to ESA was purely as an example of the sort of > tool I could need, I know in this situation it's far from being a good > choice. >
Hi again Loom - not sure that any of this is going to do you any good but here goes: 1/ United States Patent 4327258 might be of some marginal interest. 2/ From your description it sounds as though you already have the phase of the tone relative to an unmodulated tone derived from the same source and experiencing a constant relative delay compared with your path under test - what am I misunderstanding here? With respect to what do you want to determine phase? 3/ If you are looking for incidental modulation of this clean tone and hoping to infer the mechanisms for this modulation it will probably help greatly if you also display your phase modulation as frequency modulation and calculate the amplitude modulation too , cross-correlation between AM and PM or FM might reveal what you are looking for, if you also have a copy of the signals that you suspect are causing the incidental modulation then cross-correlation between these signals, their harmonics and the incidental modulation on your tone might also give clearer indications than just working on observed phase deviation at the output of your path under test. If you are exopecting to observe things like potentially intelligible AM/PM transfer from other carriers carrying coded signals then knowing the code and code phase will be very very helpful. Best of Luck - Mike