Biasing opamp the right way (was: Opamp CAP in series to ground)

[AccountRemovedPerUsersRequest]

Hi. I found this circuit and it does what I need to but just wondering is this a good practice (using C2 to remove offset)? Basically it amplifies the IN signal but the offset of OUT signal follows the IN offset (about 0.8V in this case).
AC input of the IN signal is not shown in circuit since it is not related to the question.

A

[Kleinstein]

The AC part of the signal is amplified, but the DC part is not. This can be usefull, e.g. for audio, where there is a low limit for the signal frequencies and a possible DC offset should not be amplified. A similar circuit is very common with audio power amplifiers.

[David Hess]

The capacitor means that at DC, gain is reduced to 1 minimizing the output offset voltage.  This is a common technique in most AC coupled applications.

In audio applications where the divider impedance is low to reduce noise, the capacitor must be large to achieve a low enough cutoff frequency however if an electrolytic or high dielectric constant ceramic capacitor is used, it will likely cause distortion in the transition band.  The solution for this is to use much more capacitance than necessary moving the transition band even lower.

[AccountRemovedPerUsersRequest]

Thanks. Got it.
Oversizing the capacitor value noted. Good point.

A

[TimFox]

Note that an input coupling capacitor should be added from the source to your node marked “In”, which has a DC bias of +0.8 V, as you mentioned.  The value is normally computed from the low-frequency requirement and the input resistance (83 k here), but a larger value is often required for noise reduction.  Although the capacitor is not a noise source, the input noise current of the op amp with a high source impedance (capacitive reactance at low frequencies) will increase the input noise voltage.

[AccountRemovedPerUsersRequest]

Hello,

Thanks TimFox and others.
The coupling capacitor is in place - was not shown in the attached schematic. For 200kHz input I have simulated that 220pF should be pretty close to optimal. Or... is it in the name of noise (?)

However, I'm a bit stuck how to determine the optimal resistor values (R1 and R2) to get the optimal driving impedance for the opamp. I think I can calculate what is the impedance using 100k+100k (=50k) and I think that the datasheet says that the input impedance for LTC6268 is >1000GR. And how much R1 and R2 affects to noise....

A

[David Hess]

However, I'm a bit stuck how to determine the optimal resistor values (R1 and R2) to get the optimal driving impedance for the opamp. I think I can calculate what is the impedance using 100k+100k (=50k) and I think that the datasheet says that the input impedance for LTC6268 is >1000GR. And how much R1 and R2 affects to noise....

The operational amplifier's input impedance is not what is important because it will be orders of magnitude higher than practical circuits will use.

The limit on the input impedance is usually about controlling noise, and sometimes about controlling drift or offset.  The input current noise through the input impedance produces a voltage noise and at high resistance, this will be greater than the input voltage noise of the amplifier itself.  The input bias current and change of input bias current also produce input offset voltage through the input resistance.  The input resistance by itself creates voltage noise which can become significant at higher resistances.

So the input voltage noise divided by the input current noise is just an estimate for the maximum input impedance that the amplifier should see for good noise performance.  If a higher or lower input impedance is used, then an amplifier with a different ratio of input voltage noise to input current noise will provide lower noise performance.

[AccountRemovedPerUsersRequest]

Hello. Thanks. In that case the driving impedance is about 1/10 of the estimated level. I definitely want to control drift as well. Which direction (in terms of driving impedance) I should go to balance between noise level and drift?

A

[David Hess]

Hello. Thanks. In that case the driving impedance is about 1/10 of the estimated level. I definitely want to control drift as well. Which direction (in terms of driving impedance) I should go to balance between noise level and drift?

The numbers usually end up about the same.  The additional drift from the source impedance comes from the change of input bias current over time and temperature compared to input offset voltage drift so it is the same type of calculation.

The exception to this comes from the load impedance on the output of the operational amplifier causing self heating of the amplifier, which is why the highest precision circuits use a higher load impedance.  Low impedance networks can also suffer from self heating.  If you have to drive a low impedance from a precision amplifier, then an external buffer may be required.

[AccountRemovedPerUsersRequest]

Hello,

Thanks. Comparing OPS836 and LTC6268 (which are my current candidates) the input impedance is very different when using the formula 'Input Voltage noise' divided by 'Input Current noise'. TI's own (at least one OPA836) white paper uses 2M resistor for biasing while they should be <20k resistors. This is making me confused hence continuing discussion here... On the other hand, one example in the data sheet uses 2k resistors.

A

[magic]

LTspice has a noise simulation mode and the bundled models of LT opamps include noise data consistent with datasheet specifications. Run this and you will quickly get an idea how resistors affect things. (You will need to specify an "input" source, set it to any dummy voltage source in the circuit, it doesn't matter for simple uses).

As usual with simulation, it is recommended to understand why the results are what they are. For that, familiarize yourself with specifications for input referred voltage noise and current noise density. Input current noise flows through external resistors and converts into voltage noise, input voltage noise just exists, both are amplified by closed loop gain. The sim also includes Johnson noise of resistors.

[TimFox]

The ratio of V_noise/I_noise is the source resistance that gives the lowest noise figure (not lowest noise) for the entire circuit, including the source resistance.  That optimal source resistance is typically lower for BJT-input amplifiers than for FET-input amplifiers.  In a circuit where the input is capacitively coupled to the source, at mid-frequency (where the series capacitors have negligible reactance compared with the resistors), the total source resistance is the parallel combination of the source's resistance and any bias resistors at the amplifier input.  (For a typical non-inverting op-amp circuit, you must also include the resistance of the feedback network.)
Note that you never improve the noise by adding a resistor (either in series or parallel with the input) to the circuit.  The noise figure represents the noise added to the inherent (thermal) noise of the source resistance.

[AccountRemovedPerUsersRequest]

Thank you TimFox. That makes sense now. And thanks for magic. I think I managed to figure out good enough resistor values by simulating. I do not know if a few uV noise is good in general but for this application it is low enough. Temperature drift is much bigger issue, but it is an another story.

A

[TimFox]

If you are worrying about DC drift, then everything before the input coupling capacitor does not matter--but the biasing/pull-down resistors at the op-amp input will affect the dependence on input bias current drift.

[AccountRemovedPerUsersRequest]

Yes exactly.
The amplifier is feeding signal to ADC thru a filter. I am pretty sure that every single component has its own temperature drift. But... it is another story.
What comes to the question of resistor values looks like simulated values works just fine in practice too. I just didn't want to this a way that "lets pick either 100k or 1M resistors and its fine".

A

[David Hess]

The ratio of V_noise/I_noise is the source resistance that gives the lowest noise figure (not lowest noise) for the entire circuit, including the source resistance.

That is right.  Lower impedance will give lower noise because the contribution from current noise and Johnson noise will be reduced but the ratio of V_noise/I_noise is a good place to start as an upper limit.

Thanks. Comparing OPS836 and LTC6268 (which are my current candidates) the input impedance is very different when using the formula 'Input Voltage noise' divided by 'Input Current noise'. TI's own (at least one OPA836) white paper uses 2M resistor for biasing while they should be <20k resistors. This is making me confused hence continuing discussion here... On the other hand, one example in the data sheet uses 2k resistors.

They are two very different parts, with the bipolar OPA836 having 150+ times higher current noise than the LTC6268.  2M would be pretty insane for the OPA836; maybe that was a typo.

[AccountRemovedPerUsersRequest]

Maybe a typo, but not mine. Attached image is from TI's dev board "tidrys5".

A

[TimFox]

That biasing arrangement (two 2 M resistors in a voltage divider, large bypass capacitor, and medium resistor to amplifier) is sometimes called “noiseless biasing”, since the capacitor C45 filters AC noise (thermal, 1/f, and supply ripple).  The resulting AC noise and input impedance is due to R66 = 24.9 k, but the DC drift depends on the total resistance (1.02 M).

[David Hess]

I thought it might be something like that.  The DC drift and low frequency noise is very high with such a high biasing resistance but the amplifier's low frequency gain is reduced by the coupling capacitor in the feedback network.

[AccountRemovedPerUsersRequest]

Thanks. I just read from somewhere that this setup would give 1M+24k9 impedance and I was so confused. Since that is misinformation (in terms of the input impedance) the math starts to work.
Using simulator I ended up having 56k and 47k resistors. Now comparing that to 24k9 I think I got this.

A

[AccountRemovedPerUsersRequest]

After a second thought, I should not have stumbled to that hole. It is clear that strong enough capacitor makes the resistor divider a virtual ground for the R66 resistor. So, from the impedance matching point of view it is just a common biasing circuit in the end and nothing more complicated than that.
Understanding better opamp noise behavior will help me picking more suitable opamps next time I design something...

A