Active filters noise issues

[Analog Frontend Designer]

Hi,
Currently I am involved in quite challenging AFE design for myography sensor application.

One of the impotrant issues for SNR improvement is filtering stages noise.
For example here is useful article explained the influence of high Q stages to overall output noise https://e2e.ti.com/blogs_/archives/b/precisionhub/posts/noise-from-active-filters-an-unwelcome-surprise

When I have tried to simulate the noise behaviour of LPF by means of LTspice, actually the lower Q section produces less output noise,
But I was a little bit confused that noise performance doesn't depend obviously on OpAmp type even if noise simulation is mentioned in Spice lib description.

For example, below are two sets of simulation results, first for more noisy (according to datasheet) TLV9042 Opamp and second set for much less noise MAX44260 Opamp. In both cases resistors and caps are ideal ones, so only Opamp noise should affect the final result.
But plots show that MAX44260 has more output noise, which is unexpected literally.

Is it simulation bug or incorrect LTspice work with third-party spice models?
Hope that in real life noise performance depends on Opamp type ans less noisy opamp will produce the less total output noise.

[MasterT]

I think, huge GBW difference plays role.

[Analog Frontend Designer]

I think, huge GBW difference plays role.

Might be, but some doubts about GBWP, because noise peaking located at the same low frequency (about 500Hz).
But I believe thet actual noise performance should be dependent on opamp voltage and current noise

[Benta]

Hehehe.
You've just been bitten in your behind by the Sallen-Key filter. It has no good features, except being easy to calculate.
Throw it in the waste bin, and try with a MFB filter instead.

[magic]

In both cases resistors and caps are ideal ones, so only Opamp noise should affect the final result.

In SPICE resistors have Johnson noise, so you see a combination of that and opamp noise (if it is included in your models).

Test the models in simple voltage follower application to see if they match their corresponding datasheets. Try an ideal opamp (this is noiseless) and see how it works, this will give you the amplified resistor noise alone. Calculate noise gain vs frequency by means of AC simulation and see if your result is consistent with noise density specs times noise gain of the circuit.

[Analog Frontend Designer]

Hehehe.
You've just been bitten in your behind by the Sallen-Key filter. It has no good features, except being easy to calculate.
Throw it in the waste bin, and try with a MFB filter instead.

Actually, MFB has worse noise behaviour.
look at plots below. This is the same filtred based on MFB topology

[Benta]

TBH, it's not clear what you're measuring where.
I can only see that you're biasing the input to 3.3/2 V, and injecting a noise signal.

[Analog Frontend Designer]

In both cases resistors and caps are ideal ones, so only Opamp noise should affect the final result.

In SPICE resistors have Johnson noise, so you see a combination of that and opamp noise (if it is included in your models).

Test the models in simple voltage follower application to see if they match their corresponding datasheets. Try an ideal opamp (this is noiseless) and see how it works, this will give you the amplified resistor noise alone. Calculate noise gain vs frequency by means of AC simulation and see if your result is consistent with noise density specs times noise gain of the circuit.

You are right.
Looks like spice models from TI doesn't works correct in LTspice noise analysis.
TLV9042 was replaced by ideal opamp and plots compared. Very similar to ideal.

[Analog Frontend Designer]

TBH, it's not clear what you're measuring where.
I can only see that you're biasing the input to 3.3/2 V, and injecting a noise signal.

it's just a noise analysis in LTspice. V3 serve as exitation (in specified frequency range) and noise measured at OUT node/

[Kleinstein]

Many Op amp models don't include noise.

[Benta]

TBH, it's not clear what you're measuring where.
I can only see that you're biasing the input to 3.3/2 V, and injecting a noise signal.

and noise measured at OUT node/

Nothing in your schematic indicates that. If LTSpice defaults to the "OUT" label, nice for you. But we don't know. It might also default to the next node in your circuit.
Always be explicit.

[Analog Frontend Designer]

Many Op amp models don't include noise.

Yes, and actually it was a reson of misunderstanding.
Below two plots of MAX44260 and TLV9042 pure noise simulation as unity-gain followers.
In case of MAX results completely corresponds to datasheet figure. For TLV, noise plots doesn't match to specified in datasheet.

[Terry Bites]

See if TINA-TI models this opamps noise correctly.
also see www.analog.com/en/resources/media-center/videos/5579265856001.html

[Analog Frontend Designer]

See if TINA-TI models this opamps noise correctly.
also see

Tina-TI gives the same noise plot as LTspice (for unity gain configuration). And both are wrong, much more better, than datasheet figure show.
So, one extra reason to check everything you can, even manufacturers models.

[Vojtech]

Reduce values of the resistors significantly, R2+R3 alone give about 180 nV/sqrtHz of noise, Op Amp isn't so big a problem.

[Analog Frontend Designer]

Reduce values of the resistors significantly, R2+R3 alone give about 180 nV/sqrtHz of noise, Op Amp isn't so big a problem.

About 70 nV/sqrtHz R2+R3 give.
According to my understanding this resistors noise geometrically added to a
Opamp input noise and gained  (Q and f dependent ).
Also, resistors current noise also present, but it depends on resistor type (MELF will produce the lowest possible current noise among others widely used types).
About resistors values, there should be compromise. To my mind, as lond as you can use NP0 capacitors, you should use the lowest possible resistors values. Because X7 series has much more capacitance deviation across the temperature range, worse initial tolerance and piezo effect. Although, i have never observed the last one (sufficient electrical nouse generated by mechanical impact)

[Sensorcat]

Conclusion: The Sallen-Key filter is already too complicated to start studying something new (effect, method, ...). If you want to find out or learn something new, always use the most simple item, structure, or system possible. Once the new thing is understood, move on.

[Analog Frontend Designer]

Conclusion: The Sallen-Key filter is already too complicated to start studying something new (effect, method, ...). If you want to find out or learn something new, always use the most simple item, structure, or system possible. Once the new thing is understood, move on.

However, now it isn't looks complicated . The reason was in erroneous spice models from TI. To understood this, sure some separate simulations was required.
Sallen-Key filters still better solution when noise might be a concern, because noise gained only near the cutoff frequency region. Filtering stages should utilize the lowest possible Q filter sections.
And another valuable conclusion, for noise-sensitive applications filtering stages should be placed after preamplifier stage, in order to minimize the filter noise contribution into the final output signal.

[Benta]

Keeping the noise part separate, you have a different problem with the Sallen-Key.
It will turn into a HP filter at some point. This is inherent in the topology and needs attention. MFB does not have the problem to the same degree and is much easier to "tame".
I suggest you read this TI app note (pay attention to p. 14...17):
https://www.ti.com/lit/an/sloa049d/sloa049d.pdf

[Sensorcat]

However, now it isn't looks complicated .

Everything looks easy when solved.

To understood this, sure some separate simulations was required.

Yes, and this is my point. If you had started with the most simple configuration possible, instead of the circuit you wanted to have in your application, you could have found out yourself what is going on. But it's your choice if you want to learn a principle from this episode, or if you want to hit the same wall next time, next project, next issue, again.

[Analog Frontend Designer]

Keeping the noise part separate, you have a different problem with the Sallen-Key.
It will turn into a HP filter at some point. This is inherent in the topology and needs attention. MFB does not have the problem to the same degree and is much easier to "tame".
I suggest you read this TI app note (pay attention to p. 14...17):
https://www.ti.com/lit/an/sloa049d/sloa049d.pdf

Topology selection is always tradeoff among different factors and design limitations. Mentioned SKF drawback also known as "stopband leakage" and thats why so important to have correct opamp model for simulation.
Actually, in most cases it isn't so dramatic. At first, you can "move" the leakage point by choosing opamp GBW and output impedance. And second, powerful method to reduce the stopband leakage - use the odd-order filtres with passive first-order stage.
For example, Bode plot of 3-rd order LPF as shown in first post. At least -79dB attenuation at maximum stopband leakage point (and @350kHz, which is opamp GBW). For my application it's more than enough.
But I agree, in some cases MFB might be a better solution, especially when unwanted frequencies are broadband and cover several decades above cutoff.

[Benta]

As long as you're aware of the issue, it's your design choice as engineer. I'm fine with that.

But 95% of designers think the Sallen-Key is great (which it's not!). That bothers me a bit.

[mawyatt]

Besides its simplicity, an often overlooked characteristic of the Sallen-Key LPF is the almost total lack of sensitivity of filter DC Gain to the Op-Amp and all components characteristics in the Unity Gain LP Configuration.

This is one reason we often used the SK for Low Pass Noise Filtering precision voltage references.

Best,

[mawyatt]

Keeping the noise part separate, you have a different problem with the Sallen-Key.
It will turn into a HP filter at some point. This is inherent in the topology and needs attention. MFB does not have the problem to the same degree and is much easier to "tame".
I suggest you read this TI app note (pay attention to p. 14...17):
https://www.ti.com/lit/an/sloa049d/sloa049d.pdf

Topology selection is always tradeoff among different factors and design limitations. Mentioned SKF drawback also known as "stopband leakage" and thats why so important to have correct opamp model for simulation.
Actually, in most cases it isn't so dramatic. At first, you can "move" the leakage point by choosing opamp GBW and output impedance. And second, powerful method to reduce the stopband leakage - use the odd-order filtres with passive first-order stage.
For example, Bode plot of 3-rd order LPF as shown in first post. At least -79dB attenuation at maximum stopband leakage point (and @350kHz, which is opamp GBW). For my application it's more than enough.
But I agree, in some cases MFB might be a better solution, especially when unwanted frequencies are broadband and cover several decades above cutoff.

Many Most op-amp models lack fidelity, especially in the output section, which is the fundamental cause of the dreaded "stop band leakage". We discussed this here awhile back (and way way back in an EDN article).

Improving Stop Band performance does take away from the simplicity of the SKF as it requires added components to lower the effective op-amp output impedance, either a buffer or output bias "bleed" resistor.

One advantage of locating a 1st order low pass section at the overall LP filter input is this helps attenuate the higher frequency components of the input signal from entering the active op-amp and reduces the potential IMD issues.

When designing high order active filters one always wants to locate the highest "Q" filter section at the overall filter end if possible, this maximizes the overall filter DR wrt waveform limiting.

Best,