phase noise analyzer and phase noise measurement

[wanxiangwan]

phase noise analyzer and phase noise measurement:
https://www.qsl.net/bg6khc/

[MegaVolt]

1. DMTD?
2. Noise Allan deviation graph?
3. Price?

[jhenderson0107]

I purchased one of these for $780 to supplement existing E4440A spectrum analyzer -based phase noise measurements.  I've only used it for a short while, but it works well.  The developer is helpful and responsive via email. 

The biggest limitation I've encountered thus far is that its analog front-end is bandwidth limited to  < 1 GHz.  Even signals above 200 MHz are significantly attenuated, mitigating performance. 

The device can be used to measure frequency differences (examples provided), but I don't use that feature. 

[bozidarms]

Does analyzer have independent input and reference frequencies?

[jhenderson0107]

Yes, I've attached a basic block diagram. 

[bozidarms]

Thanks jhenderson0107

If I understood right, exist only matlab scripts  for post-data manipulation
or raw data.
Is it possible to use a Miles TimeLab software, or any other software?

[jhenderson0107]

To my knowledge, Miles' Timelab software cannot be used directly.  This is unfortunate since Timelab provides an excellent user interface and analysis display.  However, since Timelab can accept a TCP stream, it may be possible to write a utility which consumes the file data produced by the PN2060A and relay it to Timelab.  But, I need to investigate the Timelab TCP packet format to verify. 

For phase noise measurements, the PN2060A ships with a Windows x86/x64 executable which is minimalistic but fast, accurate and easy to use. 

[jhenderson0107]

The PN2060A developer vendor has released server software which allows use of the instrument in conjunction with Timelab.  It works well and provides a significant functionality upgrade. 

[nctnico]

Any word on noise floor where it comes to time difference measurements (Allan deviation)? And how about Linux software?

[jhenderson0107]

Both the vendor -supplied software and Timelab are Windows-only, to my knowledge.  The typical phase noise floor of the instrument is shown in the manual, but the time deviation floor is not.  Attached is a typical time deviation plot from my SMA100B. 

[nctnico]

Problem is that I don't run Windows natively. For a slow process like Timelab that isn't much of a problem but feeding a USB3 datastream into a VM looks like asking for trouble. I prefer to avoid Windows based software as Windows doesn't fit my computer usage.

Anyway, the time deviation graphs don't look that stellar compared to a standalone DMTD unit I have under development.

[cncjerry]

the developer referenced Andrew Holme's work.  I built Andrew's system and it works great.   I got lots of help from Andrew and Adrian Rus who went over the top to assist me. 

I needed to use batteries for the runs in addition to low noise LDO regulators.  Even with clean power supplies (and the LT3045 regulators) I still had some noise.  So I'm surprised the performance is so good on this unit with USB power. Andrew's system only uses USB for control of the FPGA and then 1Gig ethernet for the data feeding the PC plotting system. 

The expense is in the references, filters before the ADCs, the DAC oscillator and the shielding and positioning of the equipment in the lab. I had the best results using Dewalt 20v batteries stepped down with the LT regulators.  Adrian made some mods to his ADC board that worked better for him.  The other factor that came into play was the interaction of the ADC clock and the DUT as well as the reference oscillators.  Adrian developed a spreadsheet that calculated the interaction of the reference, DUT and ADC clock but the bottom line is you want the harmonics and mixing to not produce a spur within the measurement range.  I think that is an adequate explanation.   the last reference oscillators I used were at 26Mhz made by Wenzel with Minicircuits filters at 27Mhz.  Then I remember getting designs from someone for non-reflective filters.  Most of this is on Andrew's site that which I highly recommend.

Adrian then sent me some other oscillator links that I was yet to implement though I acquired them.  Given you don't need (or want) references at 10Mhz in many cases, you can move onto other odd frequencies like I did at 26Mhz.  But I think the 26Mhz interacted with my ADC clock, so I then acquired one that is phased locked to 10Mhz but generates 118Mhz and change.  You also have to setup the ADC to accept the correct clock (in Andrew's implementation) and it must be clean with low jitter.  The limit on the ADC Andrew used was 125Mhz. 

One other point, in order to get the maximum performance you need to bring the DUT and references up to the ADC limits.  I ended up spending a lot of money on QB188 amps and attenuators. I used fixed attenuators as well as variable controlled by my HP70K system.

The bottom line is that to get to a clean -185 noise floor takes a lot of screwing around.  But considering you can do it for substantially less money (under $2,000 for Andrew's) implementation assuming you get the SP605 spartan6 FPGA used, and it compares to units costing 10x that number, I thought it was a good deal.  Andrew's unit didn't do ADEV work or interface directly with Timelab the last time I updated it.  But you can dump the data into Timelab, matlab, Stable32, etc.  I use another entire system for ADEV work anyway.

I never thought I could get that level of performance for such a modest price.  You can get it going for under $750 USD and then build it up from there.

Jerry

[alan.bain]

Did you use the Spartan 6 eval board as in AH's write up?

I got the impression that supply chain problems (or rather fab capacity) had effectively forced Spartan 6 out of the market but possibly this was only supply affecting production of new designs rather than eval boards.

I read AH's excellent write up and found the approach very interesting, especially the cross-correlation trick with two independent input chains to improve the measurement noise floor. Part of me likes the idea of a 3048 with all those nice HP boxes, but probably the digital approach will get better results.

[cncjerry]

Did you use the Spartan 6 eval board as in AH's write up?

I got the impression that supply chain problems (or rather fab capacity) had effectively forced Spartan 6 out of the market but possibly this was only supply affecting production of new designs rather than eval boards.

I read AH's excellent write up and found the approach very interesting, especially the cross-correlation trick with two independent input chains to improve the measurement noise floor. Part of me likes the idea of a 3048 with all those nice HP boxes, but probably the digital approach will get better results.

The 3048 has many challenges but I know people who struggled through it.  I guess the software is the biggest problem.  But I've never seen a noise floor approaching that of Andrew's on a 3048.  Maybe I've just not seen it, but it is considered an old technology that can't compete with the cross-correlation approach.  But it is cool.

Use this: https://www.ebay.com/itm/204033950500

Jerry

One other point, I purchased the kitted-out reference oscillators from a 3048 system and they barely made it to -155 at 100k, IIRC. 

I checked my plots and the A5 reference was -155 or so and the A6 was -165 with my noise floor at -185 so the analyzer was the limit.  What I don't understand is all the pictures I've seen on the web have the floor for the 3048 at around -170 if not higher yet I have a friend who built two systems, using the HP references, and he reported deeper than that, down around -180.  I would look at replacing the references as you can get Wenzel sprinter series (500's and 501's) at 10Mhz that are 10db better if not 20.  You have to check as the Wenzel 501 series are standard parts and you can get their PN specs.  the 500 series you have to test and I have some that have very good performance for less than $50USD per, not as good as their ULN series but 1/10th the price.

One last comment, it doesn't take much equipment to measure PN at one frequency as outlined in the HP memory project here:

https://www.hpmemoryproject.org/technics/bench/3048/pn_measrmt_single.htm

Granted you aren't going to get a nice plot without some GPIB automation but still, a couple amps, a low pass filter or two, a reference and Bob's your uncle.

Jerry

[jhenderson0107]

The biggest limitation I've encountered thus far is that its analog front-end is bandwidth limited to  < 1 GHz....

Following up on my earlier comment: 

I've eliminated the above bandwidth restriction restriction by assembling two 10 MHz Wenzel Streamline oscillators, a Wenzel LNAF amplifier-low pass filter and M/ACOM M8TC mixer into a small chassis.  The oscillators provide very good low-phase noise references to the PN2060A.  The clock output of my R&S generator is used as the LO to the mixer+LNAF, which down-converts the DUT to 10 MHz without any significant phase noise performance reduction.  Consequently, the bandwidth of the system is now extended to 3.4 GHz, which meets my current needs. 

[Gerhard_dk4xp]

Did you use the Spartan 6 eval board as in AH's write up?

I got the impression that supply chain problems (or rather fab capacity) had effectively forced Spartan 6 out of the market but possibly this was only supply affecting production of new designs rather than eval boards.

I read AH's excellent write up and found the approach very interesting, especially the cross-correlation trick with two independent input chains to improve the measurement noise floor. Part of me likes the idea of a 3048 with all those nice HP boxes, but probably the digital approach will get better results.

Recently, Spartan6 has been guaranteed at least to 2030 by AMD/Xilinx.

I saw 2 engineers working with a E5052B some years ago and one of them
said with a happy face: "Do you remember how ***** it used to be with the 3048??"

Gerhard
(still working on that stereo & dual conversion down converter from Xband to Timepod.
Slow progress b/c of daytime job.)

[wanxiangwan]

vendor has a new board.

[Jackob]

Do you have measurement results already - did the noise floor improve?

Br, Jackob

[wanxiangwan]

Do you have measurement results already - did the noise floor improve?

Br, Jackob

Vendor needs some time for coding, keep on watch websites.

Br, Wxw

[hpw]

the vendor has a new board.

Yes, read on his web as Development of PN2060B: hopefully the noise floor can be lower than -185dBc (or maybe -190dBc).

More interesting would be @1Hz and @0.1Hz. Also input sensitivity as levels are IMHO not given.

Hp

[Oscill8]

Can someone here point me to the Andrew Holmes documentation?
Thanks!

[3apw]

Can someone here point me to the Andrew Holmes documentation?
Thanks!

http://www.aholme.co.uk/PhaseNoise/Main.htm

[drew23]

Hi all.

I am new to phase noise measurement but bought a PN2060A to try it. Awesome device.

I have made a blog post about it.

I was in contact with Yanjun today. He is currently on summer holiday. However, he did say he has some of the PN2060A for US$480, two of the PN2060B for US$780 and is working on PN2060C/D. To quote, "Currently I am focused on the development of PN2060C/D. Hopefully, It can be finished within 1-1.5 months. For PN2060C/D, one is with 2 ADC chips, and the other one is with 4 ADC chips."

Regards Drew VK4ZXI

http://vk4zxi.blogspot.com/2023/07/an-economical-way-to-measure-phase.html

[glenenglish]

Hi Drew
After my 3048 died. I have  built my own PNA from FPGA bits and boards I have lying around  (I do that for a living) , it worked OK , I just fed raw samples to python and that did the rest.
However , compared to my mass of bits and leads,  the PN2060 is just too easy and compact so I just bought one.

Apparently this is PCB Rev 1.1 so we'll see how it goes.

glen

[alan.bain]

I'd be intrested to hear what you used for the front end given the AD DC1525A-A eval board used by Andy Holme is hard to find now (the FPGA part is more straightforward to substitute).  I've got a rather non-standard 3048 setup (HP3563A as DSA not 3561) which works well, although at times the DSA is painfully slow. At times I would be interested in a direct digital comparison option!

[hpw]

I am new to phase noise measurement but bought a PN2060A to try it. Awesome device.

working on PN2060C/D. To quote, "Currently I am focused on the development of PN2060C/D. Hopefully, It can be finished within 1-1.5 months. For PN2060C/D, one is with 2 ADC chips, and the other one is with 4 ADC chips."

As PN2060C with now V1.1 PCB.

What I am missing:

-  the additional HW parts as required to start with a simple measurement, So to organize as DIY  At the end of this parts, measurement & validation could start but not before.

- The USB 3 connections are not isolated and draw high currents. No battery solution (galvanic isolation) seen so far.

- the real given required SW parts as still DIY and SW maintenance is unknown ??

- this means, this looks IMHO not a ready to go solution.

or I am wrong?

Hp

[sky2city]

Hi all,
Does someone tested phase noise on 800MHz directly by PN2060C?
I want to by one, but I don’t know how the performance on 800MHz.  How much it will be about  performance degrade.

Thank a lot.

[jhenderson0107]

Based on the published specifications (https://qsl.net/bg6khc/pn2060c_phase_noise_analyzer.htm), direct phase noise measurements of the PN2060C are limited to <= 200 MHz.  To extend measurement range beyond 200 MHz, an external mixer and high-quality LO are required. 

[tjadwy]

Anyone can share some test results? 

[hpw]

Anyone can share some test results? 

Best to ask the developer....

and keep in mind as long as it is NOT a TURN KEY solution given or as an optional, even as for 200MHz and above...

that's a pitta, as do not like to search for ALL required parts for simple below 200MHz operation. 

So hidden unknown costs to deal with.

Just my 2 cents

[drew23]

Hi.

I have the earlier PN2060A. It works like the revised one, besides having a lower noise floor.

The PN2060C is pretty straightforward to use. You need a couple of reference OCXOs, but these don't need to be very expensive, that is the point of the cross-correlation technique. The frequency of the reference oscillators depends on the DUT. I have a pair of 10 MHz and a pair of 100 MHz OCXOs. Filters and amplifiers are needed but they too are dictated by the DUT. Battery packs are used to reduce correlated noise.

I have a blog post from a couple of months ago on measuring in the sub-200 MHz range.

I am currently finishing a post on using it with inexpensive downconverters to extend the range into micro GHz. The instrument seems to work well.

Regards Drew VK4ZXI

https://vk4zxi.blogspot.com/2023/07/an-economical-way-to-measure-phase.html

https://vk4zxi.blogspot.com/2024/02/pn2060a-ghz-phase-noise-measurement.html

[KE5FX]

From https://vk4zxi.blogspot.com/2024/02/pn2060a-ghz-phase-noise-measurement.html : "I don't know why AN3899 uses a single reference oscillator for the analyser as two would seem better for cross-correlation."

Since the downconverters are being used to measure a microwave source, the noise at each IF output is determined by the downconversion LO for that channel or the DUT, whichever is worse.  We can assume that the DUT will ultimately dominate, given that the LOs will tend to average out over time, and given that the microwave DUT is usually going to be much noisier than a decent 100 MHz reference OCXO. 

So when the IF is measured by the 53100A, the analyzer typically needs only a single reference OCXO, not a pair.

If you were using a pair of dividers, which allow the DUT noise to actually be improved by the downconversion setup, the story might be different.  E.g. in figure 5.  If dual references hadn't been used there, the measurement error at offsets >1 kHz would have been even higher.

[Gerhard_dk4xp]

Hi, John,

As you know, I'm building a down converter to be used pairwise
to measure X-Band signals with my timepod. It is dual conversion,
1st IF is 900-928MHz with SAWs. It has 2 bands: nervous DC - 3 GHz and
3 GHz-15 or 20 GHz, depending on the mixers and synths  that
happen to be available $TODAY.

Expected DUTS are PLLs or DDS&mixers.
Downconverter REF will be a 100 MHz oven locked to 10 MHz house ref
with adf4002. In a prev. version, the 100->10 MHz divider contaminated
the 100 MHz with its counting spikes, therefore the ADF4002. I hope
the dirt will stay in the PLL chip.

In the pic:

top row from the left:
100 *3 * 3 MHz == 900 MHz 2ndLO, SPI-Interface, High band down
converter, Low noise LT3042 regulator with external power trans.

middle row:
 10-> 100 MHz PLL, 100 MHz xtal oven,  LMX2594/95 synthesizer

bottom row:
5-28 MHz output/2nd down mixer, PIN mux, low band down converter SW..3GHz


Timepod REF will be 1, 4, 8 or 16 MTI-260, sloooowly locked to GPS and
Wilkinsoned together since I'm #inputs-challenged with the timepod.
The guy who sold the Lucent GPS receivers on ebay some years ago
made me an offer for 20 redundancy oven units that I could not ignore :-)
Not removed with hammer and sickle after a life time on a Chinese
telecom tower but still shrink-wrapped.

Do you think I committed systematic errors or oversights?

Cheers, Gerhard,  DK4XP

[KE5FX]

That sounds like a conservative approach, assuming that (1) you use a very low loop BW to lock the 100 MHz OCXO with the ADF4002 and that (2) you use two of them.  Spend a lot of time in ADISimPLL to make sure you are happy with the loop response, as ADF4002s aren't super quiet in-band.  You need both downconverter channels to run at the same frequency for the TimePod's benefit, and that's a good idea in any case, but you want them to behave independently at the Fourier offsets of interest.

As noted above, there is likely to be no need for an exotic reference source at the end of the chain (meaning, at the TimePod REF IN jack).  If you can find an HF OCXO with a white-noise floor near -170, that would probably be fine.

[drew23]

Hi.

The OCXO reference for each local oscillator is also the reference for the analyser. I thought this would reduce the number of noise sources by two.

In the downconverter, I think the main source of noise is the VCO in the PLL which is maybe 20 dBc/Hz worse than the OCXO. However, the two downconveters are not correlated and much of the noise is removed by cross-correlation.

The local oscillator references I have are 10 MHz, just on what was available. The local oscillator signal generators are only about US$160. I am not aware of any other microwave signal generator that is locked to an OCXO at anywhere near the price.

The cost of the whole microwave phase noise measurement system is about US$1000, an order of magnitude cheaper than anything else and within the budget of an amateur like me.

Regards Drew VK4ZXI

[hpw]

The local oscillator references I have are 10 MHz, just on what was available. The local oscillator signal generators are only about US$160. I am not aware of any other microwave signal generator that is locked to an OCXO at anywhere near the price.

The cost of the whole microwave phase noise measurement system is about US$1000, an order of magnitude cheaper than anything else and within the budget of an amateur like me.

As I wrote, I like to have a Turn Key solution (as finished with may different performance options) and NOT a DIY project more. As looks OXCO, mixer/splitter and any other parts to purchase.

This means, install the required SW & connect the HW and press the measurements at the SW. As getting what the seller promises. Nothing less or more.

Hp

[drew23]

Hi HP

The only phase noise analyser that is completely self-contained for GHz measurements is the R&S FSPN. They cost upwards of US$100k depending on top frequency.

Even the FSPN provides for external references and has low-noise power supplies for external amplifiers, depending on the measurement.

The other phase noise analysers are similar to the PN2060C, requiring external oscillators, mixers, amplifiers and power supplies. The other analysers are around US$10k, whereas the PN2060C is US$780.

Regards Drew VK4ZXI

https://qsl.net/bg6khc/pn2060c_phase_noise_analyzer.htm

[tautech]

Hi HP

The only phase noise analyser that is completely self-contained for GHz measurements is the R&S FSPN. They cost upwards of US$100k depending on top frequency.

SSA5000A models with PN option provide for this too starting under $20k.
https://int.siglent.com/products-overview/ssa5000a/
Latest App note:
https://int.siglent.com/resource-detail/65/

[switchabl]

That is a spectrum analyzer, it does not perform cross-correlation phase noise analysis. And it doesn't even have particularly low PN. The difference in sensitivity is on the order of 50 dB (!). I know many value the support you provide for Siglent products here. But I do wish that sales people everywhere would just learn to say "sorry, we don't have that" instead of automatically trying to sell you something that is obviously unsuitable.

The Keysight E505xA series is another turn-key microwave phase noise measurement system. It is also extremely expensive. The cheapest option would probably be the older E5052A. I'd expect a used 7 GHz model to be somewhere in the same ballpark as a new 53100A.

[3apw]

The only phase noise analyser that is completely self-contained for GHz measurements is the R&S FSPN.

•   AnaPico   
https://www.anapico.com/products/phase-noise-analyzers/apph-models-up-to-40-ghz/
APPH  - Fully Integrated Cross-Correlation Phase Noise Analyzer from 1 MHz to 7 / 26 / 40 GHz

•   Berkeley Nucleonics
https://www.berkeleynucleonics.com/phase-noise-testers
Model 7000 Series :  to 7 / 26 / 40 /50 /65 GHz  Phase Noise Tester - Signal Source Analyzer

•   Holzworth Instrumentation
https://holzworth.com/products/phase-noise-analyzers
HA7062 SERIES PHASE NOISE ANALYZERS : to 6 / 26 / 40 GHz

•   Keysight Technologies
N5511A : to 40 GHz
E505XA  : to 26 / 44 / 54 GHz

•   NoiseXT
https://noisext.com/solutions/phase-noise-analyzer/
PN9000 : to  1.8 / 6 / 40 / 140 GHz
NXA-6  : to  6 / 26 / 50 GHz
DCNTS  : to 50 GHz

•   Rohde&Schwarz   
https://www.rohde-schwarz.com/products/test-and-measurement/phase-noise-analyzers_254899.html
FSWP : to 8/26.5/50 GHz
FSPN :  to 8 / 26.5 GHz
FSUP :  to 8 / 26 / 50 GHz

[tjadwy]

There is also a comparison among them .

https://kvg-gmbh.de/en/product/ocxo-beispiel/

[KE5FX]

Nice rock!  That's about the right noise floor for the 53100A, but I wonder where that artifact came from.  Normally you see that sort of thing if the DUT or one of the references is still warming up.  More typically you'd see something like this when measuring something below the unit's own spec floor:



(noise/spur responses near 1 Hz and 2 kHz are real)

[Gerhard_dk4xp]

https://synergymwave.com/articles/2013/04/full_article.pdf

[KE5FX]

The spurs shown at 4:30 do not originate in the PN test set, but are typical of the various CSAC models.  They aren't especially-well optimized for spectral purity, nor were they meant to be.  The major design goals for those parts are long-term stability and low power consumption.

E.g. if you do a Google image search for csac phase noise, you'll see a few references like this one that depict similar spurs with different instrumentation.  Unexpected spurious responses on the 53100A are often remedied by letting the measurement run a bit longer, but that's not an issue with the test shown in the video.

(Obligatory disclaimer: I work on the 53100A and related stuff, but I don't speak for Microchip.)

Edit: this was in response to a post that seems to have been deleted, which mentioned the spurious responses visible on a chip-scale atomic clock measurement in a 53100A demo video. 



Cool video, I hadn't seen it before.

[tjadwy]

E.g. if you do a Google image search for csac phase noise, you'll see a few references like this one.  Unexpected spurious responses on the 53100A are often remedied by letting the measurement run a bit longer, but that's not an issue with the test shown in the video.

Thanks for the information. The misleading information has been remove.  So, it will not misguide others.

[KE5FX]

No worries!

[sky2city]

From https://vk4zxi.blogspot.com/2024/02/pn2060a-ghz-phase-noise-measurement.html : "I don't know why AN3899 uses a single reference oscillator for the analyser as two would seem better for cross-correlation."

Since the downconverters are being used to measure a microwave source, the noise at each IF output is determined by the downconversion LO for that channel or the DUT, whichever is worse.  We can assume that the DUT will ultimately dominate, given that the LOs will tend to average out over time, and given that the microwave DUT is usually going to be much noisier than a decent 100 MHz reference OCXO. 

So when the IF is measured by the 53100A, the analyzer typically needs only a single reference OCXO, not a pair.

If you were using a pair of dividers, which allow the DUT noise to actually be improved by the downconversion setup, the story might be different.  E.g. in figure 5.  If dual references hadn't been used there, the measurement error at offsets >1 kHz would have been even higher.

Another question is raised in my mind,
I don't know why AN3899 didn’t uses image filter between DUT and mixer.
The white noise floor would be 3dB higher than the actual in this case.

[KE5FX]

Ideally, yes, you'd like some additional filtering, but it gets expensive to provide that in a multichannel test set with broadband coverage.  (Even the E5053A doesn't bother with that, which surprised me as well when I first noticed it.)  The use of image-reject mixers like the HMC8191 is the next best thing.

Also, this setup isn't intended for measurements near the kTB floor.  Besides the expense of maintaining low noise figures and high image rejection, there are other complications that arise when making cross-correlated measurements at lower noise levels.  The solutions in the app note are intended to be a good fit for the devices actually being measured in the app note.  For more demanding measurements, different approaches are called for. 

(Just to be clear, I'm the author of AN3899.)