Cryptography Reference
In-Depth Information
11.2.3 Turbo CDMA
Several turbo CDMA type techniques have been proposed to jointly process
multi-detection and channel decoding:
•
Varanasi and Guess [11.53] have proposed (hard estimation) decoding and
immediately recoding each user before subtracting this contribution from
the received signal. The same operation is performed on the residual signal
to decode the information of the second user, and so on, until the final user.
•
Reed and Alexander [11.46] have proposed to use an adapted filter bank
followed (in parallel) by different decoders before subtracting, for each
user, the multiple access interference linked to the
K
−
1
other users.
•
Wang and Poor [11.56] have proposed a multi-user detector that involves
implementing in parallel the MMSE filters associated with each user, fol-
lowed by the corresponding channel decoders.
These two elements ex-
change their extrinsic information iteratively.
•
Tarable
et al.
[11.48] have proposed a simplification of the method pre-
sented in [11.56]. For the first iterations, an MMSE type multi-user detec-
tor is used, followed by channel decoders placed in parallel. For the final
iterations, the MMSE filter is replaced by an adapted filter bank.
Turb o SIC detector
In this section, channel decoding is introduced into a new successive interference
cancellation (SIC) structure. Figure 11.23 remains valid, only units
ICU
m,k
change. Each interference cancellation unit
ICU
m,k
,relativetothe
k
-th user
and at iteration
m
, is given in Figure 11.26. The originality lies in the way in
which this unit is designed: the residual error signal
e
m,k
is despread (by
s
k
)then
deinterleaved (
π
−
k
) before adding the weighted estimation of the
b
m−
1
,k
data
of the same user calculated at the previous iteration. The signal thus obtained,
y
m,k
, passes through the channel decoder that provides the
a posteriori
log
likelihood ratio, conditionally to the whole observation, of all the binary elements
(both for the information bits and the parity bits):
LLR
(
b
k
/y
m,k
)=log
P
[
b
k
=+1
/y
m,k
]
P
[
b
k
=
(11.71)
−
1
/y
m,k
]
This ratio is then transformed into a weighted estimation of the binary ele-
ments:
b
m,k
=
E
[
b
k
/y
m,k
]=tanh
1
2
LLR
(
b
k
/y
m,k
)
(11.72)
The soft estimation of user
k
at iteration
m
is given by
b
m,k
=
A
k
b
m,k
.The
difference
(
b
m,k
−
b
m−
1
,k
)
is interleaved by
π
k
before spreading by
s
k
.Theresult
