Continue to Site

Welcome to EDAboard.com

Welcome to our site! EDAboard.com is an international Electronics Discussion Forum focused on EDA software, circuits, schematics, books, theory, papers, asic, pld, 8051, DSP, Network, RF, Analog Design, PCB, Service Manuals... and a whole lot more! To participate you need to register. Registration is free. Click here to register now.

Multiple Feedback Filter - Reducing feedthrough to improve stopband attn

Status
Not open for further replies.

moddinati

Newbie level 6
Joined
Jan 6, 2012
Messages
11
Helped
0
Reputation
0
Reaction score
0
Trophy points
1,281
Activity points
1,419
Hello,

I'm designing a low-pass Butterworth filter using the multiple feedback topology. After building my prototype of my filter, there proved to be insufficient stopband attenuation due, to what I have been able to determine, to feed-through in the op-amp caused by parasitic capacitances between the op-amp input & the output. (The filter doesn't follow the theoretical -20 db/dec/pole in high frequencies and flattens out before sufficient attenuation has been reached.)

I'm trying to determine how I can reduce the effects of this feed-through capacitance. The best solution I have seen so far is to include a real, RC pole after the filter to provide the needed final attenuation. However, this solution will not work in this case as the values of the RC pole do not meet the requirements to properly drive the SAR ADC that follows.

I'm trying to determine which filter characteristics have the greatest effect on this feed-through and how it can be reduced (or pushed out to a farther frequency).

My question is:

1) Is it possible through proper selection/ addition of passive components to reduce the parasitic capacitance from Vin- to Vout?
2) What op-amp characteristics (maybe open-loop gain vs frequency?) determine where this feed-through zero is placed? How do I select an op-amp that will provide the proper attenuation to the frequencies desired?


If you have any ideas, or could point me to any resources, that may help, your help is appreciated.
 

Telling at least a few quantitative parameters of your filter would possibly allow to follow your assumptions about the nature of the problem. Why not posting the actual circuit?

Unsuitable dimensioning or wrong OP selection would be my first guess about why not achieving the specification.
 

Attached is the filter design and a plot of the measured vs. theoretical frequency response.

Disregard the green line in the plot.

I'm not sure I understand what you mean by unsuitable dimensioning. Also, besides checking the gain-bandwidth/slew rate, what other op-amp parameters should I be considering? Thanks.
 

Attachments

  • filter1.pdf
    15.1 KB · Views: 105
  • FreqResp.png
    FreqResp.png
    10 KB · Views: 126

Assuming you're actually using a 15MHz op amp to build a 60KHz lowpass, like in the schematic, it should be performing better than what you're getting. "Feedthrough capacitance" is almost certainly not your issue. If it is, then scaling your components to give smaller Rs and larger Cs should mitigate that effect. I'm betting it's a layout issue, or the op amp GBW is too low.

You haven't described your measurement procedure. In general measuring large attenuation like -60dB can be quite difficult without careful methodology or good instruments. You may just be hitting the noise floor of your measuring device.
 
Hi moddinati,
I believe, the explanation for the effect observed by you is as follows: "tail" effect.
A small part of the input signal goes directly through the feedback components (without amplification) and appears across the finite opamp output resistance.
(As another example, this happens also in a common emitter stage that has a feedback resistor between collector and base).
For rising frequencies the (open-loop) gain of the opamp reduces continuosly - resulting in an output signal that more and more is governed by this "spuriouis" signal if compared with the "correct" opamp output signal. This effect is enhanced by the rising opamp output resistance (due to a loop gain that decreases continuosly).
This is a known disadvantage of the multi-feedback topology.
There are other filter structures (e. g. positive Sallen-Key) which do not suffer from these unwanted effects.
 
I agree about your analysis of possible feedforward problems with MFB structure and OP output impedance. Plugging OP462 real parameters in the filter design tool however suggests, that output resistance has a neglectable effect in this case. There must be something different.

My guess is cross-talk involved with the quad OP package. It's mentioned in the datasheet, but not specified for the frequency range of interest, if I understand right. An additional critical point related to crosstalk are possibly the bias compensating resistors, that should be bypassed just in case.

Measuring frequency characteristics of individual biquads versus complete filter should clarify the origin of the problem.
 
Modinatti,

what is the value of Vref?

Did you check the dc operational point (opamp dc output voltages) ?

---------- Post added at 14:00 ---------- Previous post was at 13:13 ----------

I think, single supply operation for three dc coupled stages (with noninverting gain of 2) is rather problematic.

---------- Post added at 14:48 ---------- Previous post was at 14:00 ----------

My recommendation: page 16 of

https://www.google.de/url?sa=t&rct=...sg=AFQjCNE8sHPxKBTbjOKKKSlj-ZbcIUJpIA&cad=rja
 
what is the value of Vref?

Did you check the dc operational point (opamp dc output voltages) ?
Good point. I was in fact expecting, that the circuit is correctly biased to mid-supply. In this case, there should be no problem with overall DC-coupling and a gain of 1.25, I think. A not so obvious point is impedance of Vref, which is working as a virtual ground. It must have very good bypassing to avoid signal coupling through this path.

Anyway, the frequency characteristic from post #3 looks very much like a small crosstalk capacitance from circuit input to the input of the last OP stage, either inside the OP package or on the PCB. About 0.1 pF are sufficient to get the effect.
 
Thank you for taking the time to help me.

Assuming you're actually using a 15MHz op amp to build a 60KHz lowpass, like in the schematic, it should be performing better than what you're getting. .....I'm betting it's a layout issue, or the op amp GBW is too low.
The prototype board was built exactly as in schematic, using OP462 15Mhz op-amp. Could in part be layout, this was a quick prototype board is all.

"Feedthrough capacitance" is almost certainly not your issue. If it is, then scaling your components to give smaller Rs and larger Cs should mitigate that effect.

The reason I believe it is feed-through capacitance is because of the Texas Instruments application note SLOA049A - “Active Low-Pass Filter Design” pg. 16. (see attached). I just found the book "Op-Amps For Everyone" recommends using capacitor values of 1nF to a couple uF for filter design (pg. 425) to minimize the effects of parasitic capacitances, so I will probably change this.

You haven't described your measurement procedure. In general measuring large attenuation like -60dB can be quite difficult without careful methodology or good instruments. You may just be hitting the noise floor of your measuring device.

I'm using the Omicron Bode 100, which should have a noise floor around -100 dB.

I believe, the explanation for the effect observed by you is as follows: "tail" effect.

Do you have any recommended resources for learning more about this phenomenon?

This is a known disadvantage of the multi-feedback topology.
There are other filter structures (e. g. positive Sallen-Key) which do not suffer from these unwanted effects.


Hmmm...I specifically chose the multiple feedback (MFB) configuration for its (supposed) better attenuation at high frequencies as compared to Sallen-Key (SK). The Texas Instruments application note SLOA049A - “Active Low-Pass Filter Design” (attached) compares the response of a filter implemented in both MFB and SK topologies (pgs. 11-13). It discusses the reason for these non-idealities on pgs. 14-16.
Plugging OP462 real parameters in the filter design tool however suggests, that output resistance has a neglectable effect in this case.

Out of curiosity, what filter design tool are you using?

My guess is cross-talk involved with the quad OP package. It's mentioned in the datasheet, but not specified for the frequency range of interest, if I understand right. An additional critical point related to crosstalk are possibly the bias compensating resistors, that should be bypassed just in case.

:???: that doesn't sound good. To be clear, when you suggest adding a bypass capacitor you are talking about the 0.1uF capacitor in Figure 4 on pg. 9 in **broken link removed** (discussed in Appendix A).

what is the value of Vref?

Vref is 2.5V. It is provided by a DAC and buffered using an OP462 voltage follower.

Did you check the dc operational point (opamp dc output voltages) ?
No, but I will.

I think, single supply operation for three dc coupled stages (with noninverting gain of 2) is rather problematic.

???? Could you please expound?

My recommendation: page 16 of

Weiterleitungshinweis


So you are suggesting I use a Sallen-Key implementation? What's your opinion on the SLOA049A - “Active Low-Pass Filter Design” app note I mentioned previously?
 

Attachments

  • sloa049.pdf
    167.3 KB · Views: 139

To summarize (and make sure I understand correctly), the following suggestions have been made:

1) Increase capacitor sizes and decrease resistors to minimize the effects of parasitic capacitances.
2) Ensure DC Operational point is correct.
3) Add bypass capacitors to R6,R10, and R13. (I assume to minimize parasitic capacitances on the input).
4) Build each stage using a single op-amp chips (OP162) instead of the quad package (OP462) and measure the frequency response in order to determine if crosstalk is the problem.
5) Consider using the Sallen-Key topology.

Did I miss anything?
 

I used Nuhertz Filter Solutions to check the effect of OP output resistance and GBW on the filter characteristic. Even with a tenfold output resistance, the effect is still rather small and can't explain the observed frequency characteristic. Of course you can check it also easily with a SPICE simulator. Other than the suspected feedforward effect, "ordinary" crosstalk mainly depends on the filter capacitance and not the topology.

So understanding the effect correctly is also understanding for defeating it.
 
The reason I believe it is feed-through capacitance is because of the Texas Instruments application note SLOA049A - “Active Low-Pass Filter Design” pg. 16. (see attached). I just found the book "Op-Amps For Everyone" recommends using capacitor values of 1nF to a couple uF for filter design (pg. 425) to minimize the effects of parasitic capacitances, so I will probably change this.
Yeah, it's possible for capacitance between the input and output to have that effect, but that capacitance isn't part of the op amp itself, but rather stray capacitance in the signal traces.
I'm using the Omicron Bode 100, which should have a noise floor around -100 dB.
What measurement bandwidth are you using? That will determine the noise floor.

Internal crosstalk within the multiple opamp package is also another possibility. You should test each stage independently to check.
 

To summarize (and make sure I understand correctly), the following suggestions have been made:

1) Increase capacitor sizes and decrease resistors to minimize the effects of parasitic capacitances.
2) Ensure DC Operational point is correct.
3) Add bypass capacitors to R6,R10, and R13. (I assume to minimize parasitic capacitances on the input).
4) Build each stage using a single op-amp chips (OP162) instead of the quad package (OP462) and measure the frequency response in order to determine if crosstalk is the problem.
5) Consider using the Sallen-Key topology.

Did I miss anything?

To 2) Yes, absolutely necessary. For this purpose (multifeedback topology) capacitive decoupling beween the 3 stages is required, see link in posting #7

To 5) In this case and if the unity gain S&K configuration is used decoupling capacitors are not necessary (due to unity positive dc gain) - however this configuration is much more sensible to the "tail" effect as mentioned in my posting #5. I suggest at first to check for a single stage via simulation the maximum attenuation you can achieve (of course based on a realistic opamp macro model).

---------- Post added at 12:35 ---------- Previous post was at 11:16 ----------

Attached is the filter design and a plot of the measured vs. theoretical frequency response.
............................
.

Hi moddinati,

from your circuit diagram I have derived that you are going to design a lowpass of higher order (n=7 or n=8).
With htis message I like to point out that for such filter orders it can be advantageous to utilize some methods for direct filter synthesis rather than the cascade approach as preferred by you (up to now).

Such a direct filter realization is based on an active realization of a passive RLC reference structure - and has one major advantage if compared with the cascade approach: It is less sensible to opamp limitations (open-loop gain and bandwidth) and less sensitive to componenet tolerances of the passive parts.
The disadvatages are: More active and passive parts and omission of fine tune capabilities (as possible for each single stage within the cascade).
However, the second disadvantage seems to be not so important because of the reduced tolerance influence.
Two different methods can be used: (a) FDNR technique and (b) leapfrog topology (leapfrog-like coupling of integrator stages)
Example: A leapfrog lowpass (n=3) consists of 4 opamps, 3 capacitors and 9 resistors.

Final question: Are you required to use single supply?

Regards
LvW
 
Thank you all for helping me. This has been really informative so far.

Two different methods can be used: (a) FDNR technique and (b) leapfrog topology (leapfrog-like coupling of integrator stages)

For this particular design, these techniques are not desirable due to the increased component count. This design is highly limited on board space and power consumption.

Are you required to use single supply?

Yes.

To 2) Yes, absolutely necessary. For this purpose (multifeedback topology) capacitive decoupling beween the 3 stages is required, see link in posting #7

This may be a basic question, but I don't fully understand why this is needed? Help me in my ignorance.


So far, from what I'm understanding, the deviation from theory is most likely caused by 1) The feedforward/"tail" effect (maybe) or 2) Crosstalk within op-amp packaging (likely). I guess the next step is to re-build the circuit and analyze each stage individually.
 

This may be a basic question, but I don't fully understand why this is needed? Help me in my ignorance.
.

The answer is simple: For single supply the operational point (dc bias point) at the output should be at 50% of the supply voltage (in your case 0.5x5=2.5 volts)
The dc feedback factor due to the two and equal resistances in the feedback path is 0.5 - equivalent to a non-inverting dc gain of "2" that applies to the Vref.
Thus, Vref=1.25 volts (1.25x2=2.5 volts as required).
This calculation requires a dc voltage of ZERO at the filter input of each stage.
Only in this case it is possible to bias both opamp inputs with a dc value (2.5 volts), which also appears at the output. This ensures proper operation of the opamp within its linear range and with equal headroom to both sides.
 
Last edited:

I assumed so far, that the circuit purpose is filtering for a DC - 60 kHz ADC channel, and that overall DC coupling is required. At least many data acquisition and measurement systems need it. As I previously mentioned, I don't see a problem to operate the MFB filter circuit as appended in post #3 this way. The circuit has an overall DC gain of 1.25 (don't know why this particular value was choosen, but I assume it's on pupose). The single supply circuit uses a virtual ground of 2.5 V (Vcc/2), so the input voltage should swing around Vcc/2 as well. Except for a small offset voltage introduced, there's no problem with overall DC coupling, if we assume, that you are able to bypass the virtual ground sufficiently.

Particularly I don't understand where you see a DC gain of 2 for the present circuit.
 
  • Like
Reactions: LvW

    LvW

    Points: 2
    Helpful Answer Positive Rating
Hi FvM, yes - you are right. In principle, dc coupling is possible with a reference voltage at the pos. opamp input Vref=2.5 volts.
I have made a bloody error. Sorry.
LvW

Explanation: I have overlooked for a moment that the filter stage works for dc as a simple differential amplifier:
The voltage Vref is multiplied by a factor of 2 (non-inv. gain) which gives 2.5x2=5 volts at the output. But at the same time a dc voltage of 2.5 volts at the filter input is multiplied by a factor of "-1" (inv. gain) thereby substracting -2.5 volts at the output. As a result we have Vout (dc)=2.5 volts (as desired).
 
Last edited:

One more point about Vref respectively virtual ground. I previously overlooked that it's buffered by an OP buffer. This means, that you'll need to add a series resistor (a few ohms) to bypass Vref without bringing up buffer stability problems. If you are worried about DC errors introduced by the resistor, use individual RC filters for critical Vref nodes.
 
@FvM - Your assumptions in post #16 are correct. (the gain of 5/4 was chosen on the last stage because the input voltage to the op-amp is 4 V and the full scale range of the ADC is 5 V).

Thank you both for your help, it has been greatly appreciated.

I am in the process of re-building my test circuit, I will report back with what I find.

Thank you!
 

Status
Not open for further replies.

Similar threads

Part and Inventory Search

Welcome to EDABoard.com

Sponsor

Back
Top