The Designer's Guide Community
Forum
Welcome, Guest. Please Login or Register. Please follow the Forum guidelines.
Mar 28th, 2024, 2:40am
Pages: 1
Send Topic Print
Spectrum Analyzer-Phase Noise flattens out! (Read 8593 times)
iVenky
Senior Member
****
Offline



Posts: 101
Silicon Valley
Spectrum Analyzer-Phase Noise flattens out!
Aug 07th, 2017, 2:55pm
 
Hi,

I am testing a free-running quadrature VCO that I taped out. I observe a weird phase noise plot on the spectrum analyzer. Note that this is a free running oscillator, which means there is no PLL.

The phase noise flattens out close to 100 kHz instead of rising at 30 dB/dec. This trend continues even till 100 Hz. I don't know what could cause the spectrum analyzer to measure it wrong, as this is not possible. The spectrum analyzer doesn't show this behavior with other oscillators (like signal generators). Any suggestions would be greatly appreciated. Thanks.
Back to top
 
 
View Profile   IP Logged
Ken Kundert
Global Moderator
*****
Offline



Posts: 2384
Silicon Valley
Re: Spectrum Analyzer-Phase Noise flattens out!
Reply #1 - Aug 7th, 2017, 8:24pm
 
It would help if attached the phase noise plot.

Are you looking at the spectrum of the voltage, or are you passing the signal through a phase detector? If you are observing voltage, you are probably seeing the line width.

-Ken
Back to top
 
 
View Profile WWW   IP Logged
iVenky
Senior Member
****
Offline



Posts: 101
Silicon Valley
Re: Spectrum Analyzer-Phase Noise flattens out!
Reply #2 - Aug 7th, 2017, 10:52pm
 
Hi Ken,

I have attached the plot. I am using the phase noise feature of the spectrum analyzer and I believe it is implemented using the phase detector to measure the exact phase noise. I think the carrier is close to 4.5GHz (not visible in the image attached). Thanks a lot Smiley
Back to top
 

Webp_net-compress-image.jpg
View Profile   IP Logged
Ken Kundert
Global Moderator
*****
Offline



Posts: 2384
Silicon Valley
Re: Spectrum Analyzer-Phase Noise flattens out!
Reply #3 - Aug 7th, 2017, 11:11pm
 
The results are in dBc/Hz. That is normalized noise power, not phase noise. I think you are seeing the line width of your signal.

-Ken
Back to top
 
 
View Profile WWW   IP Logged
iVenky
Senior Member
****
Offline



Posts: 101
Silicon Valley
Re: Spectrum Analyzer-Phase Noise flattens out!
Reply #4 - Aug 7th, 2017, 11:35pm
 
Hi Ken,

Thanks for the reply. What I observe in most of the papers is that this dBc/Hz plot generally has a 30 dB/dec and then 20 dB/decade and then flat spectrum vs frequency, as the frequency increases. Going by your argument, it looks like the line width of my oscillator is large that I don't see the 30 dB/decade (though it's not exactly phase noise plot, I thought this slope was observable even in voltage noise). On the other hand, though the line width is bad, the phase noise at 1 MHz offset is decent and is close to -122 dBc/Hz (close to that number), which is contradictory?. In that case, do you think the phase noise measurement is valid? (I mean can it be considered a standard phase noise while reporting in papers for example?)
Back to top
 
 
View Profile   IP Logged
Ken Kundert
Global Moderator
*****
Offline



Posts: 2384
Silicon Valley
Re: Spectrum Analyzer-Phase Noise flattens out!
Reply #5 - Aug 8th, 2017, 12:14pm
 
You won't see the flicker noise (-30dB/dec slope) in the voltage if the corner frequency is less than the line width. So if I were you I would do a calculation to confirm that the line width you are seeing is expected for the noise you have, and if it is then you should stop expecting to see the flicker noise.

If you really want to see it, you can run a simulation with SpectreRF's PNoise analysis. They have added line width emulation since I worked on the simulator, but I think you can turn off the emulation, or perhaps you can just ignore it. In that case you should see the flicker noise if you simulate to a low enough frequency.

-Ken
Back to top
 
 
View Profile WWW   IP Logged
iVenky
Senior Member
****
Offline



Posts: 101
Silicon Valley
Re: Spectrum Analyzer-Phase Noise flattens out!
Reply #6 - Aug 8th, 2017, 1:16pm
 
Hi Ken,

Thank you very much for your reply. The default pss/pnoise set up shows the flicker noise -30 dB/dec slope without any linewidth effect. My question is regarding the Figure of Merit (FoM) that is generally defined for oscillators.

FOM= -L(f) + 20 log (fo/Δf) - 10 log (Power in mW).

Here L(f) is phase noise in dBc/Hz.  'fo' is the oscillation frequency and 'Δf' is the offset where the phase noise is measured. Intuitively, if the line width is large, it indicates that the phase noise is large. On the other hand, for the plot attached, FOM is large as Δf approaches low value, as the voltage noise spectrum is flat. In that case, I shouldn't consider the offset frequency below the linewidth, as it looks misleading?

Question 2: How do you generally verify if the linewidth of the oscillator matches what you observe on spectrum analyzer? You said it's possible to verify it using pss/pnoise? I found something in the options window of the PNOISE setup. Should I just check "yes" box on the lorentizian option? (attached the image)

Question 3: What do papers generally report when they say phase noise? (the real phase noise or voltage noise?). All papers talk about phase noise in dBc/Hz. In that case, they are actually measuring the voltage noise and not the phase noise exactly?
Back to top
 

Capture_014.PNG
View Profile   IP Logged
Ken Kundert
Global Moderator
*****
Offline



Posts: 2384
Silicon Valley
Re: Spectrum Analyzer-Phase Noise flattens out!
Reply #7 - Aug 11th, 2017, 10:27am
 
1. Yes, you should compute the FOM using frequencies outside the linewidth.
2. Yes, Lorenzian is another word for linewidth.
3. People use the normalized noise (dBc/Hz) and refer to it as the phase noise. This is generally okay because oscillator noise virtually all phase noise except in very unusual situations.

-Ken
Back to top
 
 
View Profile WWW   IP Logged
iVenky
Senior Member
****
Offline



Posts: 101
Silicon Valley
Re: Spectrum Analyzer-Phase Noise flattens out!
Reply #8 - Aug 11th, 2017, 12:15pm
 
Hi Ken,

Thank you very much for your reply. Since you mentioned I should consider the offset frequency above the linewidth, I have the following question.
Question 1:
I have included an image with phase noise plots for two oscillators. Oscillator 1 has smaller linewidth compared to Oscillator 2, but the phase noise value of Oscillator 1 at 1MHz is more than that of Oscillator 2. In this case, FoM is better for Oscillator 2 though the linewidth is large for that. Is it fair to say that the Oscillator 2 has lower phase noise than Oscillator 1?
Question 2:
What exactly determines the line width? I mean like the LC tank bandwidth (for an LC oscillator) or it is directly related to the phase noise? (though LC bandwidth or Q factor determines the phase noise).
The reason why I am asking this question is because for the plot obtained on the spectrum analyzer the line width is close to 100kHz though the phase noise at 1MHz is really good. I am wondering if it's reasonable to obtain such large linewidths.


Thanks!

Back to top
 

linewidth.PNG
View Profile   IP Logged
Ken Kundert
Global Moderator
*****
Offline



Posts: 2384
Silicon Valley
Re: Spectrum Analyzer-Phase Noise flattens out!
Reply #9 - Aug 11th, 2017, 3:06pm
 
You are misunderstanding linewidth. In phase noise the total power in the signal is independent of the noise level. The total area under the curve must equal the total power of the signal. If the phase noise increases, the entire curve cannot raise, that would violate the total power constraint. So what happens is the noise at low offset frequencies does not go to infinity, it levels off, and the corner frequency, known as the linewidth, adjusts to keep the total area under the curve constant. The higher the noise level, the higher the linewidth.

To understand this, consider a bicycle wheel. If the wheel rotated a degree or so then a spot on the tire might move by a centimeter or so. The rotation of the wheel corresponds to a shift in phase in the oscillator and the height of the spot corresponds to a variation in voltage in an oscillator. If you then increased the rotation by 10, so the wheel were rotating 10-20 degrees, the deviation of the height of the spot would also increase a factor of 10. But now increase by another factor of 10. Now the wheel is going through a half rotation. The deviation in the height of the spot would not go up another factor of 10 because it is constrained by the radius of the wheel. Now increase another factor of 10. Now the wheel would be going through something like 5 complete rotations, but the height of the spot would never increase beyond the radius of the wheel.

The same effect occurs in oscillators. The phase can drift without bound, but the voltage is constrained to fall within the amplitude of the signal. The Q of the oscillator provides inertia which limits the noise at high frequencies, so the largest noise is found at low frequencies. The result is that the amplitude of the noise at low frequencies flattens out as the offset frequency goes to zero.

You can have an infinite variation in phase if you wait forever, but you cannot have an infinite variation in voltage regardless of how long you wait. Thus, as the offset frequency goes to zero, the phase noise increases without bound but the voltage noise always flattens out. The corner frequency is the linewidth. It is adjusts with the noise level to keep the area under the curve equal to the total power of the signal. If you have a different oscillator with higher noise but lower linewidth, then it must have a larger amplitude.

This is why people do not use the absolute noise level of an oscillator as a figure of merit, they always use the relative noise level. It is the relative noise level above the linewidth that reveals the phase noise. However, the linewidth also represents another figure of merit for an oscillator. People designing lasers generally use linewidth as their figure of merit whereas RF designers use normalized noise.

-Ken
Back to top
 
 
View Profile WWW   IP Logged
iVenky
Senior Member
****
Offline



Posts: 101
Silicon Valley
Re: Spectrum Analyzer-Phase Noise flattens out!
Reply #10 - Aug 13th, 2017, 2:41pm
 
Hi Ken,

Thanks for the reply.  I understand the concept of linewidth, as V(t)= A cos(ωt+Φn(t)). Even though Φn(t) goes to infinity, V(t) can't.

But, it's possible to obtain phase noise plots which I attached in the previous answer, right? Please note that I have written phase noise in dBc/Hz so it's scaled with respect to the total power (or carrier power) already. In that case, can you say that Oscillator 2 has lower phase noise than Oscillator 1 though the line width is larger for Oscillator 2? Sorry if I misunderstood your answer. Thank you in advance.

Back to top
 
 
View Profile   IP Logged
Ken Kundert
Global Moderator
*****
Offline



Posts: 2384
Silicon Valley
Re: Spectrum Analyzer-Phase Noise flattens out!
Reply #11 - Aug 14th, 2017, 7:43pm
 
No, I do not believe your graph is physical.

-Ken
Back to top
 
« Last Edit: Aug 15th, 2017, 7:37pm by Ken Kundert »  
View Profile WWW   IP Logged
iVenky
Senior Member
****
Offline



Posts: 101
Silicon Valley
Re: Spectrum Analyzer-Phase Noise flattens out!
Reply #12 - Aug 16th, 2017, 1:51pm
 
Hi Ken,

Thanks for the reply. In that case, the phase noise of -120 dBc/Hz (approximately) that I observe on the spectrum analyzer at 1MHz is still valid, right, as it is after the corner frequency?

Regards,
Venkat
Back to top
 
 
View Profile   IP Logged
iVenky
Senior Member
****
Offline



Posts: 101
Silicon Valley
Re: Spectrum Analyzer-Phase Noise flattens out!
Reply #13 - Aug 19th, 2017, 11:50pm
 
Hi Ken,

An update. I measured the QVCO phase noise on another spectrum analyzer and found the source of the problem. It looks likes the first spectrum analyzer that I used was choosing the wrong center frequency for some unknown reason and there was a difference between the actual center frequency and the center frequency that the analyzer was considering because of which the phase noise plot was weird.

With the second spectrum analyzer, I wasn't facing this issue. In order to verify the above reasoning, I used the manual option that was available for entering the center frequency on the 2nd analyzer and found that I was able to replicate the phase noise plot shown in the beginning of this post by entering the wrong center frequency. I have attached the phase noise plot with the second analyzer.

The new problem that I now face is that the expected phase noise number is close to -128 dBc/Hz based on the pss simulations but I observe -113 dBc/Hz while measuring.

Question 1:

If I reduce the frequency from 5 GHz to 4 GHz using varactor, I was expecting a reduction in the phase noise from -113 dBc/Hz to some lower value. Instead, when I reduce the frequency, I see an increase in the phase noise to about -108 dBc/Hz at 4 GHz, which was unexpected based on simulation results.


Thanks!
Back to top
« Last Edit: Aug 20th, 2017, 7:55am by iVenky »  

fig1.PNG
View Profile   IP Logged
iVenky
Senior Member
****
Offline



Posts: 101
Silicon Valley
Re: Spectrum Analyzer-Phase Noise flattens out!
Reply #14 - Aug 24th, 2017, 7:32am
 
Hi Ken,

Is there a way to measure the exact phase noise instead of measuring the voltage noise on a spectrum analyzer for example?

Back to top
 
 
View Profile   IP Logged
Pages: 1
Send Topic Print
Copyright 2002-2024 Designer’s Guide Consulting, Inc. Designer’s Guide® is a registered trademark of Designer’s Guide Consulting, Inc. All rights reserved. Send comments or questions to editor@designers-guide.org. Consider submitting a paper or model.