My first attempting to design a RF noise generator

[neilhao]

It seems Zener could generate flat noise signal far beyond 6Ghz, this exceeded my expectation. I plan to design a flat gain LNA for this noise generator.

However, it seems RF noise generator is useless if it could not be used for noise figure measurement. It seems there is no feasible method to measure the ENR. Calibrating the device by using a calibrated RF generator should not be considered as a feasible method

Anyway, zener impressed me by generating wideband noise.

[Joel_Dunsmore]

You just take your PNA-X with option 029, calibrate it for a noise figure measurement, then hook your noise source to port 2 and select the "ENR" measurement trace, and Bob's your uncle, you have a calibrated, traceable, ENR measurement.  (OK, maybe not for the hobbyist, but that's how we do it). If you can measure the noise power in a known BW, you can compute the ENR directly from that as well, for example from a spectrum analyzer noise marker reading.  You convert the noise power with the noise-source turned on in dBm/Hz to equivalent Kelvin temperature, then use ENR=10*log10((K-290)/290).  To compensate for SA noise floor, measure it with a load, convert it to Kelvin and subtract it from the Kelvin temperature with the noise source on.

[neilhao]

If you can measure the noise power in a known BW, you can compute the ENR directly from that as well, for example from a spectrum analyzer noise marker reading.  You convert the noise power with the noise-source turned on in dBm/Hz to equivalent Kelvin temperature, then use ENR=10*log10((K-290)/290).  To compensate for SA noise floor, measure it with a load, convert it to Kelvin and subtract it from the Kelvin temperature with the noise source on.

Thanks Joe, this is definitely a feasible method  

I don't have access to PNA-X right now, I even did not know this powerful model before, thanks for the info again!

[RoV]

Hi neilhao, can you share something more about the design? I couldn't find it in your web page.
By looking at the small photo on the top right of your attachment, it seems much like the double-zener design appearing in IEEE Microwave and Wireless Components Letters, vol. 28, no. 4, April 2018 "A broadband Microwave Noise Generator Using Zener Diodes and a New Technique for Generating White Noise", Arslan and Yıldırım. Is it inspired by that?
Personally, I have tried zeners as noise generators, but I have been a bit disappointed by the low-pass character. Sure it depends on diode selection: small voltage diodes are based on zener effect and are not particularly noisy (low ENR), while above 6 V avalanche effect prevails and noise increases; however, avalanche tends to happen as small bursts of electrons followed by a relaxation time, which is an intrinsically low-pass phenomenon. Besides, all kinds of zener diodes tend to have large junctions in order to be able to dissipate power, so capacitance is not small, again an high frequency limitation. Best diodes are probably in the 6-12 V zone, minimal power. At least, this is my personal experience, but I'd like to hear other people thoughts.

[neilhao]

Hi neilhao, can you share something more about the design? I couldn't find it in your web page.
By looking at the small photo on the top right of your attachment, it seems much like the double-zener design appearing in IEEE Microwave and Wireless Components Letters, vol. 28, no. 4, April 2018 "A broadband Microwave Noise Generator Using Zener Diodes and a New Technique for Generating White Noise", Arslan and Yıldırım. Is it inspired by that?

Yes I was inspired by this article, I thought two zeners could be used for turning flatness noise. However in the experiments, it seem one zener was good enough for flatness. Based on my little experience,  using a current source to drive the zener and setting the current to be 10ma to 15ma could lead to the best result. The current point was just a little bit beyond the Zener point, higher current will suppress the noise.

I did not write the article for this design yet, since this is just the first attempting without LNA.

[G0HZU]

A fairly crude way to get an idea of the ENR is to make a couple of test gain blocks based on a MiniCircuits GALI-39 MMIC amplifier and use these with a spectrum analyser.

https://www.minicircuits.com/pages/s-params/GALI-39+_VIEW.pdf

This GALI-39 MMIC amp has fairly flat gain and low noise figure and very good return loss at the input and output up to 1300MHz if you look at the data or the graphs.

You can use the first GALI-39 as a fixed preamp for something like an old school HP spectrum analyser to get the analyser noise figure down to maybe 5dB. A typical classic HP analyser will have an MDS of about -150dBm in a 1Hz BW with 0dB attenuation. This means the analyser noise figure will be about 24dB. The preamp could improve this to about 5dB typical up to about 1GHz and it also offers the bonus of a decent input VSWR for the preamp. This can help to minimise uncertainty when you go on to start measuring the noise figure of test amplifiers. You can then try the noise floor method suggested by Joel to get a first approximation of ENR for the combined noise source + GALI-39 preamp and the analyser.

Then use the second GALI-39 as a test/calibration amplifier and see if your above estimate at ENR results in the correct gain and noise figure shown on the MiniCircuits datasheet for the GALI-39 after you complete a corrected Y Factor measurement on the second GALI-39. If you don't calculate the right gain and noise figure from the Y factor then tweak the ENR until it is about right across several frequency points.

I'd recommend making the noise source with a 50 ohm RF attenuator at the output (made with 0805 SMD resistors?) in order to improve the hot/cold source match to help reduce measurement uncertainty. I've made several noise sources over the years and in my opinion it is also worth it to try and find a NoiseCOM noise diode to base the noise source on. These diodes can be expensive but they output a high ENR and this means the hot/cold source impedance can be improved by adding a fixed attenuator that gets the ENR down to something in the range of 6dB to 13dB depending on your requirements.

The NoiseCOM diodes are also very stable and repeatable over temperature as well as providing a fairly flat response out past about 1GHz. Some low cost Schottky diodes can output high noise too although they won't be able to compete with a NoiseCOM diode over temperature or for flatness or for repeatability.

[0culus]

If you can measure the noise power in a known BW, you can compute the ENR directly from that as well, for example from a spectrum analyzer noise marker reading.  You convert the noise power with the noise-source turned on in dBm/Hz to equivalent Kelvin temperature, then use ENR=10*log10((K-290)/290).  To compensate for SA noise floor, measure it with a load, convert it to Kelvin and subtract it from the Kelvin temperature with the noise source on.

Thanks Joe, this is definitely a feasible method  

I don't have access to PNA-X right now, I even did not know this powerful model before, thanks for the info again!

All I can say is get ready to open your wallet. LOL 

[jadew]

I plan to design a flat gain LNA for this noise generator.

I have designed a flat gain amplifier for this exact purpose, be prepared for some hair pulling. I sell mine here: https://cojotech.com/3ghz-29dbm-flat-gain-amplifier

I also have the noise source ready (was ready even before the amplifier), but I'm still working on getting a proper ENR calibration setup going before I can start selling it. Without proper calibration it's close to useless, because it wouldn't give you any better results than measuring the noise directly.

I managed to get very good flatness on my mine, but that's the result of tinkering with the design for the past two years or so.



Depending on what your intended use is, flatness might not be an important factor, nor proper calibration. For example, it would work very well as a poor man's TG. If however you intend to use it for NF measurements like you mentioned, then you either need good flatness, or high resolution in your specification. The required accuracy of the characterization will also depend on the devices you intend to measure with it.

[G0HZU]

That NS + LNA combo looks impressively flat to 2.9GHz!

In my case I salvaged an old NC502 noise source and took the noise diode out of the metal can package. I then surface mounted it into a small screened PCB and it has a decent attenuator after it to help give a very low VSWR in hot/cold states.

I haven't revisited the ENR table for it for nearly two years and I didn't really try that hard at the time. It performs fairly well up to about 2GHz and the ENR is typically 13.5dB. The list below is the ENR table I use with it. I'm not sure how accurate the ENR is below 10MHz, I recall I did the measurements below 10MHz very hastily at a later date compared the the ENR above 10MHz.

1000000, 13.1200
2000000, 13.2000
5000000, 13.6500
10000000, 13.6000
20000000, 13.6000
100000000, 13.5500
200000000, 13.5000
300000000, 13.4000
500000000, 13.2000
700000000, 13.0000
850000000, 12.9000
1000000000, 12.8000
2000000000, 12.0000

[G0HZU]

It's also worth buying or making a decent screened test container once any formal amplifier testing takes place. I was very lucky to be able to buy a surplus screened enclosure from MCS Test a couple of years or so ago.

It really does make a difference, especially when testing LNAs designed on open PCBs. Once the door closes the screening is very impressive and the noise figure results are much more consistent.

I have the Ramsay STE2900 model here:
http://www.ramseyelectronics.com/product.php?pid=5

I paid just over £100 delivered for a used example from MCS Test. It was worth every penny. I've used this for lots of other applications as well including VCO/VFO design.

[jadew]

Yep, those NC diodes are quite nice. You got very good flatness too.

The enclosures look really nice, and it looks like the back panel can be modified/changed. The whole series appears to be available at very low prices on ebay. Thanks a lot for the tip!

[G0HZU]

Thanks. I just plotted the return loss for my homebrew 'NC520J' noise source. See the plot below. I did try quite hard to get the s11 to be good up to 2GHz. Note that NC520J is my own made up part number based on the diode being extracted from an old NC520 metal can noise source. The J is named after me

This s11 plot was measured using an E5071B VNA and a N4431B-60006 Ecal.

I also measured s11 for an Agilent 346A noise source at the same time. This noise source has about 5.7dB ENR. The s11 for this is even better as you would expect.

If I get time tonight I'll compare them when measuring a MMIC amplifier for gain and noise figure up to 2GHz. I haven't done this for a while. The NC520J noise source ENR may have drifted a bit since...

I just realised I should have colour coded the plots the other way around. Only the colours are wrong, the logo is correct. The hot response is the blue trace and cold is red in both cases.

[G0HZU]

I also made a really cheap noise source using a sub $1 Schottky diode about a year ago. If I can find it I'll post up the results. It wasn't in the same class as the NC520 but it was useable for some tasks.

[G0HZU]

Here's an image of the NC520J noise source. I'm afraid my mechanical skills are a bit lacking but I made this in a hurry many years ago on a simple PCB and it is tightly seam soldered all around. Functionally it is great, it just looks really ugly and cheap. One bonus the compact size and light weight. It is also compatible with the 346A in that it has a 28V interface and can be used with an Agilent noise figure analyser. I normally attach an SMA to BNC adaptor on the 28V (pulsed) side to give it the same BNC interface connection as the 346A.

Next to it is an equally tired looking 346A noise source from Agilent. This has seen a lot of use!

[jadew]

The small enclosure looks quite nice. I had mine machined, so I can only take credit for the design

I can attest however to the importance of a good enclosure. It's very easy to disturb a NF measurement.

[neilhao]

I thought how to calibrate the noise source as a hobbyist would be more funny and challenging, you may already have a plan for this .

For the LNA, According to the experience on CMOS, I think we could control the gain exactly as we expected by using matrix distributed amplifier.

I also have the noise source ready (was ready even before the amplifier), but I'm still working on getting a proper ENR calibration setup going before I can start selling it. Without proper calibration it's close to useless, because it wouldn't give you any better results than measuring the noise directly.

[srce]

based on the diode being extracted from an old NC520 metal can noise source... This noise source has about 5.7dB ENR.

What did you do to reduce the ENR?

[jadew]

based on the diode being extracted from an old NC520 metal can noise source... This noise source has about 5.7dB ENR.

What did you do to reduce the ENR?

There's an attenuator on the output.

[srce]

based on the diode being extracted from an old NC520 metal can noise source... This noise source has about 5.7dB ENR.

What did you do to reduce the ENR?

There's an attenuator on the output.

Was after specifics

[G0HZU]

On the NC520J noise source I added a resistive attenuator internally using SMD chip resistors and then fitted a decent 10dB SMA attenuator on the output. This gives it an ENR of about 13.6dB typically across HF-VHF.

I had a go at measuring the ENR of the decent Agilent 346A using a spectrum analyser and I did it two ways. Once with an external preamp and the analyser set to 10dB internal attenuation. The other was with the analyser's own internal preamp and the internal attenuation set to 0dB. I managed to get a similar result in both cases. The correct ENR at 100MHz is 5.73dB as stamped on the side of the 346A noise source.

The results are as below. The results were very close but this method heavily relies on the accuracy of the spectrum analyser. Even a tiny error will throw off the result and I had to run the full (automated) internal cal alignment before making the measurements. Very few analysers will be able to replicate this process without introducing lots of uncertainty. Even with my analyser there will be a fair bit of uncertainty.

I had to fiddle with an existing noise figure spreadsheet to get the one below where it works backwards to work out the ENR from the Y factor and the noise figure of the test gear. I hope I've done it all OK. It's really easy to mess up something like this when trying to correct for everything. So the spreadsheet results below may contain nuts.

The bonus of the Y factor method is it lets the performance of the noise source ENR minimise the uncertainty as it removes the need for the analyser to make accurate measurements of absolute noise power. It just has to make relative measurements to compute the Y factor. So doing it the other way around as below to try and get the ENR based on absolute noise measurements from a spectrum analyser is asking a lot of the spectrum analyser unless you are happy with something like +/-1dB accuracy for the calculated ENR.

I got close with the results below but this is partly because of the old Agilent PSA analyser I used. This was a very good spectrum analyser in its day, possibly the best available 15-20 years ago.

[RoV]

Zener or avalanche of some common low power cheap RF BJT BE / BE / BC can be used for RF noise also.

I tested several hi-freq BJTs: BFR93A, AT42035, BFP420. They do generate noise and quite wideband, but the level is relatively low, so it's difficult to get a good ENR after an attenuator of at least 15 dB, which is necessary to obtain a nice RL.
My impression is that hi-freq BJTs, having a very low BE reverse breakdown voltage (4 V or less), don't enter avalanche breakdown, but remain in zener mode, which is only moderately noisy. Besides, their base is fragile and current must be kept low.
Older classics like 2N2369 or 2N2222 have a higher breakdown voltage, perhaps >7 V, and probably generate a much larger noise, but certainly not flat in the GHz range.

[Joel_Dunsmore]

Flat to +-1 dB not an easy feat. I was tasked with the same job in 1983 and it took me 2 years to get it right also.  That design ended up in the HP 8347A (as well as the HP 8753 A,B,C and updated version in the D, and E).  I'm impressed with your good return loss. But wouldn't might higher P1dB (my goal was +24 at the amplifier output, which means burning a lot of linear power).  I'm curious what went into your design? Mine had 2 bipolar stages (HP custom BJTs) followed by 2 discrete FETs (NEC9002) stages; back then no such thing as RFIC.  Back then we also had very detailed service manuals for self-maintainers.  Here's a link to the 8347A: https://www.keysight.com/us/en/assets/9018-05192/user-manuals/9018-05192.pdf?success=true  Turns out I did the RF design but the rest of the amplifier motherboard, supply etc, was done by Chris Day as one of his first projects; now he's a director at Analog Devices.

[jadew]

Hi Joel. That's impressive and humbling.

My design is nowhere near as complicated as I imagine yours is, because like you mentioned, we now have access to nicer components. It's a two stage amplifier (two GaAs MMIC gain blocks), that already provide very good performance. The difficulty was in preserving that performance

[G0HZU]

I had a rummage through my fairly extensive stash of microwave diodes and quickly made a basic microstrip test board using a scalpel to cut the PCB shapes for a basic noise source.

So far the best diode gives about a 0.9dB noise slope across LF through 3GHz as in the plot below.

The PSA analyser has the preamp on so the noise floor is typically -169dBm/Hz when the input is terminated. The noise source is delivering about -157dBm/Hz via a 10dB SMA attenuator at its output. So plenty of ENR available even with the 10dB attenuator as you can see in the plot. I still have a few more diodes to try but it looks like my test PCB has a resonance because there is a sharp little dip in the response at just under 3GHz as you can see in the plot. This may be cause by a capacitor or the diode parasitics interacting with the PCB.

I think I'm going to have to try again with a nicer PCB material with a thinner dielectric and try some other caps. That little dip in the response can hopefully be made to vanish with a different PCB design and better components.


The plot below is 1dB/div and the internal attenuation is 0dB and the internal preamp is on. The analyser noise figure is just over 4dB down at UHF but it climbs a bit by 3GHz. If I turn the noise source off the noise level drops about 10-12dB.

[srce]

Would be interested to hear some part numbers for the diodes - the NoiseCom ones are pricey!