# Definition of amp noise

Discussion in 'Electronic Design' started by [email protected], Aug 9, 2005.

1. ### Guest

Hi,

Could someone so kindly tell me exactly what 1.3nV / Sqrt(Hz) @ 1KHz
means on a differential amp spec? What would the noise be between
37.00 Hz and 37.01 Hz? Would it be 1.3nV / Sqrt(37.00 - 37.01) or
1.3nV / Sqrt(37)

Thanks,
Paul

2. ### Guest

Sorry, that should have been 1.3nV * Sqrt(37.00 - 37.01) or 1.3nV *
Sqrt(37)

3. ### DaveGuest

Neither.

Since the units in the specs are V/Sqrt(Hz), you have to *multiply* by
the sqrt(bandwidth) to get the RMS voltage.

V = 1.3 * sqrt(0.01) nV
= 0.13 nV RMS

Since you only know what it is at 1 kHz, which is not too close to 37Hz,
it is a bit of an assumption to make that they will be the same, but
it is not too far off.

4. ### Winfield HillGuest

wrote...
That's 1.3nV times sqrt 0.01Hz. But one has to ask how the 0.01Hz
bandwidth is obtained; it's likely that other factors are involved.

5. ### Don LancasterGuest

Linear Technology has some outstanding tutorials on this.
Note that op amp noise depends very strongly on frequency band and
source impedance.

Bottom line: 1.3 is "pretty good" but not spectacular.

--
Many thanks,

Don Lancaster
Synergetics 3860 West First Street Box 809 Thatcher, AZ 85552
voice: (928)428-4073 email:

Please visit my GURU's LAIR web site at http://www.tinaja.com

6. ### Guest

Source impedance, very interesting. Thanks for the information Don.
I'll have to read up on that. I'll be dealing with high impedance.

Paul

7. ### Guest

Thanks, it makes sense. So at 0.01 Hz bandwidth it comes to 130pV.
Where does the noise come from? Is it purely thermal or just a
byproduct of the amplification process?

Here's my problem. I am trying to get a rough guestimate of the values
to see if it is even plausible. I am estimating that a signal will be
picking up ~ 1pV of RF at 37Hz on a coil. I guess it would take a
differential amplifier fed into another differential amplifier. If the
gain of each amplifier is 1E+6 than that's a total gain of 1E+12. Wow!
The device will be out and away from the city and will contain a
second nearby coil to cancel as much RF noise. So there would be two
coils-- the RF coil and the noise cancellation coil. I believe that if
the output from noise cancellation coil is properly adjusted /
calibrated that it can cancel out the RF noise from the output coil.
So if we can get the RF noise low enough then what is left? Thermal
noise? Also I would know the phase of the 37Hz signal. Could that be
an aid? The end goal is to design a device that can tell if the 37Hz
1pV signal is on or off. Is this doable? I think time is on our side.
That is, if we build a filter with high enough Q then it could detect
the signal. The problem is that the Q would probably have to be so
high that it may take a long time for the signal to rise above the
noise. Hopefully some clever math technique, possibly taking advantage
of the known signal phase, can bring the detection time under a minute.
I've worked with FT functions before, but off hand I don't see how the
phase would help that much. What if the 37Hz signal could be modulated
at say 3.7Hz? Could that help somehow?

I really appreciate any tips,
Paul

8. ### Don LancasterGuest

Sensitivity of a rf pickup system is swamped by terrestral noise below

Unless they are apallingly and bad and mesmerizingly awful, you will
observe no difference whatsoever in noise between a high noise and a low
noise amplifier picking up rf at audio frequencies.

One approach to narrow bandwidths is to use phase lock loop or
synchronous demodulator technologies.

Sounds to me like the fundamental physics of what you are trying to do
is not very well thought out.

--
Many thanks,

Don Lancaster
Synergetics 3860 West First Street Box 809 Thatcher, AZ 85552
voice: (928)428-4073 email:

Please visit my GURU's LAIR web site at http://www.tinaja.com

9. ### Rene TschaggelarGuest

These 1.3nV/rtHz apply to 100 Ohms or so. It'll be much
higher with high impedance.

Rene

Beware of the noise figures in the data sheet. They may not apply at
such low frequencies where 'flicker noise' often predominates. (A
Nuvistor can sometimes out-perform modern semiconductors for low LF
noise)

If the option is open to you, it is often more cost-effective to improve
the level of signal pick-up, rather than trying to get ever nearer to
perfection in the amplifier.

11. ### Don LancasterGuest

By a conservative estimate, the manmade and terrestrial noise at this
frequency will be something like 170 decibels ABOVE the ktb noise.

It does not matter in the least which device he uses for his input stage.

Unless everything is in an anechoic fully shielded room.

--
Many thanks,

Don Lancaster
Synergetics 3860 West First Street Box 809 Thatcher, AZ 85552
voice: (928)428-4073 email:

Please visit my GURU's LAIR web site at http://www.tinaja.com

12. ### Guest

Hey everyone thanks for the great information. I found one site that
listed terrestrial noise down to 100KHz

Don, I get a power ratio of 1e+17 and voltage ratio of 3e+8 for 170 dB.
What exactly does it mean to have a 170 dB above ktb noise? I
understand that ktb noise is receiver input noise. I see the noise
graph in above link also uses that terminology; e.g. over ktb. So 170
dB over ktb means the terrestrial noise is 3e+8 times greater than
receiver input noise? The noise graph shows dB increasing as frequency
decreases. Shouldn't it be the opposite? Perhaps those numbers are
negative; e.g., -120dB. According to the graph your 170 dB seems
reasonable. Have you seen any graphs that come close to 37Hz?

Last, how do they figure the terrestrial noise values? Is it based on
a dipole antenna at a fixed length or what? For example, wouldn't a
200-meter dipole pick up more noise than say a 10 cm radius 100-turn
coil?

Thanks,
Paul

His budget and effort would be best put into making the aerial more
directional.

At that frequency, spaced aerials and a correlation technique might be
the way to go.

14. ### Don LancasterGuest

A parabolic reflector with a 60 decibel gain might do the trick.

Tiny problem with zoning, though.

--
Many thanks,

Don Lancaster
Synergetics 3860 West First Street Box 809 Thatcher, AZ 85552
voice: (928)428-4073 email:

Please visit my GURU's LAIR web site at http://www.tinaja.com

15. ### Rene TschaggelarGuest

LOL. A 60dB parabolic at 37Hz .... Lambda is around 10'000km ...
Drop a wire into the ocean.

Rene

16. ### Ken SmithGuest

For those who are very interested in low noise at lowish frequencies:

Take a look at the company called "interfet" in Texas. They make JFETs
with a noise of about 1/4 nV/sqrt(Hz).

You can also get that level by putting 16 LSK170s in parallel.

17. ### Ken SmithGuest

Yes, using the 1PPS pulse from a GPS you could keep the systems in sync to
a fraction of a degree. If you spinkle the recievers over the entire
surface of the earth, the result would be fairly directional and reject
local noise quite well.

18. ### Guest

Don, after further studying, I believe that I can cancel out nearly all
the terrestrial noise with a second nearby coil since the 37Hz signal
would be at a local small area. The second coil, that would be away
from the 37 Hz signal, would be the terrestrial noise cancellation
coil. I would expect the terrestrial noise to cancel out fairly well
since the RF wavelength at 37 Hz is over 8 million meters.

Thanks Ken. I found a great pdf file on noise,
http://eesof.tm.agilent.com/docs/iccap2002/MDLGBOOK/7DEVICE_MODELING/6NOISE/NOISEdoc.pdf

In considering the JFETs as an amplifier, wouldn't I have to use
several resistors? It seems that resistors are a definite source of
noise, which would add to the 1/4 nV/sqrt(Hz) JFET.

Yes, great idea. I was thinking about that yesterday. Although I
didn't know that it was also the sqrt() of total parallel devices.
Isn't paralleling equal to multiple sampling? Consider a computer
program that does averaging. Say we take 10 samples, sum up all 10
samples, and take average signal. I thought that 10 averages would
decrease random noise by a factor of 10. BTW, the program would be in
sync with the 37Hz signal so that averaging would not cancel it out
like the noise.

Does anyone have any input on the following idea? According to the
above pdf, Inductors are considered fairly noise free. Perhaps that
pertains to well made L's. I have not verified that. If that is true
then a transformer should also be fairly noise free-- say a nice
toroid. Why not amplify the signal to say a few hundred times with the
transformer and then use a differential amplifier? That's presuming
that I am able to cancel out enough of the terrestrial noise. For
example, lets say we have a 200pV 37Hz signal with no noticeable noise.
After the transformer, we get a 20nV 37Hz signal with no noticeable
noise. Now our amp should have no problem with the 20nV signal. Lets
say the amp adds 1nV of noise. So that's a 20:1 signal to noise ratio.
Without the transformer it would be a 0.2:1 signal to noise ratio.

Thanks for the input,
Paul

19. ### Guest

Don, after further studying, I believe that I can cancel out nearly all
the terrestrial noise with a second nearby coil since the 37Hz signal
would be at a local small area. The second coil, that would be away
from the 37 Hz signal, would be the terrestrial noise cancellation
coil. I would expect the terrestrial noise to cancel out fairly well
since the RF wavelength at 37 Hz is over 8 million meters.

Thanks Ken. I found a great pdf file on noise,
http://eesof.tm.agilent.com/docs/iccap2002/MDLGBOOK/7DEVICE_MODELING/6NOISE/NOISEdoc.pdf

In considering the JFETs as an amplifier, wouldn't I have to use
several resistors? It seems that resistors are a definite source of
noise, which would add to the 1/4 nV/sqrt(Hz) JFET.

Yes, great idea. I was thinking about that yesterday. Although I
didn't know that it was also the sqrt() of total parallel devices.
Isn't paralleling equal to multiple sampling? Consider a computer
program that does averaging. Say we take 10 samples, sum up all 10
samples, and take average signal. I thought that 10 averages would
decrease random noise by a factor of 10. BTW, the program would be in
sync with the 37Hz signal so that averaging would not cancel it out
like the noise.

Does anyone have any input on the following idea? According to the
above pdf, Inductors are considered fairly noise free. Perhaps that
pertains to well made L's. I have not verified that. If that is true
then a transformer should also be fairly noise free-- say a nice
toroid. Why not amplify the signal to say a few hundred times with the
transformer and then use a differential amplifier? That's presuming
that I am able to cancel out enough of the terrestrial noise. For
example, lets say we have a 200pV 37Hz signal with no noticeable noise.
After the transformer, we get a 20nV 37Hz signal with no noticeable
noise. Now our amp should have no problem with the 20nV signal. Lets
say the amp adds 1nV of noise. So that's a 20:1 signal to noise ratio.
Without the transformer it would be a 0.2:1 signal to noise ratio.

Thanks for the input,
Paul

20. ### Ken SmithGuest

Resistors up to a few Gohms can be bought. They introdude very little
noise current so they can be used to bias the gate. The resistor in the
drain circuit will have less effect on the noise because the signal is
already amplified at that point. The resistor in the source, should be
bypassed with a very large capacitor so that it drops out of the picture
well below the 37Hz.

Try this:

Take one coin call heads +1 and tails -1. Flip it a bunch of times and
take the RMS. (Or not bother because we know it comes out as one)

Take two coins and do the same adding the values.

On the average 1 time in 4 you get -2 and one time in 4 you get +2. The
other 2 times you get zero. Therefor the RMS will come out to:

sqrt ( (2^2 + 2^2 + 0^2 + 0^2)/4 )

sqrt(2) = 1.414..etc

Any time you add random stuff together you get this sort of squarerooty
thing happening.

The signal, however adds linearly so your signal to noise is improved by
N/sqrt(N) or simply sqrt(N).

Ideal inductors are noise free. The resistances of an inductor make noise
just like resistors. Magnetic cores make a thing called Barkhausen(sp)
noise that sounds like snapping and poping on headphones. Inductors also
tend to pick up AC magnetic fields and are also magnetic.
Your frequency is kind of low but take a look at Triad's or Tamura's small
audio transformers.
Think more along the lines of getting up to 10 to 100 times not hundreds.
Hundreds will not be easy to do and 10 or so should be enough. The
amplifier does not need to be differential BTW.

You can make you pick up a tuned circuit to get some signal increase and
filtering right at the start.