Viterbi detector that associates multiple data dependent noise prediction filters with each possible bit pattern
View Patent ↗A Viterbi detector includes a plurality of possible bit patterns that correspond to branches of a detector trellis and a plurality of data dependent noise prediction filters, with multiple filters of different orders being associated with a given bit pattern. A method of decoding includes applying observables to a Viterbi detector that associates a plurality of data dependent noise filters with a given possible bit pattern that corresponds to a branch of the detector trellis, calculating the composite maximum likelihood branch metric by incorporating the results of filtering the observables through the associated plurality of filters, calculating the composite maximum likelihood branch metrics in the same manner for other possible bit patterns, and so forth, and associating soft output values with detected bits in the observables based on the calculated branch metrics.
1. A Viterbi detector including:
a plurality of possible bit patterns that correspond to branches of a detector trellis; and
a plurality of data dependent noise prediction filters, with multiple filters of different orders being associated with a given bit pattern, and the multiple filters including lower order filters.
2. The detector of claim 1 wherein the multiple data dependent noise prediction filters includes a highest order filter and at least a next lower order filter.
3. The detector of claim 2 wherein the multiple data dependent noise prediction filters includes all lower order filters.
4. The detector of claim 3 wherein the respective filters are associated with prediction error variances.
5. The detector of claim 4 wherein the possible bit patterns are further associated with composite maximum likelihood branch metrics that are each calculated by incorporating the results of filtering observables through the associated multiple filters.
6. The detector of claim 5 further calculating a composite maximum likelihood branch metric for a given possible bit pattern as:
Λ
CML
(
n
k
k
-
L
,
γ
)
=
1
2
log
(
det
C
(
γ
)
)
+
1
2
(
n
k
k
-
L
(
γ
)
)
T
C
-
1
(
γ
)
n
k
k
-
L
(
γ
)
where Y is the possible bit pattern, C(Y) is a Y-dependent covariance matrix and n k k-L (Y) is a Y-dependent noise or observables vector.
7. The detector of claim 6 further including a determinant processor that produces a determinant of a covariance matrix associated with a given bit pattern by determining a Cholesky factor and squaring the factor.
8. The detector of claim 7 further including a look-up table addressable by the bit patterns for storing the determinants of the covariance matrices.
9. The detector of claim 7 further including one or more processors for determining an inverse of the covariance matrix using an inverse of the Cholesky factor.
10. The detector of claim 5 further calculating a composite maximum likelihood branch metric for a given pattern as:
Λ
CML
(
n
k
k
-
L
,
γ
)
=
1
2
∑
i
=
0
L
[
log
(
σ
i
2
(
γ
)
)
+
(
[
-
b
i
T
(
γ
)
1
]
n
k
-
L
+
i
k
-
L
(
γ
)
)
2
σ
i
2
(
γ
)
]
where −b i T (Y) for i=0, 1 . . . L are tap vectors for the respective L+1 DDNP filters corresponding to bit pattern Y and σ i 2 (Y) are the associated error variances.
11. A method of detecting bits including
applying observables of a received signal to a Viterbi detector that associates a plurality of data dependent noise prediction filters of different orders with a given possible bit pattern that corresponds to a branch of a detector trellis, the plurality of data dependent noise prediction filters being associated with prediction error variances;
calculating a composite maximum likelihood branch metric incorporating the results of filtering the observables through the associated plurality of filters;
repeating the calculating step for other possible bit patterns; and
associating soft output values with detected bits in the received signal based on the calculated branch metrics.
12. The method of claim 11 further including the step of assigning data values to the detected bits based on the associated soft output values.
13. The method of claim 11 wherein the plurality of data dependent noise prediction filters includes a highest order filter and at least one lower order filter.
14. A method of detecting bits including
applying observables of a received signal to a Viterbi detector that associates a plurality of data dependent noise prediction filters of different orders with a given possible bit pattern that corresponds to a branch of a detector trellis, the plurality of data dependent noise prediction filters includes all of the including lower order filters;
calculating a composite maximum likelihood branch metric incorporating the results of filtering the observables through the associated plurality of filters;
repeating the calculating step for other possible bit patterns; and
associating soft output values with detected bits in the received signal based on the calculated branch metrics.
15. The method of claim 14 wherein the plurality of data dependent noise prediction filters are associated with prediction error variances.
16. The method of claim 14 wherein the plurality of data dependent noise prediction filters include all lower order filters.
17. The method of claim 14 further including the step of assigning data values to the detected bits based on the associated soft output values.
18. The method of claim 14 wherein the plurality of data dependent noise prediction filters includes a highest order filter and at least one lower order filter.
19. The method of claim 14 wherein the decoder is a MAP decoder.
20. The method of claim 14 wherein the decoder is a LDPC decoder.