Datasets:
kkboy1
/
LeAudio

Dataset card Data Studio Files
xet
 Community
1
text
stringlengths
1
112
ms khz sampling
supported sampling rate khz khz khz khz khz and khz
supported bitrates bytes per frame for each audio channel the
bitrate used is specified or recommended by the bluetooth
le audio profiles
supported bits per audio
sample and the algorithm allows most intermediate
values but these are the recommende d ones
number of audio channels unlimited by the specification in practice limited by the
profile implementation resources and airtime
table lc features
table reproduces the key parameters from table of the lc specification
for those looking for a less complex explanation there is an introductory video at
wwwbitlylcvideo
section the lc codec
the bluetooth sig has commissioned extensive audio quality testing from independent test
labs to quantify the subjective performance of the lc codec these show that at all sample
rates the audio quality exceeds that of sbc at the same sample rate and provides equivalent
or better audio quality at half the bitrate the practical benefit is that the total size of lc
encoded packets is around half of the size of those for sbc for the same audio stream
implementers can use that to their advantage as it reduc es the total airtime for transmission
saving battery life or they can use it to increas e the audio quality it gives them more scope
to play with parameters particularly in power constrained devices like earbuds and hearing
aids the power savi ng also allows scope for adding extra functionality into earbuds such as
more advanced audio algorithms or physiological sensors whilst retaining a long battery life
the lc encoder
figure provides a high level view of the lc encoder
figure high level overview of the lc encoder
the first element of the encoder is the low delay modified discrete cosine trans form
module ld mdct ld mdct is a well established method of performing time to
frequency transformations in perceptual audio coding it is low delay hence the ld but it
still takes time to convert the audio input sample into spectral coefficients a nd group the
corresponding energy values into bands its where a fair proportion of the codec delay
comes from
one of the modules fed from the ld mdct is the bandwidth detector which detects
incoming audio signals which have previously been sampled at different coding rates it can
detect the commonly used speech bandwidths in voice communication ie nb narrow
band khz wb wide band khz sswb semi super wide band khz swb
super wide band khz and fb full band khz if it detects a mismatch it
signals to the temporal noise shaper tns and noise level modules to forestall an d avoid
any smearing of noise into any empty upper spectrum
the main path for the frequency components generated by the ld mdct is into the
spectral noise shaper sns where they are quantised and processed the job of the sns is
chapter lc latency and qos
to maximise the perceptual audio quality by shaping the quanti sation noise so that the
eventual decoded output is perceived by the human ear as being as cl ose as possible to the
original signal
the remaining modules in the encoder are largely responsible for controlling artefacts the
most difficult sounds for a codec to handle are ones with a sharp attack such as percussion
instruments those transients are such a difficult thing for codecs to deal with that
castanets glockenspiel and triangles are key test sounds which are used for assessing a
codecs performance part of the problem with sharp attack transients is that overall these
sounds show a fairly flat spectrum the attack detector signals their presence to the
spectral noise shaper so that it can inform the temporal noise shaping module tns of
their presence the tns then reduces and potentially eliminates the artefacts for sign als
which have severe transients
the next stage is to determine the number of bits required to encode the quantised
spectrum which is the job of the spectral quantiser it can be considered as an intelligent
form of automatic gain control it also work s out which coefficients can be quantised to
zero which the decoder can interpret as silence this process risks introducing some coding
artefacts which are addressed by the noise level module using a pseudo random noise
generator to fill any gaps ens uring that everything is set to the proper level for the decoder
it also uses the input from the bandwidth detector to ensure that the encoded signal is
restricted to the active signal region once that is done the spectral coefficients are entropy
encoded and multiplexed into the bitstream
one other component of the resulting bitstream is a resampled input performed at a fixed
rate of khz this is passed through a long term post filter lptf for low bit rates
this reduces coding noise in any f rames which contain pitched or tonal information the
long term postfilter ltpf module perceptually shapes quantization noise by control ling a
pitch based postfilter on the decoder side
an encoded lc frame does not contain any timing information su ch as time stamps or
sequence numbers it is up to the system using the lc to control the timing of packets
which we saw when we looked at the core
section the lc codec
the lc decoder
the decoder is shown in figure and essentially reverses the process
figure overview of the lc decoder
the bandwidth information is used to determine which coefficients are zero with the noise
filling model inserting information for those which are inband the temporal noise shaper
and spectral shaper process these before the inverse ld mdct module trans forms them
back to the time domain the long term post filter is then applied using the transmitted
pitch information to define the filter characteristic
before the received packets for each frame are decoded the controller generates a bad
frame indica tion flag bfi if it detects any errors in the payload along with a payload size
parameter for each channel if the bfi flag is set the decoder will skip the packet and signal
that a packet loss concealment plc algorithm should be run to replace missing data in
the output audio stream any errors detected during the decode will also trigger the plc
choosing lc parameters
from a bluetooth le audio design viewpoint the closest most developers will come to the
lc specification is the parameters whi ch they use to configure it in their applications these
are a subset of the features of table and are shown in table
lc parameter values
sampling rate khz khz khz khz khz or khz
bits per sample or