RF Noise source-"Whiter" noise by summing multiple sources?

[Xnke]

Looking at finishing up a noise source project for sweeping filters with an SDR. I know it's only got 39dB of real, usable range, but that's enough to tell me if I got the filter's passband in the right spot, and saves time running into town late at night to borrow the expensive gear from the test lab.

I'm looking for something flat out to 1.5Ghz or so, +/-1dB or so is great, 2db is still usable. BFR93A's seem to be the ticket, but I don't have any and can't seem to get any without going to china. They are cheap, though.

So, given that white noise is essentially random, two devices should not produce the same spectrum at any time. Am I crazy by thinking instead of a transistor I don't have, that abusing a plain ol' MMBT2222 transistor will probably get me a few hundred Mhz, but maybe using 2 or 3 as sources, summed, would make for a flatter, whiter noise spectrum?

Maybe using a zener and a transistor?

What can be done (if anything) to compensate for the 1/f noise that skyrockets under 5-10 Mhz? A slow rolling HPF that knocks the low end down?

Anyone have any experimental data to contribute? I am not well skilled in doing measurements on the R&S in the test lab, but I will gladly give out any results I get with this stuff.

[tggzzz]

A lot will depend on the frequencies and frequency resolution you are interested in. There's a lot of difference between 1.5MHz and 1.5GHz.

For "higher" frequencies you could consider using a 10MHz-18GHz DC block from minicircuits.

For "lower" frequencies you could consider using a PRBS.

If you are just looking at the noise on a spectrum analyser, is the 1/f noise visible when the noise source is disconnected? (It could be a normal artefact of the spectrum analyser)

[OwO]

When you sum two noise signals A and B, the resulting spectrum at any given frequency point is equal to sqrt(A**2 + B**2). Summing two sources with identical spectrum will produce the same spectrum.

[T3sl4co1l]

Noise statistics are independent of spectral output, given that the autocorrelation over time scales corresponds to dips in the spectrum.

That is, consider a noise source that doesn't change very quickly: if its value is sampled relatively infrequently, it will always be a new, random value, according to the PDF of the source.  The PDF can be whatever (e.g., even distribution if from an LFSR; normal if from most analog sources).

If sampled too frequently, the value won't have changed much from its previous value, thus there is a strong autocorrelation for short time scales.

This corresponds to the reduction in bandwidth (noise power at frequency) that comes with the slow rate of change.

Tim

[Xnke]

Nice to see 7TL around still, GZ's been deafeningly silent lately, Tim.

My thoughts were more device-to-device averaging, so if a Zener diode (5.6v or less) and an avalanche diode (6.8v or more) were to be put in series, would their characteristics tend to work toward less 1/f and a flatter output or not. More research tends to point toward the few dB of tilt or ripple come more from the amplifiers following the noise source than the source itself.

At this point I just need to build an amplifier chain and substitute noise sources until I find what I want.

[Gerhard_dk4xp]

I had good success with a 240 Ohm resistor's thermal noise (2 nV/rtHz) amplified
by 2 pcs. LMH6702 to generate some broadband noise to drown an oscillator in
phase noise. It is flat from AF to at least 100 MHz and I'm not  sure if one
really sees the amplified noise of the resistor or the own noise of the first LMH6702.
Doesn't really matter, I needed only comparison levels.

The noise source is the first 10 mm from the right side of the box; the reed relays
switch the level.
<  https://www.flickr.com/photos/137684711@N07/30908039486/in/album-72157662535945536/lightbox/   >

A true Zener diode (<5V) won't produce much noise; in fact, above 10 Hz or so, a BZX84-C2V7
is not worse than 2 nV/rtHz. In the same Flickr album, there are calibrated noise density plots from
1 Hz to 1 MHz.  1nV/rt Hz is the INPUT noise of an LT1028 or AD797, it takes some effort to better that.
Note how the noise density gets worse when we get close or even above 5V.

Before I had real HP346 broadband noise sources, I used the BE junction of a BFR93A RF transistor
for broadband noise. It zeners at 6.5V or so, and being part of a RF transistors, it probably has less
capacitance than a fat Zener diode.  It worked to 2 GHz with > 22 dB excess IIRC.
It may take an additional MAR-? or ERA-X to have enough power as a sweeper replacement.
Dynamic range won't be great since over all the bandwidth, you pump a lot of RF power into your
DUT and it may saturate earlier than expected.

regards, Gerhard