This approach creates non-junk dimensions that employ bit columns, each of which represents selected code values in a given code set. This results in a fixed-length, very-rarely-changing dimension (it would only change when new codes were added to a respective code set).
At ETL-time, you would simply select and store the applicable ID of the dimension row that represents the combination of code values for the given fact.
The zero row is indeed used for unknowns.
Hope this helps!