Hello,
I worked the whole day on calculating the noise of a instrumental amplifier. My goal is to have a analytical model of the noise equations, so i can automaticly find the "best" possible resistor combination (and get a little bit more knowlage how to deal with noise).
I'll do my calculation with Matlab and wrote a little script to try a lot of combinations.
The interisting Question is: should i do amplification with the differential Amplifier and the first stage?
Should the resistors of the first stage be low ore in the differential Amplifier?
Additional I'd like to cross check a few combinations with LT-Spice.
And There my Problem begin:
All "manual" Noise calculations are spot on in the first stage were the opamp sees only a resistor at one input (if the source impedance is zero). But when i do the calculations of the differential Amplifier the "manual" calculations are always worse than the calculations with LT-Spice.
Do anyone know if LT Spice calculates input Noise cancelation?
In the Datasheet of the LT 1028 there is a diagramm of this feature. For the LME 49990 input Noise cancelation is'n named at all?
Is there a way of calculating it?
I worked the whole day on calculating the noise of a instrumental amplifier. My goal is to have a analytical model of the noise equations, so i can automaticly find the "best" possible resistor combination (and get a little bit more knowlage how to deal with noise).
I'll do my calculation with Matlab and wrote a little script to try a lot of combinations.
The interisting Question is: should i do amplification with the differential Amplifier and the first stage?
Should the resistors of the first stage be low ore in the differential Amplifier?
Additional I'd like to cross check a few combinations with LT-Spice.
And There my Problem begin:
All "manual" Noise calculations are spot on in the first stage were the opamp sees only a resistor at one input (if the source impedance is zero). But when i do the calculations of the differential Amplifier the "manual" calculations are always worse than the calculations with LT-Spice.
Do anyone know if LT Spice calculates input Noise cancelation?
In the Datasheet of the LT 1028 there is a diagramm of this feature. For the LME 49990 input Noise cancelation is'n named at all?
Is there a way of calculating it?
Sorry I have nothing more specific, but the LME49990 model is pretty basic and somewhat false. For example, the amp's 5 terminals do not obey Kirchoff's current law - any output current developed at the output is not somehow sourced from the other 4 terminals (most specifically the power supply terminals).
However, the LT1028 seems to be a pretty realistic model, far beyond a simple Boyle op amp model. I would trust it to a much higher degree, but also try to verify it with specific, pathological circuits that zero in on this behavior. In the end, verification and comparison to actual breadboarded circuits will guide your model a lot more accurately, albeit more slowly.
However, the LT1028 seems to be a pretty realistic model, far beyond a simple Boyle op amp model. I would trust it to a much higher degree, but also try to verify it with specific, pathological circuits that zero in on this behavior. In the end, verification and comparison to actual breadboarded circuits will guide your model a lot more accurately, albeit more slowly.
Am I right when I assume that you build the usual instrumentation amplifier
from 3 op amps?
If you are interested in noise, the first stages should have gain,
since then the noise of the second stage will play a lesser role.
It's analogous to the Frijs'schen Formel as its known from RF.
A voltage gain of 10 gets you most of the advantage. The thermal
noise of the feedback network counts also and is added geometrically
to the noise of the other inputs. Small resistors rule, but make sure
that the opamps can drive them. With the small voltages involved
this won't be much of a problem.
The Opamp input stage is run with large current, so the bases will
require a base current that cannot be ignored. LT made sure that
the bias currents of both inputs come from the same source and
thus are correlated. Thus the noise effects on the + and - input cancel.
That works only if the resistance on both inputs is the same, since
noise current * resistance = effective noise voltage at the input.
So if you minimize the resistance of your signal source to get less noise,
you may get punished from an unexpected side if you cannot do the
same for the feedback network.
Other solutions would be to need less bias current such as in FETs or
super-beta-transistors, but FETs usually have more noise and the
LT1028 may be quite super-beta-ish already.
I have seen in the data sheet that they now confess the noise peak
at 400 KHz, that might hurt you. The AD797 and the ADA4898-2
no not have that. The 4898-2 may be interesting for you since you
need several opamps anyway.
Nevertheless, you can get your 1 nV/sqrt HZ from a LT1028 any time.
feel free to email me.
Grüße aus Stuttgart
Gerhard
from 3 op amps?
If you are interested in noise, the first stages should have gain,
since then the noise of the second stage will play a lesser role.
It's analogous to the Frijs'schen Formel as its known from RF.
A voltage gain of 10 gets you most of the advantage. The thermal
noise of the feedback network counts also and is added geometrically
to the noise of the other inputs. Small resistors rule, but make sure
that the opamps can drive them. With the small voltages involved
this won't be much of a problem.
The Opamp input stage is run with large current, so the bases will
require a base current that cannot be ignored. LT made sure that
the bias currents of both inputs come from the same source and
thus are correlated. Thus the noise effects on the + and - input cancel.
That works only if the resistance on both inputs is the same, since
noise current * resistance = effective noise voltage at the input.
So if you minimize the resistance of your signal source to get less noise,
you may get punished from an unexpected side if you cannot do the
same for the feedback network.
Other solutions would be to need less bias current such as in FETs or
super-beta-transistors, but FETs usually have more noise and the
LT1028 may be quite super-beta-ish already.
I have seen in the data sheet that they now confess the noise peak
at 400 KHz, that might hurt you. The AD797 and the ADA4898-2
no not have that. The 4898-2 may be interesting for you since you
need several opamps anyway.
Nevertheless, you can get your 1 nV/sqrt HZ from a LT1028 any time.
feel free to email me.
Grüße aus Stuttgart
Gerhard
I suggest buying this book. You can find them used for less than $20. I bought a copy over twenty years ago.
http://www.amazon.com/Low-Noise-Electronic-Design-C-Motchenbacher/dp/0471619507
http://www.amazon.com/Low-Noise-Electronic-Design-C-Motchenbacher/dp/0471619507
- Status
- Not open for further replies.