Reply by Aitezaz October 31, 20112011-10-31
I have the following outputs at the various iterations in Turbo
Equalization.
Iteration No. 1 -> Error at the output of BCJR = 16, Error at the
output of ECC= 105
Iteration No. 2 -> Error at the output of BCJR = 8,   Error at the
output of ECC= 51
Iteration No. 3 -> Error at the output of BCJR = 6,   Error at the
output of ECC= 88
Iteration No. 4 -> Error at the output of BCJR = 6,   Error at the
output of ECC= 48
Iteration No. 5 -> Error at the output of BCJR = 6,   Error at the
output of ECC= 94

Now, in the third iteration, the error at the output of BCJR is lower
than previous iteration but error at the output of ECC is higher? My
question is whether the iterations in the turbo equalization should
ALWAYS improve the performance or we can have such sequences of error
count in which the output as shown above can happen. I am debugging my
code and don't know whether I should track it down to a bug or it is
an expected output.

The implementation involves a thresholding of LLRs at the output of
BCJR as well as ECC. I am thresholding the LLRs to remain in between
10 and -10. The BCJR implementation uses Jacobian method of
approximating log(a+b).

Can anyone direct me to a tutorial on equations that need to be
implemented in BCJR?
Thanks