Tracking Generator Alternative: 1GHz Wideband Noise Source

[alterbaron]

This was a small project I've wanted to build for awhile, and finally got around to it this weekend. 

Background:

Ideal white noise contains an equal amount of power at all frequencies all at once. White noise is a very handy test signal to have available in the lab, since you can use it to quickly measure the (scalar) frequency response of devices. Just send the noise through your device and measure the transmitted power spectrum! Sort of reminiscent of a tracking generator.

Since I don't have a tracking generator, the noise source will be especially valuable for giving real-time feedback while messing around with filters and other RF stuff. (More precise measurements can be made using a frequency synthesizer + spectrum analyzer, but it's slow going in comparison.)

Noise source boards can be found on ebay for around $20 USD (EXAMPLE) but I figured I could save some money and end up with a more polished product by building my own. (Plus, it's more fun!)

Here's a picture of the finished noise source:



The enclosure is the Hammond 1455C801, $12 CAD on Digikey.
Including all components, the total cost was <$20 CAD.

Here's the horrible board hidden inside (sorry about the blurriness -- my camera doesn't focus properly on up-close stuff):



The source of the noise is the reverse-biased base-emitter junction of an RF BJT (NE85633, the black blob in the bottom left.) The signal then passes through three MMIC amplifiers (SNA-586), each providing around 16dB gain. A Pi attenuator right before the SMA edge-mount connector serves to improve matching and present a more controlled impedance to the last MMIC amplifier stage.

The top half of the board consists of a series diode (reverse power protection), power switch, power filtering, and an indicator LED.

This was a difficult board to fabricate at home! Easily over 100 vias, each drilled out using a regular drill with fragile 1/32" bits, and then filled with solid core wire and soldered. I really should have stitched all the ground planes together, but the idea seemed a whole lot less appealing once I realized what I had already gotten myself into. Next time I do a project like this, I'll just get the board made in a fab!

The thing takes +15V/200mA and gets pretty warm to the touch when it's running. The generated power spectrum is around 40dB above the noise floor of my spectrum analyzer, and is flat within 5dB or so up to 1GHz.



For a quick test, I used the noise source to measure the response of my homemade FM bandstop filter.

Test setup (the unmarked box is the FM bandstop filter):



1GHz span response:



50-150MHz response:



I can post schematics if anyone's interested, but I'll have to revise them a bit, since I made some changes while debugging the thing.

Any and all feedback is welcome! I'm still new to RF construction, so if you think I'm doing something the wrong way, please let me know. 

[daqq]

Nice project! And a surprisingly flat output.

I'd love to see the schematic.

I can't think of any criticisms, aside from the positioning of the switch - you may wish to add some distance between the yellow cables leading to the switch and the RF part - ideally position the switch right next to the power supply to avoid/minimize any possible pickup.

Try measuring the spectrum you get at the power input to see if anything is leaking out.

David

[tggzzz]

For those not fortunate enought to have a pukka spectrum analyser, with care an RTL SDR dongle can be used as a cheap 'n' cheerful alternative.

[alterbaron]

The schematic is attached.

Try measuring the spectrum you get at the power input to see if anything is leaking out.

I'll see if I can take a peek at this tonight. Should be interesting.

I'm thinking I might want to give this project another go and try to get more output power and bandwidth (ideally up to 2.5GHz). RF chokes between the supply and each amplifier should add a few dB of gain and improve isolation. And a professionally fabricated PCB should improve performance and stability. (I had to fiddle around with the board for awhile to prevent it from oscillating. Stitching the ground planes together would definitely help.)

For those not fortunate enought to have a pukka spectrum analyser, with care an RTL SDR dongle can be used as a cheap 'n' cheerful alternative.

I'll have to try this out with my RTLSDR and see how well it works. Stay posted!

[tggzzz]

Here are some simple results of using a noise source and RTL SDR dongle to measure filters' frequency response up to 1.5GHz. Not perfect, but very good value for money!
and

FFI, see https://entertaininghacks.wordpress.com/2015/08/10/measuring-rf-filters-with-a-homebrew-1-5ghz-scalar-network-analyser-costing-3248/

[alterbaron]

Here are some simple results of using a noise source and RTL SDR dongle to measure filters' frequency response up to 1.5GHz.

Those are some pretty nice results! The dynamic range looks to be about 30dB or so.

I gave this a shot tonight (using the handy RTLSDR Scanner software), and got the following:

20MHz to 1.5GHz sweep of the noise source:



This lines up quite nicely with what I was seeing on the spectrum analyzer.

Here's the FM bandstop filter analyzed with the noise source & RTLSDR, demonstrating a dynamic range of >35dB:



And a coax stub (great test case, btw):



Pretty good for a $30 scalar network analyzer. 

It would be nice, though, if there was some way to speed up data acquisition. The 1.5G span sweep took a few minutes to run, compared to a few 10s of milliseconds with a real analyzer.

[tggzzz]

Here are some simple results of using a noise source and RTL SDR dongle to measure filters' frequency response up to 1.5GHz.

Those are some pretty nice results! The dynamic range looks to be about 30dB or so.

Thanks!

My receiver has ~35dB dynamic range too, but I use attenuators to ensure the receiver isn't saturated. Subtracting the "calibration trace" also has the effect of reducing the dynamic range a little.

I gave this a shot tonight (using the handy RTLSDR Scanner software), and got the following:
20MHz to 1.5GHz sweep of the noise source:

This lines up quite nicely with what I was seeing on the spectrum analyzer.

The similarities and differences with my equivalent measurement (discussed here: https://entertaininghacks.wordpress.com/2015/06/24/rtl-sdr-dongles-anomalous-frequency-domain-response/ ) are interesting.



There are the same abrupt transitions below 300MHz, but I measure larger amplitude variations. The 15dB rolloff to 1.5GHz is similar, but your measurement has a knee at about 880MHz. The wiggle at 1.1GHz is missing from your measurements, which makes me wonder if it is an artefact of my noise source - I have no way of telling.

Here's the FM bandstop filter analyzed with the noise source & RTLSDR, demonstrating a dynamic range of >35dB:

When I make myself another 50/75ohm pad I'll measure the response of my "at800 4G filter" https://at800.tv/industry-trade/approved-filters/ Their low frequency edge is specified to be sharp and accurate.

And a coax stub (great test case, btw):

It is, not only because it is simple to build, but also its predictable response can be used for calibrating both frequency and time domain measurements.

Pretty good for a $30 scalar network analyzer. 
It would be nice, though, if there was some way to speed up data acquisition. The 1.5G span sweep took a few minutes to run, compared to a few 10s of milliseconds with a real analyzer.

The trick is to do a "dry run" sanity check at 10MHz intervals (or wider). That's sufficient to set levels and is surprisingly adequate for some unusual types of measurement (which I'll document soon).

[alterbaron]

The similarities and differences with my equivalent measurement (discussed here: https://entertaininghacks.wordpress.com/2015/06/24/rtl-sdr-dongles-anomalous-frequency-domain-response/ ) are interesting.

Hmm. Those variations below 300MHz are pretty dramatic. Hard to say how much of the roll-off is falling sensitivity and how much is due to the noise source's power distribution without calibration of the noise source with a spectrum analyzer.

In my case, the dip around 550MHz is due to the noise source, and the peak around 900MHz is a strong local radio signal that likely leaks a bit into the noise source enclosure and is amplified along with the noise. The spectrum of the noise source falls off by about 10dB by 1.5GHz (see below), whereas the dongle sees a fall-off of about 15dB. So (at least for my dongle), the sensitivity only seems to fall by 5dB or so at 1.5GHz.

Try measuring the spectrum you get at the power input to see if anything is leaking out.

Gave this a try this afternoon. Spectrum analyzer input connected to a DC block, connected by coax to the power supply outputs. The power supply is connected to the barrel-jack input of the noise source by a 2-conductor power cord. The analyzer picks up a lot of radio signals through the power supply. I turned on averaging to try to reduce the influence of fluctuating background radio signal intensity on the results.

The first image shows the received spectrum when the power supply is on, but the noise source is turned off. The second shows the spectrum when the noise source is turned on. I don't see any differences in the spectra that aren't easily attributed to local radio signals, so I don't think the noise source is injecting power into the line.

Off:



On:



tggzzz: I have a second RTLSDR sitting around. I thought it'd be interesting to have it measure the noise spectrum and compare the result with the first. Here it is (note, vertical scale is different from the earlier plot):



Pretty much an identical response. Both dongles are RTL820T2 based.

[TheSteve]

Both of my spectrum analyzers have tracking gens but this still looks like a fun project to build, please update us if you get a board made.

[tggzzz]

Both of my spectrum analyzers have tracking gens but this still looks like a fun project to build, please update us if you get a board made.

There's no need to get a board made; everything I used is available off-the-shelf. Even the connecting leads are ready-made.

[alterbaron]

Both of my spectrum analyzers have tracking gens but this still looks like a fun project to build, please update us if you get a board made.

Sure thing! I do plan on getting a small batch of boards made soon, and I think this would make a pretty good kit.
I'll keep you posted.

[TheSteve]

Both of my spectrum analyzers have tracking gens but this still looks like a fun project to build, please update us if you get a board made.

Sure thing! I do plan on getting a small batch of boards made soon, and I think this would make a pretty good kit.
I'll keep you posted.

Thanks! I'd much prefer a proper board then to do all the via stitching by hand that you did. I also make projects using hammond boxes quite often so the project just seems like a good fit.

[tggzzz]

Both of my spectrum analyzers have tracking gens but this still looks like a fun project to build, please update us if you get a board made.

Sure thing! I do plan on getting a small batch of boards made soon, and I think this would make a pretty good kit.
I'll keep you posted.

Thanks! I'd much prefer a proper board then to do all the via stitching by hand that you did. I also make projects using hammond boxes quite often so the project just seems like a good fit.

I just used off-the-shelf (ahem, off the EBay) SMA connectors. I wouldn't call that "stitching". Such simplicity makes it easy to change the attenuators to make maximum use of the limited dynamic range, which is especially important for my dongle-based TDR.

[G0HZU]

It's worth trying to find a real noise diode to use in something like this because you can get a much flatter noise source. I think it's worth the extra expense for the diode and you can sometimes buy these diodes quite cheaply anyway

I have a couple of homebrew noise sources here that use such diodes. One is a precision high level noise source I made in the 1990s that puts out about -85dBm/Hz up to 180MHz and is flat to about +/- 0.2dB across this full range and really flat across most of the range.

I also made a noise source for receiver testing that is flat to about +/- 0.3dB to just over 1GHz. I used a packaged noise source diode rate to 500MHz but I actually depackaged it to get it to give it the (above) flat response up to 1GHz.

[LaserSteve]

I ordered one of the EBAY units for 20$.  I'll take it to a friend's RF lab and see if I can get some real numbers.  My AVCOM PSA-37 is by no means flat enough for a meaningful measurement.  However I can measure total noise power very accurately.

Will let you guys know what I find..

Steve

[alterbaron]

I ordered one of the EBAY units for 20$.  I'll take it to a friend's RF lab and see if I can get some real numbers.  My AVCOM PSA-37 is by no means flat enough for a meaningful measurement.  However I can measure total noise power very accurately.

Will let you guys know what I find..

Awesome, I'm really curious to see how it performs.

I have a couple of homebrew noise sources here that use such diodes. One is a precision high level noise source I made in the 1990s that puts out about -85dBm/Hz up to 180MHz and is flat to about +/- 0.2dB across this full range and really flat across most of the range.

I also made a noise source for receiver testing that is flat to about +/- 0.3dB to just over 1GHz. I used a packaged noise source diode rate to 500MHz but I actually depackaged it to get it to give it the (above) flat response up to 1GHz.

That's a pretty damn good result! How do you measure the flatness of the noise source to that amount of accuracy?

I'll keep my eyes open for a noise diode. What's a typical price?
Also, it would be fun to compare and contrast the performance of different RF transistors as noise sources. Hmm . . .

[G0HZU]

I measured them on several Agilent PSA spectrum analysers at work. Both noise sources are designed to have a very good source match so this minimises measurement uncertainty. I actually see a flatter result on the analyser display than I quoted but I allowed a bit extra for various contributors to uncertainty etc. But it's really difficult to measure stuff accurately across such a wide bandwidth. The PSA is still one of the best analysers on the market in terms of response flatness vs frequency.

There was (allegedly) a time when you could get noise diodes for a special low price (a few dollars) from Noisecom if you were a ham/hobby experimenter. I/m not sure if this is still the case though... However, I salvaged mine from scrap equipment so got them all for free.

[alterbaron]

Update: Board layout nearly done, here's a preview:

[LukeW]

I put this together recently... haven't tested it yet.

The power supplies are on a separate board, one 30V supply for the avalanche bias (constant current) and a 12V supply for the MMICs.

[tggzzz]

I put this together recently... haven't tested it yet.

Translation: "I did work".

I will be interested in seeing the results of your work, and hearing the pain points and lessons you learned.

[cncjerry]

If you copied the circuit for the noise source off ebay, there was a person that reworked that circuit because it was drawing way to much current at the suggested power supply voltage.  If you google around you should be able to find the mods.  I have one, made some of the changes but the one in particular where you change the noise source diode I never got around to since I have an SA + TG.  I did use it in hot mode and I thought it worked pretty well.  If you have a scope with Math + FFT you can put the display in a long retention mode and get a pretty nice plot.

[alterbaron]

I put this together recently... haven't tested it yet.

The power supplies are on a separate board, one 30V supply for the avalanche bias (constant current) and a 12V supply for the MMICs.

Very pretty! Curious to see how it performs. Do you have a schematic?
What MMICs did you use, and what did you use for the noise source proper?

[AF6LJ]

Yes this looks really interesting.

[LukeW]

I put this together recently... haven't tested it yet.

The power supplies are on a separate board, one 30V supply for the avalanche bias (constant current) and a 12V supply for the MMICs.

Very pretty! Curious to see how it performs. Do you have a schematic?
What MMICs did you use, and what did you use for the noise source proper?

Basically as shown here:
http://www.edn.com/design/test-and-measurement/4358938/Zener-diode-and-MMICs-produce-true-broadband-noise

It's an ordinary 24V zener, not a dedicated noise diode. A constant-current source is used to bias the diode.

Three MiniCircuits MMICs are used as per the schematic.

I've added an 8V regulator to supply the MMICs, and changed their bias resistors accordingly, and moved the "high" voltage boost converter circuit for the diode off to a separate board.

This allows experimentation with different voltages, as well as keeping power dissipation in the bias resistors fairly small.

Hopefully the parasitic inductances etc are OK with 0603 passives used throughout... I didn't think it was worth bothering with 0402. (A couple of 0805s are used for MMIC biasing due to power constraints.)

[JeffEurope]

Hello!

I have a question. I try the HP346B noise source on the spectrum analyzer - as the "tracking source". After connecting to +28V pulse DC voltage, the noise level on spectrum analyzer has not changed. Is it normal?

Thank you for explanations!