. Setting the
bandwidth ... ...
Sampling Frequency (Fs)
For irregularly-sampled data, the sampling frequency does not have straight-forward meaning.
Here take a representative value as Fsmed =1/(median(RR_interval))
Nyquist frequency (Fn)
Let's call our effective Nyquist frequency Fnmed=Fsmed / 2
Intuitively 0.5*Fs .. but this will not 'cover' all frequencies as some components will be >Fsmed
.
In this demo, we conservatively limit the upper bandwidth of interest to 0.45*Fnmed
Plotting the Lomb Scargle PSD
Plots are over the bandwidth [0 ... 0.45*Fnmed] Hz
Why use an upper frequency cut-off for HRV?
Bandwith limiting (anti-aliasing filter for resampling)
Why detrend the HRV?
Good point. Maybe to deal with non-stationarity? See the many references in the Literature.
.
Using the web Gui demo (with example data provided: all freq in Hz) ... ...
Gaussian noise:
This is from MatLab randn() and has been low-pass filtered at 1Hz and
sampled back onto a
random irregular time axis: Fsmed is 4Hz
Set bandwidth [0.001 ... 1.0]: all the signal appears in the PSD
Set bandwidth [0.001 ... 0.5]: low-pass effect
Set bandwidth [0.01 ... 0.5]: band-pass effect
Normal:
Set bandwidth [0.001 ... 0.3]: low frequencies dominate (PSD uninteresting)
Set bandwidth [0.09 ... 0.3]: typical LF HF distribution
after detrending
Premature baby:
Note typical high heart rate
Set bandwidth [0.001 ... 0.8]: low frequencies dominate (PSD uninteresting)
Set bandwidth [0.03 ... 0.8]: a typical LF HF distribution
after detrending: resp rate high
Synthetic sines (3-peak, 3-peak plus Brownian 'drift', textbook 2-peak
LF and HF)
Frequencies of interest: <0.01 ... 0.04 ... 0.15 ... 40 Hz
Can be used to illustrate detrending and bandwidth selection
|