This message is from the T13 list server.
On Wed, Mar 17, 2004 at 03:09:30PM -0700, Pat LaVarre wrote: > >If there is anything I would > >add to the standard it is more data on the error detecting capability > >of the CRC (since that is a rather specialized area of technical > >knowledge). > > Yes I'm curious, for one, thank you. How strong is the UDma CRC? Does > the UDma CRC defeat any stuck data pin? What kind of noise does the > UDma CRC overcome? > > Does my distrust of the CRC have a foundation in something real i.e. > CRC mathematics? > > Or are we discussing negligibly near zero probabilities of undetected > error? Answers would depend on your error model. But simplistically: one computes a 32-bit CRC. Random data has a probability of 1 in 2^32 to be accepted. In a stream where most blocks are accepted only few errors occur. So the question is whether a low weight word can be divisible by the CRC polynomial. I think the polynomial is 040460216667 (0x104c11db7) of weight 15. It will detect all single-bit errors. It will detect all double bit errors in blocks of less than 2^32-1 bits. It will detect any stuck data pin in blocks of less than 2^32 - 33 bits. It will detect all burst errors of length at most 32. But it will not detect multiples of itself, the smallest nonzero multiple having 33 bits.
