Forums

Amplitude-dependence of 1/f noise

Started by Phil Hobbs October 1, 2015
I was reading a paper by Rubiola & Brendel on oscillator noise, and he 
made the interesting comment that since 1/f noise is multiplicative in 
nature (e.g. conductance fluctuations in thick-film resistors), whereas 
other noise sources are additive, the 1/f noise corner of an oscillator 
depends on the amplitude.

He also says that circuit simulators all get this completely wrong, 
because they assume that the 1/f corner is fixed.

Any comments from RF and/or SPICE folks?

The paper is at http://arxiv.org/pdf/1004.5539 .

Cheers

Phil Hobbs

-- 
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics

160 North State Road #203
Briarcliff Manor NY 10510

hobbs at electrooptical dot net
http://electrooptical.net
On Thursday, October 1, 2015 at 3:40:23 PM UTC-4, Phil Hobbs wrote:
> I was reading a paper by Rubiola & Brendel on oscillator noise, and he > made the interesting comment that since 1/f noise is multiplicative in > nature (e.g. conductance fluctuations in thick-film resistors), whereas > other noise sources are additive, the 1/f noise corner of an oscillator > depends on the amplitude. > > He also says that circuit simulators all get this completely wrong, > because they assume that the 1/f corner is fixed. > > Any comments from RF and/or SPICE folks? > > The paper is at http://arxiv.org/pdf/1004.5539 .
(I'm not an RF or spice guy.. can I still comment? :^) I assume you are talking about section 3. Does this also applies to amps as well as oscillators? (I assume it does.) For amps the 1/f noise is a device specific parameter... I mean each opamp has it's own amount of 1/f noise. (I've got lotsa data, at least for the opa134.) Given the above, it seems rather obvious to me that the corner freq depends on the size of the 1/f noise... given that the "white" noise in opamps is pretty much the same from device to device. George H.
> > Cheers > > Phil Hobbs > > -- > Dr Philip C D Hobbs > Principal Consultant > ElectroOptical Innovations LLC > Optics, Electro-optics, Photonics, Analog Electronics > > 160 North State Road #203 > Briarcliff Manor NY 10510 > > hobbs at electrooptical dot net > http://electrooptical.net
On Thursday, October 1, 2015 at 3:40:23 PM UTC-4, Phil Hobbs wrote:
> I was reading a paper by Rubiola & Brendel on oscillator noise, and he > made the interesting comment that since 1/f noise is multiplicative in > nature (e.g. conductance fluctuations in thick-film resistors), whereas > other noise sources are additive, the 1/f noise corner of an oscillator > depends on the amplitude. > > He also says that circuit simulators all get this completely wrong, > because they assume that the 1/f corner is fixed. > > Any comments from RF and/or SPICE folks? > > The paper is at http://arxiv.org/pdf/1004.5539 .
This may be a duplicate.. google puked as I posted. (I'm not an RF or spice guy.. can I still comment? :^) I assume you are talking about section 3. Does this also applies to amps as well as oscillators? (I assume it does.) For amps the 1/f noise is a device specific parameter... I mean each opamp has it's own amount of 1/f noise. (I've got lotsa data, at least for the opa134.) Given the above, it seems rather obvious to me that the corner freq depends on the size of the 1/f noise... given that the "white" noise in opamps is pretty much the same from device to device. George H.
> > Cheers > > Phil Hobbs > > -- > Dr Philip C D Hobbs > Principal Consultant > ElectroOptical Innovations LLC > Optics, Electro-optics, Photonics, Analog Electronics > > 160 North State Road #203 > Briarcliff Manor NY 10510 > > hobbs at electrooptical dot net > http://electrooptical.net
On Thu, 1 Oct 2015 15:40:11 -0400, Phil Hobbs
<pcdhSpamMeSenseless@electrooptical.net> wrote:

>I was reading a paper by Rubiola & Brendel on oscillator noise, and he >made the interesting comment that since 1/f noise is multiplicative in >nature (e.g. conductance fluctuations in thick-film resistors), whereas >other noise sources are additive, the 1/f noise corner of an oscillator >depends on the amplitude. > >He also says that circuit simulators all get this completely wrong, >because they assume that the 1/f corner is fixed. > >Any comments from RF and/or SPICE folks? > >The paper is at http://arxiv.org/pdf/1004.5539 . > >Cheers > >Phil Hobbs
Disclaimer: I'm no expert on noise. But I suspect the author is assigning _excess_ noise to 1/f ?? "Fluctuation" is separate from 1/f. But I stand ready to model it in Spice if someone can make a case for it ;-) ...Jim Thompson -- | James E.Thompson | mens | | Analog Innovations | et | | Analog/Mixed-Signal ASIC's and Discrete Systems | manus | | San Tan Valley, AZ 85142 Skype: skypeanalog | | | Voice:(480)460-2350 Fax: Available upon request | Brass Rat | | E-mail Icon at http://www.analog-innovations.com | 1962 | I love to cook with wine. Sometimes I even put it in the food.
On 10/1/2015 3:40 PM, Phil Hobbs wrote:
> I was reading a paper by Rubiola & Brendel on oscillator noise, and he > made the interesting comment that since 1/f noise is multiplicative in > nature (e.g. conductance fluctuations in thick-film resistors), whereas > other noise sources are additive, the 1/f noise corner of an oscillator > depends on the amplitude. > > He also says that circuit simulators all get this completely wrong, > because they assume that the 1/f corner is fixed. > > Any comments from RF and/or SPICE folks? > > The paper is at http://arxiv.org/pdf/1004.5539 . > > Cheers > > Phil Hobbs
Makes sense. Analogous to an IMD measurement but with internal sources. If "noise" was introduced as an actual voltage and current waveform at each source component in a time domain simulation SPICE would get it right. An FFT of the TD data would show the spectral distribution of nonlinear mixing products. I'm thinking that the "noise" should be discrete frequencies (stepped, perhaps) as this would make it fairly simple to determine the order of nonlinearity and relative sensitivity to amplitude of each noise contributor. -- Grizzly H.
Am 01.10.2015 um 22:28 schrieb George Herold:
> On Thursday, October 1, 2015 at 3:40:23 PM UTC-4, Phil Hobbs wrote: >> I was reading a paper by Rubiola & Brendel on oscillator noise, and he >> made the interesting comment that since 1/f noise is multiplicative in >> nature (e.g. conductance fluctuations in thick-film resistors), whereas >> other noise sources are additive, the 1/f noise corner of an oscillator >> depends on the amplitude.
Drawning a 1/f signal in lots of white noise would reduce the 1/f corner, and that makes the 1/f corner somehow ill-defined. I think Rubiola has a proposal against that with his noise coefficients; I wished I had the time to read his book and not only to sample a few pages from time to time.
>> >> He also says that circuit simulators all get this completely wrong, >> because they assume that the 1/f corner is fixed.
Suppose, we have a circuit with +49R and -51R in series. Do the -51R have thermal noise? I guess: yes. Will the average simulator get that right? I guess: probably not. The size of the conductivity matrix is extremely costly in terms of execution time, so the simulator will probably try its best to eliminate redundant nodes and simply add the resistors and skip the junction. Yes, I'm pessimistic and I create more questions than I answer. Last weekend, I have measured the noise of some LEDs and Zeners. The preamplifier averaged 20 ADA4898 op amps, 220pV/sqrt Hz, I think I have described that here already. FFT analyzer is an Agilent 89441A vector signal analyzer, one FFT per decade, controlled from a Linux PC, combined and plotted with gnuplot. One measurement takes 6 minutes of wall clock time, most for the lowest decade. Setup: preamp with shorted input to see the noise floor, input terminated with 47R wire resistor, 47R + 2K to 14V from 10 NiMH cells. No additional noise visible from the battery: < https://picasaweb.google.com/lh/photo/vaqmennDl6tz7sdnM2T1_dMTjNZETYmyPJy0liipFm0?feat=directlink > LEDs: fed from 14V with 1K or 2 K, red & blue Osram SMD LEDS, Toshiba 0603 LED red. < https://picasaweb.google.com/lh/photo/60hTra_W104ZGNyU_PZ1TtMTjNZETYmyPJy0liipFm0?feat=directlink > BZX84: sot-23 fed from 14V via 1K < https://picasaweb.google.com/lh/photo/yrfwcAKyw0Re1h9MD52w49MTjNZETYmyPJy0liipFm0?feat=directlink > Regarding the LEDs, a well-known chip designer oracled about Lorentzian noise and lots of G/R noise but I have no idea if that invalidates the measurement or is an interesting discovery. Nevertheless, the BZX84-xVy 1/f corner is low, as is their noise level. The preamp has 3 poles at 0.1Hz and is currently overcompensated, so flatness > 300 KHz is not given. The limitations are visible. regards, Gerhard
On Thursday, October 1, 2015 at 4:59:42 PM UTC-4, Jim Thompson wrote:
> On Thu, 1 Oct 2015 15:40:11 -0400, Phil Hobbs > <pcdhSpamMeSenseless@electrooptical.net> wrote: > > >I was reading a paper by Rubiola & Brendel on oscillator noise, and he > >made the interesting comment that since 1/f noise is multiplicative in > >nature (e.g. conductance fluctuations in thick-film resistors), whereas > >other noise sources are additive, the 1/f noise corner of an oscillator > >depends on the amplitude. > > > >He also says that circuit simulators all get this completely wrong, > >because they assume that the 1/f corner is fixed. > > > >Any comments from RF and/or SPICE folks? > > > >The paper is at http://arxiv.org/pdf/1004.5539 . > > > >Cheers > > > >Phil Hobbs > > Disclaimer: I'm no expert on noise. But I suspect the author is > assigning _excess_ noise to 1/f ?? > > "Fluctuation" is separate from 1/f. > > But I stand ready to model it in Spice if someone can make a case for > it ;-)
I'm no noise expert either. (But it's fun to measure.) It seems to me a model that defines the corner freq. (where white noise = 1/f noise, or something like that.) Is just fine... as long as that is a parameter I can change in the model. George H.
> > ...Jim Thompson > -- > | James E.Thompson | mens | > | Analog Innovations | et | > | Analog/Mixed-Signal ASIC's and Discrete Systems | manus | > | San Tan Valley, AZ 85142 Skype: skypeanalog | | > | Voice:(480)460-2350 Fax: Available upon request | Brass Rat | > | E-mail Icon at http://www.analog-innovations.com | 1962 | > > I love to cook with wine. Sometimes I even put it in the food.
On Thursday, October 1, 2015 at 6:17:26 PM UTC-4, Gerhard Hoffmann wrote:
> Am 01.10.2015 um 22:28 schrieb George Herold: > > On Thursday, October 1, 2015 at 3:40:23 PM UTC-4, Phil Hobbs wrote: > >> I was reading a paper by Rubiola & Brendel on oscillator noise, and he > >> made the interesting comment that since 1/f noise is multiplicative in > >> nature (e.g. conductance fluctuations in thick-film resistors), whereas > >> other noise sources are additive, the 1/f noise corner of an oscillator > >> depends on the amplitude. > > Drawning a 1/f signal in lots of white noise would reduce the > 1/f corner, and that makes the 1/f corner somehow ill-defined. > I think Rubiola has a proposal against that with his noise > coefficients; I wished I had the time to read his book and > not only to sample a few pages from time to time. > > > >> > >> He also says that circuit simulators all get this completely wrong, > >> because they assume that the 1/f corner is fixed. > > Suppose, we have a circuit with +49R and -51R in series. > Do the -51R have thermal noise? I guess: yes. > Will the average simulator get that right? I guess: > probably not. The size of the conductivity matrix is > extremely costly in terms of execution time, so the simulator > will probably try its best to eliminate redundant nodes and simply > add the resistors and skip the junction. > > Yes, I'm pessimistic and I create more questions than I answer. > > > > > Last weekend, I have measured the noise of some LEDs and Zeners. > The preamplifier averaged 20 ADA4898 op amps, 220pV/sqrt Hz, > I think I have described that here already. FFT analyzer is an > Agilent 89441A vector signal analyzer, one FFT per decade, > controlled from a Linux PC, combined and plotted with gnuplot. > One measurement takes 6 minutes of wall clock time, most for > the lowest decade. > > Setup: > preamp with shorted input to see the noise floor, input terminated > with 47R wire resistor, 47R + 2K to 14V from 10 NiMH cells. > No additional noise visible from the battery: > > < > https://picasaweb.google.com/lh/photo/vaqmennDl6tz7sdnM2T1_dMTjNZETYmyPJy0liipFm0?feat=directlink > >
I like it! Do the batteries couple to local sound vibrations? I'm not sure I understand the other plots. (I mostly need someone to draw pictures for me.) When I went to look for shot noise in forward biased pn junctions, I had to put a cap across the bias resistor, (feed by a voltage source.) else the diode (as noise source) had too low a source resistance, to "let the noise out". George H.
> > > LEDs: > fed from 14V with 1K or 2 K, red & blue Osram SMD LEDS, > Toshiba 0603 LED red. > > < > https://picasaweb.google.com/lh/photo/60hTra_W104ZGNyU_PZ1TtMTjNZETYmyPJy0liipFm0?feat=directlink > > > > > BZX84: > sot-23 fed from 14V via 1K > > < > https://picasaweb.google.com/lh/photo/yrfwcAKyw0Re1h9MD52w49MTjNZETYmyPJy0liipFm0?feat=directlink > > > > Regarding the LEDs, a well-known chip designer oracled about > Lorentzian noise and lots of G/R noise but I have no idea if > that invalidates the measurement or is an interesting discovery. > > Nevertheless, the BZX84-xVy 1/f corner is low, as is their noise level. > > The preamp has 3 poles at 0.1Hz and is currently overcompensated, > so flatness > 300 KHz is not given. The limitations are visible. > > regards, Gerhard
On Thu, 1 Oct 2015 16:58:33 -0700 (PDT), George Herold
<gherold@teachspin.com> wrote:

>On Thursday, October 1, 2015 at 4:59:42 PM UTC-4, Jim Thompson wrote: >> On Thu, 1 Oct 2015 15:40:11 -0400, Phil Hobbs >> <pcdhSpamMeSenseless@electrooptical.net> wrote: >> >> >I was reading a paper by Rubiola & Brendel on oscillator noise, and he >> >made the interesting comment that since 1/f noise is multiplicative in >> >nature (e.g. conductance fluctuations in thick-film resistors), whereas >> >other noise sources are additive, the 1/f noise corner of an oscillator >> >depends on the amplitude. >> > >> >He also says that circuit simulators all get this completely wrong, >> >because they assume that the 1/f corner is fixed. >> > >> >Any comments from RF and/or SPICE folks? >> > >> >The paper is at http://arxiv.org/pdf/1004.5539 . >> > >> >Cheers >> > >> >Phil Hobbs >> >> Disclaimer: I'm no expert on noise. But I suspect the author is >> assigning _excess_ noise to 1/f ?? >> >> "Fluctuation" is separate from 1/f. >> >> But I stand ready to model it in Spice if someone can make a case for >> it ;-) >I'm no noise expert either. (But it's fun to measure.) >It seems to me a model that defines the corner freq. >(where white noise = 1/f noise, or something like that.) >Is just fine... as long as that is a parameter >I can change in the model. > >George H. >>
[snip] See InputReferredNoise.zip on the Device Models & Subcircuits Page of my website. ...Jim Thompson -- | James E.Thompson | mens | | Analog Innovations | et | | Analog/Mixed-Signal ASIC's and Discrete Systems | manus | | San Tan Valley, AZ 85142 Skype: skypeanalog | | | Voice:(480)460-2350 Fax: Available upon request | Brass Rat | | E-mail Icon at http://www.analog-innovations.com | 1962 | I love to cook with wine. Sometimes I even put it in the food.
On Thursday, October 1, 2015 at 3:17:26 PM UTC-7, Gerhard Hoffmann wrote:
> Regarding the LEDs, a well-known chip designer oracled about > Lorentzian noise and lots of G/R noise but I have no idea if > that invalidates the measurement or is an interesting discovery.
Very "enlightening" -- especially that the Zeners are so competitive with the LEDs, and that the Toshiba LED performs so much worse than the Osram part. Looks like it's a bad idea to mindlessly replace Zeners with LEDs in low-noise bias applications, or to consider all red LEDs interchangeable. Questions: - Why is there such a non-monotonic relationship between noise and Zener voltage? Have you tried an avalanche-mode Zener for comparison (i.e., one that's rated at more than ~7V)? - What happens if you expose the LEDs to other sources of ambient light -- particularly a fluorescent lamp or other modulated source -- while measuring them? I wonder if they remain sensitive to other wavelengths while they're emitting their own light. - Where do you get those nice metal castings you're using for your boards in those photos? -- john, KE5FX