8
\$\begingroup\$

I'm doing a project at work that has a couple of esoteric requirements, particularly for a 0.1 Hz (0.07 Hz really, due to part availability limitations) Highpass filter (in the data acquisition system).

Right now, I'm using a 22uF film cap, and 100K resistor, and the whole affair works quite well. However, the film cap is enormous (1.240" L x 0.532" W), and the resultant PCB is really a very large (there are many channels).
I really don't want to go too much higher for the R in the filter, since it's going into an op-amp. With the existing system (OP27, need the really low 1/f knee), +-10nA bias current, you get \$10*10^{-9}A*100,000\Omega = 0.001V,\$ or 1mV of offset due to bias current.

WIMA used to manufacture some compact 22uF 16V film caps, but they have EOLed them without a replacement.

Unfortunately, the application is a bit extreme. The caps need to be able to handle extremely low temperatures, and hard vacuum, which I think means electrolytics are out.

Does anyone manufacture large, low voltage film caps (voltages in question are +-5V, nothing major)? Alternatively, does anyone know how electrolytics fare in vacuum?

\$\endgroup\$
4
  • 1
    \$\begingroup\$ What's the bandwidth, what's the required roll-off below 0.1Hz, allowable passband ripple? \$\endgroup\$
    – stevenvh
    Jul 27, 2011 at 12:46
  • 1
    \$\begingroup\$ low temp + hard vacuum = space app? \$\endgroup\$
    – stevenvh
    Jul 27, 2011 at 15:23
  • \$\begingroup\$ @stevenvh - Stratosphere/Mesosphere, so... sorta? \$\endgroup\$ Jul 27, 2011 at 21:43
  • \$\begingroup\$ Desired roll-off is 6db/octave or better, not important. Allowable passband ripple is as little as possible. \$\endgroup\$ Jul 27, 2011 at 21:59

5 Answers 5

7
\$\begingroup\$

I'd forget analogue techniques and use DSP. At 0.1 Hz virtually any MCU could be used, but I'd use a dsPIC as I have the MDS dsPIC filter design utility. It actually writes the code for me. It'll be cheaper, smaller, and operate in a vacuum without any problems.

\$\endgroup\$
3
  • \$\begingroup\$ You can run into significant accuracy issues with bi-quad coefficients when the corner frequency is very close to zero unless you bring your sampling rate way down. \$\endgroup\$
    – Mark
    Jul 27, 2011 at 21:50
  • \$\begingroup\$ Not a viable option. We're looking at a very small AC signal of interest, modulated onto a very large, (slowly) variable DC signal. There is a gain of 100-1000 after the highpass to make the signal large enough for the data-acquisition system to process. \$\endgroup\$ Jul 27, 2011 at 22:06
  • \$\begingroup\$ Basically I need the filter to let me do DSP stuff on the filter output. \$\endgroup\$ Jul 27, 2011 at 22:07
5
\$\begingroup\$

Why does it have to be a film cap? Why not a ceramic? I'm no expert on vacuum, but I think they should be able to handle that fine.

According to my calculation, you only need 16 µF with 100 kΩ to get a 100 mHz rolloff. In any case, a couple of 10 µF 20V ceramics in parallel with good dielectrics should work. Using them over a small part of their voltage range keeps the capacitance reasonably constant.

\$\endgroup\$
3
  • \$\begingroup\$ It's really a 0.07234 Hz filter. The design spec was 0.1 Hz, but available film caps reduced it to 10uF or 22uF. \$\endgroup\$ Jul 27, 2011 at 21:54
  • 3
    \$\begingroup\$ @Fake: Caps aren't that accurate anyway. There is no way you're going to nail the rolloff frequency to 4 digits! Your two values are only 1/2 octave apart, which doesn't change the answers either way. So is there any reason you can't use ceramic caps? They would be small and cheap. \$\endgroup\$ Jul 27, 2011 at 23:14
  • \$\begingroup\$ I really should have put a "~" in front of the frequency. Just sweeping the thing to confirm the filter works takes forever. I wouldn't want to have to measure the knee with any accuracy. It would take hours!. \$\endgroup\$ Jul 28, 2011 at 5:11
3
\$\begingroup\$

Instead of using an RC highpass filter, why not use an RL highpass filter - but instead of using a real inductor use an inductance gyrator. You can use active components (and a couple much smaller capacitors) to simulate a massive inductor connected to ground to give you your low frequency cutoff point, and it will save you a lot of board space and the other problems of using a large capacitor. Here are some notes on gyrators.

Edit: Here's a gyrator RL filter design for a cutoff frequency of 0.1 Hz, using 2 opamps, resistors and a 0.1uF capacitor to simulate a 1000 H inductor. The gyrator design is based on the one here by Jim Thompson.

\$\endgroup\$
4
  • \$\begingroup\$ Hmmmm. What are the functional differences between a gyrator lowpass and a RC lowpass? \$\endgroup\$ Jul 28, 2011 at 5:15
  • 1
    \$\begingroup\$ @Fake Name Highpass or lowpass, so long as the input signal is within the bandwidth and input voltage range of the opamps used in the gyrator there is essentially not much functional difference. The simulated inductor can't store energy, or create a back EMF, but for filters this is not really an issue. \$\endgroup\$
    – MattyZ
    Jul 28, 2011 at 5:35
  • \$\begingroup\$ To do a gyrator lowpass filter, one would need a floating gyrator or a "functionally dependent negative resistor." It's actually fewer components to do a highpass filter because the simulated inductors are grounded. \$\endgroup\$
    – MattyZ
    Jul 28, 2011 at 5:49
  • \$\begingroup\$ @Fake Name On second glance the gyrator circuit I posted earlier may not be the best solution for filtering the 1/f noise - it works fine for small signals but high amplitude, very low frequency pulses will cause U2 to rail and the filter won't behave properly. The component values would have to be adjusted based on the maximum input voltage excursions. \$\endgroup\$
    – MattyZ
    Jul 29, 2011 at 7:53
3
\$\begingroup\$

Many EKG and EEG machines have "a very small AC signal of interest, modulated onto a very large, (slowly) variable" unwanted near-DC signal -- the "baseline wander". Since the heartbeat can drop as low as 40 Hz, we typically want a linear-phase highpass to cut off everything below around 0.5 Hz. a b

Perhaps you could use the same techniques they use for their highpass filter:

  • Servo loop: Instead of passing the signal through the capacitor of a high-pass passive RC filter, they use an active filter that integrates the DC component and subtracts it from the signal ("servo loop"). A active lowpass filter somehow tweaks the main signal chain to produce a highpass effect. My understanding is that this approach can be scaled up to extremely high resistances -- say, 10 MOhm and 1 uF to get a roughly 0.015 Hz high-pass corner frequency -- without the noise that such high resistance values normally cause.

  • digital filtering: some people say that baseline wandering is easier to filter out in software than in hardware.a b c

The Imac Engineering people claim they have a hipass corner frequency of 0.03 Hz. (See the "Highpass Filter Simulation" page -- how do I link directly to that page?)

The INA322 datasheet in Fig. 9 "Simplified ECG Circuit for Medical Applications" uses a servo loop driving the REF input to produce a highpass effect.

Figure 37 of the INA333 datasheet has another servo loop.

Figure 69 of the AD8420 datasheet has another servo loop: 0.5 Hz high-pass.

Figure 70 of the AD8295 datasheet has another servo loop.

Figure 5 of "Getting the most out of your instrumentation amplifier design" has another servo loop.

The ECG prototype from Matthew Shieh has another servo loop.

\$\endgroup\$
0
\$\begingroup\$

EPCOS have Metallized Polypropylene Film (MKP/MFP) series capacitors listed on their website. Digikey have these capacitors up to 110µF!!

\$\endgroup\$
1
  • \$\begingroup\$ I know about these. However, they are very big - the 22uF is 37.5 mm between PC Pins, which is the problem I have with the ones I have. \$\endgroup\$ Jul 27, 2011 at 21:45

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.