Why does an operation on a NaN of type double convert it to a different bit representation with the Softing CAN-AC2 in xPC Target 3.4 (R2008a)?

I am working with the Softing CAN-AC2 in xPC Target 3.4 (R2008a).

For the decoding I am using an embedded MATLAB function for the decoding. In this function I use a typecast to split the 64-bit double into two uint32s.

Most of the time the decoding runs without any issues. However, sometimes the CAN data is formatted such that it actually represents a NaN if interpreted as a double.

Specifically a sNaN (first 12-bits are 1’s, and a one of the remaining bits is also 1). When this sNaN double is typecast to two uint32s, the sNaN is silently converted to a qNaN, which changes 1 bit of the data, thus corrupting it.

And my decoding does not work anymore.

Answer 1

[MathWorks Support Team]

This enhancement has been incorporated in xPC Target 4.2 (R2009b). For previous product releases, read below for any possible workarounds:

Any sort of operation on a double that is a NaN (assignment, cast, arithmetic, etc.) can convert the NaN to a different bit representation that is also a NaN.

To work around this issue you would need to use the CAN bit-packing/unpacking blocks which use memcopy to avoid this issue.