Episode 1542 I show the classic method using a spectrum analyzer Keysight phase noise paper: keysig.ht/xRh2h1 old HP paper: www.hparchive.com/seminar_note... Be a Patron: / imsaiguy
Excellent job introducing us to the PN! It's so clear now! LOL Now downloading (and hopefully studying) all the nice Application Notes pertaining PN from HP, Keysight, Analog and other great sources!
The HP App note 150 series gets into measuring noise and explains the components of the 2.5 dB and the 1.2 factor for the resolution bandwidth. My HP 8563e has a marker noise function that will read the noise power directly and takes the aforementioned into account. I think your 8591 has it as well. I learned how to make the noise measurements on older analog HP spectrum analyzers so still do the old fashioned way often, but the marker noise function is handy as a time saver. But we didn't usually use the spectrum analyzer for phase noise measurements because the typical oscillators we employed had such low phase noise that you can't measure it directly on the spectrum analyzer - same problem you showed. Good phase noise performance is really important though in a lot of digital communications circuits and there have been a number of times when either poor phase noise or poor oscillator isolation from mechanical shock or vibration has resulted in data link problems that were kind of hard to pin down. These can be very hard to measure in the field . . .
The windowing applied to the signal prior to taking the FFT (to constrain leakage from the start and end of the time sequence) is critical when you are interested in measuring very large dynamic ranges. The window function will result in a maximum sidelobe level plus a falloff per bin. If a Hann window is being applied, it has a pretty high first sidelobe but a falloff of 18 dB. Other windows are available with lower first sidelobe levels but slower falloff. It would be interesting to see if you got better results by reducing your RBW. The Signlent analyzer looks like it might be near its noise floor, but the Keysight unit might benefit...
Excellent video. I only understood the concept of phase noise. I didn't know anything about how it was actually measured. My business partner had a million dollar phase noise rack in his lab that he had spent a great deal of time characterizing, and I would just look at it. I was afraid to touch anything. Black magic. :)
I have been making phase noise measurements using a HP 70K series SA. For the most part I can get measurements that are comparable to or exceed the phase noise measurements listed on various signal sources. A HP 3335A @ 20MHz is listed @
the same autocorrelation method can be used for 1/f noise. There is a NIST paper where they had to run the autocorrelation for over a week to measure the 1/f spectrum of a NiCd battery
Another limitation of this measurement technique is that it also includes the amplitude noise in the measurement that amounts to additional error that's not phase noise.
Now, that I think about it a little more, you have similar problem as jammer signal in receivers. A strong signal next to a weak signal. To increase dynamic range tune out the carrier when measuring the offset, by changing the span. You probably could use notch filter on the carrier when measuring the offset power. Measuring anything with dynamic range of more the 90 dB is challenge. Another indirect way is to measure RMS jitter, with the help of GPSDO. Then you can calculate phase noise from jitter, using Shannon's law. I am interested on correlater circuit. I guess they also use it in CDMA communication to detect signals below noise level.
R&S have some great videos about phase-noise measurements on their channel. I can jsut remember during my time in a RF R&D lab (20 years), the curse of measuring phase noise at better than -174dBc at 1% offset. (Heck, we laughed at products with a -155dBc phase noise)…Not something you would want to do every day, unless you are in the design-phase. Had to use cystals for a notch filter and quite a bit of other circuitry, and lots of experimentation to come up (hopefully) with a correct answer. Somehow the cross-correlator solution never came up in our lab-discussions, but maybe our solution was good enough, or maybe that was just too expensive to implement, can’t remember…
thanks, I learned from your presentation. But since your filter is the same for the peak and offset, it will cancel out and you do not need to correct for it. Also, you need to correct for your environmental noise ( which depends on the temperature.), and wires ( which act as Ant.) , and return loss, etc.
I knew from the outset that those OCXOs are going to be significantly better than any run-of-the-mill master oscillator in any instrument that is not specifically designed and built for such measurements. You might as well be measuring a voltage reference with a DVOM. Look inside the instrument, hopefully you don't find a TCXO which are known to potentially have phase noise.
You just answered several questions I have had for years regarding PN measurements. For giggles did you try your HP8563 (??) SA? I have a HP8566B and a third part TG but with the TG my DR is limited to ~80dB. I'd be very interested in more PN vidoes if you are so inclined...THANKS es 73
You referred to the LOs as "IF" a couple times, but we know what you meant 🙂. The dual-LO 'Correlation' approach would provide some reduction, but there's always some RANDOM correlation between two random sources, and so you would not get a 'very high' rejection from the correlation algorithm alone (maybe 6dB?
So what you are really telling me is "Don't put too much faith in your instrument." OK, got it. I consider this a simple illustration of the kind of problem the people at NIST must love to work on, otherwise they would go crazy.
It may indeed be a bit higher, but likely still well below the measuring capabilities of either spectrum analyzer. In other words it might be something like -132dBc/Hz at 100Hz instead of -140, but both of those are quite far below the minimum phase noise measurement capabilities of all but a handful of very expensive spectrum analyzers.
Siglent has noise marker function, why not use it? Also the detector type - should it be positive peak or rather sample? When delta f for phase noise is above few khz can use the crystal oscilator noth filter to limit the carrier and decrease the noise floor by reducing attenuator value?
Always difficult to know exactly what were the measuring conditions from the datasheet. Even more difficult to get the same measurements, even for standard methods. My experience is that switching from Keysight to Rhode & Schwarz will give a different result... A R&D engineer from Analog Devices told me one day : take the datasheet with care, they are edited by the marketing guys 🤣
One thing that's bugging me: why should one use a cross correlation method ? The signal source analyzers are quite expensive. If the noise of the LO of the spectrum analyzer dominates the phase noise of DUT, can we not put a low noise amplifier in between the DUT and the spectrum analyzer, assuming we know the noise of the amplifier, and it has a flat band over the BW we are trying to measure ? This will scale the noise of the DUT, bring it higher than that of the spectrum analyzer, and we could back track to see the noise in the DUT. Thanks, will really appreciate the response.
th-cam.com/video/vWhFNgG_bl4/w-d-xo.html Gregory just released this one or two days after your great phase noise demostration. He shows the cross corrolations method using two Spectrum Analyzers.
At 1:12 you mention that the spur on the upper lobe is an artifact from the SA. How did you determine that and is that really expected for such a $$$ spectrum analyzer? I would have expected a $35k Keysight to not show any false spurs or artifacts. What's going on there?
it is common for spectrum analyzers to have spurs. some have a spur removal option to delete them but it slows down the sweep. Yes this is expensive and you might not expect it, but this is a small portable unit so not perfect. there are ways to test if it is a spur, setting attenuator settings for changing the freq of the signal. I had a second analyzer that did not show the spur
Excellent job introducing us to the PN! It's so clear now! LOL Now downloading (and hopefully studying) all the nice Application Notes pertaining PN from HP, Keysight, Analog and other great sources!
The HP App note 150 series gets into measuring noise and explains the components of the 2.5 dB and the 1.2 factor for the resolution bandwidth. My HP 8563e has a marker noise function that will read the noise power directly and takes the aforementioned into account. I think your 8591 has it as well.
I learned how to make the noise measurements on older analog HP spectrum analyzers so still do the old fashioned way often, but the marker noise function is handy as a time saver.
But we didn't usually use the spectrum analyzer for phase noise measurements because the typical oscillators we employed had such low phase noise that you can't measure it directly on the spectrum analyzer - same problem you showed.
Good phase noise performance is really important though in a lot of digital communications circuits and there have been a number of times when either poor phase noise or poor oscillator isolation from mechanical shock or vibration has resulted in data link problems that were kind of hard to pin down. These can be very hard to measure in the field . . .
The windowing applied to the signal prior to taking the FFT (to constrain leakage from the start and end of the time sequence) is critical when you are interested in measuring very large dynamic ranges. The window function will result in a maximum sidelobe level plus a falloff per bin. If a Hann window is being applied, it has a pretty high first sidelobe but a falloff of 18 dB. Other windows are available with lower first sidelobe levels but slower falloff.
It would be interesting to see if you got better results by reducing your RBW. The Signlent analyzer looks like it might be near its noise floor, but the Keysight unit might benefit...
Excellent video. I only understood the concept of phase noise. I didn't know anything about how it was actually measured. My business partner had a million dollar phase noise rack in his lab that he had spent a great deal of time characterizing, and I would just look at it. I was afraid to touch anything. Black magic. :)
I have been making phase noise measurements using a HP 70K series SA. For the most part I can get measurements that are comparable to or exceed the phase noise measurements listed on various signal sources. A HP 3335A @ 20MHz is listed @
the same autocorrelation method can be used for 1/f noise. There is a NIST paper where they had to run the autocorrelation for over a week to measure the 1/f spectrum of a NiCd battery
I always wondered about if you should measure the upper or lower side of the signal for measurements like this. Now I know! Thanks!
Thank you for the explanation
Wonderful video, thanks!!
Another limitation of this measurement technique is that it also includes the amplitude noise in the measurement that amounts to additional error that's not phase noise.
Now, that I think about it a little more, you have similar problem as jammer signal in receivers. A strong signal next to a weak signal. To increase dynamic range tune out the carrier when measuring the offset, by changing the span. You probably could use notch filter on the carrier when measuring the offset power.
Measuring anything with dynamic range of more the 90 dB is challenge.
Another indirect way is to measure RMS jitter, with the help of GPSDO. Then you can calculate phase noise from jitter, using Shannon's law.
I am interested on correlater circuit. I guess they also use it in CDMA communication to detect signals below noise level.
R&S have some great videos about phase-noise measurements on their channel. I can jsut remember during my time in a RF R&D lab (20 years), the curse of measuring phase noise at better than -174dBc at 1% offset. (Heck, we laughed at products with a -155dBc phase noise)…Not something you would want to do every day, unless you are in the design-phase. Had to use cystals for a notch filter and quite a bit of other circuitry, and lots of experimentation to come up (hopefully) with a correct answer. Somehow the cross-correlator solution never came up in our lab-discussions, but maybe our solution was good enough, or maybe that was just too expensive to implement, can’t remember…
thanks, I learned from your presentation.
But since your filter is the same for the peak and offset, it will cancel out and you do not need to correct for it.
Also, you need to correct for your environmental noise ( which depends on the temperature.), and wires ( which act as Ant.) , and return loss, etc.
Very interesting. Thanks.
Interesting. Thank you.
I knew from the outset that those OCXOs are going to be significantly better than any run-of-the-mill master oscillator in any instrument that is not specifically designed and built for such measurements. You might as well be measuring a voltage reference with a DVOM. Look inside the instrument, hopefully you don't find a TCXO which are known to potentially have phase noise.
Interesting!! Thanks!
You just answered several questions I have had for years regarding PN measurements. For giggles did you try your HP8563 (??) SA? I have a HP8566B and a third part TG but with the TG my DR is limited to ~80dB.
I'd be very interested in more PN vidoes if you are so inclined...THANKS es 73
I always learn something ;)
Great!
You referred to the LOs as "IF" a couple times, but we know what you meant 🙂. The dual-LO 'Correlation' approach would provide some reduction, but there's always some RANDOM correlation between two random sources, and so you would not get a 'very high' rejection from the correlation algorithm alone (maybe 6dB?
So what you are really telling me is "Don't put too much faith in your instrument." OK, got it.
I consider this a simple illustration of the kind of problem the people at NIST must love to work on, otherwise they would go crazy.
More specifically to know the limitations of your specific instruments.
Thank you for this informative video. I am also curious if the oscillator data sheet claims -140dbc but is actually higher.
It may indeed be a bit higher, but likely still well below the measuring capabilities of either spectrum analyzer. In other words it might be something like -132dBc/Hz at 100Hz instead of -140, but both of those are quite far below the minimum phase noise measurement capabilities of all but a handful of very expensive spectrum analyzers.
Siglent has noise marker function, why not use it? Also the detector type - should it be positive peak or rather sample?
When delta f for phase noise is above few khz can use the crystal oscilator noth filter to limit the carrier and decrease the noise floor by reducing attenuator value?
Common Mode Rejection - now in the frequency domain :-)
The master time source at NIST is a cesium fountain. I wonder how they characterize the phase noise of an oscillator that is a laser light beam.
Always difficult to know exactly what were the measuring conditions from the datasheet. Even more difficult to get the same measurements, even for standard methods.
My experience is that switching from Keysight to Rhode & Schwarz will give a different result...
A R&D engineer from Analog Devices told me one day : take the datasheet with care, they are edited by the marketing guys 🤣
What is the phase noise performance of the FieldFox?
I was waiting for you to measure on your 8591E for a comparison. Did you try that of line by any chance using the HP math for analog SA?
I agree. He's leaving our garage capabilities behind with his new $$$$ equipment. Some perspective would be appreciated.
One thing that's bugging me: why should one use a cross correlation method ? The signal source analyzers are quite expensive.
If the noise of the LO of the spectrum analyzer dominates the phase noise of DUT, can we not put a low noise amplifier in between the DUT and the spectrum analyzer, assuming we know the noise of the amplifier, and it has a flat band over the BW we are trying to measure ? This will scale the noise of the DUT, bring it higher than that of the spectrum analyzer, and we could back track to see the noise in the DUT.
Thanks, will really appreciate the response.
I have no idea, I suggest reading the paper I link in the description
I make 194 bpm drill'n'bass. How do I optimise my phase?
click track
🌟🌷🌟
th-cam.com/video/vWhFNgG_bl4/w-d-xo.html
Gregory just released this one or two days after your great phase noise demostration. He shows the cross corrolations method using two Spectrum Analyzers.
At 1:12 you mention that the spur on the upper lobe is an artifact from the SA. How did you determine that and is that really expected for such a $$$ spectrum analyzer? I would have expected a $35k Keysight to not show any false spurs or artifacts. What's going on there?
it is common for spectrum analyzers to have spurs. some have a spur removal option to delete them but it slows down the sweep. Yes this is expensive and you might not expect it, but this is a small portable unit so not perfect. there are ways to test if it is a spur, setting attenuator settings for changing the freq of the signal. I had a second analyzer that did not show the spur