Microphones with OpAmp and de-emphasis feedback network

[Itsme]

I would need something that looks like this:



That way we would get a more realistic behavior over frequency. Even some crude estimations about robustness against RFI would be possible..

 

[Itsme]

You might wanna consider using "AC 1" for AC analyses - that way, the graph will be referenced to 1V = 0dB.

Thanks for the suggestion. I will use it for specific adjustments for sure. While working on concepts, I prefer to have the actual levels of the internal nodes. I also usually switch to a log scale for that purpose.
Btw. everybody is invited to use the simulations and to play around with them

 

[Екатерина]

Perhaps you can take a look at these two, hoping they will be helpful to you

 

[Khron]

While working on concepts, I prefer to have the actual levels of the internal nodes. I

For transient simulations, sure. For AC ones there's literally no point, unless you insist on making the math difficult for yourself (and whoever else you're showing those graphs to).

In those screenshots, the "peak" level is somewhere around -35dB with a 1mV (Vpeak? RMS? Vpp?) input signal - whatever the hell does that mean? With "AC 1", you'll get clearer numbers (6dB's gonna mean a gain of 2, plain and simple, etc)

As i mentioned, you can set "AC 1000", and youll get the same smooth graph, just huge dB numbers, that's all it affects. But suit yourself...

 

[Itsme]

Perhaps you can take a look at these two, hoping they will be helpful to you

Not complete but showing the test circuit and some curves would possibly allow to tweak the model until it reproduces the shown curves... I'll give it a try.

 

[Itsme]

For transient simulations, sure. For AC ones there's literally no point, unless you insist on making the math difficult for yourself (and whoever else you're showing those graphs to).

In those screenshots, the "peak" level is somewhere around -35dB with a 1mV (Vpeak? RMS? Vpp?) input signal - whatever the hell does that mean? With "AC 1", you'll get clearer numbers (6dB's gonna mean a gain of 2, plain and simple, etc)

As i mentioned, you can set "AC 1000", and youll get the same smooth graph, just huge dB numbers, that's all it affects. But suit yourself...

I was referring to my to how I use simulations in my day job as an EE. I'm very confident that the way I do simulations make a lot of sense to me .
As mentioned I appreciate your input for displaying the frequency response for specific mic implementations. I don't come from the audio side of engineering so other conventions apply to my kind of work.
The source level of 1mV in the first examples is almost totally random as no specific capsule is defined yet.
Once we choose a specific capsule I would prefer to use mVpp which allows to monitor any violations of max. amplitudes or not sufficient utilization of the available dynamic range.
I hope that explains the rationale of my approach at least a bit.

 

[Khron]

I was referring to my to how I use simulations in my day job as an EE.

No reason to brag, good for you, but it sounds (no pun intended) like your day-job is NOT audio-related..?

Once we choose a specific capsule I would prefer to use mVpp which allows to monitor any violations of max. amplitudes or not sufficient utilization of the available dynamic range.

For the third time - in simulations, that's relevant only for transient simulations, where you WILL be able to see / measure distortion, clipping etc.

For AC sweep simulations, take a dive and set the AC value for the signal source to "1" (not 1m), and then you'll be able to display and easily see the ACTUAL dB gains / values of the circuit - 0dB being unity-gain, +6dB being 2x amplification, etc.

(As you may or may not have noticed, you CAN (and must) set different values of the signal source, that get used in AV versus transient simulations.... Right?)

As i mentioned, you can set "AC 1000", and youll get the same smooth graph, just huge dB numbers, that's all it affects.

Which is why i said this.

With your current "AC 1m", you get something like "-35dB" peak signal in your graph in post #7 here. What does that mean??? How does one quantify that, without spending half an hour crunching through tedious math?

Just try setting the "AC 1", that's all i ask, make it easier for all of us - only takes you 5sec to try it, and if you don't like it, you can always go back

 

[Itsme]

No reason to brag, good for you, but it sounds (no pun intended) like your day-job is NOT audio-related..?

Not recently. But my background is in precision analog and mixed signal electronics, but it's not limited to that...

For the third time - in simulations, that's relevant only for transient simulations, where you WILL be able to see / measure distortion, clipping etc.

For the process of developing a circuit I still prefer an absolute scale even in the time domain. I often work with very high dynamic ranges and as an example, the change of the output voltage of a high precision voltage reference circuit might be only a few ppb. At a linear scale in a time domain simulation it be comes very cumbersome to probe different nodes and zoom in all the time. Shifts in the frequency domain at a log scale are easy to see even if a 10V signal shifts by a few nano volts...

With your current "AC 1m", you get something like "-35dB" peak signal in your graph in post #7 here. What does that mean??? How does one quantify that, without spending half an hour crunching through tedious math?

At this point of the design absolute values are meaningless - to me at least. For now I focused on the functionality of the circuit like how to influence the frequency response or making sure the circuit has sufficient phase margin and doesn't oscillate.

Just try setting the "AC 1", that's all i ask, make it easier for all of us - only takes you 5sec to try it, and if you don't like it, you can always go back

I already confirmed that I will do that...
I provided the .asc file for everybody to use it. You can easily change the settings to your liking and share it then...

 

[Khron]

At this point of the design absolute values are meaningless - to me at least.

Exactly - we're looking at that frequency response graph in dB, which is a RATIO (and thus, relative) - i think / hope we can agree on that?

how to influence the frequency response or making sure the circuit has sufficient phase margin and doesn't oscillate.

Once again, all relative (dB) numbers.

But let me try to exemplify more clearly what i mean:

Your signal source in LTspice is 1mV, right? Vpp, Vrms, the AC simulation doesn't care. But 1mV, whatever it is, is what, -60dBV (if we take 1V as the 0dB reference)?

So your resulting frequency graph peaks out at -35dB or so? That means you have a maximum gain of... -35dB - (-60dB) = 25dB or so? Why the guesswork? Why the unnecessary math?

Set the AC source to "1". Probe it, you'll get a nice flat line at 0dB. Probe the output of the circuit, you'll get whatever you get referred to 0dB.

Why the obfuscation?

As I've mentioned before, AC simulations will NOT show you where the circuit clips etc, so expecting to determine headroom or anything absolute like that is pointless. That's what the transient simulations are for, and those ignore whatever you set as the AC signal value. That's why it's in a separate section of the configuration window of the voltage/signal source.

Am i making any sense here?

 

[MicUlli]

Only for clarity:
NFB networks in OpAmp Circuits NEVER reduce SNR. But depending on the circuit topology NFB networks can be designed that they sacrifice SNR only a little bit.

 

[micolas]

Only for clarity:
NFB networks in OpAmp Circuits NEVER reduce SNR. But depending on the circuit topology NFB networks can be designed that they sacrifice SNR only a little bit.

Ok. The purpose of the NFB network in our case is to modify the FR.
If we reduce the circuit gain and increase the signal delivered by the capsule through a higher polarization voltage, can we obtain a better final SNR? Of course we establish some reasonable limits for headroom, maximum spl, distortions.

 

[Itsme]

Exactly - we're looking at that frequency response graph in dB, which is a RATIO (and thus, relative) - i think / hope we can agree on that?

Yes we can. Maybe there is a misunderstanding: although I posted results in a dB scale (referred to 1mV (which i would change to actual specs once a capsule is picked) I usually scale the y-axis in powers of 10 which in LTSpice is named 'Logarithmic' in comparison to 'Decibels'. In the 'Logarithmic' setting I can easily probe wide dynamic ranges with actual values. Switching between these two settings is conveniently done by a right click on the y-axis scale but it won't be normalized to 1V. I will do the normalization to 1V in future posts, If I don't find a different display format more meaningful in some cases...

As I've mentioned before, AC simulations will NOT show you where the circuit clips etc, so expecting to determine headroom or anything absolute like that is pointless.

Correct, it won't show clipping. That would be determined in the time domain (or by reading the data sheet of the opamp, in this case. But once I know that value, I can easily monitor the nodes relevant to that - again, that's how I like to run simulations effectively. I often have to go through dozens of iterations of a circuit in a short period of time and that's for me the most efficient way to handle my daily workload.
This is by no means necessarily the right way to simulate in general.

 

[Itsme]

Only for clarity:
NFB networks in OpAmp Circuits NEVER reduce SNR. But depending on the circuit topology NFB networks can be designed that they sacrifice SNR only a little bit.

Unfortunately that's not the case. One influential factor is resistor noise which at low signal levels can't be ignored. Other influences could be inductors that pick up RF from the environment under real-world conditions.
But in a slightly simplified consideration an amplifier amplifies the amplitude of everything exposed to it's input regardless whether it is a wanted signal or noise.
There is a difference between inverting and non-inverting amplifiers though which is due to noise gain, which at a signal gain of (-) 1 is 1 for the non-inverting- and 2 for the inverting configuration.

 

[rogs]

Unless these proposed NFB networks are going to be more sophisticated than simple first order low pass filters, they will only have a limited application.
While they maybe may be suitable as 'de-emphasis' for a specific type of Neumann capsule, a simple LPF is generally a bit too brutal to try and compensate for the high mid range lift, found in many (typically chinese) LDC capsules.
That type of FR correction requires bandpass filtering, to avoid excessive HF attenuation.

That more complex type of filter is best applied outside the microphone, IMHO. Either in a DAW for a recording, or as hardware EQ, introduced at line level into the signal path, for live performance.

 

[micolas]

Unless these proposed NFB networks are going to be more sophisticated than simple first order low pass filters, they will only have a limited application.
While they maybe may be suitable as 'de-emphasis' for a specific type of Neumann capsule, a simple LPF is generally a bit too brutal to try and compensate for the high mid range lift, found in many (typically chinese) LDC capsules.
That type of FR correction requires bandpass filtering, to avoid excessive HF attenuation.

That more complex type of filter is best applied outside the microphone, IMHO. Either in a DAW for a recording, or as hardware EQ, introduced at line level into the signal path, for live performance.

I completely agree with everything you say. You have defined your position very clearly in other posts.
Your concept assumes exactly this: capturing sources directly, as linearly as possible and subsequent processing in hardware or in DAW, depending on the application, live or for recording.
Our declared intention was to combine reasonable filtering in the microphone accompanied by post-processing.
I have not yet met a microphone that did not need any subsequent processing, usable on any source.
We really appreciate your knowledge, original ideas, experience in the field, which is why we ask you to correct us if we are going in the wrong directions.
Many thanks!

 

[Itsme]

Unless these proposed NFB networks are going to be more sophisticated than simple first order low pass filters, they will only have a limited application.
While they maybe may be suitable as 'de-emphasis' for a specific type of Neumann capsule, a simple LPF is generally a bit too brutal to try and compensate for the high mid range lift, found in many (typically chinese) LDC capsules.
That type of FR correction requires bandpass filtering, to avoid excessive HF attenuation.

That more complex type of filter is best applied outside the microphone, IMHO. Either in a DAW for a recording, or as hardware EQ, introduced at line level into the signal path, for live performance.

I also won't propose to go wild with a ratsnest of components.
the approach I chose adds, besides the three gain setting resistors, only two small, inexpensive components. Two more could be used, if desired for the low end of the frequency range. C8 is for compensation purposes and not directly for tweaks in the audible range.
C9 sets the corner frequency while R13 sets the amount of de-emphasis.