Sonifex NICA X, Reference Manual

Page: 63 / 64

NICA X Reference Manual
55
critical are highly resonant spectra. There is a potential drawback in the presence of noise-like signals as, under
such conditions, the prediction process is incapable of providing any coding gain. Fortunately, the significance of
this effect is eclipsed by the inherent noise masking which accompanies these signals.
C.8 - apt-X100 Sub-Band ADPCM
Sub-band ADPCM has been shown to be a very efficient solution to the problem of high quality audio compression.
The diversity of redundancy removal processes involved not only reduce its sensitivity to uncharacteristic signals,
but collectively take advantage of the best properties of each. This means that non-musical or discordant sounds
are treated as any other signal and will be faithfully reproduced.
The code is structured as a series of 16 bit words. The scheme has been referred to as 4 bit digital audio. More
accurately it is 4 bits per sample, as one code word is generated for every four 16 bit linear PCM words input. This
scheme uses a 4 band QMF (Quadrature Mirror Filter) tree structure where each band incorporates a backward
adaptive quantiser and predictive block.
By allowing all parameter adaptations within the encoder and encoder blocks to follow in a backward mode, no
specific side information is transmitted between them. The effect of this is to improve the immunity of the system
to transmission errors and to reduce the coding delay across the encoder-decoder process.
C.9 - Inherent Properties
REAL-TIME OPERATION. The processing power of today's high speed digital signal processing technology has
made it possible to code and decode audio in real-time.
LOW, CONSTANT CODING DELAY. By operating in the time domain to analyse the audio signal, the amount of
backward information required to optimise the predictors is relatively small, giving the advantage of a minimal
coding delay. The calculations are made accurately over a fixed 122 samples, making the total constant codex
delay 3.8mS at a sample rate of 32kHz - the time it takes audio to travel approximately 1.25 metres, or a little
over 4 feet. The effects of this delay are inaudible for live off-air foldback and compensation is unnecessary for
synchronisation with picture.
AUTOMATIC RESOLUTION TO ANY SAMPLE FREQUENCY UP TO 50KHZ. Sub-band coding is able to be resolved to
any input sample frequency. In fact, the upper band limit is caused by the speed of the processing technology
used. This means that apt -X100 can be implemented without modification into systems needing the wide
bandwidth performance of 48kHz sampling, or to squeeze high quality speech into 64kbit/s using a 16kHz sample
rate.
HIGH IMMUNITY TO BIT ERRORS. Linear PCM, by its nature, is a very fragile medium. As each word of
information is describing an analogue level, any corruption of the data can have disastrous results, causing a rapid
breakdown in the sonic performance. The effects of errors range from nasty spiky clicks to white noise at full
modulation. The information contained in the apt-X100 code describe a temporal analysis of the waveform and
therefore the effects of corrupting the data are far less severe, making the process several orders of magnitude
more robust. So resistant, in fact, that it has so far not been deemed necessary to provide any form of error
protection under normal operating conditions.