the following paper provides some insides about truncated ternary multipliers:
https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-cdt.2013.0133
, I understand the concept explained here in general but there is one detail related to the calculation of the error described there regarding a fractional 8x8 binary multiplication. The calculation of the error in ulp is done by:
8 ulp/2 + 7 ulp/4 + 6 ulp/8 + 5 ulp/16 + 4 ulp/32 + 3 ulp/64 + 2 ulp/128 + ulp/256 = 7.004 ulp
which in general is provided by the following expression:
I'm not quite following where this calculation came from, since this is a fractional value i think it is correct that the weight that each truncated bit is of the form 2^(-d), where d is the bit position, but im not sure why the values of k, (k-1), (k-2) are divided by that weight in order to calculated the ulp. probably im still missing something in this particular case.
any clues about this will be greatly appreciated.
thanks in advance.
