Flicker noise voltage distribution.

[[email protected]]

I appreciate all that info. I'm surprised to read that white noise is
not always Gaussian!  Isn't Johnson noise (that's caused by natural
thermal energy) Gaussian?  I wouldn't doubt that some forms of
simulated white noise could be a different distribution.

PL- Hide quoted text -

- Show quoted text -

Paul, I'll try and get this right. Hopefully if I make a mistake Phil
H. the noise expert will correct me.
When you talk about white and Gaussian you are talking about two
different things.
White refers to how the power is distributed in frequency space. V^2/
Hz. White just means there's equal power in each Hz of the
spectrum.
I think you could make a white noise source out of a purely digital
signal.
Then Gaussian refers to how the power is distributed in voltage
space. Hmm, As I write this I realize that I don't know how to
measure the distribution... but anyway the digital noise source would
have an amplitude distribution that non-Gaussian, but still white.
The same will apply to 1/f noise. 1/f refers to the frequency
distribution. The amplitude distribution of 1/f noise will be source
dependent. Someone else wrote the same thing much earlier in this
thread.

George Herold

 

[[email protected]]

<[email protected]>
 "Phil Allison" :
 "John Larkin"
George Herold



















** What a nice man George is  ....


** As luck would have it  -  I just located a pair of 1 watt, 100kohm
"carbon comp" resistors taken from a Fender tube guitar amp during
refurbishing.

Plus, I also have a pair of unused 0.5 watt, 150 kohm " Metal Glaze"
resistors ( aka "Cermet" )  -  so the test rig will get another go soon as I
cook dinner and watch my favourite TV shows.

It's about 6pm Saturday, here in Sydney.

Weather is mild and overcast (23C),  could be a thunder storm on its
way.....

.... rgds,    Phil- Hide quoted text -

- Show quoted text -

"so the test rig will get another go soon as I
cook dinner and watch my favourite TV shows."
Wow, That's what I call quick service. Thanks Phil.
The carbon composition resistors I used were not the carbon film type,
but the solid dark brown body. These use to be made by Allen-Bradley
(sp) in the US. I didn't record the wattage but it was either 1/4 or
1/2 Watt. We still buy these from somewhere. I see "Little Demons"
listed in my Digikey catalog. One good use these resistors still have
is that they're non-magnetic and I use the small valued ones for
terminating the RF in our optical pumping apparatus.


"Weather is mild and overcast (23C), could be a thunder storm on its
way..... "

Rainy and 50 F. (10 C) here in Buffalo NY. It's also the start of
deer hunting season (shotguns) so there's a lot of "blamming" going
on...

George

 

[Paul]

"Phil Allison"
George Herold






**  OK    -  I did a couple more resistor noise tests.

Results as follows:

1 watt carbon composition = 0.22 uV/V ( per decade of frequency)

0.5 watt *Metal Glaze (cermet) = 1.7uV/V  ( !!!!!!!! )

------------------------------------------------------------

From earlier testing:

0.5 watt  MF   =   0.09 uV/V

2 watt  MF  =  0.12 uV/V

0.5 watt carbon film  =  0.22 uV/V

( * leaded resistors made by IRH about 20 years back. )

See  ABSE  for pic of the stars of the show.

....   Phil

Great data. The cermets have a lot of 1/f noise. I'm betting certain
brands of cermets would have higher 1/f noise since there's such a
wide range of cermet materials. You could get even more 1/f noise from
certain diodes instead resistors since they're non-linear, especially
GaAs? I'm wondering what the 1/f noise record is for passive
components.

Paul

 

[[email protected]]

"Phil Allison"
George Herold






**  OK    -  I did a couple more resistor noise tests.

Results as follows:

1 watt carbon composition = 0.22 uV/V ( per decade of frequency)

0.5 watt *Metal Glaze (cermet) = 1.7uV/V  ( !!!!!!!! )

------------------------------------------------------------

From earlier testing:

0.5 watt  MF   =   0.09 uV/V

2 watt  MF  =  0.12 uV/V

0.5 watt carbon film  =  0.22 uV/V

( * leaded resistors made by IRH about 20 years back. )

See  ABSE  for pic of the stars of the show.

....   Phil

Very nice, much thanks again Phil. I have to think about how to
convert my measurments to your units...
It was striking that as I clicked the center frequency of the
(constant Q) band pass filter to lower and lower values the average
value of the noise stayed the same...You could also observe the Dicke
noise in the noise, which goes as the square root of the measurement
time times the band width. At the lower frequencies the fluctuations
in the average value got bigger and bigger.

Your cermet measurement is a bit distrubing. Is this the same thing
that my conductive plastic pots are made out of? I like the 1/2 Watt
single turn pots from Clarostat.

George Herold

 

[Paul]

Very nice, much thanks again Phil.  I have to think about how to
convert my measurments to your units...
It was striking that as I clicked the center frequency of the
(constant Q) band pass filter to lower and lower values the average
value of the noise stayed the same...You could also observe the Dicke
noise in the noise, which goes as the square root of the measurement
time times the band width.  At the lower frequencies the fluctuations
in the average value got bigger and bigger.

Your cermet measurement is a bit distrubing.  Is this the same thing
that my conductive plastic pots are made out of?  I like the 1/2 Watt
single turn pots from Clarostat.

George Herold

I read somewhere that the noise increased with smaller and smaller
carbon resistors. Too bad I can't find that web page, as it was nice
where it outlined various types of noise.

This type of low frequency noise is good for data loggers, and fed to
a FFT function to display the entire spectrum. If for instance, you
have a chopper (zero-adjust) op-amp, you could see the spike at it's
clock frequency. Most low temp drift and low Vos op-amps have chopper
circuitry.


PL

 

[Phil Allison]

"Phil Hobbs"

** Any care to guess what planet this fool is from?

Did you try different values and better resistors to rule out temperature
fluctuations?

** Is his plant in our galaxy ??

I've made those sorts of measurements too, and found no 1/f noise down to
a part in 10**8, using ~10k ohm RN55C resistors and battery power.

** Funny how the makers *specify* the excess noise of MF resistors as
being 0.1uV/V per decade.

That's 1 part in 10*7 when using only 1 volt bias.



...... Phil

 

[Phil Allison]

<[email protected]
"Phil Allison"

** OK - I did a couple more resistor noise tests.

Results as follows:

1 watt carbon composition = 0.22 uV/V ( per decade of frequency)

0.5 watt *Metal Glaze (cermet) = 1.7uV/V ( !!!!!!!! )

------------------------------------------------------------

From earlier testing:

0.5 watt MF = 0.09 uV/V

2 watt MF = 0.12 uV/V

0.5 watt carbon film = 0.22 uV/V

( * leaded resistors made by IRH about 20 years back. )

See ABSE for pic of the stars of the show.

Very nice, much thanks again Phil. I have to think about how to
convert my measurments to your units...

Your cermet measurement is a bit distrubing. Is this the same thing
that my conductive plastic pots are made out of? I like the 1/2 Watt
single turn pots from Clarostat.

** Yep.

Never good practice to have a carbon or cermet pot passing small signals
have a DC bias at the same time, unless you like noise.

But it is very unlikely that a signal voltage alone will generate enough
excess noise to be significant - after all 2ppm equates to -114 dB.


...... Phil

 

[Phil Allison]

<[email protected]

Paul, I'll try and get this right. Hopefully if I make a mistake Phil
H. the noise expert will correct me.
When you talk about white and Gaussian you are talking about two
different things.
White refers to how the power is distributed in frequency space. V^2/
Hz. White just means there's equal power in each Hz of the
spectrum.
I think you could make a white noise source out of a purely digital
signal.
Then Gaussian refers to how the power is distributed in voltage
space. Hmm, As I write this I realize that I don't know how to
measure the distribution... but anyway the digital noise source would
have an amplitude distribution that non-Gaussian, but still white.
The same will apply to 1/f noise. 1/f refers to the frequency
distribution. The amplitude distribution of 1/f noise will be source
dependent. Someone else wrote the same thing much earlier in this
thread.


** Pink noise is widely used in the world of audio - for many purposes
including room acoustics testing and speaker power handling tests. In all
cases the noise is both band limited and amplitude limited to a known "peak
to average" ratio.

For speaker power handling tests, the noise would be typically limited to a
50 to 500 Hz band and the peak to average ratio restricted to 6dB.

Typical devices sold as " Audio Pink Noise Generators " have peak to
average ratios of 12 to 15 dB - as do "pink noise" tracks recorded on many
test CDs.


..... Phil

 

[Phil Allison]

"Phil Hobbs"

"Earth to Allison....come in Allison...

** Wot a complete fuckwit .....

Sorry?

** You sure need to be.

Try ACTUALLY reading my post - pal.

See the bit about "... audio band filter " ??

Cretin.

1 part in 10**7 is a ratio.

** My, there is a very bad echo in here .........

1/f noise is generally due to conductance fluctuations

** You don't say ????

(see the papers of Paul Horn in the 1980s), so until it becomes nonlinear,
e.g. due to avalanches or excess heating in the regions of higher
conductivity, the ratio should be constant.

** You don't say ????

0.1uV/V looks like a constant ratio to me.

And the makers set an *upper limit* of 1 part in 10**7, not a spec plus or
minus so many percent.

** Some give a typical figure as well as a "max"

- normally 0.07 uV/V for MF types.

Ie 30% less.

Yawnnnnnnnnnnnnn....

Sort of like the input current spec of a CD4000-series gate, which is
probably 6 orders of magnitude above the actual median value. (You
wouldn't sit still for that sort of sophistry from someone else, you
know.)

** The mind numbingly idiotic sophistry YOU spend your time pouring all
over Usenet would choke an elephant.

Kindly **** off.



...... Phil

 

[Phil Allison]

"Phil Hobbs"

** Any care to guess what planet this fool is from?

** Wot a complete fuckwit .....

** The mind numbingly idiotic sophistry YOU spend your time pouring all
over Usenet would choke an elephant.

Kindly **** off.

You ridiculous, autistic wanker.




..... Phil

 

[Paul G.]

<[email protected]

Paul, I'll try and get this right. Hopefully if I make a mistake Phil
H. the noise expert will correct me.
When you talk about white and Gaussian you are talking about two
different things.
White refers to how the power is distributed in frequency space. V^2/
Hz. White just means there's equal power in each Hz of the
spectrum.
I think you could make a white noise source out of a purely digital
signal.
Then Gaussian refers to how the power is distributed in voltage
space. Hmm, As I write this I realize that I don't know how to
measure the distribution... but anyway the digital noise source would
have an amplitude distribution that non-Gaussian, but still white.
The same will apply to 1/f noise. 1/f refers to the frequency
distribution. The amplitude distribution of 1/f noise will be source
dependent. Someone else wrote the same thing much earlier in this
thread.


** Pink noise is widely used in the world of audio - for many purposes
including room acoustics testing and speaker power handling tests. In all
cases the noise is both band limited and amplitude limited to a known "peak
to average" ratio.

For speaker power handling tests, the noise would be typically limited to a
50 to 500 Hz band and the peak to average ratio restricted to 6dB.

Typical devices sold as " Audio Pink Noise Generators " have peak to
average ratios of 12 to 15 dB - as do "pink noise" tracks recorded on many
test CDs.


.... Phil

I tried a similiar test, except I used a 10Hz fixed bandwidth on a
HP3581A wave analyzer. I wanted to verify the Johnson noise (no DC
current), and then apply a DC current, and see if the noise spectrum
changed as I varied the center frequency. In other words, when and
where is the noise pink or white.
I tried using wirewound, metal film, and the traditional carbon
film cheapy resistors (Philips 1/4W) under those conditions. The value
was 100Kohms. With no DC, they all produced the Johnson noise
predicted by v=sqtr(4KTBR), within about 20%. The spectrum was white,
ie, as I moved the center freq. from about 20Hz - 2KHz, there was
little change in amplitude. The measured bandwidth was kept at 10Hz.
With DC current (25V across 100K), the wirewounds showed no change,
the metal films increased noise very slightly.
With DC current the carbon film resistors put out about 10 times
the Johnson noise (at 100Hz center). I checked the noise vs. center
frequency, and by about 1-2KHz the noise dropped back to the thermal
noise. At 30Hz noise was about 25-30 times greater than thermal. When
I plotted the noise vs 1/f I got more or less a straight line (up to
about 1KHz), which confirmed the pink noise nature of the current
noise. These carbon resistors give a noise distribution of white noise
above a "break frequency" of 1-2 KHz, and 1/f noise (pink) below that.
Many electronic devices (ie.,MOSFETs, BJTs) have similiar
distributions, with white and pink areas.
The metal film resistors gave me 10-20% more total noise at 30Hz
than the Johnson noise. At frequencies above that, the total noise was
the same as the Johnson noise. I wouldn't be surprised if they to
would give 1/f noise, but the corner frequency would be 10-20 Hz, too
far down for my meter.

It could be the equipment, but the amplitude distribution did not
seem to be the same for the Johnson noise as the current noise. It's
hard to clearly define it, but the current noise seemed to have much
more pronounced dips and peaks than the thermal noise. The Johnson
noise swung the meter by about 10-20% (very roughly), and for the same
position on the scale, the current noise would vary 20-40%. Both had
the same "smoothing" on the analyzer, both at 10Hz bandwidth. This is
not an authoritative claim that the current noise is non-gaussian!
There may have been an additional noise effect (shot noise, moisture
tracks?) that was showing up.


Paul G.

 

[Phil Allison]

"Paul G."
"Phil Allison"

I tried a similiar test, except I used a 10Hz fixed bandwidth on a
HP3581A wave analyzer.

** Was Noah around to witness this ?

I wanted to verify the Johnson noise (no DC
current), and then apply a DC current, and see if the noise spectrum
changed as I varied the center frequency. In other words, when and
where is the noise pink or white.
I tried using wirewound, metal film, and the traditional carbon
film cheapy resistors (Philips 1/4W) under those conditions. The value
was 100Kohms. With no DC, they all produced the Johnson noise
predicted by v=sqtr(4KTBR), within about 20%. The spectrum was white,
ie, as I moved the center freq. from about 20Hz - 2KHz, there was
little change in amplitude. The measured bandwidth was kept at 10Hz.
With DC current (25V across 100K), the wirewounds showed no change,
the metal films increased noise very slightly.
With DC current the carbon film resistors put out about 10 times
the Johnson noise (at 100Hz center). I checked the noise vs. center
frequency, and by about 1-2KHz the noise dropped back to the thermal
noise. At 30Hz noise was about 25-30 times greater than thermal. When
I plotted the noise vs 1/f I got more or less a straight line (up to
about 1KHz), which confirmed the pink noise nature of the current
noise. These carbon resistors give a noise distribution of white noise
above a "break frequency" of 1-2 KHz, and 1/f noise (pink) below that.
Many electronic devices (ie.,MOSFETs, BJTs) have similiar
distributions, with white and pink areas.

** The "break frequency" is a function of resistor type, value and the
applied DC voltage.

Lower R values will have higher break frequencies for the same DC voltage.

Noisier types similarly have higher break frequencies for the same DC
voltage.

As my tests show, there is a circa 1 to 20 ratio between MF and cermet for
1/f noise.

For example, a 1 kohm cermet resistor with 25 volts DC bias ( ie 42uV per
decade of frequency ) would have its break frequency at about 100MHz.



...... Phil

 

[[email protected]]

Phil,

Did you try different values and better resistors to rule out
temperature fluctuations?  1 part in 10**7 is only 5 millikelvins for a
200 ppm/degree resistor, which is easily believable, but temperature
drift is not the same as 1/f noise.   A stirred oil bath or ultralow
tempco resistors would help.

Hmm, I measured noise at 100 Hz and higher. Hard to imagine the
resistor changing temperature that fast. Isn't there some "thermal
time constant" for a cube of side 'a'? I could always stcik it in some
bigger mass.

I've made those sorts of measurements too, and found no 1/f noise down
to a part in 10**8, using ~10k ohm RN55C resistors and battery power.

Cheers,

Phil Hobbs- Hide quoted text -

- Show quoted text -

OK. what frequency range and bais voltage was that? When biased at
27/2 Volts and sampled with a Q=1 two pole band bass filter at f=100Hz
I say about twice the Johnsone noise of the 10k ohm resistor. So
something like 130-200 nV if I divide by the applied voltage, that's
1-2 X10^-8 V/V. But my band width was not a decade? Do I need to
multiply by 3? (sqrt 10)? Oh these were cheap Xicon MF resistors.

But I really only did these measurements as checks on the much larger
excess noise of the carbon resistors.

It could have been noise in the batteries.

George Herold

 

[Mike Monett]

I measured it using a 9V battery, a low noise preamp (homemade,
about 650 pV/sqrt(Hz)) and an HP 3562 dynamic signal analyzer.

Phil Hobbs

Phil,

Getting an amplifier with flicker noise that low near DC is pretty
good. Any chance to post a schematic?

Thanks,

Mike Monett

 

[[email protected]]

"Phil Hobbs"



**  Wot a complete fuckwit  .....







** You sure need to be.

 Try  ACTUALLY  reading my post  -  pal.

 See the bit about "... audio band filter " ??

 Cretin.

 >>> I've made those sorts of measurements too, and found no 1/f noise down
to



 **   My, there is a very bad echo in here .........


 ** You  don't say  ????


 **  You don't say  ????

  0.1uV/V  looks like a constant ratio to me.


**  Some give a typical figure as well as a "max"

 -  normally 0.07 uV/V for MF types.

  Ie  30% less.

 Yawnnnnnnnnnnnnn....


**  The mind numbingly idiotic sophistry  YOU   spend your time pouring all
over Usenet would choke an elephant.

   Kindly **** off.

.....  Phil- Hide quoted text -

- Show quoted text -

Geesh Phil, Give Phil H. a break. He certainly knows more than me
about noise. (Which isn't saying much, I only claim to be able to
measure it at some level.) Isn't there some chance that the
resistors he looked at had lower noise?

George Herold

 

[[email protected]]

I measured it using a 9V battery, a low noise preamp (homemade, about
650 pV/sqrt(Hz)) and an HP 3562 dynamic signal analyzer.

I'm not claiming that anyone's measurements are wrong, only that there
are lots of other sources of low frequency junk that are not flicker
noise.  Temperature drift is one very strong candidate.

Millikelvin temperature changes at low audio frequencies aren't that far
fetched.  The forcing is fairly wideband and the thermal mass is small.
     I have a bunch of really nice 1 ppm/K resistors that I'll try out
in the next few days--bug me if I forget.

Cheers

Phil Hobbs- Hide quoted text -

- Show quoted text -

Yeah, well my junk was just a 4nV/rtHz op-amp... and the rest home
built electronics. And I'll claim nothing about my battery noise.
But say can't I estimate the resistors thermal time constant. I've
done this for pieces of metal and at least got within a factor of two
or so. I guess I need to know the heat capacity and thermal diffusion
coef. I have no idea what materials they make resistors out of.

George H.

 

[[email protected]]

<[email protected]>
 "Phil Allison"
 "Phil Hobbs"





Geesh Phil, Give Phil H. a break.

** What part of him  should I break first  ??


** The demented cretin is full of absurd crap.

(Which isn't saying much, I only claim to be able to
measure it at some level.)  Isn't there some chance that the
resistors he looked at had lower noise?

** I posted the details of my test set-up and the resistors I tested.

All the results are  *entirely consistent* with maker's published data and
the laws of physics.

The direct opposite is true for Hobbs.

Planet Hobbs is not in this universe.

......   Phil

Phil, I'm certainly not questioning your results. But if there are
lower niose MF resistors out there don't you want to know of them. A
resistor with lower excess noise does not break any law of physics.

George H.

 

[Phil Allison]

"Phil Hobbs"




Geesh Phil, Give Phil H. a break.

** What part of him should I break first ??


** The demented cretin is full of absurd crap.

(Which isn't saying much, I only claim to be able to
measure it at some level.) Isn't there some chance that the
resistors he looked at had lower noise?

** I posted the details of my test set-up and the resistors I tested.

All the results are *entirely consistent* with maker's published data and
the laws of physics.

The direct opposite is true for Hobbs.

Planet Hobbs is not in this universe.

Phil, I'm certainly not questioning your results.

** Yawwwnnnn - not relevant to anything I wrote.


But if there are
lower niose MF resistors out there don't you want to know of them.

** Yawwwnnnn - not relevant to anything I wrote.


A resistor with lower excess noise does not break any law of physics.


** Yawwwnnnn - not relevant to anything I wrote.

Go away.

Plane Hobbs is waiting for you.



...... Phil

 

[Mike Monett]

Mike Monett wrote:

That was several years ago. I probably still have it kicking
around in a drawer in my lab.

It used iirc 3x Toyo-Rohm low noise transistors (2SD786) in
parallel.

The 1/f corner was around 1 Hz or a bit below (good BJTs are
amazing).

Phil Hobbs

Heh, I'm having problems finding a datasheet in Engish. There's some
tantalizing fragments, such as

en = 0.55nV / sqrt(Hz) (at 10Hz, 10mA)

http://doc.chipfind.ru/pdf/rohm/2sd786.pdf

There's also some discussions of the low noise at:

http://sci.tech-archive.net/Archive/sci.electronics.design/2005-
12/msg02183.html

http://sci.tech-archive.net/Archive/sci.electronics.design/2005-
12/msg02317.html

There's probably more but I didn't take time to check.

The 2SD786 seems to be available for $26.98, but there's no date so
it's not clear if that price is still valid:

http://littlediode.com/components/2SD786_Transistor.html

The Analog Devices MAT02 and SSM-2210 are less expensive at about
$6.00/1k:

http://www.analog.com/en/other/matched-
transistors/MAT02/products/product.html

The Mat02 and SSM-2210 are also supposed to have a fairly low corner
frequency, depending on the collector current. But it doesn't look
to be as low as 1Hz for the 2SD786:

http://www.analog.com/static/imported-files/data_sheets/MAT02.pdf

http://www.analog.com/static/imported-files/data_sheets/SSM2210.pdf

The SSM-2210 datasheet describes three devices in parallel, giving
500pV/SQRT(Hz) and a corner of 1.5Hz. It uses a red LED as a bias
current reference voltage. This is also described in the Analog
Devices app note AN-102:

http://www.elektronik.ropla.eu/pdf/stock/adi/temp/pdf/2611.pdf

So it would be very interesting to take a peek at your
implementation if you have time to post it.

Thanks,

Mike Monett

 

[[email protected]]

I measured it using a 9V battery, a low noise preamp (homemade, about
650 pV/sqrt(Hz)) and an HP 3562 dynamic signal analyzer.

I'm not claiming that anyone's measurements are wrong, only that there
are lots of other sources of low frequency junk that are not flicker
noise.  Temperature drift is one very strong candidate.

Millikelvin temperature changes at low audio frequencies aren't that far
fetched.  The forcing is fairly wideband and the thermal mass is small.
     I have a bunch of really nice 1 ppm/K resistors that I'll try out
in the next few days--bug me if I forget.

Cheers

Phil Hobbs- Hide quoted text -

- Show quoted text -

Say Phil H. I was thinking about this last night. What is the current
noise from your low noise pre-amp? (I assume the typical 1 to 2 pA/
rtHz.) When looking at a 10 kOhm resistor isn't this going to swamp
the wonderful low voltage noise?

George Herold