Storage-Efficient Representation of Decimal Data
Usually n decimal digits are represented by
4n bits in computers.  Actually, two BCD digits 
can be compressed optimally and reversibly into 7 bits,
and three digits into 10 bits, by a very simple 
algorithm based on the fixed-length combination of two
variable field-length encodings.  In over half 
of the cases the compressed code results from the conventional
BCD code by simple removal of redundant 
0 bits.  A long decimal message can be subdivided into
three-digit blocks, and separately compressed; 
the result differs from the asymptotic minimum length
by only 0.34 percent.  The hardware requirement 
is small, and the mappings can be done manually.
CACM January, 1975
Chi Chen, T.
Ho, I. T.
