How should I normalize time series data that shows high volatility in order to be used in an ANN? Let's take the price of Bitcoin for example.

I have the following doubts:  

 - As stated in [this paper][1] it is not sufficient to normalize dataset. This makes sense since the Bitcoin price range is from 0.06USD to 19,343USD. I think if data is fed to an ANN as a sliding window then each window has to be normalized.   
 - Methods like Min-Max and Z-Score are good for pattern comparison and also standardize range. A pattern between 1000USD and 1200USD becomes exactly the same as the same pattern between 1000USD and 1400USD so valuable information is lost. Of course it's possible to create an additional feature to represent volatility.  

So what is the best way to normalize volatile time series data like the price of Bitcoin?


  [1]: http://www.cs.ucr.edu/~eamonn/SIGKDD_trillion.pdf