The only analogue filter you should apply in real-time (at aquisition) is an antialiasing filter, as it is not possible to remove aliasing afterwards. All other filters can (and should) only be applied afterwards. This assures you have the most original signal possible. Once you apply anotther filter at acquisition time (say you apply a low-pass at 35 Hz as you are hinting) that smooths the ECG and seriously attenuates high frequencies, an effect that cannot be reversed afterwards if you (or a colleague) wants to use your ECG recordings to do something else where frequencies above 35 Hz might be useful.

Best wishes.

It's a tricky question. 

First of all, what your system is?

For biopac, you have only choices between  1 to 500 and 10 to 500Hz for EMG. It's analogue filters work before your signals are digitized. As Potter & Bolls mentioned in their book, (2012, p.56) these 'online' filtering is already applied in your signal before it digitized and recorded to your computer system. In this case, applying another band-pass filter can be 'recursive' and reduce both the amplitude of signal and noise (Cacioppo, Tassinary, & Bernston, 2007, p.848). Colbourn devices are similar to biopac.

In my knowledge, AD Instruments provides digital filters (not analogue). So, theoretically, there should be no difference between real-time (online) filtering and post-processing with band-pass filters.

Some researchers like Blascovich and colleagues suggest software filtering in their book (Blascovich, Vanman, Mendes, & Dickerson, 2011), but the choice is depend on your device and preference.

One point you should keep in mind is some signals(i.e. EMG) have the bandwidth overlaps with electric noise (50Hz or 60Hz). So, when you apply notch filter it affect not only the noise but also to your signal.

There are several helpful materials mentioning the problem.

Handbook of Psychophysiology

(Cacioppo, Tassinary, & Bernston, 2007, p.848-849)

Social Psychophysiology for Social and Personality Psychology

(Blascovich, Vanman, Mendes, & Dickerson, 2011).

Psychophysiological Recording

(Stern, Ray, & Quigley, 2000, p.224-225) 

The only analogue filter you should apply in real-time (at aquisition) is an antialiasing filter, as it is not possible to remove aliasing afterwards. All other filters can (and should) only be applied afterwards. This assures you have the most original signal possible. Once you apply anotther filter at acquisition time (say you apply a low-pass at 35 Hz as you are hinting) that smooths the ECG and seriously attenuates high frequencies, an effect that cannot be reversed afterwards if you (or a colleague) wants to use your ECG recordings to do something else where frequencies above 35 Hz might be useful.

Best wishes.

I totally agree with the aforementioned replies. If it's possible try to keep the raw data signals and perform off-line signal enhancement. Two exceptions are the application of anti-aliasing filtering, as Fernando said, and (if necessary) low-pass filtering to avoid saturation of the preamplifier or signal clipping in the ADC due to large artifacts.

As for your filter settings, the selected edge frequencies seem reasonable. Keep in mind that especially for the ECG either an FIR filter (constant group delay) or zero-phase filtering should be applied to avoid high-frequency components (such as parts of the R wave) spreading into the ST segment or T wave.

If you only want to detect the QRS complex a bandpass from 1 to 35 Hz is applicable, but if you want to perform a more comprehensive signal analysis consider using a notch filter to cancel the power line frequency (50 or 60 Hz) since the complete spectrum of an ECG lies between 0.2 and 150 Hz. If you get any power line interference in your EMG recordings also consider additional notch filtering there.

Best,

Dennis

Dear all

Thank you very much for your comments. They are really helpful and I appreciate it.

Best, Sibylle

I too must agree with Fernando, but keep in mind the importance of your signal-to-noise ratio at acquisition. Any filtering of noise that contaminates the physiological range can and will distort the signal you want to evaluate. If you can minimize noise and amplify your signal as close to the recording site, your post-processing filter options will be easier to determine.

Another vote of agreement with Fernando.  We are taught to acquire and store data from the signal.  This gives you the most flexibility and options for post-processing. If it is acquired filtered, that is difficult to go back and undo.