Very sensitive audio amplifier problems

[gxti]

I'm building a white noise generator for feeding a computer's random number generator, so I need good quality noise. Apparently the best for the money is the zener and/or avalanche noise from a zener diode, but the noise levels are very small. How small exactly I'm not sure, so I fabbed a PCB to determine it experimentally.

I made a simple circuit using two LM833 in series with a total of 100 gain which gives an output of 1.58 volts peak-to-peak. If I short the diode to check the (non-diode) noise levels, I get 0.344Vpp. This is probably good enough for practical purposes, but I want to see if I can do better. Suggestions on how I can get the (bad) noise down?

Schematic: http://partiallystapled.com/~gxti/circuits/2010/12/08-rng_experiment.png
PCB font: http://partiallystapled.com/~gxti/circuits/2010/12/08-rng-front.png
PCB back: http://partiallystapled.com/~gxti/circuits/2010/12/08-rng-back.png

[allanw]

Try the tips in here:

First thing to try is to lower the resistor values, but that might not be the main source of noise.

edit: I'm not sure how helpful this video is since the noise you might be experiencing is 1/f noise instead of higher frequency noise. Perhaps try using a chopper amp? Here's a lot more readings to try:

http://www.google.com/search?sourceid=chrome&ie=UTF-8&q=low+noise+op+amp#sclient=psy&hl=en&q=noise+op+amp+filetype:pdf&aq=f&aqi=&aql=&oq=&gs_rfai=&pbx=1&fp=4488cb887e893d25

[TopherTheME]

If the intrinsic noise is thermal noise from those high value resistors you are using, the noise can probably be considered "good" as thermal noise is approximately white.

Do you have a scope with FFT or some means of sampling the signal at a high rate?

[gxti]

Thanks for the video, something to keep in the toolbox in any case. While now blindingly obvious, at the time I didn't realize that higher resistor values would cause problems.

I do have a DSO, I tried to get some screenshots this morning but it wasn't getting along with my SD card so I'll try to put some up tonight. The output doesn't look as "white" as I'd like it to, it seems to have an intrinisic frequency in the tens of kilohertz range and I can see it as a (small) peak on the FFT. The noise with the diode shorted is also much higher than I would expect from the calculations in the video, hundreds of millivolts actual versus tens of microvolts expected.

Would adding a low-pass filter somewhere along the chain help? I'm sampling at at most 10KHz but I'm not sure whether rejecting higher frequencies would help or hurt performance.

[jimmc]

Looking at the data sheet for a typical zener (eg BZX79C6v8 http://www.supplyframe.com/partsearchservlet/partnerWormhole.action?id=2345676&partnerName=DSA)
The dynamic impedance (Zz) is 30 to 80 ohm at 1mA so the attenuation of noise / rubbish on the 12v supply will be around 40dB ( ~Zz/R1) so unless the 12v supply is very clean you will have a problem.
Similarly noise on the output of the 5v regulator will be amplified by the full gain of the circuit (10,000).

To avoid these problems...

First decouple the supply to the zener very heavily by splitting R1 and adding a large capacitor from the junction of the two parts to ground.
Second  loose the 5v supply completely.
To do this DC couple the circuit by removing R2 & R3, shorting C4 and C5 and replacing the 5v supply to the lower end of R6 & R7 with a capacitor from each to ground (using 100nF will give a 16Hz LF roll off as before).

This will give the amplifiers unity DC gain so their outputs will sit a the zener voltage (rather than 5v + 100 times the offset voltage)

Jim

[gxti]

Here are some DSO shots with a 10V zener and a 12V switchmode wall wart (not ideal I know, but it needs to be resilient even if I have to spend more on the filtering than the amplifier):
http://partiallystapled.com/~gxti/circuits/2010/12/09-rng-fft-1.png
http://partiallystapled.com/~gxti/circuits/2010/12/09-rng-fft-2.png

The FFT peaks at 300kHz. I would blame it on the power supply, except for the results on battery coming out so similar.

Here's the result using a 5.6V zener and a 9V battery:
http://partiallystapled.com/~gxti/circuits/2010/12/09-rng-bat-fft-1.png
http://partiallystapled.com/~gxti/circuits/2010/12/09-rng-bat-fft-2.png

Note how on battery the signal is only 22mVrms compared to hundreds of millivolts for the line supply, so the supply noise is apparently causing major problems. That said, there are still distinctive peaks on the spectrum, one at 70kHz and one again at 300kHz. Something about the circuit, perhaps the 5V regulator as jimmc suggests, is resonating around that value.

Thanks a lot for the decoupling input jimmc, I went with two stages of 10 gain each because of offset problems but perhaps this will be better. Is there a name for this topology with the feedback loop coupled to ground through a capacitor instead of directly? I'm currently reading an IEEE essay on single-supply amplifiers that uses it (don't know how it escaped me before).

[apex]

Hey, cool project!
Reminds me of http://translate.google.de/translate?js=n&prev=_t&hl=de&ie=UTF-8&layout=2&eotf=1&sl=de&tl=en&u=http%3A%2F%2Fwww2.kielnet.net%2Fhome%2Fjulien.thomas%2Ftech%2FXR232WEB.htm.

You could try out his schmematic for the amp.

apex