RALCWI Vocoder
CMX608/CMX618/CMX638
5.
General Description
The CMX608/CMX618/CMX638 are a pin-compatible family of near toll quality voice encoders/decoders
(Vocoders). The voice encoding/decoding is performed by a new, low data rate algorithm which is based
on Robust Advanced Low Complexity Waveform Interpolation (RALCWI) technology. The data rate is
selectable between 2400bps or 2750bps (all without FEC) or 3600bps (with FEC). The frame length is
selectable from 20ms, 40ms, 60ms or 80ms (all without FEC) or from 60ms or 80ms (with FEC). The
integral Viterbi FEC decoder can optionally use "soft decision" metrics to improve its decoding ability if the
signal can be applied in a 4-bit digital representation of the received/demodulated analogue signal. The
FEC can also be used on its own, so that data can be decoded/error-corrected then re-encoded and
forwarded on. This allows use in a digital voice repeater.
RALCWI technology uses unique proprietary signal decomposition and parameter encoding methods,
ensuring high voice quality at high compression ratios. The voice quality of RALCWI-class Vocoders, as
estimated by independent listeners, is similar to that provided by standard Vocoders running at bit rates
above 4000 bps. The Mean Opinion Score (MOS) of voice quality for this Vocoder is about 3.5-3.6. This
value was determined by a paired comparison method, performing listening tests of developed and
standard voice Vocoders.
The RALCWI Vocoder operates on a “frame-by-frame” basis. The 20ms source voice frame consists of
160 samples of linear 16-bit PCM sampled at 8kHz. The Voice Encoder performs voice analysis at the
high time resolution (8 times per frame) and forms a set of estimated parameters for each voice segment.
All of the estimated parameters are quantized to produce 48- or 55-bit frames, using Vector Quantization
(VQ) of different types. All of the vector quantizers were trained on a mixed multi-language voice base,
which contains voice samples in both Eastern and Western languages.
Voice parameters are coded by the Voice Encoder with high efficiency. Sensitivity to errors is estimated
for each output bit and a bit frame is split into two parts: “more sensitive to errors” and “less sensitive to
errors”. To protect the voice parameters against channel errors without reducing voice quality, the “more
sensitive” bits are protected. Such protection is provided by the FEC Encoder/Decoder mechanism. The
Voice Decoder de-quantizes the parameters and re-synthesizes the voice.
Control of the CMX608/CMX618/CMX638 is over a C-BUS serial port from the host µC. On the
CMX618/CMX638 only, the CSEL pin allows the selection of an internal or external CODEC, by connecting
the CSEL pin to either IOVDD or VSS respectively. If the internal CODEC is selected, the SDI, STRB and
SCLK pins should be connected to Vss. Synchronisation to the host µC is by use of either the (rising-edge
active) SYNC input or by writing to the SYNC register ($02). All major data paths are buffered in each
direction with FIFO registers. The SSP port, which is used for the external ADC serial input and DAC serial
output, is buffered in this way, as is the data to and from the CODEC (CMX618/CMX638 only) and also the
vocoded data which is streamed into and out of the C-BUS interface. The management of these FIFOs is
handled automatically and does not require any user intervention, other than the supply or consumption of
data at the appropriate rate.
On the CMX618/CMX638 only, a voice CODEC is included, consisting of a microphone input to an ADC
and a DAC with an earpiece output. A variable gain stage is associated with each converter. The
differential input stage includes 20dB of switchable gain, for use as a microphone pre-amplifier. The input
gain blocks are provided to allow for inputs from different microphone or other audio devices. At the very
highest gain combinations, the input noise may start to be significant. In this case, users may wish to
consider an external low-noise preamplifier prior to the differential input stage. For most applications this
will not be necessary. The differential output stage includes 6dB of switchable gain. By using the output
stage in a differential configuration, a further 6dB of gain can be achieved. Both input and output stages
include a high-order digital channel filter, to constrain the input and output signals to an audio bandwidth of
4kHz. See Figure 20 and Figure 21 for further details. This avoids the necessity of adding external third
(or higher) order filters, thus saving external components. A small number of external components, as
shown in Figure 5, Figure 6 and Figure 7 is all that is required to implement the analogue interfaces to the
CMX618/CMX638 voice CODEC.
2014 CML Microsystems Plc
13
D/608_18_38/11
CMLMICRO [ CML MICROCIRCUITS ]
