Soft-decision Viterbi decoder

I'm having trouble getting the output of a soft-decision decoder to be the same as a hard-decision decoder. The SNR of my signal in question is high enough that the outputs of the two should be the same. I've attached a matrix of 4-bit quantized signals-under-test (each signal is a column) and a script I'm using to compare the decoded outputs.

Even if I try to quantize at higher than four bits or used the unquantized decoder, I still don't get the same output.

Any help would be appreciated.

Thanks, Scott

[Scott]

OK, in making a specific example generic enough to be useful, I did forget a couple details.

Of the 5604-byte payload, the last 36 bytes are different packet-to-packet. The remaining 5568 bytes are the same.

The problem I'm seeing is that the first byte is decoded in error with the soft-decision decoder. With the traceback depth I'm using I wouldn't expect differences after byte 5568 to affect the very first (and only the very first) byte of some packets.

Thanks, Scott

[Scott]

I think I've found that the encoder operated such that the internal states were non-zero. How would I configure my Viterbi to start at a known, non-zero state?