The method uses the Calkin-Wilf tree.
Here is a very brief resume of the method:
If this method can be applied, than the fraction with a bit mark is
written in the compressed file.
Otherwise the bits string is transformed with XOR rule - each 2
consecutive bits are taken in order to obtain form a string with many
possibilities of 0101010101010... an more uniform string - something
like the delta transform from the RLE method.
On this string is applied again the method: write the CW fraction and
in case the remaining bits from the string, with some bits marks also.
http://bocut-new-compress-algos.blogspot.com/search/label/Calkin%20Wilf%20Compress