=max(x.y)+f(x-yl) (4.36) where=In(+e is a correction function.In practice.it can be realized using a lookup table. For a set of real numbers,by noting that =e,we have n2rnle+++e+er+e） =lne+e+.+e+e） =lne4+e+.tem+e- =In+ter.ww]）】 =E可，E(6,.,E(6,-,E(6-,6,月 (4.37) This shows that In(e+++)can be computed recursively.Suppose that ln(eA+es+.te-）has been computed.Let △=ln(es+es+.+e-) Then ln(es+es+.+e）=lne+e） =max{a,dg}+f0△-d,) (4.38) 4.3.6.1 Max-Log-MAP Decoding Algorithm It is a suboptimal version of the aforementioned MAP algorithm,which is devised to and decoding computational complexity In the Max-Log-MAP algorithm,we use the following Max-Log approximation: n(位e=a} (4.39) With this approximation,we have the following derivation.For (4.24),we have 4,(s)兰lna(s) =ln∑asxs,-ln∑∑a.66,4-30   max( , ) | | x y f xy c   (4.36) where     | | | | ln 1 x y cf xy e    is a correction function. In practice, it can be realized using a lookup table. For a set of real numbers 1 2 { , ,., }q    , by noting that ln e    , we have   1 2 2 1 1 ln ln i q qq q i e ee e e e                        1 1 2 2 ln ln q q q e e ee e e                        2 1 1 2 , ln q E q q ee e e               , 2 1 1 2 ln ln q E q q e e ee e                         = EE E E     12 2 1 , , , ,( , )     q qq     (4.37) This shows that   1 2 ln i ee e       can be computed recursively. Suppose that   1 2 1 ln i ee e        has been computed. Let   1 2 1 ln i ee e          Then     1 2 ln ln i q ee e ee               max , | |   qc q  f    (4.38) 4.3.6.1 Max-Log-MAP Decoding Algorithm It is a suboptimal version of the aforementioned MAP algorithm, which is devised to provide an efficient trade-off between error performance and decoding computational complexity. In the Max-Log-MAP algorithm, we use the following Max-Log approximation:   1 1 ln max i q i i q i e              (4.39) With this approximation, we have the following derivation. For (4.24), we have ( ) ln ( ) Ak k s s   1 1 ' ' ln ( ') ( ', ) ln ( ') ( ', ) kk kk s ss   s ss s ss                 