I am currently dealing with low level signal(voltages in milli-volt from the sensors) for the first time. I was advised to use filtering circuit moreover RC filtering which I have already implemented. But as I am new to low level signal filtering, I do not have sufficient knowledge on capacitor selection.

For the filtering purpose, should I use normal ceramic capacitor or a capacitor something like metalized polyester capacitor? What capacitors are best for low level signal filtering?

Concern is on PCB size,i.e if I use ceramic capacitors (0805,X5R,0.1uF) pcb size would decrease by significant amount but if I use metalized polyster film capacitor like this it would require more PCB space. How bad a capacitor selecton would affect for the needs of such low level signal filtering?

enter image description here

  • 1
    \$\begingroup\$ What do you want to filter? \$\endgroup\$ – User323693 Jun 27 '15 at 9:48
  • \$\begingroup\$ schematic diagram please? \$\endgroup\$ – Techydude Jun 27 '15 at 10:02
  • \$\begingroup\$ What's the range of frequencies? You have to be more specific on the kind of filtering you want to do. \$\endgroup\$ – Leo Jun 27 '15 at 19:01
  • 1
    \$\begingroup\$ The table on the following site (wima.de/EN/characteristics.htm) gives you a nice overview of the characteristic of different dielectrics. Depending on your application you have to decide which parameters are important and which are not. \$\endgroup\$ – christoph Jul 20 '15 at 7:22
  • 1
    \$\begingroup\$ @KubaOber OK, using X2Y capacitor on input is good for EMI/RFI (high frequencies) but on attached schematic there's a low pass RC filter with very low cut-off! The best solution is EMI/RFI LC filter (ferrite bead and only a few nF) on input of amplifier and then big RC filter between amplifier and ADC. \$\endgroup\$ – Jakub Rakus Jul 20 '15 at 20:00

All electrical circuits require balance between the components. One component affects another. The answer to your question about the passive filter relies on information about the active components. We'll walk through it here.

The strain gauge load cell data sheet will have information about the sensitivity and the frequency response of your specific cell.

Take this one for example: http://www.phidgets.com/documentation/Phidgets/3132_0_Datasheet.pdf

The first step is to balance the Cell and the ADC. This is a prerequisite to passive selection.

This cell sensitivity is 8mv/V. This means at maximum deflection (780g of force) you should see 8mv/V * 5V = 40 mv of output across the terminals. The first thing you should confirm is that your ADC with 128x internal gain will scale this signal appropriately. That gain should convert your signal to about 5120 mv under the worst case. If the ADC converts 5V to full scale digital, then you are about where you want to be (because you are probably not trying to measure full deflections) If the ADC had only 3.3V of input range, you would want to make a modification. If your loads are so small that you know there is a smaller working voltage than the ADC can scale on it's own, you should look at a differential op amp to get the voltage into the correct range for the ADC. Let's assume that's not needed for now.

The next thing to plan is your ADC sampling frequency. A simple rule of thumb for sampling is to sample at twice the highest frequency that you might want to measure. This is sometimes called the Nyquist theorem.

This particular cell doesn't have a listed maximum frequency, but lets assume that it is rated at 100 Hz. This means the mechanical components are small enough that they can oscillate at 100 Hz or less without going non linear. I suspect most strain gauges out there are bigger and therefore have lower response.

So, if you want to record information about 100Hz signals you need to sample at 200 Hz and you need to make sure that there is no signal content above 100 Hz. Signal above 100Hz can alias to other frequencies. In this case 150 Hz would look like 50 Hz after sampling.

So it would be good to design a filter that has a cutoff frequency (meaning half the energy, 3 dB, is removed) at 100 Hz. The Wikipedia page on RC Time constant shows the cutoff frequency formula Fc = 1 / 2*piRC

So this says that 100 = 1 / 2*piRC so R*C = 1 / 2*pi*100 = 1.59e-3

So in this case you could choose 0.1 uF and 15.9 kOhms. However, you should make sure that your series resistance is less than 10 percent of your source impedance. In this cell the source impedance is only 1000 Ohms, so you should try and keep series impedance below 100 ohms. Dropping R to 15.9 Ohms and raising C to 100 uF will keep the same RC and cutoff frequency.

So now you know the ideal characteristics of the parts and you can go shopping :)


These are all the ceramic caps at DigiKey that have a 100 uF rating. Your selection will be driven by rated voltage. The larger the cap, the better the working voltage. You need a 5V working voltage, since the signal will average about 2.5 V coming out of the leads from the gauge, and a 2x margin is not a bad idea. The rating is for 6V, and as you can see that puts you in a 1210 package. That's not a bad size.

Resistor shopping is easier.

You can search Digikey for all the thin film resistors at 16K with 1% tolerance.

These come as small as 0402. Here you are checking for power dissipation. These are rated at up to 1/16 W before burning up and opening the circuit. Your high impedance source is probably feeding a really high impedance ADC so your currents are going to be in the microamps range. P=I^2 * R shows that you have nothing to worry about here. You could go down to 0201 (which are almost impossible to work with, so I don't recommend it.)

It is true that passives have non linear properties that manifest at high frequencies, but if your strain gauge and ADC are working in the 100's to 1000's of Hz you are not going to be encountering them.

It looks to me like you can go ceramics and thin film resistors all the way on this one.

  • \$\begingroup\$ It's worth noting that the Nyquist criterion only guarantees absence of aliasing. The closer you get to fs/2, the worse the amplitude accuracy becomes, and at fs/2 the worst case amplitude error is 100%, i.e. you measure no signal. So, if you actually want to measure a 100Hz signal, then 300Hz would be the minimum sampling rate to use, and you'd need a digital reconstruction filter afterwards to recover the amplitude. Yes: the raw data samples can have an amplitude error, but the reconstructed signal can have higher amplitude, reflecting true input. \$\endgroup\$ – Kuba hasn't forgotten Monica Feb 28 '20 at 16:47
  • \$\begingroup\$ But when sampling at 3x bandwidth, steep input filters are a must, and they are not practical for low frequencies unless you're designing a high order switched capacitor filter within an IC. For the 300Hz sampling case, you'd need a filter that goes from passband to bandstop between 100Hz and 150Hz. The behavior of that filter has direct influence on amplitude accuracy of the reconstructed signal. For 16 bit sampling that would call for a rolloff of about 200dB per octave. That calls for a mems quartz filter, unless the sigma-delta ADC has a steep digital filter (many do - it's easy). \$\endgroup\$ – Kuba hasn't forgotten Monica Feb 28 '20 at 16:56

This might help you decide. These are the three main types: -

enter image description here

In extreme applications you need to be aware that capacitors (like any components) are impure; they have an effective series resistance (ESR) and an effective series inductance. They also have an internal discharge mechanism that slowly dissipates the charge stored and some are better than others.

Other things to watch out for in ceramic capacitors, C0G/NP0 is the best type for filters because they have very good temperature stability. X7R are OK if the filtering isn't needed to be kept tight but X7R and X5R and other ceramics with bigger dielectric constants (more farads per cubic inch) have a tenendcy to change their capacitance with applied voltage and so in filters this can be a bad situation because for a changing DC voltage on the signal to be filtered could result in a retune of the filter. Worst still is that for a big signal this voltage dependancy can distort the shape of the signal.


The metalized polyester capacitor option added to the question is not good for making comparisons with because it is rated at 100 volts (hence large 2220 package) and this circuit just does not need a cap of that voltage rating. Comparing this choice with that of a generic ceramic 0805 is no real choice at all - it would be foolish to not choose the 0805 in this application because the voltage it sees is dc stable and even if the capacitance drifted with temperature a bit it's not going to make that big of a deal.

I'll also point out that the resistors shown (R1 and R2) are probably not needed if the "bridge" is within a few inches of the ADC because the bridge itself will have an effective value of series resistance due the bridge resistors. Any low frequency filtering point is dictated by bridge resistance and the extra series resistors so why bother with R1 and R2?

  • \$\begingroup\$ I am currently using four load cells (50 kg half bridge strain gauge load cell) which are placed at the four corners of a square bench. Electronic circuit is at the center of the bench indicating that load cells are like 2 feet away from the center. But yes, at the point where all the wires (three wires from each load cell) meet forms the bridge and this bridge is merely 20 mm away from the RC filtering and this RC filter itself is not more than 10 mm away from the ADC. I get the point of not using the Resistors but If I do not use these resistors I do not get smooth readings. \$\endgroup\$ – Arjun Jul 20 '15 at 10:04
  • \$\begingroup\$ Where do Tantalum capacitors fall into that infographic? \$\endgroup\$ – vicatcu Jul 25 '15 at 22:02
  • \$\begingroup\$ @vicatcu - they are clearly not there but, if you'd care to add a little substance to your question, maybe I can answer why. \$\endgroup\$ – Andy aka Jul 25 '15 at 22:18
  • \$\begingroup\$ @Andyaka sorry I meant are they lumped in with any of those classifications shown, or are they fundamentally different? What applications are they used in? \$\endgroup\$ – vicatcu Jul 25 '15 at 22:26
  • \$\begingroup\$ @vicatcu tantalums tend to be used in power supply decoupling circuits because their tolerance isn't so good to make them usuable in filters much but sometimes they are. \$\endgroup\$ – Andy aka Jul 26 '15 at 20:03

A strain gauge is usually a low frequency signal in the order of perhaps a few hundred hertz at most (there are, as always, exceptions, but this is true for the majority of interfaces).

For a low frequency filter, the cut-off characteristics are probably not critical, so X7R ceramics will probably be fine assuming your local temperature is not going to vary very much. If the local temperature will vary significantly, you would probably be better to use a C0G (NPO) type that has effectively zero temperature coefficient.

When using X7R (which has a class 2 dielectric - see link above), you should ensure that the rated voltage of the device is at least twice the voltage you would expect to see across it. This avoids problems due to the effects of DC bias that changes the capacitance (down) with increased DC bias.

Note that ceramics can also exhibit microphonic effects, but provided the excitation frequency is low, they should not interfere significantly with the signal. I recently did a strain gauge interface into active filters with C0G ceramics and I had no problems at all.

If the location of your circuit is in a high vibration environment, then a polymer type device may be more suitable as the vibration can cause the ceramic to induce a signal (noise, effectively) into the very small voltage produced by the strain gauge.

In this particular case, where there is little ripple and low frequencies, you might consider a tantalum or niobium device. These have high capacitive densities and although they have their quirks, this application seems almost perfect.

Without knowing the details of your ADC, I cannot comment on whether this filter needs to provide an anti-aliasing function.

The ADS1230 is a Delta-Sigma converter with extensive filtering, so your filter is simply the removal of unwanted artefacts such as spurious noise, and does not need to provide anti-aliasing. The datasheet recommends the use of a capacitor across specific pins to achieve anti-aliasing (see page 10, figure 18 of the datasheet).

  • \$\begingroup\$ I am currently using ADS1230 20 bit ADC configured at gain of 128. \$\endgroup\$ – Arjun Jul 20 '15 at 9:49

In general for filtering your question should be why shouldn't I use ceramic caps? (As they are the cheap, reliable and small solution).

Are you using excessively high power or need excessively high capcitance?

Then consider using aluminum electrolytic capacitors. (It doesn't look like you are so you should be fine).


Are you concerned about mechanical stress (vibrations) which can cause (generally very small) voltage to be introduced across the capacitor or do you need a super linear frequency response?

Then consider film capacitors. (It looks like your frequency range is low enough that neither of these should be a problem).

Ceramic vs. Film Capacitor: Which one is preferred in audio circuits?

Otherwise stick to ceramics, they are general purpose for a reason.


This looks to me to be more of a DC application, such as measuring slowly changing strain in a beam or for a scale, rather than oscillations (because of the capacitor arrangement you're proposing).

If that is the case, and you're trying to reduce noise or EMI potential in the circuit then I would not add R1,R2 because the values could be different and could change differently with temperature and humidity, and I would also not add C1 and C2 as they could introduce noise from the ground traces, and simply go with C3. C3 should be a very low leakage capacitor because the leakage appears as a resistance across the bridge, and if that changes it may affect the measurement. A capacitor like PTFE, metallized polyester film, but this may be overkill for your application.

Murata gives leakage resistances for ceramics around 500M ohms, which may be plenty good for your application. (Consider it as a resistor and see if the extreme values, eg. 1Mohm swinging up to 1000Mohm, of the resistance contribute at all to the measurement). If an extreme swing in the leakage resistance doesn't affect your output, then ceramics are a good choice, provided that the whole thing isn't vibrating.

Metallized film - leakage resistances around 15000M ohms http://www.kemet.com/Lists/ProductCatalog/Attachments/109/F3301_R82.pdf

Bad choices for C3 would be polarized capacitors such as electrolytic or tantalum, because the polarity seen by C3 could switch.

Ferrite beads are a good way to reduce EMI coming in on wires and such, and shielding with metal/lized tapes or cages is good for PCB protection. The 5V supply is likely your biggest source of noise, so characterizing and reducing that noise is important.

If you have the luxury of measuring the value many many times then you can eliminate the effect of noise through averaging, or more sophisticated digital filtering techniques that will allow you to pick the corner frequency through software, and of course, the frequency is limited to the Nyquist frequency of the ADC, so half of the sampling rate. If you sample a 1kHz, then you can only resolve 500Hz oscillations in the signal, less-than-excessive amounts of RF interference may not be significant.

Lastly, you should consider the resolution that you're going to get out of your ADC, because that may be the biggest limiting factor to the accuracy of the measurement. You have to have a voltage reference for the ADC, and if that varies over time (relative to the supply for the bridge) then you won't be able to compare measurements taken at different times.

Your gain on the ADC is 128, so 5V/128 = 39mV input swing before you bang into the top rail. Assuming you have a 10 bit ADC, then you can resolve 5V/1024 = 4.88mV per step, which will translate into an accuracy in terms of strain or force. If your strain gauge puts out a larger swing for the range of forces that you're looking to measure you may have to reduce the gain, and consequently the precision.

You may also consider using an instrumentation op amp with an integrator as the primary input stage, and let it do the lifting and the ADC simply measure the output from the integrator.



Use ceramic caps. They are dirt cheap, don't wear out or age (if used within spec), and have no obvious bad properties.

They don't exist in large sizes though, which is their main limitation However for filtering low current sensor signals, this is often not a problem (I'm assuming your low voltage signals are also low current).

It's probably wise to choose caps rated higher than your nominal voltage with some margin.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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