Cryptography Reference
In-Depth Information
Figure 11.17 - Performance of the adaptive MMSE turbo equalizer for BPSK trans-
mission on the Proakis C channel, with a 4-state rate
R
non-recursive non-
systematic convolutional code and a 16384 bit pseudo-random interleaver.
=1
/
2
During the processing of the 16384 information symbols, the turbo equalizer
operates in a decision-directed manner. The equalizer filters each have 21 coef-
ficients (
F
=
G
=10
). The coecients are updated using the LMS algorithm.
The step size is set to
μ
=0
,
0005
during the training period, and then to
μ
=0
,
000005
during the tracking period. Simulation results are given in Fig-
ure 11.17, considering 10 iterations at reception. We observe a degradation of
the order of only 1 dB compared to the ideal situation where the channel is as-
sumed to be perfectly known. We note that when the channel is estimated and
used for the direct computation of the coecients of the MMSE equalizer, losses
in performance will also appear, which reduces the degradation in comparison
to the ideal situation of Figure 11.16. Note also that, to track the performance
of Figure 11.17, we have not taken into account the loss in the signal to noise
ratio caused by the use of training sequences.
In the light of these results, we note that the major difference between adap-
tive MMSE turbo equalization and that which uses direct computation of the
coecients from the estimate of the channel lies in the way the filter coe-
cients are determined, since the structure and the optimization criterion of the
equalizers are identical.
To finish, we point out that, in the same way as for the turbo MAP equalizer,
we can use EXIT charts to predict the theoretical convergence threshold of the
