I've been reading some EEG papers where the authors use a bandpass filter at, say, 1Hz and 30Hz. I've also seen some accelerometer papers where they bandpass at 0.5Hz and 10Hz

What is the difference between a bandpass (1Hz, 30Hz) and just a regular low pass filter at 30Hz?

It seems like they're trying to remove really low frequencies, but does it really make that much of a difference? Is there a particular reason why this is done?

  • 2
    $\begingroup$ the only difference between the two filters of your example, is the BPF is also blocking DC. there might be some drift or really slow "wow" (drifting up and down) that the LPF allows but the BPF blocks. otherwise they both do the same thing above 30 Hz. $\endgroup$ – robert bristow-johnson Mar 25 '17 at 4:30
  • $\begingroup$ What does it mean to block DC? $\endgroup$ – Simon Mar 25 '17 at 12:06
  • $\begingroup$ DC is "Direct Current" and has a frequency of 0 and blocking DC means that the filter has gain of 0 (or $-\infty$ dB) at that frequency of 0. $\endgroup$ – robert bristow-johnson Mar 25 '17 at 13:31

It seems like they're trying to remove really low frequencies, but does it really make that much of a difference? Is there a particular reason why this is done?

The practical reason is going to be due to the instrumentation between the transducer and the software processing the information. As a practical matter it is difficult to remove all DC offsets in the electronics. (IE, it costs money with extras parts or calibration procedures). If the signal has no meaning at DC, it is much cheaper to not spend as much removing all of the DC signal in the electronics, and instead remove it in the signal processing.

As a practical matter, actual systems are rife with parasitics. Things we do not understand or do not want to bother to control for. Since it is almost as easy to implement a band pass as a low pass, and there is no useful information at these low frequencies, it is good design practice to filter the data, so that design parasitics in those frequency ranges do not need to be otherwise dealt with.

What does it mean to block DC?

DC is going to appear in the data as an offset. This can cause various design headaches. One is scaling. A significant offset can cause data to saturate in various processing stages. It may therefore be necessary to increase the dynamic range of the processing stages to deal with large DC (offset) components. And if these signals are not useful, then removing them can make things easier.

  • $\begingroup$ That makes sense. Is it more typical to filter at 0.5Hz or 1Hz to remove DC? $\endgroup$ – Simon Mar 26 '17 at 6:19
  • $\begingroup$ The lower frequency bound will be entirely dependent on the system in question. The closer the cutoff is to the signal of interest, the sharper the cutoff needs to be. (IE, harder to implement) However the farther the cutoff is from the frequency of interest, the longer it takes to remove the undesired signal. These are the design trade offs that makes this vocation so interesting. There is no right answer. However, over time a particular field will often settle on a standard cutoff, because it has been proven to work in that specific problem domain. $\endgroup$ – Stephen Rauch Mar 26 '17 at 6:27

A bandpass filter allows signals between two specific frequencies to pass, but that discriminates against signals at certain frequencies.

A lopass filter is a filter that passes signals with a frequency lower than a certain cutoff frequency and minimizes signals with frequencies higher than the cutoff.

There's a slight distinction. What's useful depends on the situation.


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.