Correct determination of adc_gain and baseline in to_wfdb

[bemoody]

hen exporting waveform data to WFDB format, we want to preserve the resolution of the original source tracks as best we can.

This is important for analysis of the data (avoiding artificial noise created by re-scaling the sample values), converting the results into other formats without degradation, and also has benefits for data compression.

(wfdb-python doesn't currently support saving compressed signals, but I'm working on it. You can convert to compressed WFDB format by using wfdb.wrsamp and then using snip to compress the result.)

If the original source tracks in the .vital file are stored in integer format, it's easy - we can just write out those integer values and set the gain and baseline accordingly.

For tracks stored in floating-point formats, I don't know whether the original measurement resolution is known. So far, I've only looked at the file 0001.vital, which contains two floating-point tracks (BIS/EEG1_WAV and BIS/EEG2_WAV). Both of those have trk.gain = 1.0 although the actual resolution appears to be 0.05 µV. The code given here works for these two examples, but the algorithm is not great.

All other issues aside, wfdb-python currently has a bug (MIT-LCP/wfdb-python#418) that is screwing up conversion using wfdb.wrsamp with p_signal. So here I've changed the code to use d_signal instead.

This is just a first attempt; comments welcome!

[bemoody]

The code given here works for these two examples, but the algorithm is not great.

Actually, sorry, I made a mistake and the code doesn't work! In line 779:

    values = np.isfinite(values)

what I meant was values = values[np.isfinite(values)].