Speech Coding

Because the DTC is digital, it is necessary to convert
the user speech from analog form to digital. In other words, the handset
(and the network) must include a digital speech-coding scheme. In
IS-54, the speech-coding technique uses Vector Sum Excited Linear Prediction
(VSELP). This is a linear predictive coding (LPC) technique that
operates on 20-ms speech samples at a time. For each 20-ms sample, the
coding scheme itself generates 159 bits. Thus, the coder provides an effective
bit rate of 7.95 Kpbs.
The RF interface, however, is an error-prone medium. Therefore, to
ensure high speech quality, it is necessary to include mechanisms that mitigate
against errors caused in RF propagation. Consequently, the 159 bits
are subject to a channel-coding scheme designed to minimize the effects of
errors. Of the 159 bits, 77 are considered class 1 bits (of greater significance
to the speech perception) and 82 are considered class 2 bits. As shown in Figure
3-17, the 77 class 1 bits are passed through a convolutional coder, which
results in 178 bits. These 178 bits are combined with the 82 class 2 bits to
give a total of 260 bits, and the 260 bits are allocated across the time slots used by the subscriber. Thus, each 20 ms of speech gives rise to a transmission
of 260 bits, resulting in a gross rate of 13 Kbps over the air interface.