Any idea to improve this electronic filter?

In my view, the circuit looks like a bridged-T inverting bandpass with the output taken at the (2nd order) lowpass node. With equal caps the Q-factor is the root of the resistor ratio (if I got that right), no wonder that it rings severely with a large ratio like 22:1.
The output load has to be accounted for especially if it's not fully resistive (when it is resisitve its value is simply paralleled to the input resistor). I haven't checked what happens with a capacitive load but I would guess it totally throws off things...

See https://www.diyaudio.com/community/...range-oscillator.205304/page-485#post-6907924
 
It is probably possible to get around that limitation and reduce the settling time by using carefully located non-linear components like diodes, but the leakage of a real 470µ will remain problematic. Something to think about.
If this path doesn't lead to a practical, workable solution, too bad, but that's life, that's engineering, and something else will need to be found.
It is good to know that based upon a deterministic, rational analysis, the problem doesn't have a satisfactory solution. It avoids wasting time on a sterile project.
Many thanks for your insight: it saves me a lot of time
 
I'm not sure if the leakage of a typical (rather than worst case) low-leakage electrolytic capacitor will be a big problem for a version with symmetrical supply. The settling time will be much shorter because a 0 initial voltage is already close to the final value. Leakage with a near zero voltage across the capacitor will also be far less than with half the supply voltage across it. If it is a problem, maybe some sort of capacitance multiplier circuit with a film capacitor can solve it - although you probably need three op-amps then.

Anyway, you're welcome. It was fun to do the calculation!
 
It may be interesting to note that CL is inversely proportional to the third power of C, also known as C3.

With C = 33 pF, the capacitance of the whole circuit drops from 300 nF to 71.181818... nF (still much more than if you had just connected your high-voltage capacitors in parallel), but CL drops from 470 uF to 4.7 uF. There are reasonably small 4.7 uF MKT film capacitors available nowadays. RC becomes 56 kohm.
 
Last edited:
I have tested both versions of the filter, and they worked (almost) as expected.
For the 470µ, I used a non-polar cap, and I had to short it manually after the supplies were applied, because the output of the opamp seemed stuck at the + rail. After ~10 seconds, it began to return slowly towards 0V.
It would probably have been able to recover by itself, but I was a little impatient.
I noticed something harmless, but odd: a low-level oscillation at a fraction of Hz was permanently present. It didn't prevent the circuit from operating normally, and it was completely blocked at the output due to 1.5nF capacitor. The amplitude was low enough (<1V) to keep the opamp in its linear region, and it had thus no practical impact.
The sim shows a quirky behavior between 0.2 and 0.4Hz; it is probably related:

1676200940772.png


The less extreme version behaved exactly as expected.

I wonder whether it would be possible to use the + input in some constructive (and creative) way to boost the performances further, and/or use less extreme components values?
 
Anyway, good to know that the 71.181818... nF version works as intended!

For the 300 nF version, I'll try to come up with alternatives for the inductive term that neither require gigahenry inductors, nor require electrolytic capacitors, nor make the op-amp's output DC voltage excessively sensitive to input bias current, nor cause 1 hour settling times.
 
Damn...I have a feeling this page is going to be another important page where I won't be able to understand a single phrase so I'm writing here one that I can understand. This way I'm gonna be able to claim I was also part of history 🙂 just like Forrest Gump caught in the middle of Watergate scandal ...
Is there any way that mere mortals can understand this calculus?
Maybe this will begin to help -- a handout PDF from MIT: https://web.mit.edu/2.14/www/Handouts/PoleZero.pdf
 
  • Like
Reactions: dreamth
A gentle bump is inevitable due to the zeros in the transfer. Good to know that this one also works in reality!

How long do you have to wait after switch-on?
As I said, I cheated: after 15s or so I shorted the cap and had to wait a further <10s for settling. It probably depends on the offset of the opamp, and the initial conditions. Adding a leakage resistor across the cap might help, and I don't think it would impair the operation of the filter: after all, an Ecap even bipolar is pretty lossy, and it has a 2K7 series resistor in circuit anyway. Replacing it with a 2K4 would result in the same loss level, and allow a clean start under all conditions?
 
I understood from your posts that you cheated when you tried it with the wrong series resistor, that you then simulated it with the correct series resistor, and then tried it again with the correct resistor. I didn't know if you cheated again after changing the series resistor, but maybe I'm just misinterpreting something.

A resistor in parallel with the capacitor will increase the closed-loop DC gain and increase the sensitivity of the settled DC output voltage to op-amp offset, bias current and high-voltage capacitor leakage. To keep those effects under control, the parallel resistor shouldn't be too much smaller than 10 Mohm. The effect on the damping will then be negligible, so the series resistor can remain 2.7 kohm.
 
For example, with 1 Mohm in parallel with the capacitor and not taking into account capacitor leakage, the DC current to voltage transfer changes from -32 Mohm to -251.4075995 Mohm and the voltage gain from 1 to 10.9730727. The series resistor can then just stay 2.7 kohm.

If you would use a parallel resistor of a few dozen kiloohm, you would probably have to correct the series resistor, but the effect on the gain for offset, bias current and high-voltage capacitor leakage would then be quite dramatic.
 
Last edited:
When I made subsequent tests with the correct resistor, I simply kept the cap shorted until after the power was applied, to minimize the startup time.
I am going to test it without any intervention of any kind, to have an idea of the real startup time (using this particular TLO72, supplies, etc.)
I am also going to test the effect of a parallel resistor. I don't think it will have a detrimental effect: the cap will always see the offset voltage of the amplifier, and the output will have this voltage multiplied by the gain, but since the output is AC-coupled, it will only begin to have an impact if it eats up considerably on the dynamic range.
For example, with 220K, the offset, say 2mV, will be multiplied by ~50, resulting in a 0.1V shift. Not really problematic.
In circuits where the only discharge path for the caps is through semiconductor junctions, a residual voltage of 100~200mV can linger on some nodes for hours, sometimes day, and adding a deterministic leakage path can be useful