Precision active filters û real-life experience sought

[benta]

I’m consulting on a project where an active low-pass filter with a cutoff around 40…50 kHz is needed. Filter order/type would be 4…6th, maximally flat group-delay (no, this does not mean Bessel, but almost). The filter is to be used for reference and calibration, which means that it must be very precise (in terms of adherence to ideal transfer function and gain). A digital filter (A/D->DSP->D/A) is not an option in this application.

These topologies for cascaded 2nd order filters were evaluated:

“Positive finite gain” aka “Sallen-Key”. This was dropped immediately due to its extremely poor performance (rejecting this topology is a “knee-jerk” reaction with me).

“Infinite gain” aka “Multiple Feedback/MFB”. Better: with a sensible design, component sensitivities can be kept reasonably low (which is essential for a precision filter). However, high-frequency feed-through is also inherent. And does infinite gain really exist?

“Negative finite gain”. This is where my real questions start. This is a topology that I’ve not seen in the wild, white papers or application notes are basically non-existent (I’ve searched, but perhaps the filter is known under another name). A mathematical analysis of the filter shows that it is extraordinarily attractive, as with careful design all component sensitivities can be kept well below 1. Also, the resistive feedback means that high-frequency feed-through (a problem in Sallen-Key and MFB filters) is probably negligible. Downside is that two opamps are needed for a 2nd-order stage (high-impedance buffer plus negative-gain stage), which I don’t see as a problem with modern high-performance opamps.

So, did anyone here work with negative gain filters?

Can anyone say anything about component-parasitic effects on this filter?

Can anyone give an opinion on opamp effects on this filter, eg, gain drop, phase shift, output impedance rise with frequency etc.?

Does anyone have other ideas or other filter topology suggestions?

A negative gain filter topology schematic is here:

Any feedback or insight is welcome, Thanks in advance,

Benta.

 

[VEBill]

Thirty-some years ago, I was tasked to build a 3-channel, low pass, switched-capacitor filter for 10kHz underwater acoustics. The scientist-in-charge wanted all three channels to be as exactly the same as possible.

The first thing I did was to borrow a calibration-grade DVOM with 'I-can't-even-remember' how many digits (many!). I then signed out the complete inventory (hundreds) of the required value discrete components, and spent a couple of days "binning" them all by their exact measured value. You even had to be careful how you held them while measuring.

If (for example) I needed a quantity of "10k" nominal value resistors, then what I ended up with was a quantity (for example) 10,007.3 +/- 0.1 ohm resistors (if that happened to be the value of the 'bin' with the required quantity).

I also took great care with the board layout to minimize uncontrolled stray capacitance, and keep everything identical channel-to-channel.

Even the ICs had to be hand-selected from the available inventory.

The above steps were all in addition to the usual care taken with any electronics design/build.

When it was complete, all three channels had exactly the same characteristics. The 'scope traces of the outputs couldn't be separated even with extreme settings.

Manual component binning - a very relaxing way to spend a few days.

 

[biff44]

An interesting topic! At 50 KHz, I would suggest you only use op amps with GBW product in to 20 to 100 MHz range, and keep the circuit gain low. You do not want the poles and zeros of the op amp itself screwing up your precise filter response.

That is all I've got for you. Post again if you find a solution.

http://www.MaguffinMicrowave.com

Maguffin Microwave wireless design consulting

 

[zappedagain]

Benta,

Did you ever get any farther on this? I knew something looked odd about that attachment and it finally dawned on me today. The amplifier stage is shown with a gain of 'K=-1', so there are gain components around that op-amp that aren't shown. If you get your component values wrong you'll have a nice RC oscillator instead of a filter.

Z

 

[benta]

Zappedagain, as it seems nobody has any experience with this filter topology, I'm doing some simulation and I'll build a filter to do some measurements on.
You're right, the gain block is simplified on the schematic. I'll use a non-inverting buffer to get high input impedance, followed by a negative gain stage.

Like I said, mathematically the topology is extremely attractive, but real-world experience will show if that's really the case.

Cheers,

Benta.