5–69
1
2
3
4
5
6
ït
çi
G3 FACSIMILE COMMUNICATIONS
(2) Soft-judgment Vitterbi decoding
Recommendation V.17 stipulates only 8-state trellis coding, and the fact
that no stipulations whatsoever are made regarding decoding for this cod-
ing scheme is as already mentioned above. For this reason, the following
describes an instance where soft-judgment Vitterbi decoding is used as a
decoding method for trellis coding in a V.17 MODEM described above.
The basic repetitive calculation of the trellis algorithm is as already
described in item 5.5.1. Also, the branchmetric or pathmetric that is calcu-
lated in this algorithm shall be the one calculated by Euclidean distance.
Now, signals that are transmitted after having passed through a trellis
decoder at time n are demodulated by a QAM demodulator on the receiv-
ing side MODEM, and are shown in the signal spatial diagram in Fig. 5-
56.
The received signals are expressed as points other than the 128 points as
the receiver signals also generally contain line noise components, for
example. Due to the fact that the trellis coder in a recommendation V.17
MODEM has the repetitive structure shown in Fig. 5-55, the branchmetric
and pathmetric in the Vitterbi algorithm perform calculation on the eight
states A to H.
For this reason, the branchmetric is calculated. In soft-judgment Vitterbi
decoding that uses the Euclidean distance, this branchmetric is calculated
as the Euclidean distance between the received signals and each signal
point. At this stage, a single branch having the smallest distance is deter-
mined in the 16 signal point group for each of A to H. For example, the
signal points for A are 16 points A1 to A16. Of these 16 points, a certain
point is determined. In other words, the points for four bits that are not
coded in Vitterbi decoding are of no concern to later calculations
Next, the sum of the branchmetric of each calculated state and the path-
metric of each state in time n-1 corresponding to this branchmetric is cal-
culated, and that sum is stored together with the survivor path as the
pathmetric at time n. Decoded data is output after this repetitive calcula-
tion is executed for a certain fixed period of time. At this point, the prob-
lem of how long this period of time should be occurs. According to
Ungerboeck’s research, it was reported that the error correction capability
of Vitterbi decoding is sufficiently demonstrated by storing over a period
of time six times the number of bits that are stored to a convolutional
coder. In other words, in V.17 MODEM coding, the survivor path and
pathmetric are calculated and stored extending over 18 (3
×
6) modulation
sectors, and decoded data is output by backtracking over the past 18 modu-
lation sectors. When a pathmetric that extends like this over 18 modulation
sectors is considered, it is anticipated that its value will increase consider-