High end A/D converter shootout
Gordon Nay
Currently under consideration are the following:
dB dB 44-96
AD122-96 MKII
Mytek 8x96
SEK'D 2496 DSP
Lucid AD9624
AD8824
Sonorus Modular 8
Swissonics AD 8 (PRO) 8ch 96KHz
AD96 4ch 96KHz
Prism Dream AD-2
If there are any other A/D's that you feel are in the same class and that we
should include, please Email me with Make, Model and why you feel they
qualify. Price will not be a consideration. We are especially interested in
8ch devices, but we need to compare them to the best of the 2ch devices.
24bit is a must, 96KHz is preferable.
I will be posting conclusions results when complete.
Scott Hughes
proaudio101
Also Tango 24. Please consider doing some brass intruments with a
Coles 4038, and some really loud Marshall guitar stuff with an SM-57.
In article <391A5D92...@usa.net>,
Sent via Deja.com http://www.deja.com/
Before you buy.
Fletcher
>
> We are getting ready to do the next generation of A/D converter shootouts.
>
> Currently under consideration are the following:
>
> dB dB 44-96
> AD122-96 MKII
> Mytek 8x96
> SEK'D 2496 DSP
> Lucid AD9624
> AD8824
> Sonorus Modular 8
> Swissonics AD 8 (PRO) 8ch 96KHz
> AD96 4ch 96KHz
> Prism Dream AD-2
>
> If there are any other A/D's that you feel are in the same class and that we
> should include, please Email me with Make, Model and why you feel they
> qualify. Price will not be a consideration. We are especially interested in
> 8ch devices, but we need to compare them to the best of the 2ch devices.
> 24bit is a must, 96KHz is preferable.
>
> I will be posting conclusions results when complete.
Well...depending on when you do it, the Crane Song 'Spider' [see:
http://www.cranesong.com for details] should probably be included, as
the Apogee AD-8000-SE.
You say 'price' is an issue, but you havn't mentioned what that means.
If you can be clearer, perhaps other recommendations will emerge.
--
Fletcher
Mercenary Audio
TEL: 508-543-0069
FAX: 508-543-9670
http://www.mercenary.com
Kevin Kelly
we stack up.
KK
Richard Rives
Gordon Nay <insi...@mediaone.net> wrote in message
news:9psS4.855$J7.6...@typhoon.mw.mediaone.net...
Arny Krueger
"Gordon Nay" <insi...@mediaone.net> wrote in message
news:9psS4.855$J7.6...@typhoon.mw.mediaone.net...
> We are getting ready to do the next generation of A/D converter shootouts.
>
> Currently under consideration are the following:
>
> dB dB 44-96
> AD122-96 MKII
> Mytek 8x96
> SEK'D 2496 DSP
> Lucid AD9624
> AD8824
> Sonorus Modular 8
> Swissonics AD 8 (PRO) 8ch 96KHz
> AD96 4ch 96KHz
> Prism Dream AD-2
If you want to make your test scientific, hook 'em all up to the same high
quality signal, ( adding a few level-match tones and a click at start and
finish for timing) record the output of each one into .wav files and
compare the results using the free DBT software you can download from
www.pcabx.com.
You could also get real gutsy and put the .wav files for people to audition
themselves.
Paul Gitlitz
L. David Matheny
"Kevin Kelly" <kell...@aol.com> wrote in message
news:20000511091811...@ng-fe1.aol.com...
Rick Krizman
Kevin Kelly wrote:
> How about including the MOTU 1296 so those of us who drive Chevy's will see how
> we stack up.
>
You absolutely should include one or two low priced entries, just to keep the big
boys honest.
Rick Krizman
KrizManic Music
Jedd Haas
"RME ADI-8 PRO Hi-Quality AD converter" (see http://www.sfb.net/ad-da08.htm).
Or is one of the converters already mentioned this one under another name?
--
Jedd Haas - Artist, Writer, Guitarist. http://www.gallerytungsten.com
http://www.antijazz.com
Gordon Nay
We are only interested in what sounds best!
Fletcher <Flet...@mercenary.com> wrote in message
news:391A8D...@mercenary.com...
> Gordon Nay wrote:
> >
> > We are getting ready to do the next generation of A/D converter
shootouts.
> >
> > Currently under consideration are the following:
> >
> > dB dB 44-96
> > AD122-96 MKII
> > Mytek 8x96
> > SEK'D 2496 DSP
> > Lucid AD9624
> > AD8824
> > Sonorus Modular 8
> > Swissonics AD 8 (PRO) 8ch 96KHz
> > AD96 4ch 96KHz
> > Prism Dream AD-2
> >
> > If there are any other A/D's that you feel are in the same class and
that we
> > should include, please Email me with Make, Model and why you feel they
> > qualify. Price will not be a consideration. We are especially interested
in
> > 8ch devices, but we need to compare them to the best of the 2ch devices.
> > 24bit is a must, 96KHz is preferable.
> >
> > I will be posting conclusions results when complete.
>
>
>
Gordon Nay
We have decided to include the 1296. Motu didn't do so well up against the
best in our previous test, but you never know. The feature set in these
units make them very appealing. We need to know if there new converter is a
surprise, like the Sonorus was in our last evaluation.
Kevin Kelly <kell...@aol.com> wrote in message
news:20000511091811...@ng-fe1.aol.com...
> How about including the MOTU 1296 so those of us who drive Chevy's will
see how
> we stack up.
>
> KK
Gordon Nay
couldn't get one in on time, then. I will add it to my list again and we
will see.
I will post again the final list we actually can procure.
Scott Hughes <hug...@usa.net> wrote in message
news:391A5D92...@usa.net...
> The Weiss A/D?
Gordon Nay
it this time. Since we are only interested in finding what we think is the
best, I am including only the best (Sonorus A/D24) from the last shootout
for reference. We certainly expect to find something better now.
proaudio101 <proau...@yahoo.com> wrote in message
news:8fdrs5$c74$1...@nnrp1.deja.com...
> I didn't see Apogee on the list...I guess that might just be a given.
> Also Tango 24. Please consider doing some brass intruments with a
> Coles 4038, and some really loud Marshall guitar stuff with an SM-57.
>
>
>
>
>
> In article <391A5D92...@usa.net>,
> hug...@usa.net wrote:
Gordon Nay
include it, if available in time. It does not look like it, though.
As soon as you can get your hands on one of these, let me know.
The AD8000(special edition) with 8ch DA was included on our last shootout.
Thank You
Fletcher <Flet...@mercenary.com> wrote in message
news:391A8D...@mercenary.com...
> Gordon Nay wrote:
> >
> > We are getting ready to do the next generation of A/D converter
shootouts.
> >
> > Currently under consideration are the following:
> >
> > dB dB 44-96
> > AD122-96 MKII
> > Mytek 8x96
> > SEK'D 2496 DSP
> > Lucid AD9624
> > AD8824
> > Sonorus Modular 8
> > Swissonics AD 8 (PRO) 8ch 96KHz
> > AD96 4ch 96KHz
> > Prism Dream AD-2
> >
> > If there are any other A/D's that you feel are in the same class and
that we
> > should include, please Email me with Make, Model and why you feel they
> > qualify. Price will not be a consideration. We are especially interested
in
> > 8ch devices, but we need to compare them to the best of the 2ch devices.
> > 24bit is a must, 96KHz is preferable.
> >
> > I will be posting conclusions results when complete.
>
>
>
roc...@my-deja.com
Gordon Nay
Very impressive sound for a $400 card. The only problem was that we had to
install internal pads to calibrate I/O levels. We still use that card in an
offline editing room. We haven't tried the AX-88 and are not likely to
include it in this shootout, but I thank you for the suggestion.
Paul Gitlitz <pa...@pointbob.net> wrote in message
news:boolhso3b1j984vhl...@4ax.com...
> If your going to include chevys then I vote for the ADB AX-88.
> On 11 May 2000 13:18:11 GMT, kell...@aol.com (Kevin Kelly) wrote:
>
David L. Rick
You must certainly include the latest dCS box. It uses a
unique conversion architecture and sounds very good. Also,
the Meitner is a fine-sounding converter, although it only
output DSD last I knew. Finally, since you are interested
in eight channel boxes, consider the newest offering from
Benchmark Media. These folks seem to wring unusually good
specs from the same old Crystal Semiconductor chipset by means
of careful system design.
Do you plan to run all the converters at once? If they must
be synchronized, what will be the clock master? Very different
results may obtain when testing ADCs with external clocks than
are heard with internal clocking.
David L. Rick
Seventh String Recording
dr...@hach.com
--------------------------------------------------------
In article <9psS4.855$J7.6...@typhoon.mw.mediaone.net>,
"Gordon Nay" <insi...@mediaone.net> wrote:
> We are getting ready to do the next generation of A/D converter
shootouts.
>
> Currently under consideration are the following:
>
> dB dB 44-96
> AD122-96 MKII
> Mytek 8x96
> SEK'D 2496 DSP
> Lucid AD9624
> AD8824
> Sonorus Modular 8
> Swissonics AD 8 (PRO) 8ch 96KHz
> AD96 4ch 96KHz
> Prism Dream AD-2
>
> If there are any other A/D's that you feel are in the same class and
that we
> should include, please Email me with Make, Model and why you feel they
> qualify. Price will not be a consideration. We are especially
interested in
> 8ch devices, but we need to compare them to the best of the 2ch
devices.
> 24bit is a must, 96KHz is preferable.
>
> I will be posting conclusions results when complete.
>
>
Carl Rusk
in part due to their honesty. It is certainly no frills but I believe it has
the lowest jitter when slaving to external clock of any converter on the
market (9 pico seconds)
Carl Rusk.
> From: Scott Hughes <hug...@usa.net>
> Organization: EarthLink Inc. -- http://www.EarthLink.net
> Reply-To: hug...@usa.net
> Newsgroups: rec.audio.pro
> Date: Thu, 11 May 2000 07:13:45 GMT
> Subject: Re: High end A/D converter shootout
>
> The Weiss A/D?
>
gm
resolution" in A/D converters.
Everyone here does realize, don't they, that real questions exist as to whether
"real" 24 bit performance can really be directly measured? And that there is
every possibility that this rather chimerical spec is probably not being
reached by ANY of these products?
Do we REALLY have to tolerate "marketing bits" for ANOTHER 10 years?
C'mon guys and girls, can we please demand some backup from the manufacturers
who make unsupportable claims?
Kindly,
George Massenburg
gml...@ix.netcom.com
gm
Interesting that you should put it this way.
I rant so little that I felt that I could get away with it. Ahhh, well, I
suppose not. Any thoughts about where I should go to rant?
Gordon, have you established a testing protocol yet? Are you planning to do
pair testing by A-B-X or A-B-C-HR?
Do you mind our asking who "us" is?
By the way, I don't have a favorite converter. Maybe the ones I build for my
own use, but they're not products and you probably wouldn't like them anyway.
Thanks,
George
Gordon Nay wrote:
> I haven't seen any evidence from any of these manufacturers claiming noise
> or distortion 24bits down. And, I think that everyone in this thread that I
> have seen post so far has complete understanding of this. This rant should
> go someplace else. It is not helpful to what we are trying to do. That is:
>
> Find the best sounding converter, ALL CLAIMS ASIDE.
>
> We may not even do any test to determine what units meet what claims, our
> focus will be on how these units SOUND.
>
> Please respond, George, with what you think is the best sounding converter,
> regardless of claims, and why. Your experience could be far more valuable in
> this way.
> If what you recommend seems resonable to us, we may include your choice in
> our test.
>
> Thank You
>
> gm <gml...@ix.netcom.com> wrote in message
> news:391B5BB9...@ix.netcom.com...
Harvey Gerst
>Dear Gordon,
>
>Interesting that you should put it this way.
>
>snip
>
>By the way, I don't have a favorite converter. Maybe the ones I build for my
>own use, but they're not products and you probably wouldn't like them anyway.
>
>Thanks,
>George
Now that's funny!!!
Harvey Gerst
Indian Trail Recording Studio
http://www.ITRstudio.com/
Redtenor
gm wrote:
> Dear Gordon,
>
> Interesting that you should put it this way.
>
> I rant so little that I felt that I could get away with it. Ahhh, well, I
> suppose not. Any thoughts about where I should go to rant?
>
> Gordon, have you established a testing protocol yet? Are you planning to do
> pair testing by A-B-X or A-B-C-HR?
>
> Do you mind our asking who "us" is?
>
> By the way, I don't have a favorite converter. Maybe the ones I build for my
> own use, but they're not products and you probably wouldn't like them anyway.
>
> Thanks,
> George
>
Terry Demol
Pacific Microsonics Model 2
and dCs
Gordon Nay wrote in message
<9psS4.855$J7.6...@typhoon.mw.mediaone.net>...
Gordon Nay
Thank You
EggHd
Rant?
<< It is not helpful to what we are trying to do. >>
Define "we". You have one of the most respected designers and engineers
offering advice and you have a problem?
Nice....
Michal
I share your concern, especially with market flooded with cheap 24 bit stuff.
However, who's there to blame for "marketing bits"? I think these are the users who
think "24 bits is 6x24=144dB so this must be great!" . Very few dare to even look
at the specs before making a decision. It is no different than marketing of
anything else, consumers like catchy words.
Educated user should realise that 24 bits should be followed by good electrical
performance and susequently by good sonic properties. I'm glad there is another
shootout on the way.
As for if 24bit is manufacturers' lie- it isn't. All 24 bit converters are 24bit in
a sense that they output 24 bit words. The differences are in what the lower bits
carry - be it crap or information.
The fact that none of them measures 24bit DR ie 144dB is not a lie either. DR
measurement of 120dB in the best case doesn't tell you that only 20 bits are
useful. Cut bottom 4 bits and the measurement will worsen. There is a valuable
information in lower 4 bits that may not be measurable through FFT but does impact
the sound.
In conclusion, I believe that 24 bit conversion can extremely good and much of
differences we are hearing have to do with standard design issues, same as in any
other piece of gear be this analog or digital.
24 bit marketing bits? Irrevelant. Educated buyer will now that the "sound" is
what we all look for. Knowing that at least 24 bits are toggled there will help.
Regards
Michal @ Mytek
*********************************************
http://www.mytekdigital.com >>>>>>
24 bit 48/96 k Digital Audio Converters
DAW 9624 (tm) , D-Master Recorder (tm)
*********************************************
Gordon Nay
Ah, I see, perhaps you are upset that we haven't included your converters in
our evaluation. Maybe you could label them "32 bit Super Deluxe" and we will
all snap them up. I have tested your equipment in my labs and set up A/B
comparisons using them for clients in the past. I won't tell you the results
now because in all likelihood you wouldn't believe me.
Seriously George, you accused the members who posted in this newgroup and
the manufacturers of the best converters of something that simply isn't
true. I saw no one or any advertising that claimed 24bit quality, only 24bit
resolution. I do think almost all here know the difference.When someone
doesn't understand it, we should wait until they actually post something to
indicate it, not treat them all like there just dumb. That is where and when
you should rant George, that is if you want to treat newcomers that way.
If you go back and review the post, George, I think you will find that I am
right on this and that you owe the group an apology. If you do, you have
fully restored my respect. If not, then consider that a person's reputation
can be a lot like the advertising you abhor. When you indicate that only you
understand 24bits and that the rest of us shouldn't even talk about it
(because it upsets you that we may not understand it like you), then you are
arrogant and full of false advertising.
The fact is, George, some manufacturers have moved past 16bits of quality
and need to output more bits. The next logical step was 24bits, even if all
of the bits are not valid. You may believe that we need a different way to
distinguish between resolution and the quality (THD+N, SNR, IMD, #BITS OF
MONOTINICITY, ETC.), but what we have does work for giving it numbers. The
problem is that no matter what numbers we use, we still need to do our own
listening test and that is the purpose of what I am trying to do here.
Now, get a hold of your heart George, because of the architecture of
computers, it makes sense to deal with 32bit audio within the computer (many
applications do). This may hurt, but it wouldn't surprise me to see 32bit
converters on the market someday and never see 24bit quality.
Thanks George for ending what was a very constructive thread for my
purposes. Next time would you please start a new post when it is a new
subject. Every post and Email that I received up until your post was someone
trying to be helpful. I value there input whether I can actually act on
there advice or not.
To all of you that gave me your input, I sincerely appreciate it. I will
endever to get back to you with results on your converters along with our
methodology when it is complete.
Take care George
Us, will always love you!
gm <gml...@ix.netcom.com> wrote in message
news:391B70A5...@ix.netcom.com...
> Dear Gordon,
>
> Interesting that you should put it this way.
>
> I rant so little that I felt that I could get away with it. Ahhh, well, I
> suppose not. Any thoughts about where I should go to rant?
>
> Gordon, have you established a testing protocol yet? Are you planning to
do
> pair testing by A-B-X or A-B-C-HR?
>
> Do you mind our asking who "us" is?
>
> By the way, I don't have a favorite converter. Maybe the ones I build for
my
> own use, but they're not products and you probably wouldn't like them
anyway.
>
> Thanks,
> George
>
>
> Gordon Nay wrote:
>
> > I haven't seen any evidence from any of these manufacturers claiming
noise
> > or distortion 24bits down. And, I think that everyone in this thread
that I
> > have seen post so far has complete understanding of this. This rant
should
> > go someplace else. It is not helpful to what we are trying to do. That
is:
> >
> > Find the best sounding converter, ALL CLAIMS ASIDE.
> >
> > We may not even do any test to determine what units meet what claims,
our
> > focus will be on how these units SOUND.
> >
> > Please respond, George, with what you think is the best sounding
converter,
> > regardless of claims, and why. Your experience could be far more
valuable in
> > this way.
> > If what you recommend seems resonable to us, we may include your choice
in
> > our test.
> >
> > Thank You
> >
> > gm <gml...@ix.netcom.com> wrote in message
> > news:391B5BB9...@ix.netcom.com...
> > > I have deeply mixed feelings regarding all of the claims I see of
"24-bit
> > > resolution" in A/D converters.
> > >
> > > Everyone here does realize, don't they, that real questions exist as
to
> > whether
> > > "real" 24 bit performance can really be directly measured? And that
there
> > is
> > > every possibility that this rather chimerical spec is probably not
being
> > > reached by ANY of these products?
> > >
> > > Do we REALLY have to tolerate "marketing bits" for ANOTHER 10 years?
> > >
> > > C'mon guys and girls, can we please demand some backup from the
> > manufacturers
> > > who make unsupportable claims?
> > >
> > > Kindly,
> > > George Massenburg
> > > gml...@ix.netcom.com
> > >
> > >
> > >
> > >
j.taylor
MOTU converters. More people will be buying those than any in this list.
And you should see how they stack up, especially against 96kHz stuff.
Will Hunt
Gordon Nay wrote:
> I said "price will not be a consideration"
> We are only interested in what sounds best!
>
> Fletcher <Flet...@mercenary.com> wrote in message
> news:391A8D...@mercenary.com...
> > Gordon Nay wrote:
> > >
> > > We are getting ready to do the next generation of A/D converter
> shootouts.
> > >
> > > Currently under consideration are the following:
> > >
> > > dB dB 44-96
> > > AD122-96 MKII
> > > Mytek 8x96
> > > SEK'D 2496 DSP
> > > Lucid AD9624
> > > AD8824
> > > Sonorus Modular 8
> > > Swissonics AD 8 (PRO) 8ch 96KHz
> > > AD96 4ch 96KHz
> > > Prism Dream AD-2
> > >
> > > If there are any other A/D's that you feel are in the same class and
> that we
> > > should include, please Email me with Make, Model and why you feel they
> > > qualify. Price will not be a consideration. We are especially interested
> in
> > > 8ch devices, but we need to compare them to the best of the 2ch devices.
> > > 24bit is a must, 96KHz is preferable.
> > >
> > > I will be posting conclusions results when complete.
> >
> >
> >
EggHd
right on this and that you owe the group an apology. >>
This is one of the most ridiculous posts I have ever seen.
I do not know George Massenberg (I don't particularly like his compressors) but
based on what he posted in this thread, I can't figure out why you seem so
freaked by him.
Do your test and be happy. Get over it.
Harvey Gerst
of our little community. What he's done in this industry speaks for itself,
it doesn't need any defense, certainly not from me.
Gordon, excuse my ignorance, but who the hell are you? What do you do? What
have you done? I can't find any albums you've done, any products you've
designed, or even any posts from you, before this thread. What "labs"? What
"clients" are you talking about?
>"Gordon Nay" <insi...@mediaone.net> wrote:
>Dear George,
>
>Ah, I see, perhaps you are upset that we haven't included your converters in
>our evaluation. Maybe you could label them "32 bit Super Deluxe" and we will
>all snap them up. I have tested your equipment in my labs and set up A/B
>comparisons using them for clients in the past. I won't tell you the results
>now because in all likelihood you wouldn't believe me.
>
>Seriously George, you accused the members who posted in this newgroup and
>the manufacturers of the best converters of something that simply isn't
>true. I saw no one or any advertising that claimed 24bit quality, only 24bit
>resolution. I do think almost all here know the difference. When someone
>> > gm <gml...@ix.netcom.com> wrote:
Harvey Gerst
Gordon Nay
far as they can perceive it.
We are going to include the Motu 1296, but for the most part we are only
going to include units that have specs that warrant consideration as "the
best." I'm sorry if this does not help those considering a mid priced unit,
but no one has offered up a budget for there evaluation. We are including
other mid priced units as they have surprised us in the past with there
performance and specs and we will need them as a reference to how the new
converters have progressed.
We will not try to do a complete ranking of these units, as I said, my
client will determine what he likes best and why. You'll just have to wait
and see what I can post later on. I will detail the methodology then.
I have not seen an up to date posting of such a qualitative evaluation that
was not from a biased or doubtful source. That is why we must determine for
ourselves the answer to this extremely important question. All of the
original music from my client will be forever archived through this
converter, it must be the best available, not just the best hype. We did
this in the last two years, and in all likeliness, we will do it again next
year.
Take care!
j.taylor <jejt...@mediaone.net> wrote in message
news:391B8AEF...@mediaone.net...
> You should try the Paris converters. And the Pro24 converters. And the
> MOTU converters. More people will be buying those than any in this list.
> And you should see how they stack up, especially against 96kHz stuff.
>
> Will Hunt
>
> Gordon Nay wrote:
>
> > I said "price will not be a consideration"
> > We are only interested in what sounds best!
> >
> > Fletcher <Flet...@mercenary.com> wrote in message
> > news:391A8D...@mercenary.com...
Lynn Fuston
There was a DC offset at -55 dB FS, just off the Tascam's meters, so no one
knew it was there. But on my 16-bit, 96 dB metering it was dreadfully obvious.
What does THAT do for your 24-bit, 144 dB headroom system?
Not good things I assure you.
Don't even ask me how well they used the available 55 dB of headroom!!
Lynn Fuston
3D Audio Inc
Music Mixing and Mastering
On a scenic hilltop outside of historic
Franklin, Tennessee
http://www.3daudioinc.com
email:go3d...@aol.com
Tonebarge
"computer geek goes audio" here. Ponder this then:
A Gordon for me, A gordon for me,
If you're no a Gordon, you're nay use to me,
The Seaforth are braw, the Black Watch an' a',
But the plucky wee Gordon's the pride o' them a'.
This thread is killing me. Here's "somebody" who is "testing" chevies and fords and
"price is no consideration." I'm dyin'. Air. Gimme more air. If we don't have a
winner for funniest thread this one should be nominated. Boy oh boy, he sure gets
my respect for his condescending chutzpah toward one of my all time fucking heroes.
Is this guy related to the schlub who "designs" circuits but can't hear the
difference between tubes, fets and chips? But then, if a really good one were sent
to him for "testing," wouldn't that be casting your pearls before swine? Mmmm,
mmmmmm, I'm sure looking forward to the "test" results. Bwwaaaahaaaahaaaaa.
ROTFLMAO
TB
David Morgan (MAMS)
Gordon Nay wrote in message ...
>Find the best sounding converter, ALL CLAIMS ASIDE.
Then why are you limiting your selection of the "best" to only those that
claim to be 24 bit ? I'm just stupid enough to say that I'd hold the 16 year
old 16 bit converters in my Mitsubishi up against a WHOLE LOT of todays
so called 24 bit whiz boxes.
GM is quite right about the tenacity of 24 bit claims. I whole heartedly agree
that most of them are "marketing' bits.
>We may not even do any test to determine what units meet what claims,
That seems as though it should be the first order of business. Those that
have obviously lied or exagerated their products worthiness should be thrown
out of the running at the starting gate.
>our focus will be on how these units SOUND.
Then you should include a lot of 16 and 20 bit converters. Unless you've
already run such a test on converters below the 24 bit claim, you may be
quite surprised.
>If what you recommend seems resonable to us, we may include your choice in
>our test.
You should consider yourself lucky that he spoke to you again at all, foo.....
David Morgan (MAMS)
Morgan Audio Media Service
Dallas, Texas (972) 622-1972
__________________________________________
Main Street Studio, Garland, TX (972) 487-4045
Arny Krueger
> I have deeply mixed feelings regarding all of the claims I see of "24-bit
> resolution" in A/D converters.
>
> Everyone here does realize, don't they, that real questions exist as to
whether
> "real" 24 bit performance can really be directly measured? And that there
is
> every possibility that this rather chimerical spec is probably not being
> reached by ANY of these products?
>
> Do we REALLY have to tolerate "marketing bits" for ANOTHER 10 years?
> C'mon guys and girls, can we please demand some backup from the
manufacturers
> who make unsupportable claims?
I think that I just might hate "marketing bits" and unsupportable claims as
much as anybody.
However if nobody makes me pay too much extra for them, (they don't seem to
be optional on a lot of new gear so I guess that means they cost "nothing
extra") they seem harmless enough.
Handling them in a computer context need not be especially painful or
costly, either. Remember, with SP/DIF and AES/EBU 24 bits of what can be
audio data are shipped per sample, no matter what. For most CPU's there is a
16 bit option and a 32 bit option and not a lot in-between.
If you want a really clean 16 bit end product, it is generally agreed that
you have to do all the processing leading up to the final product with more
than 16 bits.
For a lot of the audio processing I've been doing lately, I've found that
the tools at hand provide more reliable, predictable results if I simply do
the work in 32 bit floating point (24 bit mantissa). Sometimes test results
have shown that I need to add a healthy dose (up to 2,205,000 samples per
second) of upsampling. That's what the numbers say! I just praise God for
tools that make it all possible, and get my job done. ;-)
Arny Krueger
> Dear Gordon,
>
> Interesting that you should put it this way.
>
> I rant so little that I felt that I could get away with it. Ahhh, well, I
> suppose not. Any thoughts about where I should go to rant?
>
> Gordon, have you established a testing protocol yet? Are you planning to
do
> pair testing by A-B-X or A-B-C-HR?
Just another friendly reminder that a free ABX Comparator that has been
tested up to 24/96 (and has no format limits built into it) is readily
available at www.pcabx.com. There is also a decent-sized stash of original
and progressively-degraded musical test signals that can at least be used
for training listeners with the test format.
Arny Krueger
"Gordon Nay" <insi...@mediaone.net> wrote in message
news:mkHS4.1015$J7.7...@typhoon.mw.mediaone.net...
> We tried an ADB mult!wave 24bit card (no longer available) 2 shootouts
ago.
> Very impressive sound for a $400 card. The only problem was that we had to
> install internal pads to calibrate I/O levels. We still use that card in
an
> offline editing room. We haven't tried the AX-88 and are not likely to
> include it in this shootout, but I thank you for the suggestion.
Other obvious candidates are:
Aardvark Direct Pro 24/96
http://www.pcavtech.com/soundcards/DP2496/index.htm
Event Layla http://www.pcavtech.com/soundcards/layla/index.htm
DAL CardD Deluxe http://www.pcavtech.com/soundcards/CardDDeluxe/index.htm
Lucid Technology PCI32 + ADA1000
http://www.pcavtech.com/soundcards/ADA1000/index.htm
and Lynx Studio Technology LynxONE
http://www.pcavtech.com/soundcards/LynxONE/index.htm .
Those are the best 5 sound cards I've tested so far at www.pcavtech.com.
All work at standard levels of -10 and +4 and have balanced I/O.
Fletcher
>
> I haven't seen any evidence from any of these manufacturers claiming noise
> or distortion 24bits down. And, I think that everyone in this thread that I
> have seen post so far has complete understanding of this. This rant should
> go someplace else. It is not helpful to what we are trying to do. That is:
>
While you're at it, would you discover the 'best shade of blue' and the
'prettiest girl'? What the hell, seeing as you want to make subjective
opinions "fact"...when you find the nicest shade of blue, be a sport and
keep us in the loop.
I guess I'm starting to get the picture...you have a client, and are
doing some kind of sales/consulting kinda thing. Is that it?
While "that rant" may not be immediately useful to *your* current goals
and aspirations, it is certainly a *valid* point of discussion for a
usegroup collection of folks that are interested in such things.
Threads meander from time to time, and while you seem to have a specific
need, that will not deter the other participants in this forum from
discussing amongst themselves...the topic: "Is 24 bit really 24 bit?"
No offense intended...but that debate/Q &A is *way* more interesting to
me [and I would suspect a few other people] than your quest for the
"ultimate converter". Hence, while you may not feel George's response
was germain to your reality, it's a damn interesting 'diving board' for
other discussion.
Best of luck with your search.
Fletcher
>
> I have deeply mixed feelings regarding all of the claims I see of "24-bit
> resolution" in A/D converters.
As a 'digital dummy'...no, I have no idea of what you're talking
about...would you be so kind as to elaborate for people like me who are
sucked into the "marketing bits".
I'm just starting to try to understand this stuff...so I have a bit [no
pun intended] of a 'learning curve' ahead of me. Any help will be
greatly appreciated.
Due to my lack of education, I'm most suseptable to 'marketing
horseshit' at this point. As I understand it, 'bit depth' effects "word
length", which effects not only 'dynamic range', but 'resolution' as
well?.
How do the 'increased sampling rates' effect resolution? And why/how is
it different?
Gordon Nay
me [and I would suspect a few other people] than your quest for the
"ultimate converter". Hence, while you may not feel George's response
was germain to your reality, it's a damn interesting 'diving board' for
other discussion.>>
It sure is, it is worthy of a new post and should be repeated often. I have
written to George and asked him to do just that.
Fletcher <Flet...@mercenary.com> wrote in message
news:391BD6...@mercenary.com...
Richard Rives
Dave Martin
This is a completely different goal than finding the 'best' converter.. I've
read discussions between respected mastering engineers concerning
converters, and those guys will choose converters based on the needs of a
particular project, because they hear a difference. Not a better/worse
difference, but more of a chocolate/strawberry/vanilla difference. One isn't
better than another, except how it relates to the program material.
There is no 'best' microphone, there's no 'best' compressor, and I suspect
that there's not a 'best converter'. And if you've managed to convince a
client to pay you money to fine the best, good for you. But in reality,
you're looking for something that isn't there. And, based on your response
to gm, you're being fairly obnoxious about it.
--
Dave Martin
DMA, Inc.
Nashville, TN
"Gordon Nay" <insi...@mediaone.net> wrote in message
news:njNS4.1741$J7.8...@typhoon.mw.mediaone.net...
Mike Rivers
> I have deeply mixed feelings regarding all of the claims I see of "24-bit
> resolution" in A/D converters.
>
> Everyone here does realize, don't they, that real questions exist as to whether
> "real" 24 bit performance can really be directly measured? And that there is
> every possibility that this rather chimerical spec is probably not being
> reached by ANY of these products?
Why, of course it can be directly measured, as long as you accept the
definition of "24-bit" that the semiconductor manufacturers have
provided to us. If it spits out 24-bit data, it's 24-bit. Just hook
up a scope and count. <g>
Sort of like comparing a pint of Good Humor ice cream with a pint of
Ben and Jerry's. Both are a pint, but one will do a much better job
of clogging your arteries and putting a smile on your face than the
other.
> Do we REALLY have to tolerate "marketing bits" for ANOTHER 10 years?
Sigh! I guess so. And I suppose it won't be long before we have the
32-bit converter.
> C'mon guys and girls, can we please demand some backup from the manufacturers
> who make unsupportable claims?
Well, many of the serious manufacturers are now putting a
signal-to-noise specification, or a dynamic range specification (which
takes into account the noise-not-intelligence). This always raises
questions from the uninitiated like "What's wrong with this converter?
It must not be 24-bit. I thought that it was supposed to be 144 dB
with 24 bits".
Bits is bits, but noise and dynamic range is what gets the work done
(better or worse).
--
I'm really Mike Rivers (mri...@d-and-d.com)
Mike Rivers
In article <20000512030012...@ng-cg1.aol.com> go3d...@aol.comsnot writes:
> I just had a 24-bit DAT master come in for mastering last week and guess what?
> There was a DC offset at -55 dB FS, just off the Tascam's meters, so no one
> knew it was there. But on my 16-bit, 96 dB metering it was dreadfully obvious.
> What does THAT do for your 24-bit, 144 dB headroom system?
>
> Not good things I assure you.
>
> Don't even ask me how well they used the available 55 dB of headroom!!
55 dB of headroom? That ought to be 50 dB more than they really need.
<G>
John La Grou
<insi...@mediaone.net> wrote:
>Seriously George, you accused the members who posted in this newgroup and
>the manufacturers of the best converters of something that simply isn't
>true. I saw no one or any advertising that claimed 24bit quality, only 24bit
>resolution. I do think almost all here know the difference.
I don't. "24 bit resolution" tells me that all 24 bits have been
accurately resolved. Maybe they're out there, but I don't know of any
24-bit ADC that can linearly resolve an entire 24 bits.
JL
fs...@televar.com
a difficult time coming up with a hard standard by which to measure
converter performance. I mean, what exactly do we measure to determine
the resolution of a converter. Passband noise? Weighted or
Unweighted? Linearity? What?
Years ago I built converters that used UltraAnalog parts. That companys
stated that these parts were 20 bit parts. They measured very well--
about 19+ bits of resolution (and they sound great, imo). So those
parts were very, very close to their advertised claims; close enough
for me to not be offended by their 20 bit claim. However when those
parts are used with a poorly designed analog section, the overall
performance drops and could no longer be considered 20 bit performance,
if that is, noise floor is the criteria.
My point is (and sorry for the rambling) that I have not measured many
converters that meet their advertised claims when the broadband noise
floor is used as a brickwall reference point. I have measured
converters that approach their advertised claims when linearity was
measured (22-23+ bits). So what's the point? The point is that the
noise floor is not a brick wall in digital anymore than it is in
analog. Audio information is obtainable from data below the noise
floor of the converter.
So I am still unclear of how to determine the actual resolution of a
converter. But then again, I'm an analog kinda guy and I don't really
don't know what I'm talking about with this digital stuff.
Still, I share George's concern.
Fred Forssell
Forssell Technologies
www.forsselltech.com
----
In article <391B5BB9...@ix.netcom.com>,
gml...@ix.netcom.com wrote:
> I have deeply mixed feelings regarding all of the claims I see of "24-
bit
> resolution" in A/D converters.
>
> Everyone here does realize, don't they, that real questions exist as
to whether
> "real" 24 bit performance can really be directly measured? And that
there is
> every possibility that this rather chimerical spec is probably not
being
> reached by ANY of these products?
>
> Do we REALLY have to tolerate "marketing bits" for ANOTHER 10 years?
>
> C'mon guys and girls, can we please demand some backup from the
manufacturers
> who make unsupportable claims?
>
> Kindly,
> George Massenburg
> gml...@ix.netcom.com
Sent via Deja.com http://www.deja.com/
Before you buy.
Scott Dorsey
>I have deeply mixed feelings regarding all of the claims I see of "24-bit
>resolution" in A/D converters.
>
>Everyone here does realize, don't they, that real questions exist as to whether
>"real" 24 bit performance can really be directly measured? And that there is
>every possibility that this rather chimerical spec is probably not being
>reached by ANY of these products?
>
>Do we REALLY have to tolerate "marketing bits" for ANOTHER 10 years?
>
>C'mon guys and girls, can we please demand some backup from the manufacturers
>who make unsupportable claims?
I think you're complaining about something that is merely a symptom of a
more fundamental problem.
Specsmanship is rampant throughout the whole industry. There are companies
out there which are making truly professional microphones, high grade
devices, which are listing "frequency ranges" without tolerances on their
data sheet. There are more data sheets than I can count which have frequency
response plots "derived from third octave measurements" without noting
that fact.
The fact that there are lots of 24 bit converters out there which only have
16 significant bits is comparatively minor. In the last five years or so,
I have seen a huge increase in this kind of thing, and it makes me feel like
I'm in a consumer electronics store in the seventies, with PMPO ratings
on everything.
I regard this as a terrible thing, because the more meaningless and fudged
specifications appear, the more people begin to develop the illusion that
all specifications are meaningless, and that's a step on the road to rejecting
all measurements (something which has done great damage in the audiophile
world).
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."
z_man_...@hotmail.com
Steinberg is trying to sell it with its Nuendo system using a different
faceplate. Lots of Cubase/Nuendo users seem to _really_ like the sound
of these.
Z Man
In article <jnh-110500...@192.168.0.2>,
j...@epsno.com (Jedd Haas) wrote:
> how about the
>
> "RME ADI-8 PRO Hi-Quality AD converter" (see
http://www.sfb.net/ad-da08.htm).
>
> Or is one of the converters already mentioned this one under another
name?
Alan
I must confess I haven't taken the time to read through this entire thread
yet, but please add Merging Technologies Sphynx interface to the
evaluation/shootout. I think you'll be pleasantly surprised.
http://www.merging.com
Respectfully,
Alan Thacker
Jedd Haas wrote in message ...
>how about the
>
>"RME ADI-8 PRO Hi-Quality AD converter" (see
http://www.sfb.net/ad-da08.htm).
>
>Or is one of the converters already mentioned this one under another name?
>
>--
>Jedd Haas - Artist, Writer, Guitarist. http://www.gallerytungsten.com
>http://www.antijazz.com
Lynn Fuston
>> Don't even ask me how well they used the available 55 dB of headroom!!
>
>55 dB of headroom? That ought to be 50 dB more than they really need.
><G>
Did I mention that the peak to average ratio on the unmastered version was 5.5
dB?
So who needs that extra 50 dB of dynamic range anyway, much less that other 90
below that?
Lynn Fuston
3D Audio Inc
Music Mixing and Mastering
Franklin, TN
http://www.3daudioinc.com
email: go3d...@aol.com
Aux
> By the way, I don't have a favorite converter. Maybe the ones I build for my
> own use, but they're not products and you probably wouldn't like them anyway.
Ouch!
--
Marc
Larry Lessard
that the best mesurement of the quality except for actually listening?
From my experience, I haven't heard any digital converter that sounds
better than the Otari RADAR 16 bit. IT is the only one I know of that
resolves all 16 bits (although I haven't checked the specs on a lot of
them). It's the truncating on lesser converters that causes the
degredation of the sound quality. The reason for this according to Otari
is the use of ceramics in their converters. Call Otari to get the actual
technical facts because I'm not a technical guy. I just use my ears.
I've also had great results using the Crane Song HEDD A/D for mixing. I
haven't had the pleasure of using the GML converters.
--
• • • • • • • • • • • •
Larry Lessard, Systems Consultant
Pro Audio Design, Inc., 357 Liberty St. Rockland, Ma. 02370
Authorized SSL service and factory outlet for preowned SSL consoles.
"A Full Service Professional Audio Company"
http://www.proaudiodesign.com • http://www.UsedSSL.com
phone: 781-982-2600 ext.605 fax: 781-982-2610
L. David Matheny
news:391BD6...@mercenary.com...
> Gordon Nay wrote:
> > I haven't seen any evidence from any of these manufacturers claiming
noise
> > or distortion 24bits down. And, I think that everyone in this thread
that I
> > have seen post so far has complete understanding of this. This rant
should
> > go someplace else. It is not helpful to what we are trying to do. That
is:
> >
> > Find the best sounding converter, ALL CLAIMS ASIDE.
>
> While you're at it, would you discover the 'best shade of blue' and the
> 'prettiest girl'? What the hell, seeing as you want to make subjective
> opinions "fact"...when you find the nicest shade of blue, be a sport and
> keep us in the loop.
>
With all due respect, I don't think it's quite that subjective. As I
understand it, he intends to record below -48dB, discard the high-order
bytes, and listen using (good, old, reliable) 16-bit D/As. Even without
golden ears, if samples from the bottom 16 bits of the converters sound
good, the 24-bit state of the art is a lot better than I suspect. I would
be interested in hearing the upshifted 16-bit samples.
Arny Krueger
"L. David Matheny" <dmno...@netassoc.nut> wrote in message
news:2qYS4.242$mh3.4248@news...
>
> With all due respect, I don't think it's quite that subjective. As I
> understand it, he intends to record below -48dB, discard the high-order
> bytes, and listen using (good, old, reliable) 16-bit D/As. Even without
> golden ears, if samples from the bottom 16 bits of the converters sound
> good, the 24-bit state of the art is a lot better than I suspect. I would
> be interested in hearing the upshifted 16-bit samples.
>
If you want to hear music samples with varioius amounts of "bits", please
see:
http://www.pcabx.com/technical/index.htm under
"24, 16, 15, 14, 12, 11, and 10 bit 44 KHz Samples"
Rick Krizman
Gordon Nay wrote:
>
> We will not try to do a complete ranking of these units, as I said, my
> client will determine what he likes best and why. You'll just have to wait
> and see what I can post later on. I will detail the methodology then.
With all due respect, why should we be interested in what your client thinks?
RIck Krizman
KrizManic Music
Michal
fs...@televar.com wrote:
> I agree with the general tone of George's posting here. But I also have
> a difficult time coming up with a hard standard by which to measure
> converter performance. I mean, what exactly do we measure to determine
> the resolution of a converter. Passband noise? Weighted or
> Unweighted? Linearity? What?
I think DR and distortion correlate best with AD sonic performance. I want
to underscore _correlate_, as oppose to _represent_. It's possible that a
converter with lower DR will sound preferably to better measuring one. A
good example is DSD, which sound stunning , yet measures worse than PCM.
Similarly AD797 opamp form Analog has one of the lower distortion
measurements, yet it doesn't sound good. We can only measure that much at
the moment and the methods are fairly primitive, the ear as an evaluation
device remains unchallenged.
Michal @ Mytek
*********************************************
http://www.mytekdigital.com >>>>>>
24 bit 48/96 k Digital Audio Converters
DAW 9624 (tm) , D-Master Recorder (tm)
*********************************************
CASPERCOMM
>better than the Otari RADAR 16 bit.
When my studio upgraded from RADAR I to RADAR II we had a chance to AB the two
units. I was expecting the diference to be subtle but it wasn't even close, the
RADAR II converters are a huge improvement.
Doug Joyce
with a project studio for demo's. I lurk and listen here mostly. I'm late
picking up this thread but I've got to ask, why is it that when I sometimes
log on here and George Massenberg has offered up a comment/suggestion/
observation, then inevitably he is flamed by some newbie (the guitar amp
micing
thing) or someone cops the attitude that his views must be challenged or
criticized? Why not just just let the man have his say and respect his
experience
and track record? Why, why, why WHY???? This condescending crap just pisses
me off!!! (rant, grumble, bitch, moan, etc)
Gordon Nay <insi...@mediaone.net> wrote in message
news:VcMS4.1659$J7.8...@typhoon.mw.mediaone.net...
> Dear George,
>
> Ah, I see, perhaps you are upset that we haven't included your converters
in
> our evaluation. Maybe you could label them "32 bit Super Deluxe" and we
will
> all snap them up. I have tested your equipment in my labs and set up A/B
> comparisons using them for clients in the past. I won't tell you the
results
> now because in all likelihood you wouldn't believe me.
>
> Seriously George, you accused the members who posted in this newgroup and
> the manufacturers of the best converters of something that simply isn't
> true. I saw no one or any advertising that claimed 24bit quality, only
24bit
> resolution. I do think almost all here know the difference.When someone
> gm <gml...@ix.netcom.com> wrote in message
> news:391B70A5...@ix.netcom.com...
> > Dear Gordon,
> >
> > Interesting that you should put it this way.
> >
> > I rant so little that I felt that I could get away with it. Ahhh, well,
I
> > suppose not. Any thoughts about where I should go to rant?
> >
> > Gordon, have you established a testing protocol yet? Are you planning
to
> do
> > pair testing by A-B-X or A-B-C-HR?
> >
> > Do you mind our asking who "us" is?
> >
> > By the way, I don't have a favorite converter. Maybe the ones I build
for
> my
> > own use, but they're not products and you probably wouldn't like them
> anyway.
> >
> > Thanks,
> > George
> >
> >
> > Gordon Nay wrote:
> >
> > > I haven't seen any evidence from any of these manufacturers claiming
> noise
> > > or distortion 24bits down. And, I think that everyone in this thread
> that I
> > > have seen post so far has complete understanding of this. This rant
> should
> > > go someplace else. It is not helpful to what we are trying to do. That
> is:
> > >
> > > Find the best sounding converter, ALL CLAIMS ASIDE.
> > >
> > > We may not even do any test to determine what units meet what claims,
> our
> > > focus will be on how these units SOUND.
> > >
> > > Please respond, George, with what you think is the best sounding
> converter,
> > > regardless of claims, and why. Your experience could be far more
> valuable in
> > > this way.
> > > If what you recommend seems resonable to us, we may include your
choice
> in
> > > our test.
> > >
> > > Thank You
> > >
> > > gm <gml...@ix.netcom.com> wrote in message
> > > news:391B5BB9...@ix.netcom.com...
> > > > I have deeply mixed feelings regarding all of the claims I see of
> "24-bit
> > > > resolution" in A/D converters.
> > > >
> > > > Everyone here does realize, don't they, that real questions exist as
> to
> > > whether
> > > > "real" 24 bit performance can really be directly measured? And that
> there
> > > is
> > > > every possibility that this rather chimerical spec is probably not
> being
> > > > reached by ANY of these products?
> > > >
> > > > Do we REALLY have to tolerate "marketing bits" for ANOTHER 10 years?
> > > >
> > > > C'mon guys and girls, can we please demand some backup from the
> > > manufacturers
> > > > who make unsupportable claims?
> > > >
> > > > Kindly,
> > > > George Massenburg
> > > > gml...@ix.netcom.com
> > > >
> > > >
> > > >
> > > >
> > > > Gordon Nay wrote:
> > > >
> > > > > We are getting ready to do the next generation of A/D converter
> > > shootouts.
> > > > >
> > > > > Currently under consideration are the following:
> > > > >
> > > > > dB dB 44-96
> > > > > AD122-96 MKII
> > > > > Mytek 8x96
> > > > > SEK'D 2496 DSP
> > > > > Lucid AD9624
> > > > > AD8824
> > > > > Sonorus Modular 8
> > > > > Swissonics AD 8 (PRO) 8ch 96KHz
> > > > > AD96 4ch 96KHz
> > > > > Prism Dream AD-2
> > > > >
> > > > > If there are any other A/D's that you feel are in the same class
and
> > > that we
> > > > > should include, please Email me with Make, Model and why you feel
> they
fs...@televar.com
OK, but DR as determinded by what? At the low end of the measurement
process we are going to hit the noise floor, and there is still useable
audio data the can be pulled from the bits below the noise floor. So
it seems to me that the passband noise floor shouldn't be the
measurement point on the low end of the scale. (I don't really like
that concept, but it seems that it could be a valid point.) The same
thing applies to distortion measurements. Do we include the passband
noise in the measurement (THD+N) or do we ignore it (THD)?
I don't know the answer to these questions. To my knowledge, there is
no *standard* by which to measure actual dynamic range in converters.
Maybe I'm wrong on that, as I said I am not a digital kind of guy. But
I have done differential linearity tests on some of the newer
converters out there, and their performance does come *close* to 24 bit
performance. Their noise floor measurements do not.
So, as I said, I agree with the general tone of Georges posting. For
me, this type of market is hype and I don't care for marketing hype. I
wish more manufactures and marketers would be honest and not push the
claim they make in their literature to limits of credibility or
beyond. I know I'm spittin' in the wind on that one.
--
Fred Forssell
Forssell Technologies
www.forsselltech.com
In article <391CB453...@erasemytekdigital.com>,
Michal <mic...@erasemytekdigital.com> wrote:
>
> I think DR and distortion correlate best with AD sonic performance. I
want
> to underscore _correlate_, as oppose to _represent_. It's possible
that a
> converter with lower DR will sound preferably to better measuring
one. A
> good example is DSD, which sound stunning , yet measures worse than
PCM.
> Similarly AD797 opamp form Analog has one of the lower distortion
> measurements, yet it doesn't sound good. We can only measure that
much at
> the moment and the methods are fairly primitive, the ear as an
evaluation
> device remains unchallenged.
>
> Michal @ Mytek
>
> *********************************************
> http://www.mytekdigital.com >>>>>>
>
> 24 bit 48/96 k Digital Audio Converters
>
> DAW 9624 (tm) , D-Master Recorder (tm)
>
> *********************************************
>
>
Sent via Deja.com http://www.deja.com/
Before you buy.
Michal
fs...@televar.com wrote:
> To my knowledge, there is
> no *standard* by which to measure actual dynamic range in converters.
>
> OK, but DR as determinded by what?
True , this is another area where numbers lie. Different manufacturers use
different definitions , often to make their numbers look good, sometimes
impossible (like claims of -120dB THD). However there is an AES standard
and a number of commonly accepted definitions. I use definitions from
Crystal:
http://www.cirrus.com/products/papers/meas/meas.html
> ) The same
> thing applies to distortion measurements. Do we include the passband
> noise in the measurement (THD+N) or do we ignore it (THD)?
>
Whatever- I meant DR or THD+N, but it doesn't really mattter since as I
said it only _correlates_ , in other words you can't tell how something
sounds just by looking at the numbers, no matter what method.
Terry Demol
fs...@televar.com wrote in message
<8fjt11$u3d$1...@nnrp1.deja.com>...
>Hi Michal,
>
>OK, but DR as determinded by what? At the low end of the
measurement
>process we are going to hit the noise floor, and there is
still useable
>audio data the can be pulled from the bits below the noise
floor. So
>it seems to me that the passband noise floor shouldn't be
the
>measurement point on the low end of the scale. (I don't
really like
>that concept, but it seems that it could be a valid point.)
>thing applies to distortion measurements. Do we include
the passband
>noise in the measurement (THD+N) or do we ignore it (THD)?
>
knowledge, there is
>no *standard* by which to measure actual dynamic range in
converters.
of guy. But
>I have done differential linearity tests on some of the
newer
>converters out there, and their performance does come
*close* to 24 bit
>performance. Their noise floor measurements do not.
The DR is usually measured with a -60dB signal and lumping
THD + noise with a 20k BW. At -60dB the converter
nonlinearities
are usually less than RMS sum of noise so its really just
the residual
noise which is being measured.
I'm not sure having 24bit noise floor is really necessary
although obviously we strive for what we can get.
I think Mical summed it up when using the AD797 as an
example. Our experiences are identical with a twist. We
tried the 797 V's other HQ audio opamps and also didn't
think it sounded very good but we did it thru very linear SS
chain
*and* SET (yes, single ended triode) chain. Differences
in opamps were still detected thru SET chain with
its measured DR of maybe 80dB and THD of maybe .2%
(at listening levels)... so much for masking theory.
I have confirmed this many times. Another example
is Marshall 30th aniversary GTR amp. DR maybe
70dB THD maybe 2% on cleanest setting. Start
swapping opamps in the effects loop send return
(only ones used) The differences can still be heard
even with .0003% THD opamps like 2604.
>
>So, as I said, I agree with the general tone of Georges
posting. For
>me, this type of market is hype and I don't care for
marketing hype. I
>wish more manufactures and marketers would be honest and
not push the
>claim they make in their literature to limits of
credibility or
>beyond. I know I'm spittin' in the wind on that one.
I suppose its getting harder and harder for the converter
manufacturers as the DR gets better slowly. Even Crystal
admits that measuring the non linearities is a challenge
because its so dependant on the analog front end used
which can even interact and cancel non linearities
of the converter.
Again I agree with Michal, measurements are a good
start but the ear is the final arbiter until better
correlation
between the two are learned.
Terry Demol
Andrew P. Mullhaupt
> gm wrote:
> >
> > I have deeply mixed feelings regarding all of the claims I see of
"24-bit
> > resolution" in A/D converters.
>
>
> about...
Every added bit means twice the amplitude resolution. That translates into
about a factor of 1000 for every 10 bits. Full scale is the level that uses
all the bits.
So here's an experiment which illustrates one way of estimating the bit
depth of a converter, as well as a thought experiment to determine how much
you care, or why some of those bits might be marketing bits.
Generate sine waves at two different frequencies. Adjust one to be twice the
amplitude of the other, sum them and then adjust the overall peak amplitude
to be full scale. Now feed that signal to the ADC. What you get is a
digitized version of the two sine waves. It is quite feasible to then
digitally filter the resulting wordstream to remove the big sine wave
component. What's left (the residual) _ought_ to look like the smaller
digitized sine wave. You can actually measure how close to a pure sine wave
the residual is in various ways. But if it fails to look like a digitized
sine wave, then you don't even have two bits of resolution.
Now perform this experiment with the ratio of amplitudes of the two sine
waves being four, eight, sixteen, etc.; each time you double the ratio of
amplitudes, you check for another functioning bit. A ratio of 2, 4, and 8,
corresponds to 1,2 and 3 bits, 1024 is for 10 bits, 1,048,576 for 20 bits,
16,777,216 for 24 bits, 1,073,741,824 for 32 bits, etc.
I glossed over the point that if you have 10 bits and run the sine waves at
an amplitude ratio of 1024, then the residual is a sine wave but only one
bit's worth of sine wave - it looks like a square wave (and vice versa) and
so grading how good that residual is not entirely obvious. In fact it's
easier to avoid that in practice. Instead, you look at the results for
ratios of 2,4,8, etc., and where the residual finally craps out and looks a
lot more random than like some sort of wave and figure that the number of
"good" bits is some number less that the number of doublings of the
amplitude.
If you actually look at the ADC circuit, you may find the thing is capable
on the blackboard of discriminating between 16,777,216 levels and outputting
all 24 bits of those. That part is likely to be called a 24 bit converter.
But what happens in practice can be quite a bit different. If you run an
amplitude doubling test then it's possible that it will crap out at 19 bits,
and that although the other 5 bits are being output by the converter, they
are essentially random thumb twiddling of digitized background noise. Now
the manufacturer may use a 24 bit wide word, and use all those output bits,
and call them 24 bit converters, but a recording engineer would be well
within his rights to regard that converter as a 19 bit converter. Those 5
extra nonsense bits are the "marketing bits" that George refers to.
In order for a converter with 24 good bits, critical parts of the circuit
have to be able to easily perform at 144dB dynamic range. Otherwise, they
add enough internal noise to screw up those last few sensitive bits before
conversion is even started.
Some points:
1. It's gotten a _lot_ easier to make better converters since people went to
"one bit" type conversion, and these can safely be "dialed up" for 24 bits.
That doesn't mean they actually get 24 bits.
2. It's not so easy to make just any analog circuit you like have a 144dB
dynamic range at room temperature.
3. Some parts of a particular type may be 24 bit wide, spec'ed as good to 19
bits, and some of them may perform to 19 bits, others to 22, due to
part-to-part variation.
4. Calling a converter 24 bit because it outputs 24 bits even though maybe
at most 21 of them are good is like saying a car can do 150 miles per hour
because the speedometer goes up that high. But would that stop marketing?
5. The "marketing" bits are the "least significant" (highest resolution)
bits. So you can't use them for headroom. Digital saturation sounds pretty
lame, so you almost always have to give away a bit or two for headroom. By
using "look ahead" (i.e. non real time conversion) you can reduce this to
not much.
> Due to my lack of education, I'm most suseptable to 'marketing
> horseshit' at this point. As I understand it, 'bit depth' effects "word
> length", which effects not only 'dynamic range', but 'resolution' as
> well?.
Bit depth and word length are pretty close; sometimes identical.
Dynamic range is different. You won't get more dynamic range than the bit
depth permits (although there are nonlinear conversion schemes that can
stretch a little that's more important when there are not many bits). But
just because your digital circuit is deep enough doesn't mean that you get
that much dynamic range.
Resolution can mean either of these, depending on who's talking.
Now to put into perspective how much you care:
A Neumann U89i has a dynamic range spec for the microphone amplifier of
117dB. That corresponds to 19.5 bits _if_ you don't have any of the
specified headroom left. You probably never need all those 19.5 bits to get
the U89i output since the signal to noise ratio is spec'ed at 77dB; which is
just under 13 bits. Giving yourself 6dB of digital headroom you can
reasonably expect to get a U89i well recorded if you have 16 good bits. This
is the kind of consideration that made 16 bit words seem like a reasonable
choice for the original CD standard, and to some extent it's still a valid
argument.
Other Neumann microphones have amplifiers that seem to top out around 140
dB, and some microphone amplifiers seem to get a little more 140dB (I think
Josephson measurement microphones do that). But that tells me that 144dB is
not your average analog circuit performance - you have to have a pretty good
signal chain to get upset about not having 25 good bits. Note that even
quite good microphones don't seem to have signal to noise ratios that need
more than 16 good bits.
Now most 16 bit converters didn't give you 16 good bits. This is why a 24
bit converter that only has 20 good bits is still better than a 16 bit
converter that has 16 good bits (all other things being equal), even though
it has 4 "marketing" bits. And a converter with 24 bits
> How do the 'increased sampling rates' effect resolution?
Strictly speaking, not at all.
However, with the one-bit converters, sample resolution is achieved by
oversampling, and this can get a bit confusing.
> And why/how is it different?
Leaving aside the inner workings of one-bit or other multirate converters:
Sample rate doesn't have anything to do with resolution or dynamic range and
vice versa. You can sample 24 good bits at 22kHz, and you will get a pretty
high resolution picture of whatever you have from something under 10kHz on
down. This means your cymbals will not come out so accurate. You can sample
12 good bits at 44kHz and you will have the same amount of information about
the sound, but here you will get a pretty good frequency coverage up to
20kHz so the stuff won't sound shelved off at 10kHz, but there is likely to
be a lot of funky grainyness and I expect reverb tails and other quiet
things won't sound so good.
If you want to get enough digital quality to cover just about anything
analog can do now, you probably need 24 good bits and 48kHz, maybe a bit
more.
Now there are some technical aspects of sample rates. Filter characteristic
steepness scales with sampling frequency, so a one pole filter at 96kHz
sample rate is as steep as a two pole filter at 48kHz. But it's easier to
build a one pole filter than a two pole filter. This is why 64x or 128x
oversampling filters are popular - you can sometimes make a simple filter do
the job of a more complicated just by cranking up the sample rate, as well,
and cheaper. But these filters typically up- and down- convert so you don't
have to have 64 ro 128 times the data bandwidth through the whole digital
signal path. It's been widely thought that it's cheaper to make a lot of
simple things faster (just crank up the clock) than to make things more
complicated, so you see that a lot, and it's yet another frequency that the
marketers can put on your plate.
Later,
Andrew Mullhaupt
Monte P McGuire
L. David Matheny <dmno...@netassoc.nut> wrote:
>With all due respect, I don't think it's quite that subjective. As I
>understand it, he intends to record below -48dB, discard the high-order
>bytes, and listen using (good, old, reliable) 16-bit D/As. Even without
>golden ears, if samples from the bottom 16 bits of the converters sound
>good, the 24-bit state of the art is a lot better than I suspect. I would
>be interested in hearing the upshifted 16-bit samples.
An observation... this test method has nothing to do with how I use a
converter. Does anyone run a converter at -48dBFS? Why concentrate
on the bottom 16 bits?
There are lots of potential problems when you drive a converter near
full scale, so IMHO, high level performance is pretty darn important.
If the design skimps on power supply bypassing or has some sort of
related problem, this will get completely unnoticed at such low
levels. Also, some designs that use two or more converters and gain
ranging to get a very low noise floor might be run so that only one of
the converters is used. That's not a realistic test, since you'd miss
any artifacts of the gain ranging scheme.
Regards,
Monte McGuire
mcg...@world.std.com
Arny Krueger
<fs...@televar.com> wrote in message news:8fjt11$u3d$1...@nnrp1.deja.com...
> Hi Michal,
>
> OK, but DR as determinded by what?
Ultimately, audibility.
>At the low end of the measurement
> process we are going to hit the noise floor, and there is still useable
> audio data the can be pulled from the bits below the noise floor.
AFAIK there three technical properties of the noise floor determine the
degree to which musical signal below it are audible:
(1) The amplitude of the noise floor.
(2) The spectral content of the noise floor
(3) The Probability Density (PDF) of the noise floor.
> So
> it seems to me that the passband noise floor shouldn't be the
> measurement point on the low end of the scale. (I don't really like
> that concept, but it seems that it could be a valid point.)
Once the three things I've mentioned are known, the degree to which any
prticular musical signal below the noise floor is heard can be estimated.
Obviously the degree to which individuals hear below the noise floor varies,
but that variation has limits and can also be predicted.
>The same
> thing applies to distortion measurements. Do we include the passband
> noise in the measurement (THD+N) or do we ignore it (THD)?
It strikes me that at least, we consider both seperately and also together.
> I don't know the answer to these questions. To my knowledge, there is
> no *standard* by which to measure actual dynamic range in converters.
There are standards for comparing them with technical tests.
I've probably measured and listened to as many different converters as most
(over 100) particularly in the last 3 years, given all the tests I've done
for www.pcavtech.com.
I report SNR at www.pcavtech.com in the form of a spectral analysis, with
average levels annotated.
I find that in general, the noise floors I see have spectral response that
is at least as flat as the high-level bandpass of the convertor, which is to
say it is usually quite flat. I really don't have the tools to look at PDF's
but I suspect they don't vary a great deal.
> Maybe I'm wrong on that, as I said I am not a digital kind of guy. But
> I have done differential linearity tests on some of the newer
> converters out there, and their performance does come *close* to 24 bit
> performance. Their noise floor measurements do not.
Agreed. Often it is less than what good 16 bit performance would suggest.
> So, as I said, I agree with the general tone of Georges posting. For
> me, this type of market is hype and I don't care for marketing hype. I
> wish more manufactures and marketers would be honest and not push the
> claim they make in their literature to limits of credibility or
> beyond. I know I'm spittin' in the wind on that one.
In most cases it is easier to for me to explain what a "24 bit converter" is
(it is almost a converter that handles or creates 24 bit samples though the
API) than what a "20 or 18 bit converter" is when the "20 or 18 bit
converter" handles or creates only 16 bit samples to through the API.
Mark Plancke
I think you might want to send a copy of this to David J. for
possible inclusion in the FAQ. Very good stuff.
Mark Plancke
I don't know the secret of success, but the secret
of failure is to try to please everybody. --Bill Cosby
SOUNDTECH RECORDING STUDIOS
Windsor, Ontario, Canada
http://SoundTechRecording.com
Thomas W
> If you run an amplitude doubling test then it's possible that
> it will crap out at 19 bits, and that although the other 5
> bits are being output by the converter, they are essentially
> random thumb twiddling of digitized background noise.
I'd like to come back to Fred Forssell's point about convertor
linearity.
Those last 5 bits may be representing noise, but so long as the
convertor maintains linearity there will also be audio signal
hidden in there.
Noise reduction techniques (FFT noise reduction) can eliminate
10 dB or more of background noise, revealing signal *below* the
noise floor of the convertor.
> Those 5 extra nonsense bits are the "marketing bits" that
> George refers to.
They are only nonsense, if the convertor loses linearity. And
even marketing people have their uses :-)
This issue is also kind of related to dither vs truncation. A
good A/D should be linear below it's noise floor, which
effectively dithers the signal into the noise... non-linearity
near the noise floor will give nasty correlated distortion
artifacts similar to truncation.
Cheers,
Thomas
* Sent from RemarQ http://www.remarq.com The Internet's Discussion Network *
The fastest and easiest way to search and participate in Usenet - Free!
Doug Joyce
Andrew P. Mullhaupt <amul...@zen-pharaohs.com> wrote in message
news:8flcuf$kgb$1...@slb6.atl.mindspring.net...
Andrew P. Mullhaupt
Mark Plancke <Ma...@SoundtechRecording.com> wrote in message
news:rurths48o51m39r86...@4ax.com...
> Andrew
>
> I think you might want to send a copy of this to David J. for
> possible inclusion in the FAQ. Very good stuff.
OK. We might want to edit it a little if it's going there.
Also, I glossed over some technical details, but I was trying to get the
points across to people who are not comfortable with digital. If we want to
dot i's and cross t's then it gets more verbose.
Later,
Andrew Mullhaupt
Andrew P. Mullhaupt
Thomas W <thomasw...@trade-exchange.co.nz.invalid> wrote in message
news:2f594493...@usw-ex0102-084.remarq.com...
> Hello Andrew,
>
> > If you run an amplitude doubling test then it's possible that
> > it will crap out at 19 bits, and that although the other 5
> > bits are being output by the converter, they are essentially
> > random thumb twiddling of digitized background noise.
>
> linearity.
>
> Those last 5 bits may be representing noise, but so long as the
> convertor maintains linearity there will also be audio signal
> hidden in there.
That's an optimistic point of view. There _may_ be some signal in there;
there may not.
> Noise reduction techniques (FFT noise reduction)
FFT is one of the last things I would resort to. Adaptive filters in the
time domain make a lot more sense to me.
> can eliminate
> 10 dB or more of background noise, revealing signal *below* the
> noise floor of the convertor.
There are cases where you can eliminate loads of noise. There are cases
where you just can't.
> > Those 5 extra nonsense bits are the "marketing bits" that
> > George refers to.
>
> They are only nonsense, if the convertor loses linearity.
No, they really, really can be nonsense, linearity or not. Look's like I'll
have to prove that one.
Suppose you have a 24 bit converter which is good to 19 bits. Create a sine
wave and another signal which is a signal which interpolates a quasirandom
sequence of 32 distinct values. This second signal is basically constructed
so that when it is passed through the anti-aliasing filter for the converter
you get a quasirandom sequence with 32 different values. Now scale the
signals so that the superposition peaks at full scale and the funny signal
is only affecting the 5 least significant bits.
Ok so put this combined signal into the converter. You'll get 19 good bits
of sine wave. You'll get the rest of the bits that look random. Well,
they're supposed to - they're quasirandom. So there isn't anything spectral
you can do that will do any good - you'll leave those bits alone whatever
filtering religion you subscribe to, and no matter how linear your
converters are. They already have the right spectrum!
So if you're going to get any information from those bits, you're going to
have to do something that isn't filtering to correct them. If the 5 bit
signal comes from a cryptographically strong quasirandom number generator,
you will have to solve some extremely hard problems to find any way of
reducing the nonsense. And I can interpolate any sequence I like into a
band-limited signal by any number of methods, although trigonometric
interpolation with the FFT will work.
And if you're anti-aliasing filters aren't good enough, you're even further
up the creek without a paddle, which results from Ornstein's theorem.
> This issue is also kind of related to dither vs truncation.
Yes. It is almost the same issue. It's late, and I'm feeling a bit perverse,
so I'll let you think about what the above example has to say about
dithering....
Later,
Andrew Mullhaupt
Paul Frindle
real issues of the measurement of converters as being different to
conventional analogue equipment. The classes of error (and the profiles of
these errors with level and time) in the digital domain and conversion
equipment, renders the normal measurements of frequency response and THD+N
very poor yardsticks by which to judge sound quality. The marketing
statements that describe 'resolution and bits' are wholly misleading and
irrelevant in most cases and do nothing to fill the gap between conventional
measurements and those required in digital equipment. In fact the very
useage of the term 'resolution' in this way reveals a complete
misunderstanding of the science of digital audio.
I agree whole heartedly with GM and this post that this is the result of an
increase in 'consumeresque' marketing that is both derisive and insulting in
the professional field. And I would also agree that this kind of
pseudo-scientific hogwash is eroding the confidence and expectations of the
very customers the manufacturers are trying to impress. The 'leaking' of the
profit driven consumer market and its pseudo-science into the professional
field is most strikingly illustrated by the mystique and sophistry
surrounding the attempts to introduce new data formats into the industry.
These are directly intended to generate new revenue into the music delivery
industry, which is under threat from expiring royalties and electronic
delivery methods. Little if any of these initiatives have sound quality as
their primary motivation and there are actually no properly tested and
scientifically sustainable theories to support the claims that are mooted
for the increases in percieved sonic quality they offer, that can be
attributed directly to the formats themselves.
There is no reason to believe anything other than that wordlengths relate
only to final SNR and sampling rates relate only to allowable frequency
range. There is no such thing as a converter with a true dynamic range of
143dB and any converter that apparently sounds better when made capable of
reproducing ultrasonic frequencies is unfortunately badly designed and has
excess errors in the linearity, frequency or phase domains, at legal
sampling rates. All converters that suffer from excess linearity, frequency
response or phase errors (either signal or clock jitter derived) will sound
different. Those manufacturers that either do not understand the necessary
parameters or chose to cut corners will end up selling you converters (and
processing) that change the sound in often subtle and inconsistent ways. But
it would be a mistake to assume that all this apparent confusion denotes a
universal absence of actual science on the subject!
Sure, I'll be hacked off for ranting aswell (in the bunker)!
fs...@televar.com wrote:
> I agree with the general tone of George's posting here. But I also have
> a difficult time coming up with a hard standard by which to measure
> converter performance. I mean, what exactly do we measure to determine
> the resolution of a converter. Passband noise? Weighted or
> Unweighted? Linearity? What?
>
> Years ago I built converters that used UltraAnalog parts. That companys
> stated that these parts were 20 bit parts. They measured very well--
> about 19+ bits of resolution (and they sound great, imo). So those
> parts were very, very close to their advertised claims; close enough
> for me to not be offended by their 20 bit claim. However when those
> parts are used with a poorly designed analog section, the overall
> performance drops and could no longer be considered 20 bit performance,
> if that is, noise floor is the criteria.
>
> My point is (and sorry for the rambling) that I have not measured many
> converters that meet their advertised claims when the broadband noise
> floor is used as a brickwall reference point. I have measured
> converters that approach their advertised claims when linearity was
> measured (22-23+ bits). So what's the point? The point is that the
> noise floor is not a brick wall in digital anymore than it is in
> analog. Audio information is obtainable from data below the noise
> floor of the converter.
>
> So I am still unclear of how to determine the actual resolution of a
> converter. But then again, I'm an analog kinda guy and I don't really
> don't know what I'm talking about with this digital stuff.
>
> Still, I share George's concern.
>
> Fred Forssell
> Forssell Technologies
> www.forsselltech.com
>
> ----
> In article <391B5BB9...@ix.netcom.com>,
> gml...@ix.netcom.com wrote:
> > I have deeply mixed feelings regarding all of the claims I see of "24-
> bit
> > resolution" in A/D converters.
> >
> > Everyone here does realize, don't they, that real questions exist as
> to whether
> > "real" 24 bit performance can really be directly measured? And that
> there is
> > every possibility that this rather chimerical spec is probably not
> being
> > reached by ANY of these products?
> >
> > Do we REALLY have to tolerate "marketing bits" for ANOTHER 10 years?
> >
> > C'mon guys and girls, can we please demand some backup from the
> manufacturers
> > who make unsupportable claims?
> >
> > Kindly,
> > George Massenburg
> > gml...@ix.netcom.com
>
Steve Brown
Steve
Carl Rusk wrote:
> Benchmark is excellent, although they do not yet make a 24bit D/A, this is
> in part due to their honesty. It is certainly no frills but I believe it has
> the lowest jitter when slaving to external clock of any converter on the
> market (9 pico seconds)
>
> Carl Rusk.
>
> > From: Scott Hughes <hug...@usa.net>
> > Organization: EarthLink Inc. -- http://www.EarthLink.net
> > Reply-To: hug...@usa.net
> > Newsgroups: rec.audio.pro
> > Date: Thu, 11 May 2000 07:13:45 GMT
> > Subject: Re: High end A/D converter shootout
> >
> > The Weiss A/D?
Gordon Nay
make 24bit A/D according to there WEb site.
We are only interested in testing the A/Ds at this time and we are
considering the Benchmark.
Steve Brown <sbr...@nnbt.com> wrote in message
news:392434B3...@nnbt.com...
Thomas W
> That "if ... are linear" is one stumbling block. Suppose you
> want to get 24 bits of original signal from 16 bit quantized
> data.
You are referencing Pacific Microsonics, and their specialized
CD dither process to recover 20 bits of data out of the dither?
One might guess that you have, or currently do, work for them.
(These CDs are playable in normal players, but extra data can be
recovered since the dither signal is generated by a repeatable
algorithm and can be subtracted again with a special playback
deck.)
> Then the original quantization has to have been linear to 24
> bits. That's got to be less than 1 millionth of a percent THD
> for a full scale signal. And not just for a 1kHz tone.
Was it Fred Forssell who said he'd tested convertors for
differential linearity, and found some could maintain it nearly
24 bits down?
I'm not sure that absolute linearity is required; only enough
linearity that the thermal noise at say -110 dB doesn't convolve
signal at say -120 dB.
[FFT techniques are the standard and basic tool for noise
reduction :]
> Yeah, but I wouldn't use them. FFT is easy for some people
> because it makes the frequency domain easy to understand.
Okay.
> But the FFT as commonly (almost universally) practiced in
> audio is based on the periodogram - which has some well
> understood blemishes that can be removed by adding in some
> other tricks. But you're still pretty well stuck to stationary
> noise processes with good statistics.
That's fine to deal with thermal/ electronic noise, so long as
the harmonic distortion of the *noise* won't obscure the signal
we're retrieving below the noise floor. 20 dB should almost be
enough.
[Cool Edit provides a practical demonstration :]
> Good and best can be different.
For sure.
[Techniques rely upon having had a 'silence' sample :]
> That's not so good if the noise is nonstationary, is it?
> Especially if the noise is signal dependent (e.g. noise gated
> or keyed guitar track).
Well, I'm talking about removing the convertor noise without (at
this stage) caring about what's plugged in.
But even in those incorrect configurations, the NR will work. It
would however target the residual noise of the noise gate and
the convertors, rather than the guitar source...
This can be overcome by moving the noise gate *after* the noise
reduction. Also appropriate technique for compression; use after
NR.
[One presumes the limit of reduction, would be the error in
discriminating between noise and signal :]
> In some sense, yes.
I'll be bold and *define* the remaining noise, as the error in
discrimination of the NR process.
> And the trick is that a lot of people think that a noise
> model is the way to do this - such as that "silence sample"
> idea you mentioned above. Well, the optimal approach uses at
> least both a noise model _and_ a signal model. If you have
> signal dependent noise, (like dynamics processing upstream in
> the chain) then you may want to think about using a joint
> model.
Having a model of the signal, can surely improve discrimination
between it and noise. But doesn't sound appropriate for plug-and-
play application.
[White-noise spectrum by quasi-random generation :]
> No. And that's the point. The second signal has the same
> spectrum as white noise so you can't separate it by using a
> spectral noise model. But it's not white noise, it is entirely
> predictable, and can be removed exactly from the original sine
> wave. No, Cool Edit will not achieve it for you. But it can b
> done.
Understood that. This whole quasi-random thing is uh, relevant
to your thing but not to mine. It did confuse me for a while, as
to why you brought it in.
[Sine + quasi-random white spectra :]
>The full band component is _signal_ and _not_ noise.
Ooops, we just removed it. Well, I was never keen on that quasi-
stuff anyway :-)
[By these means we can quite drastically reduce the total power
of the noise signal :]
> But it was a signal component! The 'baby' was constructed to
> be spectrally indistinguishable from 'bathwater'.
Most basic musical inputs will not be full band, so spectral NR
should be fairly effective. Complete mixes tend more to full-
band, maybe not so useful for those.
> Not every signal process is called a filter. The ones that
> boil down to multiplying the spectrum by some function _are_
> filters by just about anyone's definition.
Understood.
> To clean up the superposed sine and quasirandom signal, you'd
> have to figure out that those quasirandom bits were _not_
> random (i.e. crack the quasirandom generator) and then you
> could get rid of _lots_ of other noise that might have crept
> in thermally. [snip]
> You can actually clean the quasirandom signal better than the
> sine wave if you can reproduce the sequence generator! But
> this process is not called filtering by most definitions of
> the word filter.
Yes, no... And you can clean that quasi-random signal rather
well, by regenerating it from scratch since you know what it is.
> The point of cryptographically hard sequence generation is to
> show that last process for cleaning up the quasirandom
> component, while possible, can be exceedingly difficult.
> In other words, there will be an escalating sequence of more
> and more clever noise removal algorithms that will sieve out
> wider and wider signal classes.
I thought you were arguing *against* noise reduction?
> I think the ladder can't go "all the way up" - in other words
> for every finite noise removal algorithm I can construct a
> signal that the algorithm will be fooled into recognizing as
> noise.
Only within the boundaries of discrimination.
> I think the old diagonalization proof for the insolubility of
> the halting problem for Turing machines works, but I didn't
> work it out carefully.
I think you're right, but in a pragmatic manner believe dinner
to be a more important problem.
> There's lots of things that reduce noise in many cases. You
> just can't do it in general. Try denoising a signal that has a
> pretty flat spectrum with these tools.
Accepted.
Regards,
Andrew P. Mullhaupt
"Thomas W" <thomasw...@trade-exchange.co.nz.invalid> wrote in message
> Hello Andrew -
>
> > That "if ... are linear" is one stumbling block. Suppose you
> > want to get 24 bits of original signal from 16 bit quantized
> > data.
>
> You are referencing Pacific Microsonics, and their specialized
> CD dither process to recover 20 bits of data out of the dither?
No. But it's not surprising that there is someone out there who
> One might guess that you have, or currently do, work for them.
I'm with a company called "S.A.C. Capital Management LLC"
> > Then the original quantization has to have been linear to 24
> > bits. That's got to be less than 1 millionth of a percent THD
> > for a full scale signal. And not just for a 1kHz tone.
>
> Was it Fred Forssell who said he'd tested convertors for
> differential linearity, and found some could maintain it nearly
> 24 bits down?
Maybe some can. But the glue circuitry, everything, has to be that linear to
expect to recover extra bits by filtering.
> I'm not sure that absolute linearity is required; only enough
> linearity that the thermal noise at say -110 dB doesn't convolve
> signal at say -120 dB.
I doubt that. What about good old unintended coupling between the digital
parts of the converter circuit and the upstream parts? Those _are_
correlated with the digitized signal.
> [Techniques rely upon having had a 'silence' sample :]
> > That's not so good if the noise is nonstationary, is it?
> > Especially if the noise is signal dependent (e.g. noise gated
> > or keyed guitar track).
>
> Well, I'm talking about removing the convertor noise without (at
> this stage) caring about what's plugged in.
And I'm pointing out that you can't get away with that. If there is leakage
of the converted signal back to the input of the converters, then a gated
signal silence sample will give you only a look at the linear part of the
converter noise, and it may well be the wrong thing to derive a matched
filter from in order to try and clean up the converter noise on a nonzero
signal.
> But even in those incorrect configurations, the NR will work.
It will do something. If you're concerned about the accuracy of the
recovered bits, then you may be screwed completely.
> Having a model of the signal, can surely improve discrimination
> between it and noise. But doesn't sound appropriate for plug-and-
> play application.
It's done all the time. Just about every modem in use today does this trick.
All cell phones have this sort of filtering in them. Those are pretty plug
and play.
> [Sine + quasi-random white spectra :]
> >The full band component is _signal_ and _not_ noise.
>
> Ooops, we just removed it. Well, I was never keen on that quasi-
> stuff anyway :-)
My German Expressionist synth-pop will not sound so good through your sound
laundry.
> [By these means we can quite drastically reduce the total power
> of the noise signal :]
> > But it was a signal component! The 'baby' was constructed to
> > be spectrally indistinguishable from 'bathwater'.
>
> Most basic musical inputs will not be full band,
There is your signal model. You didn't think you had one, but you do. I
think your signal model isn't going to do well on cymbals.
> > You can actually clean the quasirandom signal better than the
> > sine wave if you can reproduce the sequence generator! But
> > this process is not called filtering by most definitions of
> > the word filter.
>
> Yes, no... And you can clean that quasi-random signal rather
> well, by regenerating it from scratch since you know what it is.
That's actually the point. If you're recovering bits, there is _no better
way_ to do it than to reproduce the original pre-quantized signal! In other
words, the extent to which you can launder out the converters can be
strongly dependent on a good signal model.
> > The point of cryptographically hard sequence generation is to
> > show that last process for cleaning up the quasirandom
> > component, while possible, can be exceedingly difficult.
>
> > In other words, there will be an escalating sequence of more
> > and more clever noise removal algorithms that will sieve out
> > wider and wider signal classes.
>
> I thought you were arguing *against* noise reduction?
Neither for nor against. It's probably a good thing to do if done well. It's
just that you can't make claims that you can do it in general with a
spectral approach.
> > There's lots of things that reduce noise in many cases. You
> > just can't do it in general. Try denoising a signal that has a
> > pretty flat spectrum with these tools.
>
> Accepted.
Later,
Andrew Mullhaupt
Thomas W
and good evening -
[Referencing Pacific Microsonics?]
> No. But it's not surprising that there is someone out there who
>I'm with a company called "S.A.C. Capital Management LLC"
Are they developing a dither-extracting CD player also? :-)
[Differential linearity 24 bits down]
> Maybe some can. But the glue circuitry, everything, has to be
> that linear to expect to recover extra bits by filtering.
If we have differential linearity (incl the analog stages) to
that level, but not absolute linearity... what will we get when
we drop the noise bands?
[Not sure that absolute linearity is required]
> I doubt that. What about good old unintended coupling between
> the digital parts of the converter circuit and the upstream
> parts? Those _are_ correlated with the digitized signal.
I would regard the digital section as a moderately steady noise
generator. Thermal electronic noise is going to keep a few bits
flickering at all times.
> And I'm pointing out that you can't get away with that. If
> there is leakage of the converted signal back to the input of
> the converters, then a gated signal silence sample will give
> you only a look at the linear part of the converter noise, and
> it may well be the wrong thing to derive a matched filter from
> in order to try and clean up the converter noise on a nonzero
> signal.
If digital section leakage back -> analog section is
significant, then you've got a good argument for a more
sophisticated signal/ noise model.
Don't know to what extent this is the case, or whether simple
thermal electronic noise is the problem.
[Model of the signal, not appropriate for plug-and-play
application]
> It's done all the time. Just about every modem in use today
> does this trick.
> All cell phones have this sort of filtering in them. Those are
> pretty plug and play.
The modem knows it's talking to another modem, with
characteristic modulated signals. This is a different problem
from recording an arbitrary mic & musical instrument for most
accurate reproduction.
> My German Expressionist synth-pop will not sound so good
> through your sound laundry.
That was the point of the joke.
[Most basic musical inputs will not be full band]
> There is your signal model. You didn't think you had one, but
> you do. I think your signal model isn't going to do well on
> cymbals.
I knew there was a model, in that this form of NR would work
best on band-limited signals. The cymbals won't be degraded; the
NR just won't remove anything. Seems fine to me.
[Regenerate quasi-random signal:]
> That's actually the point. If you're recovering bits, there is
> _no better way_ to do it than to reproduce the original pre-
> quantized signal!
Which is useful for musical applications, in exactly which way?
[I thought you were arguing *against* noise reduction?]
> Neither for nor against. It's probably a good thing to do if
> done well. It's just that you can't make claims that you can
> do it in general with a spectral approach.
Not for every source, just for most of them :-)
My original point: spectral NR can remove masking noise in some
of those 'marketing bits', revealing information in them.
Your point: advanced signal & noise modelling will achieve
better NR results.
It's been interesting thrashing the issue. Slightly cautious
here, since you obviously have a strong grasp of theory but...
my original point still holds?
Cheers,
Andrew P. Mullhaupt
"Thomas W" <thomasw...@trade-exchange.co.nz.invalid> wrote in message
> Hi Andrew,
>
> and good evening -
>
> [Referencing Pacific Microsonics?]
> > No. But it's not surprising that there is someone out there who
> >I'm with a company called "S.A.C. Capital Management LLC"
>
> Are they developing a dither-extracting CD player also? :-)
No, we do, uh, 'other', stuff with DSP.
> [Differential linearity 24 bits down]
> > Maybe some can. But the glue circuitry, everything, has to be
> > that linear to expect to recover extra bits by filtering.
>
> If we have differential linearity (incl the analog stages) to
> that level, but not absolute linearity... what will we get when
> we drop the noise bands?
I haven't thought that one through carefully, but I don't think it matters.
> [Not sure that absolute linearity is required]
> > I doubt that. What about good old unintended coupling between
> > the digital parts of the converter circuit and the upstream
> > parts? Those _are_ correlated with the digitized signal.
>
> I would regard the digital section as a moderately steady noise
> generator.
I wouldn't. The clock, yes, but the transitions which are part of conversion
are correlated with the input.
> Thermal electronic noise is going to keep a few bits
> flickering at all times.
That too, but that's less troublesome.
> > And I'm pointing out that you can't get away with that. If
> > there is leakage of the converted signal back to the input of
> > the converters, then a gated signal silence sample will give
> > you only a look at the linear part of the converter noise, and
> > it may well be the wrong thing to derive a matched filter from
> > in order to try and clean up the converter noise on a nonzero
> > signal.
>
> If digital section leakage back -> analog section is
> significant, then you've got a good argument for a more
> sophisticated signal/ noise model. Don't know to what extent this is the
case, or whether simple
> thermal electronic noise is the problem.
It's a common enough issue in some applications that you can read about ways
to avoid it on spec sheets for A/D converters, such as this bit of a
Fairchild part sheet:
"To minimize noise injection into the analog section, VDDA may be
connected to a
separate, regulated +5V power supply. VDDD may be connected to a digital
supply....
AGND and DGND pins should be connected to a common ground plane. For
optimum
performance treat analog and digital PWB traces as transmission lines.
Route analog
connections cleanly to the TMC1175A. Segregate digital connections and
if necessary
terminate clocks to eliminate ringing. Prevent digital return currents
from flowing across
analog input sections of the TMC1175A."
and in the application notes for same:
"Capacitive coupling from digital to analog circuits may result in
poor A/D conversion.
Consider the following suggestions when doing the layout...."
> [Model of the signal, not appropriate for plug-and-play
> application]
> > It's done all the time. Just about every modem in use today
> > does this trick.
> > All cell phones have this sort of filtering in them. Those are
> > pretty plug and play.
>
> The modem knows it's talking to another modem, with
> characteristic modulated signals.
It doesn't have to.
> This is a different problem
> from recording an arbitrary mic & musical instrument for most
> accurate reproduction.
Trust me. It's very doable plug and play. It's not one of the harder
adaptive filtering problems.
> [Most basic musical inputs will not be full band]
> > There is your signal model. You didn't think you had one, but
> > you do. I think your signal model isn't going to do well on
> > cymbals.
>
> I knew there was a model, in that this form of NR would work
> best on band-limited signals. The cymbals won't be degraded; the
> NR just won't remove anything. Seems fine to me.
No, spectral signal-unaware noise reduction _can and does_ remove flat
spectrum components preferentially to narrow band components. It will not be
fine if we're talking about bits you can actually hear.
> [Regenerate quasi-random signal:]
> > That's actually the point. If you're recovering bits, there is
> > _no better way_ to do it than to reproduce the original pre-
> > quantized signal!
>
> Which is useful for musical applications, in exactly which way?
The best possible method of reproduction is musically useful. When you can
do it is the question.
> [I thought you were arguing *against* noise reduction?]
> > Neither for nor against. It's probably a good thing to do if
> > done well. It's just that you can't make claims that you can
> > do it in general with a spectral approach.
>
> Not for every source, just for most of them :-)
Well, OK, so you can do it well for sine waves and their friends.
But music does _not_ have such a simple spectrum that I would ignore it
entirely.
> My original point: spectral NR can remove masking noise in some
> of those 'marketing bits', revealing information in them.
>
> Your point: advanced signal & noise modelling will achieve
> better NR results.
No, my point is that I can give you signals such that no conceivable
practical algorithm can clean up any of the "marketing bits".
And for spectral noise removal, there are signals that sound like they have
narrow band features but which are totally unimproved by that sort of
technique. Here's an example in one bit depth:
1 0 0 1 0 1 1 0 0 1 1 0 1 0 0 1 0 1 1 0 1 0 0 1 1 0 0 1 0 1 1 0 . . .
When you take this signal as the first unresolved bit of your samples, what
happens is that spectral noise removal actually does get the signal out for
short sequences, but the longer the sequence you have, the spectral noise
removal converges to leaving the input invariant. This is essentially
because if you periodically continue the sequence at any convenient (e.g.
power of 2) length you get what looks like a narrow band spectrum. As the
sequence is not periodic, this continuation introduces an error in the
subsequent noise removal - spectral discrimination that gets the first bits
right, gets the next bits wrong. You can get an idea why this is so from the
rule that the sequence is constructed by:
1
10
1001
10010110
1001011001101001
...
that is, start with a sequence of just a single 1, then repeatedly append to
your sequence it's complement. This is why any power-of-two length periodic
continuation gets as many bits wrong as it gets right, and that should
provide the intuition that there are phase things going on in there that a
spectral method isn't going to catch.
It's a little harder to see that the sequence never repeats for any period,
but that is true, and this was used by mathematicians as early as Axel Thue
(circa 1900) and Marston Morse in his well known "infinite chess" paper
(from the 1920s if memory serves), which justified the 50-move rule in
chess. The bit we're going to do here is from Mahler and Wiener's 1927 paper
where the Fourier spectrum of this sequence was used to show that there are
spectral densities which are not made of lines and smooth parts, but of
stuff in between.
Since it's easy, we might as well show that the sequence not only doesn't
repeat, but it has a bad spectrum. For that let's remove the DC offset by
changing the 0's in the sequence to -1's.
Let the generating function of the sequence be f(z) = 1*z^0 - 1*z^1 - 1*z^2
+ 1*z^3 + ... Since the coefficients of this power series are bounded
integers, the series converges uniformly and absolutey in the unit circle
|z| < 1. We will get the Fourier spectrum by considering the limit as z
tends to the unit circle (non-tangentially, for those who want all the
bloody details).
Note now that the sequence has the interleaving property; take every other
element:
1 -1 -1 1 -1 1 1 -1 -1 1 1 -1 1 -1 -1 1 . . .
and you get the sequence back again
1 -1 -1 1 -1 1 1 -1 . . .
and every other element starting from the odd ones is the negation of the
sequence
-1 1 1 -1 1 -1 -1 1 . . .
If you expres this property in terms of the generating function, you get
f(z) = f(z^2) - z f(z^2)
or
f(z) = (1 - z) f(z^2)
Now there are some things we have to figure out about f(z) using this
property. First we need to get a hold of the behavior of f in the limit as z
tends to 1 from below. Suppose f tends to a limit f(1); then it satisfies
the equation
f(1) = (1 - 1) f(1^2) = 0 * f(1)
so the only limit that f(z) can have as z tends to 1 is 0. It might not have
a limit, though; but Abel's theorem tells us that the limit exists because
the average of the sequence exists (details available on request).
Now it's immediate that if f(1) = 0, then by the functional equation, we can
compute f(-1)
f(-1) = (1 - (-1)) f((-1)^2) = 2 f(1) = 0
and repeating this for i and -i, (the fourth roots of unity),
f(i) = (1 - i) f(i^2) = (1 - i) f(-1) = 0
f(-i) = (1 + i) f((-i)^2) = (1 + i) f(-1) = 0
and continuing, the eighth, sixteenth, etc., roots of unity are all points
at which f is zero.
This yields a dense set of points on the unit circle where the spectrum
vanishes, so the smooth part of the spectrum of the sequence is zero.
This leaves the question of whether there are lines in the spectrum of the
sequence; which corresponds to poles in f(z). Well, f(z) is analytic inside
|z|<1 and the dense set of zeros on the unit circle form a natural boundary.
Thus there are no poles on or inside the unit circle. There are no poles
outside the unit circle since the sequence has finite power (the amplitude
is bounded). So there are no lines in the spectrum, either.
This sequence has a spectral density which is called "singular".
Most people are not aware that this could happen, and most engineers believe
such signals are not of practical importance. (I suppose Manfred Schroeder
is an exception there.) But there is no easy way to show that signals that
one encounters in real life have no singular spectral component. It's like
the crazy uncle that nobody in the family talks about. Yet in one very well
understood sense - and probably the appropriate sense (Wiener Measure)
singular spectra are ubiquitous in signals - if you threw darts at the
continuous signal dartboard, you'd almost never hit a signal without a
singular spectral component.
Now what happens in practice is that the longer the piece of such a signal
you get, the smaller your estimate of the spectral density at any point.
Which means that the noise reduction by spectral techniques using only a
noise model will converge toward doing nothing to such a signal; more and
more accurately, as it turns out - the short segment noise reductions are
"fooled" by the fact that short segments of this type of sequence appear to
have lines (periodic components) which are not there.
If the truncation error is such a signal, naive (i.e. no signal model)
spectral techniques are essentially guaranteed to _add_ noise, but less and
less as they learn the signal, to the point where for a sufficiently long
signal segment, the added noise is as small as you like. But there is no
reduction of noise.
Thus, they _really, really are_ marketing bits.
The part about that there are better ways to clean signals up is only to
reinforce the point that even those better ways aren't going to always work.
> It's been interesting thrashing the issue. Slightly cautious
> here, since you obviously have a strong grasp of theory but...
> my original point still holds?
Not hardly. Those marketing bits are marketing bits. Good bits are good,
dicey bits can be unrecoverable in many sequences.
Later,
Andrew Mullhaupt