DC-100Hz flicker free noise standard

[GK]

I need a low frequency Gaussian noise generator which is amplitude stable/regulated and completely free of flicker (1/f) noise right down to DC. Here is what I've come up with after a brief late evening of scribbling. I see no obvious issues so far and my sums look sorta right.

I played around with Zener diodes and reverse-biased BJT Vbe's and uggg... too inconsistent and variable for my liking. So I amplify the noise of a 100k resistor with a BIFET op-amp having an unusually low voltage noise specification (2.9nV) and completely negligible current noise (<7fA). A flat portion of the amplified output spectrum is down-converted to DC to solve the 1/f problem.

The reason for the comparatively wide bandwidth (4kHz) is to ensure that the output spectrum bandwidth is entirely defined by the final low pass filters. However I think 2 kHz would be fine also - that is still 10 times higher than the maximum filtered output bandwidth when down-converted. This is all very preliminary ATM and all I can fit in this evening.

The final filters will have a -3dB bandwidth slightly lower than the specified bandwidths as the specified bandwidths (5, 10, 20, 50 and 100 Hz) are to be noise bandwidths. I can't remember the -3dB-frequency to noise-bandwidth constant for a 4-pole filter right now and it's too late to look it up.



   

[GK]

Yes, auto-zeroing op-amps eliminate 1/f (at least as low in f as the datasheets charts show) but you still have to amplify the output noise to a usable level, filter to the required bandwidth and stabilise the amplitude for temperature.

I'd still like to try out the down-conversion technique though as a demonstration of principle and I think it will still work marginally better. Eventually I'd like to characterise the 1/f noise to very long periods sampling the output noise with a low sampling rate DAC and my PC over days.

I have use an RMS detector for amplitude control in various applications, but I did originally get the idea from a Jim Williams application note; possibly the one that you linked to.

[GK]

I appear to have overestimated the bandwidth of the AD745 noise source. Checking the datasheet, the AD745 has huge input capacitance - 20pF differential and about the same again common mode. This corners with the 100k noise source resistor.

Working on rules of 10 I've decided to reduce the down-converted bandwidth to 2 kHz (4 kHz to 6 kHz for the 5 kHz carrier). That will still be adequate. I'd like the noise source bandwidth to be 10x my band-pass filter high-pass cut-off; so at 60 kHz. That of course includes the additive HF roll-off of all high-gain noise source amplifier stages. The AD745 is also quite pricy, will search for a cheaper unit with less input C.

I'll also drive the mixer with a 5 kHz squarewave rather than the 10% duty cycle rectangular wave from the 4017 currently shown. This means there won't be even harmonics of the carrier and associated sidebands in the output spectrum. The MC1496 is also fully balanced so there won't be any (or rather there will be little) carrier or "IF" channel in the output, which does relax the final low-pass filter requirements. 4-poles is probably overkill.

[GK]

ADA4001 looks the goods. Way lower input C than the AD745. en is higher at 7.7nV, but still very much low enough. 7.7nV is equivalent to 3.6k, which is almost 30 times lower than 100k. The 100k noise will therefore dominate to the tune of the en contributing approximately 0.16 dB to the total rms noise amplitude. en is also essentially flat over the frequency range of interest. Current input noise (in) at 3fA can be safely neglected with a source of only 100k (3fA * 100k = 0.3nV contribution)

[GK]

You don't even need a DAC - just generate and then filter a pseudorandom rectangular wave. Have done that already. Boring.

[SArepairman]

I think HP made a very low frequency noise generator powered by a vacuum tube.

I almost bought it on ebay for use with a dynamic signal analyzer a while back.
It was only like 50$, but it musta weighed 50lbs. Had a bunch of settings all the way down to sub Hz noise.
3722A I believe.. but now I see them on ebay for 300$+

[GK]

Well I have the circuitry for the bandwidth-limited noise source part finalized now:



The simulated spectral density of the source:



Zoomed into the band of importance/interest (+/-100 Hz about the 5 kHz carrier). Worst case amplitude variation (~0.02 dB) simulated for the specified 1% tolerance for the bandpass filter R's and C's:

[SArepairman]

why are you building this source? how does temperature stability look like for your circuit?

[GK]

why are you building this source?

Because.

how does temperature stability look like for your circuit?

As already detailed, the amplitude will be regulated, otherwise the amplitude would have a predictable temperature dependence governed by the formula for thermal (Johnson–Nyquist) noise in a resistor. I think the AD736 will serve nicely as the detector.