4
\$\begingroup\$

I have been designing some multiple feedback bus filters as part of an electronics project which I am setting up for my students. I have simulated the devices successfully and I have built them. However they are producing the most insane behavior... they seem to be super sensitive to moisture. Before that please see my due diligence with simulation:

Specs:

  • Using the TL3472 op amp, 4.5 MHz GBW, 13V/us slew rate, has a bit of a large input offset voltage at 1.5mV but I did also try the TL081
  • +-3.7V supplies
  • Center frequency = 13.5 kHz
  • Q = 9
  • Gain = 1

The circuit schematic: enter image description here

Simulation circuit with parasitics (I simulated 300 instances using a Monte Carlo): enter image description here

Simulation results (Perfect): enter image description here

Stability test circuit (TI's technique): enter image description here

Stability test circuit results (Great, 82 degrees PM): enter image description here

After all that I built the test circuits on a breadboard and then soldered them together. After a little while I found that the output slowly varied and that if I blew on the circuit the center frequency and gain would change significantly. After hours of debugging for ever possible human error/obvious issue I discovered moisture seems to be the culprit.

Here are some videos of me blowing on the circuit and spraying it (On a drive): https://drive.google.com/drive/folders/1ayOex94lrCbvPWqzv2LzK6VT_3tRQVVs?usp=sharing

My questions:

  1. Do you agree this is a moisture problem, or is it something else?
  2. How would I address this problem (Assuming it is moisture)
  3. Why does such a common topology have this as a problem, I am not exactly pushing any limits to the extreme.

To save some time here is all the things I have tried to get the damn filter to work (Not exhaustive):

  • Changed all test equipment and leads
  • Built 6 versions of the circuit (some 13.5kHz some 8kHz and some 5.5Khz)
  • 3 on breadboards
  • 3 soldered prototypes
  • Checked every connection repeatedly and resoldered everything
  • Replaced every component
  • Used 1 nF initial filter capacitors
  • Resulted in resistor values around 100 kΩ
  • Used 10 nF initial filter capacitors (recommended starting value)
  • Resulted in resistor values around 10 kΩ
  • Tested with two op amps
  • TL081
  • TL3472
  • Ran extensive simulations including
  • Parasitic capacitance analysis
  • Monte Carlo tolerance analysis
  • Temperature analysis
  • Stability analysis for phase margin
  • Used Texas Instruments calculators and other reference calculators
  • All agreed closely, within roughly 5% on resistor values
  • Added decoupling capacitors
  • 100 nF
  • 220 µF
  • Tried grounding every node with 1 MΩ resistors to eliminate floating nodes
  • Tried using a 200c soldering iron to check if it was temperature dependance (It did drop the gain but only after like a minute)
  • [Edit 2] I have tried adding feedback from the output back to the input (Changing from Bessel to Chebyshev, resistor divider with 200 and 15.2k resistor). This initially does seem to help, I need to do more testing

EDIT 1: So additional tests/clarifications based on responses:

  • I am currently making a single stage (1 op amp) MFB bandpass filter
  • I have used ceramic capacitors, I am using pots for the resistors to tune them exactly, although I did use resistors before to the same effect
  • I am not using a PCB, so no guard rings (Remember I am making this for undergrad students who can't make PCBs). The inverting node and output node are probably like 1mm away. I am using breadboards and these 2 types of prototype boards: enter image description here

The op amp layout (Using op amp 2): enter image description here

  • Used isopropyl alcohol on nodes (Much higher resistance than normal water), got a similar response just not quite as rapid

  • Based on Tony's response I made a filter with f0=1khz and Q=9 so: $$ \mathrm{GBW} = f_0 Q^2 = 1 \cdot 9^2 = 81\,\text{kHz} $$ This way below the GBW and tested it and concluded it is just as sensitive if not more.

  • I don't have access to a high GBW bandwidth op amp at the moment

\$\endgroup\$
9
  • 1
    \$\begingroup\$ What parts you used? Which types of capacitors for example? If impedances are high, flux residue on PCB can leak current based on moisture to high impedance nodes. How large gaps are between sensitive PCB traces? Do you use guard rings? Can you post bill of materials and PCB designs? \$\endgroup\$ Commented yesterday
  • 2
    \$\begingroup\$ I would immediately think of temperature and not moisture, though both can affect it. Have you tried some temperature-only changes? (air temperature of course changes the moisture too, so...) \$\endgroup\$ Commented 16 hours ago
  • \$\begingroup\$ Hey guys, thanks for helping brain storm. @Justme I made an edit in the post with some updated info. Pipe I did do a temperature test with a soldering iron and the response didnt change much, and when it did it matched the simulation when I increased simulation temperature. \$\endgroup\$ Commented 13 hours ago
  • \$\begingroup\$ @reecebird Yeah, there's many kinds of ceramic capacitors, stable types and unstable types, some change their capacitance based on DC bias over them and some may be piezoelectric so mechanical vibrations cause an electrical signal. You can't build a circuit with random ceramic caps without knowing what exactly you bought to trace if it is suitable or not. Same for the potentiometers, they could be moisture sensitive, and wired like that, the wiring may receive electric and magnetic disturbances from mobile phones and mains voltage wiring. \$\endgroup\$ Commented 12 hours ago
  • \$\begingroup\$ [Proof]. See link below \$\endgroup\$ Commented 12 hours ago

2 Answers 2

Reset to default
4
\$\begingroup\$

Do you agree this is a moisture problem, or is it something else?

Yes and no. You can reduce the effect by unobvious methods.

I am not exactly pushing any limits to the extreme

This is the only false assumption.

NPO or film caps are essential. Using std XR7 or similar parts will make this into a very sensitive thermometer.

Test with a dry hot air tool, hair dryer or soldering iron for 1s to IC.

I can’t explain the theory but can prove below in simulations both in Falstad and TI filter designer. So I now believe the problem is 2-fold.

  1. If Temp+ causes -ve f then you have an NTC effect or visa versa from unstable parts. Film and NP0 ceramic are essential. This will prevent a Dynamic T issue.

Tolerance of passives must be <<5% for accuracy and 4th order is too sensitive for precision if you had not pots.

  1. For the 13.5 kHz you will definitely see a static error on due to GBW

Proof Using Falstad filter sim comparing ideal Op Amp with slider GBW.

My experience dictates that the GBW of the OA is your problem and thus Rfb=R3 is very sensitive to any shift in GBW such as parasitic C up to 50 pF which has an exponential effect on Q in each stage, rather than 1st order. I estimate you need 10MHz min. GBW for the 1st stage and preferably 20MHz for both if using a Bessel filter. If using a low ripple Chebychev the GBW sensitivity will be a bit higher.

Since 1nF is adequate to reduce the sensitivity of moisture parasitic, the parts most affecting GBW are the feedback resistors which will greatly reduce GBW with <50 pF.

My formula for one stage of a 2nd order BPF is \$GBW=f_o*Q^2=13.5*9^2=1094 kHz\$ No problem? Not so fast. By cascading 2 identical stages you now have increased the Q sensitivity of the Center frequency by the square of Q=9 to 81 so almost 9 times the GBW is now required for each stage. This is a problem for error correction of negative feedback with GBW to control the Center frequency that will shift lower with parasitics on the R3 feedback resistors.

One clue that you have too sensitive of a design is the max:min R ratio of ~ >50 dB.

The way one should design a BPF with a BW of 13.5kHz/91 is to specify BW instead then choose a lower Q Bessel filter or use more stages and stagger the lower Q poles. This is the normal result. Then your R ratios will be much lower and the Q of each stage will be much lower then the parasitic sensitivity will be much lower.

The next problem is by duplicating each filter now makes the Q^2 more sensitive to parasitic.
The 1st order feedback is sensitive to phase +/- 1 decade about 45 degrees then diminished after that. But with a 4th order BPF with a theoretical Q of 81, the phase sensitivity at resonance is very sensitive to phase error due to insufficient GBW.

How would I address this problem

Your filter needs to use a Quad Op Amp with 20 MHz GBW with an 8th order filter for acceptible results.

enter image description here

\$\endgroup\$
12
  • 1
    \$\begingroup\$ Thanks for the effort in response Tony, much appreciated. Damn that is a lot of sections, I find that hard to believe that the requirements to build a simple sharpish filter are so high (Yet again I'm only 23 so I still got plenty to learn). Why are NPO or film caps essential? I just did a simulation with an ideal op amp and the TL3472 with 50p parasitics in the feedback loops, not much of a difference...? Also when cascading 2 filters the Q isnt Q^2 isn't it: $$ Q_{\text{tot}} = \frac{Q_{\text{single}}}{\sqrt{\sqrt{2}-1}} \approx 1.554\,Q_{\text{single}} $$ I measured this as well \$\endgroup\$ Commented 12 hours ago
  • \$\begingroup\$ Some of the stuff has gone a little over my head, but is there anything realisticly I can do? I can't really use anything more than 2 op amps per filter (Once again students need to build these). I tried Sallen key and it worked but when the input was off it became an oscillator... \$\endgroup\$ Commented 12 hours ago
  • \$\begingroup\$ Rather than Q because complex filters use different Q to shape a BPF max flat( gain or group delay ) USE 3dB BW specs not Q. …. NPoh means NPzero means 0 +/- 50 ppm/‘C. All other ceramic caps are several thousand ppm/‘C NTC \$\endgroup\$ Commented 9 hours ago
  • \$\begingroup\$ I did 1% BW filters as part of my remote control thesis in 1975. Are these high school students? Later in my 1st job that year I needed 0.5 % BP filters for a VLF Doppler design , so I had to choose 5 deg X crystals so I understand your issues very well. I don’t think gyrators answer sensitivity reduction \$\endgroup\$ Commented 9 hours ago
  • \$\begingroup\$ I figured you had worked on something cool like that, bit out of my pay grade haha. These are 3rd year E&E engineering students, they are making a remote control car using analogue circuitry, I am a masters student specing their project. So I understand were you are coming from but to implement a precise circuit is out of the question for these kiddos \$\endgroup\$ Commented 8 hours ago
3
\$\begingroup\$

The MFB topology is known to have a rather large active sensitivity (sensitive to non-ideal opamp gain). In contrast, the Sallen-Key topology has a prety large passive sensitivity (sensible to parts tolerances).

The best tradeoff is the 2-opamp GIC topology (known as "Antoniou block"). In this structure, both opamps cancel their nonidealities each other out up to a certain degree.

GIC: Generalizes Impedance Converter.

Second-order filter stages in this topology can be found in the literature as "GIC-Filters" or "Fliege filters", see here (for example):

https://circuitsarchive.org/circuits/filters/second_order_bandpass/filter-fliege-second-order-bandpass-non-inverting/

\$\endgroup\$
3
  • \$\begingroup\$ Thanks for the response, I might try Sallen again and see how it goes. How many op amps for a 4th order? I can't really use more than 2 per section as the students need to build it. \$\endgroup\$ Commented 11 hours ago
  • 1
    \$\begingroup\$ @ reece bird - 4th order GIC filter needs 2 opamps per section resulting in 4 opamps in total. Design formulas can be given upon request. For S&K realization ,of course, only 4 opamps in total (2 stages),, \$\endgroup\$ Commented 10 hours ago
  • \$\begingroup\$ @reecebird. If you show an example of a tight Breadboard layout with a Quad OA 8th order is easy 1% BW with no pots and calibrated parts in parallel Series to <0.1% but must be film or NPO then trim 5% parts with anything. We used to nailfile down monolithic ceramic with calibrated test gear to 0.1% \$\endgroup\$ Commented 9 hours ago

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.