High Quality White Noise Gen
in [Design]
|
From: George Herold on 2 Jun 2010 15:30 On Jun 2, 1:59 pm, whit3rd <whit...(a)gmail.com> wrote: > On Jun 2, 10:37 am, George Herold <gher...(a)teachspin.com> wrote: > > > > >> >Yup, and if the voltage asymmetry is a problem you can add the signal > > > >> >from two diodes, one biased from the positve supply and the other from > > > >> >the negative. > > Hmmm, you are right.... I still don't think that summing the voltage > > noise from a bunch of unipolarized zeners is going to get rid of the > > voltage asymmetry. > > You can use two zeners on a single power supply, in bridge > configuration; couple the output through a transformer to get > the difference. Symmetry is guaranteed if you balance the bridge > correctly. Ahh, more than one way to skin that cat. Say speaking of noise sources, a colleague put together a digital noise source. A counter steps through a look up table that feeds a DAC. The lookup table holds a whole comb of sine waves equally spaced in frequency space up to 32kHz. (I dont recall the frequency spacing but I could find out.. a few Hz or so.) The phases of all the sine waves were chosen randomly. The DAC was 12 bit (an AD7541 I think). The whole thing was clocked several times lower than Nyquist limit (~128 kHz). Now the problem we observed, (and could never cure), was intermodulation distortion above the 32 kHz cutoff. The signals above the cutoff frequency were down by only 50 dB, and my colleague was expecting something closer to 70 dB down. (Is that right for 12 bit resolution on the DAC? ) I worked on all the layout and analog portions of the circuit but could never make it any better. There was talk about clock jitter on SED recently and I wondered if this could be the source of the problem? Or maybe you have some other idea. Thanks, And If you think it might be clock jitter then can you tell me how to measure it also. (OK I can use google too.) George H. > > I've seen zeners (on a curve tracer) that were so noisy that the > breakover region was a blur.
From: Paul Keinanen on 2 Jun 2010 16:13 On Mon, 31 May 2010 12:23:10 GMT, jimslone(a)esterlux.com (Jim Slone) wrote: >What are the best options for high quality audio white noise >generation? A pseudo random noise (PRN) generator (a shift register and a few XOR gates in the feedback path) will generate quite good white noise sequences at frequencies well below the clock frequency.
From: John Larkin on 2 Jun 2010 16:25 On Wed, 2 Jun 2010 11:57:56 -0700 (PDT), George Herold <gherold(a)teachspin.com> wrote: >On Jun 2, 1:46�pm, John Larkin ><jjlar...(a)highNOTlandTHIStechnologyPART.com> wrote: >> On Wed, 2 Jun 2010 10:37:12 -0700 (PDT), George Herold >> >> >> >> >> >> <gher...(a)teachspin.com> wrote: >> >On Jun 2, 10:55�am, John Larkin >> ><jjlar...(a)highNOTlandTHIStechnologyPART.com> wrote: >> >> On Wed, 2 Jun 2010 07:32:55 -0700 (PDT), George Herold >> >> >> <gher...(a)teachspin.com> wrote: >> >> >On Jun 1, 5:51�pm, John Larkin >> >> ><jjlar...(a)highNOTlandTHIStechnologyPART.com> wrote: >> >> >> On Tue, 1 Jun 2010 11:35:59 -0700 (PDT), George Herold >> >> >> >> <gher...(a)teachspin.com> wrote: >> >> >> >On May 31, 12:56�pm, John Larkin >> >> >> ><jjlar...(a)highNOTlandTHIStechnologyPART.com> wrote: >> >> >> >> On Mon, 31 May 2010 12:23:10 GMT, jimsl...(a)esterlux.com (Jim Slone) >> >> >> >> wrote: >> >> >> >> >> >What are the best options for high quality audio white noise >> >> >> >> >generation? >> >> >> >> >> >I have been using generic diodes and reversed biased transistors. Then >> >> >> >> >someone mentioned there are special parts available with better >> >> >> >> >characteristics. >> >> >> >> >> >Can anyone please give me a pointer? >> >> >> >> >> >Jim Slone >> >> >> >> >> You can buy noise diodes from lots of people... just google <noise >> >> >> >> diode> >> >> >> >> >Does anyone know what makes a high price "noise diode" any better than >> >> >> >your garden variety Zener? >> >> >> >> Probably a very small junction area (for low capacitance, high current >> >> >> density) and maybe some doping profile. Not a power device! >> >> >> >> Regular zeners get spikey and asymmetric and sort of oscillate at low >> >> >> current. You can get noise diodes that behave at low currents. >> >> >> >> >> If you want really flat, really gaussian noise, a mathematical random >> >> >> >> stream (single-bit) or random word (dac) generator is probably best. >> >> >> >> See AoE for details. >> >> >> >> >> For audio, it doesn't matter much. A 10-volt zener biased at a few mA >> >> >> >> is fine. >> >> >> >> >Yup, and if the voltage asymmetry is a problem you can add the signal >> >> >> >from two diodes, one biased from the positve supply and the other from >> >> >> >the negative. �(Though I've never tried this trick.) >> >> >> >> Or sum the signals from a bunch of them. Central limit theorem. >> >> >> >Well that is not going to get rid of the voltage asymmetery. >> >> >> >If you need real Gaussian noise you can look at the shot noise from a >> >> >photodiode illuminated by an LED. �Gives you noise ~100 times bigger >> >> >than the johnson noise of the sense resistor. �(Assuming a 5 Volt DC >> >> >drop across R). �But this has one big drawback. �It's very sensitve to >> >> >vibrations. >> >> >> Shot noise is the ultimate asymmetric waveform. It's made of >> >> single-photon unidirectional spikes. If it manages to be Gaussian, >> >> it's because a lot of asymmetric signals are being summed. Central >> >> limit theorem. >> >> >>http://en.wikipedia.org/wiki/Illustration_of_the_central_limit_theorem >> >> >> John- Hide quoted text - >> >> >> - Show quoted text - >> >> >Hmmm, you are right.... I still don't think that summing the voltage >> >noise from a bunch of unipolarized zeners is going to get rid of the >> >voltage assymetry. �But I'd be happy to be wrong too. �Have you ever >> >tried this? �It would be simple enough to put 5 or 6 together and see >> >what the output looks like. �(As long as you don't mind my summing >> >with an opamp)... Maybe I can find some 'fun' time on Friday. >> >> >George H. >> >> The math says it must be so. Still, the sum would converge to Gaussian >> faster if half of the lopsided signals were inverted. > > >OK I'll just have to try it. I find the math more convincing if I can >see it in some experimental result. (Shot noise is a good example, >but it�s hard to get the current low enough so that you could see a >non-Gaussian distribution.) > >> >> Zener noise gets more symmetric at higher currents. 10 mA is usually >> OK for a small 10-volt zener. > >Yeah I think that is just the result of the I-V curvature. I'll try >running them down at low currents where the asymmetry is larger. This >is an experiment to show the central limit theorem and not make a good >Gaussian noise source. > >George H. >> >> I sometimes generate Gaussian-distributed numbers by summing a bunch >> of RAN() calls, which are uniform on [0,1]. Six to ten works well, and >> the crest factor is finite and known. >> >> John- Hide quoted text - >> >> - Show quoted text - Here is my elegant, peer-reviewed research paper on the subject: ftp://jjlarkin.lmi.net/Zener_Noise.pdf John
From: John Larkin on 2 Jun 2010 16:36 On Wed, 2 Jun 2010 12:30:19 -0700 (PDT), George Herold <gherold(a)teachspin.com> wrote: >On Jun 2, 1:59�pm, whit3rd <whit...(a)gmail.com> wrote: >> On Jun 2, 10:37�am, George Herold <gher...(a)teachspin.com> wrote: >> >> > > >> >Yup, and if the voltage asymmetry is a problem you can add the signal >> > > >> >from two diodes, one biased from the positve supply and the other from >> > > >> >the negative. >> > Hmmm, you are right.... I still don't think that summing the voltage >> > noise from a bunch of unipolarized zeners is going to get rid of the >> > voltage asymmetry. >> >> You can use two zeners on a single power supply, in bridge >> configuration; couple the output through a transformer to get >> the difference. � Symmetry is guaranteed if you balance the bridge >> correctly. > >Ahh, more than one way to skin that cat. > >Say speaking of noise sources, a colleague put together a digital >noise source. A counter steps through a look up table that feeds a >DAC. The lookup table holds a whole comb of sine waves equally spaced >in frequency space up to 32kHz. (I don�t recall the frequency spacing >but I could find out.. a few Hz or so.) The phases of all the sine >waves were chosen randomly. The DAC was 12 bit (an AD7541 I >think). The whole thing was clocked several times lower than Nyquist >limit (~128 kHz). Now the problem we observed, (and could never >cure), was intermodulation distortion above the 32 kHz cutoff. The >signals above the cutoff frequency were down by only 50 dB, and my >colleague was expecting something closer to 70 dB down. (Is that >right for 12 bit resolution on the DAC? ) I worked on all the layout >and analog portions of the circuit but could never make it any >better. There was talk about clock jitter on SED recently and I >wondered if this could be the source of the problem? Or maybe you >have some other idea. > The DAC quantization, and any nonlinearity, will add harmonic distortion. Plus the sines may occasionally peak together and clip the dac. Plus it's not trivial to get -70 dB distortion at these frequencies. We use random number generators and boxcar filters to generate Gaussian noise to feed into dacs. This little box does this, all in a Spartan3 FPGA... http://www.highlandtechnology.com/DSS/T346DS.html Rob cleverly, somehow, allowed the user to program the noise bandwidth from mHz to 2 MHz without affecting the RMS amplitude. John
From: Vladimir Vassilevsky on 2 Jun 2010 17:42
Paul Keinanen wrote: > On Mon, 31 May 2010 12:23:10 GMT, jimslone(a)esterlux.com (Jim Slone) > wrote: > > >>What are the best options for high quality audio white noise >>generation? > > > A pseudo random noise (PRN) generator (a shift register and a few XOR > gates in the feedback path) will generate quite good white noise > sequences at frequencies well below the clock frequency. For cheap random number generators, I prefer LCGs to LSFRs. It is very simple to compute something like: seed = (seed << 3) - seed + 5; It cycles through all states, so you don't have to worry about initialization. It also guarantees maximum period as long as the multiplier and the additive are simple numbers. Vladimir Vassilevsky DSP and Mixed Signal Design Consultant http://www.abvolt.com |