# Power Spectral Densities - Averaging over epochs and Parameter Considerations **URL:** \ **Category:** Support & Discussions\ **Tags:** eeg, epochs\ **Created:** [October 18, 2023, 2:21pm UTC](https://mne.discourse.group/t/power-spectral-densities-averaging-over-epochs-and-parameter-considerations/7694 "2023-10-18T14:21:07Z")\ **Posts on this page:** 1\ **Page:** 1
**Author:** ![annina](https://yyz2.discourse-cdn.com/free1/user_avatar/mne.discourse.group/annina/32/2901_2.png) [@annina](https://mne.discourse.group/u/annina)\ **Post date:** [October 18, 2023, 2:21pm UTC](https://mne.discourse.group/t/power-spectral-densities-averaging-over-epochs-and-parameter-considerations/7694/1 "2023-10-18T14:21:07Z")
Hi ! I have multiple PSG recordings from all different patients and I want to compute the PSD per subject but also averaged over all the subjects. I’m able to extract the psd per subject with ```python spectrum = epochs.compute_psd(method = 'welch',fmin=2.0, fmax=40.0, n_fft = int(2*epochs.info['sfreq']), n_overlap= int(epochs.info['sfreq']/2))` psds, freqs = spectrum.get_data(return_freqs=True) ``` and then I would concatenate the psds of the subjects together and then take the mean of that concatenated np array such that I get an average psd for all my subjects. But what I still don’t understand is the following: 1. What is the unit of the Power Spectral Density? Moreover, I’ve seen that sometimes the logarithmic scale has been applied to the resulting psds - is there a standard way to handle and visualize the power spectrum? 2. When computing the PSD plot, e.g. for just one subject, I would do the following: `spectrum.plot( average=True)` which averages over the channels, but what about the epochs? Do I have to average over the epochs as well? 3. Do the chosen parameter values for `n_fft` and `n_overlap` seem reasonable to you? Cheers, Annina - MNE version: 1.5.1 - operating system: Windows 11