Noise generator using device self-noise

[anotherjim]

Well, the DR55's only plus was to be the cheapest programmable DM. You had to be Ninja quick to change the batteries without losing your programs. The sound was thin and lacked one vital one for the time - claps. We used to cover Blue Monday with one and did the claps with an E&MM Synwave. Ahhh, memories...

What I have at the moment is essentially the Rene Schmitz Noise module that I posted above. Run it all off 5V. Use the DR55 noise transistor instead of the reverse BE. Fit a pot in the collector load resistance so the DC point can be tweaked and DC couple to the +in of the first op-amp. The pot is tweaked for a centred noise wave out of the second amp. This arrangement allows for LF noise too for if you want Rumble.
It might be a bit better to use a 6v regulator, but I have an idea for it as an appendage to a PT2399 based idea so it will be good to be able to use 5v.

[bool]

Blue Monday

Dssss Dssss Dssss

[Rob Strand]

Well, the DR55's only plus was to be the cheapest programmable DM. ... The sound was thin and lacked one vital one for the time - claps. We used to cover Blue Monday with one and did the claps with an E&MM Synwave. Ahhh, memories...

Youtube came to the rescue.   That sound, it's definitely an 80's sound.  Miles apart from the Roland R8.   We used the R8 when the real drummer couldn't make it to Jams.  We used to call it "Roland the drummer" (as in the guy's name).

You had to be Ninja quick to change the batteries without losing your programs.

Ha, the tricks people find

What I have at the moment is essentially the Rene Schmitz Noise module that I posted above. Run it all off 5V. Use the DR55 noise transistor instead of the reverse BE. Fit a pot in the collector load resistance so the DC point can be tweaked and DC couple to the +in of the first op-amp. The pot is tweaked for a centred noise wave out of the second amp.

Sounds like a winner.  The 5V reg will put an end to any supply related issues.
So did you only use first transistor of the DR55?

It might be a bit better to use a 6v regulator, but I have an idea for it as an appendage to a PT2399 based idea so it will be good to be able to use 5v.

5V is fine.  The noise is about 15% lower at 5V.  It's not as if the level is calibrated.

[anotherjim]

This is the breadboarded affair at the moment.
Only one transistor, but it's either the first or second of the DR55 design depending on how you look at it 

I've no idea if...
It's optimal.
It's repeatable without selecting things.
How much noise is silicon and how much is carbon.

I do know...
It is stable.
It has low supply ripple (I'm deliberately using an unregulated transformer based 9V DC supply).
The noise sounds full. Without any filtering, it sounds like a military jet engine. There is some HF content so I popped C5 in.
Tricky to know if the bias is optimal for maximum clean before clipping without scoping the output.

Although I'm not aiming for a piece of test gear, I might box it up standalone. It could be handy for checking out EQ and tone controls.

[Rob Strand]

Looks like a winner.  FYI electro caps are reversed.

It's repeatable without selecting things.

How much noise is silicon and how much is carbon.

The noise level and quality should be quite repeatable once the bias is set.  From what I can see the thing that will cause variation in the level is the transistor gain - even if the bias is set "correctly" for say mid-rail.  Higher gain transistors produce more noise.

The noise is 99% silicon.

Although I'm not aiming for a piece of test gear,

The circuit has one appealing property for test instruments.  There's very little temperature dependency.  The temperature dependence comes from beta (and minor IC shifts).  The square-root dependency on beta weakens the temperature dependency.

A first order calculation gives the output noise from the first stage as,

    Vonoise ~  (Vcc/2) sqrt(q beta / (2 IC))     [V rms / root Hz]

where, q = electron charge, beta is the transistor gain IC is the collector current, Vcc = supply voltage.
(The formula will estimate 10 to 20% higher than reality for low Vcc.  It assumes the collector is biased to VCC/2 and the connection has an unbypassed feedback bias.   For other bias arrangements the noise will double provided the bias resistors are large compared to rpi.)

You can see lower collector currents give higher noise.   Unfortunately the collector current would need to be decreased by a factor of 10000! to get rid of one gain stage which is asking a bit much; and the noise bandwidth will suffer.

It makes me wonder if a 2N7000 can be used in a similar circuit.   Those things tend to have more noise than BJT's.   IIRC the noise is excess noise and more 1/f like so it's not going to be as nice as the BJT.  Not sure where the noise bandwidth will end-up.

[anotherjim]

Duh, yes C3 & C4 drawn reverse. Built that way, any cap leakage would kill it. I went up to 10uF for the hell of it since earlier circuits would become unstable with response going that low. As it happens, those caps could be 680nF film for a 20Hz bottom.
I did wonder about using a MOSFET, and as it happens the 2N7000 is something I have.

[Rob Strand]

I went up to 10uF for the hell of it since earlier circuits would become unstable with response going that low.

I wonder what the mechanism was.  The inductance of the 10uF caps or perhaps power supply related?

As it happens, those caps could be 680nF film for a 20Hz bottom.

Probably not a bad thing for audio.  For a test instrument I should imagine you could flatten out the 1/f noise.

I'm going to build one to see how stable it is over time.    Higher voltages are going to be more stable.  The formula I gave has some deliberate  simplifications but in reality there is a VBE term in there which will vary with temperature.   

Having the variable collector resistor like in your schematic should help with repeatability as the high gain transistor will operate at higher currents.  In fact the formula has a (Beta/IC) term which is IB and since IB = (VCC/2 - VBE)/RBB,   where RBB is the 2M,  setting the collector voltage makes that a constant.   In other words by setting the collector bias to VCC/2 you should do a very good job of setting the noise output to a constant value for any transistor - which is quite a cool thing!

If we connected the RBB to VCC instead of the collector it reduces the VBE effect by a factor of two.  And since we are adjusting the collector bias voltage anyway we might trade feedback bias for better stability;  I'd need to look at what wins in the end.

One thing though it's going to be vastly more consistent than zeners and reverse VBE junctions.   I've never bothered to look at the temperature stability of those since I'm usually happy just to get a good output.     The only down side for the shot noise version is the small noise voltages which means more attention to shielding around that first stage.

Thanks a lot for posting the topic.   It's been quite interesting digging into this stuff again.

[bool]

You could actually band-limit at the first opamp (like using a 1u as C2) and have it more broadband at second opamp (w/o removing the 22pF in nfb).

[anotherjim]

You could actually band-limit at the first opamp (like using a 1u as C2) and have it more broadband at second opamp (w/o removing the 22pF in nfb).

Yes, upper and lower limits have 2 poles each available for shaping the spectrum. As non-inverting, there is always x1 full band getting thru, but in this case, it's going to negligible.

I do worry the DC bias could be the weak spot. Although I can centre the output with the bias adjust, the collector voltage is below Vcc/2, doesn't want to get much higher than 2v and adjustment also interacts with the noise amplitude.  So I might just be lucky with the way the op-amp offset error is handling that to the final stage. Possibly the last amp should have a DC bias adjust mixed into the -in.

[Rob Strand]

I do worry the DC bias could be the weak spot. Although I can centre the output with the bias adjust, the collector voltage is below Vcc/2, doesn't want to get much higher than 2v and adjustment also interacts with the noise amplitude.

It's the feedback bias!   So maybe it's better to use say 4M7 connected to VCC after all.
The level noise will go up quite a bit, and the noise bandwidth decreases to 20kHz.
Rc =4k7 and RBB =2M7 gets back a bit of bandwidth and Rc=1.8k and RBB=1M0 puts thing back how they were (more or less).

[bool]

I think you overcomplicate this. You could pragmatically and simply bias the first opamp directly from the 7805 (via +5V) with just another 1M/100n RC... Opamps will be happy with that since you plan to run on 9V, your wallet too.

[Rob Strand]

I think you overcomplicate this. You could pragmatically and simply bias the first opamp directly from the 7805 (via +5V) with just another 1M/100n RC... Opamps will be happy with that since you plan to run on 9V, your wallet too.

I liked how Jim did it.   You have to pick a bias point for the transistor collector so why not VCC/2?  then, the opamps bias up without extra parts.   With such tiny signals I suspect powering the opamps off the nice quite regulated rail is going to save some headaches; you have to think of PSRR figures.    The only thing to be careful about is many opamps don't swing rail to rail and at 5V you don't have a lot of swing left.

[anotherjim]

I tried a regulated Vref before as well as the usual voltage divider with and without an op-amp buffer. In those attempts, there was instability and excessive supply ripple.

I think it only needs amps with rail-rail output. The op-amp inputs can't get close to either rail in normal working. So any CMOS type should do with perhaps a preference for the faster types. Any requirement for higher signal levels should easily be met with a little gain in the following circuit. As I might have said earlier, the only reason I can think of for higher level is you want a full-scale random CV in a conventional synth.

[bool]

So that's a no-go then... But have you thought about using your topology with a "standard" opamp and changing the 7805 to say a 7809 (and vice-versa a higher supply). That "should work" (in theory... ).

[Rob Strand]

But have you thought about using your topology with a "standard" opamp and changing the 7805 to say a 7809 (and vice-versa a higher supply).

It certainly crossed my mind.   Then I started thinking maybe I could just use a string of BJT stages (with gain set by resistors not 're').   If it turns out to produce a stable output I'd feel compelled to put in a pink noise filter which is best done with an opamp; a minor case of requirements creep.  That would be a standard opamp on the last stage running off 9V (or whatever) like you mentioned before.  You  shouldn't have to worry about power supply noise with opamps in the later stages.

I'm really interested to see how stable it is.  At the end of the day a calibrated source is probably best done with a PRBS these days.   The output level is automatically determined from the supply rail and the output is strong enough to be immune to hum, noise and interference.

[bool]

For sure (in theory at least... ) it would be most hygienic to allocate the largest portion of signal gain to the first opamp and "filter-and-boost" with the later stage(s). That would naturally split the circuit into "clean" and "dirty" blocks, similar to how you deal with highgain circuits.

Has anybody thought of adding a soft-clip after (or inside the nfb loop) the broadband gain stage?

[anotherjim]

In earlier versions, I did try NFB diode limiting. Diode or LED and with or without series resistance. it did what I would expect. However, I don't want to restrict the dynamics in the basic generator circuit.

[bool]

Did you think of using a zener-bjt combo as a generator? In early 80's I made a noise source for a elec-percussion effect with some sort of a zener-bjt combo and a filter with iirc a 741. I forgot the details, sorry. (It ran off a 9V battery).

Another Q: in your cckt you split the c-b resistor into 2 halves (R6 & R3). What happens if you shunt the split-point of these resistors to gnd with a cap, say 100n? And what if you change the ratio? say R6 at 200k-ish, then a cap to gnd; R3 at 2meg-ish? better or worse noise?

And why did you split the resistors in the first place? I may be wrong, but "resistor noise" should add in a rms way, while one bigger resistor should make a couple dB more self-noise if total R matches the cumulative two smaller r's.

[anotherjim]

I was being very literal when I said the scheme was what was on the breadboard! The x2 1M was in lieu of a single 2M2.  The only 2M2 I have are fiddly little 1/10th watt things with really skinny wire.
I did try Zener for a noise source but it was pretty feeble. I found a few ways to increase the noise without increasing amp gain but they still need a lot of gain anyway, so it's swings and roundabouts.

[PRR]

> why did you split the resistors in the first place? I may be wrong, but "resistor noise" should add in a rms way

No. Each resistor loads the other. One physical resistor, or same-value split in 10 or 100 physical resistors, is the same hiss.

Consider that real resistors are billions of nano-resistors (specs of carbon or molecules of nichrome). That does not matter.

I don't think it really matters even if your physical resistors have "excess noise", hiss higher than the thermal noise expected. Except that as you go from 100K to 100Meg you may get into crappier conductor and higher excess hiss.