Dave Plowman (News) wrote:
In article ,
DAB sounds worse than FM wrote:
Well, I've argued this issue over and over, and have yet to be
convinced in the slightest that Radio 3 should have a 50% higher bit
rate than Radios 1 & 2. Given the following reason, I really do not
have a clue how you can actually argue that Radio 3 should have a
far higher bit rate than Radios 1 or 2:

Clue. Look at the dynamic range of the samples on your website. R1&2 -
about 3 dB. R3 - about 25.


Exactly! It is the narrow dynamic range that makes R1 and R2 more
difficult to encode than R3.

I bet that's confused ya!


--
Steve - www.digitalradiotech.co.uk - Digital Radio News & Info

Find the cheapest Freeview, DAB & MP3 Player Prices:
http://www.digitalradiotech.co.uk/fr..._receivers.htm
http://www.digitalradiotech.co.uk/da...tal_radios.htm
http://www.digitalradiotech.co.uk/mp...rs_1GB-5GB.htm
http://www.digitalradiotech.co.uk/mp...e_capacity.htm

In article , DAB sounds worse than
FM wrote:
Dave Plowman (News) wrote:
In article , DAB sounds worse
than FM wrote:
Well, I've argued this issue over and over, and have yet to be
convinced in the slightest that Radio 3 should have a 50% higher bit
rate than Radios 1 & 2. Given the following reason, I really do not
have a clue how you can actually argue that Radio 3 should have a far
higher bit rate than Radios 1 or 2:

Clue. Look at the dynamic range of the samples on your website. R1&2 -
about 3 dB. R3 - about 25.


Exactly! It is the narrow dynamic range that makes R1 and R2 more
difficult to encode than R3.

I bet that's confused ya!

It has certainly puzzled me. Can you explain your reasoning and define what
you mean by "more difficult"?

FWIW I have no experience of DAB. But with freeview the times I (think!) I
may have noticed problems with R3 are mostly when the sound levels are
quite low. e.g. Strings playing very quietly. i.e. at levels well below
what I hear on R2.

Slainte,

Jim

--
Electronics http://www.st-and.ac.uk/~www_pa/Scot...o/electron.htm
Audio Misc http://www.st-and.demon.co.uk/AudioMisc/index.html
Armstrong Audio http://www.st-and.demon.co.uk/Audio/armstrong.html
Barbirolli Soc. http://www.st-and.demon.co.uk/JBSoc/JBSoc.html

Jim Lesurf wrote:
In article , DAB sounds worse
than FM wrote:
Dave Plowman (News) wrote:
In article , DAB sounds
worse than FM wrote:
Well, I've argued this issue over and over, and have yet to be
convinced in the slightest that Radio 3 should have a 50% higher
bit rate than Radios 1 & 2. Given the following reason, I really
do not have a clue how you can actually argue that Radio 3 should
have a far higher bit rate than Radios 1 or 2:

Clue. Look at the dynamic range of the samples on your website.
R1&2 - about 3 dB. R3 - about 25.


Exactly! It is the narrow dynamic range that makes R1 and R2 more
difficult to encode than R3.

I bet that's confused ya!

It has certainly puzzled me. Can you explain your reasoning and
define what you mean by "more difficult"?


The noise to mask ratio (NMR - noise (error) energy to energy under
masking curve for each subband) gives a measure of coding head-room, and
you want it to be as low as possible (i.e. noise as far below the
masking threshold as possible). Because Radios 1 & 2 and all the pop
stations have audio processing applied then the spectrum tends to be
wide and flat, which tends to result in aa lot of remaining frequency
components after the psychoacoustic model has produced the masking
curves to throw away the inaudible subbands. The same is not true for
classical music, because its spectrum isn't as flat, and on average less
frequency components remain after masking. Therefore, for a given bit
rate, there are more bits per post-masking frequency component for Radio
3 than for Radios 1 & 2, thus the NMR is superior (lower) for Radio 3,
because the noise energy is the quantisation noise, which decreases as
the bits per frequency component encoded increases.


FWIW I have no experience of DAB. But with freeview the times I
(think!) I may have noticed problems with R3 are mostly when the
sound levels are quite low. e.g. Strings playing very quietly. i.e.
at levels well below what I hear on R2.


Dynamic range and sound level for MPEG-encoded audio are irrelevant,
because the MPEG encoder changes the sample values to floating point.


--
Steve - www.digitalradiotech.co.uk - Digital Radio News & Info

Find the cheapest Freeview, DAB & MP3 Player Prices:
http://www.digitalradiotech.co.uk/fr..._receivers.htm
http://www.digitalradiotech.co.uk/da...tal_radios.htm
http://www.digitalradiotech.co.uk/mp...rs_1GB-5GB.htm
http://www.digitalradiotech.co.uk/mp...e_capacity.htm

Returning again to the original topic of the thread, I can report
that the BBC have confirmed that "there was a problem with some of
the coders on Radio 3 but we believe that all is now fixed".

Patrick Wallace
------------------------------------------------------------------

In article , Pat Wallace wrote:
Returning again to the original topic of the thread, I can report
that the BBC have confirmed that "there was a problem with some of
the coders on Radio 3 but we believe that all is now fixed".

Ah. Useful. That seems to explain the problem at least partially.

I was wondering about the possibility of a temporary rate reduction
(as postulated by another poster). However, from listening to R3/DAB
at a time when I knew the bandwidth had been reduced, I concluded that
the fault seemed to have a bigger impact.

--
John Phillips

In article , DAB sounds worse than
FM wrote:
Jim Lesurf wrote:
In article , DAB sounds worse
than FM wrote:


Exactly! It is the narrow dynamic range that makes R1 and R2 more
difficult to encode than R3.

I bet that's confused ya!

It has certainly puzzled me. Can you explain your reasoning and define
what you mean by "more difficult"?


The noise to mask ratio (NMR - noise (error) energy to energy under
masking curve for each subband) gives a measure of coding head-room, and
you want it to be as low as possible (i.e. noise as far below the
masking threshold as possible).

OK.

Because Radios 1 & 2 and all the pop stations have audio processing
applied then the spectrum tends to be wide and flat, which tends to
result in aa lot of remaining frequency components after the
psychoacoustic model has produced the masking curves to throw away the
inaudible subbands.

Is that the case in the timescales relevant for the data reduction 'frames'
(or whatever the correct term is)? I can see that R1/2 tend to use audio
'compression' (in the old sense) and this may work to flatten the medium
term power spectrum. However that does not in itself mean the spectrum is
'white' if it has a finite number of components. Nor does it necessarily
mean that each individual processed time-frame will have a near uniform
power spectral density. Do you have some data on this relevant to R1/2?


The same is not true for classical music, because its spectrum isn't as
flat, and on average less frequency components remain after masking.

As you can see above, I can see your general point and it seems logical.
However I'm not certain of your use of terms like 'flat' here. A signal
might only contain a few components of the same level, or it might give a
spectrum with a uniform spectral density, but these would be quite
different cases. Also a spectrum may be uniform when averaged over one time
interval, but not uniform over another. (Indeed, for music this seems
desirable if we don't just want to listen to white noise. :-) )


Therefore, for a given bit rate, there are more bits per post-masking
frequency component for Radio 3 than for Radios 1 & 2, thus the NMR is
superior (lower) for Radio 3, because the noise energy is the
quantisation noise, which decreases as the bits per frequency component
encoded increases.


FWIW I have no experience of DAB. But with freeview the times I
(think!) I may have noticed problems with R3 are mostly when the sound
levels are quite low. e.g. Strings playing very quietly. i.e. at
levels well below what I hear on R2.


Dynamic range and sound level for MPEG-encoded audio are irrelevant,
because the MPEG encoder changes the sample values to floating point.

Is it the case that all MP2/3's encode the spectra as floating point
values? If so, what is the precision?

The point pun you make here is interesting as I have been wondering if
some of the artefacts I think I've noticed at low level may be due to
rounding or precision/quantisation errors and have been wondering if this
is due to the *receiver* using too low a level of precision.

Slainte,

Jim

--
Electronics http://www.st-and.ac.uk/~www_pa/Scot...o/electron.htm
Audio Misc http://www.st-and.demon.co.uk/AudioMisc/index.html
Armstrong Audio http://www.st-and.demon.co.uk/Audio/armstrong.html
Barbirolli Soc. http://www.st-and.demon.co.uk/JBSoc/JBSoc.html

Jim Lesurf wrote:
In article , DAB sounds worse
than FM wrote:
Jim Lesurf wrote:
In article , DAB sounds
worse than FM wrote:


Exactly! It is the narrow dynamic range that makes R1 and R2 more
difficult to encode than R3.

I bet that's confused ya!

It has certainly puzzled me. Can you explain your reasoning and
define what you mean by "more difficult"?


The noise to mask ratio (NMR - noise (error) energy to energy under
masking curve for each subband) gives a measure of coding head-room,
and you want it to be as low as possible (i.e. noise as far below the
masking threshold as possible).

OK.

Because Radios 1 & 2 and all the pop stations have audio processing
applied then the spectrum tends to be wide and flat, which tends to
result in aa lot of remaining frequency components after the
psychoacoustic model has produced the masking curves to throw away
the inaudible subbands.

Is that the case in the timescales relevant for the data reduction
'frames' (or whatever the correct term is)? I can see that R1/2 tend
to use audio 'compression' (in the old sense) and this may work to
flatten the medium term power spectrum. However that does not in
itself mean the spectrum is 'white' if it has a finite number of
components. Nor does it necessarily mean that each individual
processed time-frame will have a near uniform power spectral density.
Do you have some data on this relevant to R1/2?


No data; I've just looked at a lot of spectra. I know it's not white,
but it's a hell of a lot flatter and broader for R1/2 than R3. R3 tends
to tail-off quickly, whereas R1/2 tails-off significantly slower and for
the vast majority of the time it goes right the way up to the brickwall
filter.


The same is not true for classical music, because its spectrum isn't
as flat, and on average less frequency components remain after
masking.

As you can see above, I can see your general point and it seems
logical. However I'm not certain of your use of terms like 'flat'
here. A signal might only contain a few components of the same level,
or it might give a spectrum with a uniform spectral density, but
these would be quite different cases. Also a spectrum may be uniform
when averaged over one time interval, but not uniform over another.
(Indeed, for music this seems desirable if we don't just want to
listen to white noise. :-) )


I agree that it's not flat, but it is a hell of a lot flatter than for
R3.


Therefore, for a given bit rate, there are more bits per post-masking
frequency component for Radio 3 than for Radios 1 & 2, thus the NMR
is superior (lower) for Radio 3, because the noise energy is the
quantisation noise, which decreases as the bits per frequency
component encoded increases.


FWIW I have no experience of DAB. But with freeview the times I
(think!) I may have noticed problems with R3 are mostly when the
sound levels are quite low. e.g. Strings playing very quietly. i.e.
at levels well below what I hear on R2.


Dynamic range and sound level for MPEG-encoded audio are irrelevant,
because the MPEG encoder changes the sample values to floating point.

Is it the case that all MP2/3's encode the spectra as floating point
values? If so, what is the precision?


MPEG Layer I/II use 6 exponent bits (referred to as a scale factor)
which covers -118 dB to +6dB in 2dB steps and between 2 and 15 bits for
the mantissa, depending on subband and masking curve level.


The point pun you make here is interesting as I have been wondering
if some of the artefacts I think I've noticed at low level may be due
to rounding or precision/quantisation errors and have been wondering
if this is due to the *receiver* using too low a level of precision.


I think it's far more likely that you're hearing an MPEG artefact...

IME, the tracks that fair the worst on digital radio are loud electric
guitar tracks. Even within the same track the audio quality can vary
from being very good to absolutely abysmal. This can happen when the
loud electric guitar pauses and you've just got a vocal, and then the
electric guitar starts again and it is simply attrocious. This is, and
always will be, caused simply by insufficient bit rate. If VBR (variable
bit rate) and statistical multiplexing across the multiplex (as used on
digital TV) could be used then this suituation could be drastically
improved, but we can't use either, so when a track that is difficult to
encode is on then Radio 1 listeners in particular just have to suffer so
that the Radio 3 listeners don't. So, the next time you think you hear a
slight MPEG artefact, just consider that Radio 1 listeners have to put
up with most tracks consist of audio + MPEG artefacts throughout the
track.

If you can justify that to yourself as being fair then the only
conclusion I can come to is that you're extremely selfish.


--
Steve - www.digitalradiotech.co.uk - Digital Radio News & Info

Find the cheapest Freeview, DAB & MP3 Player Prices:
http://www.digitalradiotech.co.uk/fr..._receivers.htm
http://www.digitalradiotech.co.uk/da...tal_radios.htm
http://www.digitalradiotech.co.uk/mp...rs_1GB-5GB.htm
http://www.digitalradiotech.co.uk/mp...e_capacity.htm

In article , DAB sounds worse than
FM wrote:
Jim Lesurf wrote:
In article , DAB sounds worse
than FM wrote:


Because Radios 1 & 2 and all the pop stations have audio processing
applied then the spectrum tends to be wide and flat, which tends to
result in aa lot of remaining frequency components after the
psychoacoustic model has produced the masking curves to throw away
the inaudible subbands.

Is that the case in the timescales relevant for the data reduction
'frames' (or whatever the correct term is)? I can see that R1/2 tend
to use audio 'compression' (in the old sense) and this may work to
flatten the medium term power spectrum. However that does not in
itself mean the spectrum is 'white' if it has a finite number of
components. Nor does it necessarily mean that each individual
processed time-frame will have a near uniform power spectral density.
Do you have some data on this relevant to R1/2?


No data; I've just looked at a lot of spectra. I know it's not white,
but it's a hell of a lot flatter and broader for R1/2 than R3. R3 tends
to tail-off quickly, whereas R1/2 tails-off significantly slower and for
the vast majority of the time it goes right the way up to the brickwall
filter.

The difficulty is that doesn't necessarily lead to your conclusion. The
spectral components present in any time frame may extend across a wider
range, and be more unform in size. But if the *number* of components that
are resolved in the time frame are sigificantly less, then the 'weeding'
process may lose less info. Impossible to assess this without much more
specific info than simply observing a tendency for the components that are
present to have similar levels, etc.

Hence I think the point you make is certainly an important one, but it may
not establish the conclusion you draw without more specific evidence. Not
saying you are wrong. Just saying 'dunno', but 'not proven' simply from
what you have said.



Is it the case that all MP2/3's encode the spectra as floating point
values? If so, what is the precision?


MPEG Layer I/II use 6 exponent bits (referred to as a scale factor)
which covers -118 dB to +6dB in 2dB steps and between 2 and 15 bits for
the mantissa, depending on subband and masking curve level.

OK. The interesting part here seems to me to be that the matissa may be
down to just a few bits.

The point pun you make here is interesting as I have been wondering
if some of the artefacts I think I've noticed at low level may be due
to rounding or precision/quantisation errors and have been wondering
if this is due to the *receiver* using too low a level of precision.


I think it's far more likely that you're hearing an MPEG artefact...

That is certainly possible. Also quite possible that I am imagining it, or
it stems from something else in the chain...

IME, the tracks that fair the worst on digital radio are loud electric
guitar tracks. Even within the same track the audio quality can vary
from being very good to absolutely abysmal. This can happen when the
loud electric guitar pauses and you've just got a vocal, and then the
electric guitar starts again and it is simply attrocious. This is, and
always will be, caused simply by insufficient bit rate. If VBR (variable
bit rate) and statistical multiplexing across the multiplex (as used on
digital TV) could be used then this suituation could be drastically
improved, but we can't use either, so when a track that is difficult to
encode is on then Radio 1 listeners in particular just have to suffer so
that the Radio 3 listeners don't. So, the next time you think you hear
a slight MPEG artefact, just consider that Radio 1 listeners have to
put up with most tracks consist of audio + MPEG artefacts throughout
the track.

If you can justify that to yourself as being fair then the only
conclusion I can come to is that you're extremely selfish.

Can't really comment on what may be 'fair' here. Just have an interest in
when the system may show audible problems. Since I don't listen much to
R1/2 (and never on DAB) I can't pass any comment on them one way or the
other.

FWIW in terms of *video* I have certainly seen very 'odd' effects at times
on BBCTV4 via DTTV. e.g. I have DVD+R's of one prom where the 'live' sic
broadcast has a picture that 'stutters' throughout a large part of the
broadcast, but where the late-night repeat is fine. I assume this was
variable rate statmux stealing bitrate from BBCTV4 during the evening to
give to some other station(s), but a higher rate being available after
midnight for the repeat. Would not surprise me if similar 'audio' crudities
turned up on DAB at times. But as I say, I have no direct experience of
that so can only surmise. The effect I think I've heard on DTTV R3 seems to
occur on some occasions and not others even when the music is similar.

Slainte,

Jim

--
Electronics http://www.st-and.ac.uk/~www_pa/Scot...o/electron.htm
Audio Misc http://www.st-and.demon.co.uk/AudioMisc/index.html
Armstrong Audio http://www.st-and.demon.co.uk/Audio/armstrong.html
Barbirolli Soc. http://www.st-and.demon.co.uk/JBSoc/JBSoc.html

Jim Lesurf wrote:
In article , DAB sounds worse
than FM wrote:
Jim Lesurf wrote:
In article , DAB sounds worse
than FM wrote:


Because Radios 1 & 2 and all the pop stations have audio processing
applied then the spectrum tends to be wide and flat, which tends to
result in aa lot of remaining frequency components after the
psychoacoustic model has produced the masking curves to throw away
the inaudible subbands.

Is that the case in the timescales relevant for the data reduction
'frames' (or whatever the correct term is)? I can see that R1/2 tend
to use audio 'compression' (in the old sense) and this may work to
flatten the medium term power spectrum. However that does not in
itself mean the spectrum is 'white' if it has a finite number of
components. Nor does it necessarily mean that each individual
processed time-frame will have a near uniform power spectral
density. Do you have some data on this relevant to R1/2?


No data; I've just looked at a lot of spectra. I know it's not white,
but it's a hell of a lot flatter and broader for R1/2 than R3. R3
tends to tail-off quickly, whereas R1/2 tails-off significantly
slower and for the vast majority of the time it goes right the way
up to the brickwall filter.

The difficulty is that doesn't necessarily lead to your conclusion.


Common sense dictates that it does.


The spectral components present in any time frame may extend across a
wider range, and be more unform in size. But if the *number* of
components that are resolved in the time frame are sigificantly less,
then the 'weeding' process may lose less info. Impossible to assess
this without much more specific info than simply observing a tendency
for the components that are present to have similar levels, etc.

Hence I think the point you make is certainly an important one, but
it may not establish the conclusion you draw without more specific
evidence.


What I've heard over the last 3 years tell me that I'm right.


Not saying you are wrong. Just saying 'dunno', but 'not
proven' simply from what you have said.


I cannot prove this with absolute certainty, but it's beyond reasonable
doubt IMO.


Is it the case that all MP2/3's encode the spectra as floating point
values? If so, what is the precision?


MPEG Layer I/II use 6 exponent bits (referred to as a scale factor)
which covers -118 dB to +6dB in 2dB steps and between 2 and 15 bits
for the mantissa, depending on subband and masking curve level.

OK. The interesting part here seems to me to be that the matissa may
be down to just a few bits.


Yes, less bits are assigned to higher frequencies because we're less
sensitive, apparently. Having said that, it's the top-end that is the
biggest problem, IMO.


The point pun you make here is interesting as I have been
wondering if some of the artefacts I think I've noticed at low
level may be due to rounding or precision/quantisation errors and
have been wondering if this is due to the *receiver* using too low
a level of precision.


I think it's far more likely that you're hearing an MPEG artefact...

That is certainly possible. Also quite possible that I am imagining
it, or it stems from something else in the chain...

IME, the tracks that fair the worst on digital radio are loud
electric guitar tracks. Even within the same track the audio quality
can vary from being very good to absolutely abysmal. This can happen
when the loud electric guitar pauses and you've just got a vocal,
and then the electric guitar starts again and it is simply
attrocious. This is, and always will be, caused simply by
insufficient bit rate. If VBR (variable bit rate) and statistical
multiplexing across the multiplex (as used on digital TV) could be
used then this suituation could be drastically improved, but we
can't use either, so when a track that is difficult to encode is on
then Radio 1 listeners in particular just have to suffer so that the
Radio 3 listeners don't. So, the next time you think you hear a
slight MPEG artefact, just consider that Radio 1 listeners have to
put up with most tracks consist of audio + MPEG artefacts throughout
the track.

If you can justify that to yourself as being fair then the only
conclusion I can come to is that you're extremely selfish.

Can't really comment on what may be 'fair' here. Just have an
interest in when the system may show audible problems. Since I don't
listen much to R1/2 (and never on DAB) I can't pass any comment on
them one way or the other.


Well I can, and they sound ****e, and it is unfair that they sound ****e
while R3 uses a 50% higher bit rate. Only a fascist would disagree.


FWIW in terms of *video* I have certainly seen very 'odd' effects at
times on BBCTV4 via DTTV. e.g. I have DVD+R's of one prom where the
'live' sic broadcast has a picture that 'stutters' throughout a
large part of the broadcast, but where the late-night repeat is fine.
I assume this was variable rate statmux stealing bitrate from BBCTV4


Probably. The BBC never alter the stat-mux bit rate allocations, which
is incredibly lazy, IMO.


during the evening to give to some other station(s), but a higher
rate being available after midnight for the repeat. Would not
surprise me if similar 'audio' crudities turned up on DAB at times.


I thought you were talking about video?



--
Steve - www.digitalradiotech.co.uk - Digital Radio News & Info

Find the cheapest Freeview, DAB & MP3 Player Prices:
http://www.digitalradiotech.co.uk/fr..._receivers.htm
http://www.digitalradiotech.co.uk/da...tal_radios.htm
http://www.digitalradiotech.co.uk/mp...rs_1GB-5GB.htm
http://www.digitalradiotech.co.uk/mp...e_capacity.htm

Jim Lesurf wrote:
In article , DAB sounds worse
than FM wrote:
Jim Lesurf wrote:
In article , DAB sounds worse
than FM wrote:


Because Radios 1 & 2 and all the pop stations have audio processing
applied then the spectrum tends to be wide and flat, which tends to
result in aa lot of remaining frequency components after the
psychoacoustic model has produced the masking curves to throw away
the inaudible subbands.

Is that the case in the timescales relevant for the data reduction
'frames' (or whatever the correct term is)? I can see that R1/2 tend
to use audio 'compression' (in the old sense) and this may work to
flatten the medium term power spectrum. However that does not in
itself mean the spectrum is 'white' if it has a finite number of
components. Nor does it necessarily mean that each individual
processed time-frame will have a near uniform power spectral
density. Do you have some data on this relevant to R1/2?


No data; I've just looked at a lot of spectra. I know it's not white,
but it's a hell of a lot flatter and broader for R1/2 than R3. R3
tends to tail-off quickly, whereas R1/2 tails-off significantly
slower and for the vast majority of the time it goes right the way
up to the brickwall filter.

The difficulty is that doesn't necessarily lead to your conclusion.
The spectral components present in any time frame may extend across a
wider range, and be more unform in size. But if the *number* of
components that are resolved in the time frame are sigificantly less,
then the 'weeding' process may lose less info. Impossible to assess
this without much more specific info than simply observing a tendency
for the components that are present to have similar levels, etc.

Hence I think the point you make is certainly an important one, but
it may not establish the conclusion you draw without more specific
evidence. Not saying you are wrong. Just saying 'dunno', but 'not
proven' simply from what you have said.


An interesting experiment is to encode some CD material into VBR MP3 at
a given quality level. That way the MP3 encoder chooses what bit rate to
use to achieve the given level of quality on a frame-by-frame basis.

If you want to compare results with me then do the following:

* download Lame v3.90.3:

http://www.hydrogenaudio.org/forums/...howtopic=28123

* I use RazorLame as a front-end GUI, which you can find he

http://www.dors.de/razorlame/download.php

* overwrite lame.exe in the RazorLame folder with lame.exe that you
downloaded with Lame v3.90.3 above

* then in Lame Options in RazorLame use the VBR presets on he

http://www.hydrogenaudio.org/forums/...howtopic=28124

--alt-preset standard
--alt-preset- extreme

Do this by going to the Expert tab, entering one of the above into the
Custom options edit box and tick Only use custom options.

Theoretically, the higher the average bit rate the more difficult that
piece of music is to encode to a given level of quality.


--
Steve - www.digitalradiotech.co.uk - Digital Radio News & Info

Find the cheapest Freeview, DAB & MP3 Player Prices:
http://www.digitalradiotech.co.uk/fr..._receivers.htm
http://www.digitalradiotech.co.uk/da...tal_radios.htm
http://www.digitalradiotech.co.uk/mp...rs_1GB-5GB.htm
http://www.digitalradiotech.co.uk/mp...e_capacity.htm