DAC Resolution Test and Don't EVER use Digital Volume Control

[mswlogo]

Just looking at your gig, nice TV!
In my "ear" mind, I would think it's better if you can switch your listening position to achieve a more "LEDE, Live End Dead End" setup, putting your stereo on the flat wall side and your chair on the open side. As it is, you will get lots of early echo from the back wall messing up with the main sound from the front...

Thanks for the comment. I believe I'd have more problems with bass swapping things around as you suggest. I should put something to break it up on the back wall though. It sounds pretty impressive as it is though. I've upgraded rears since that photo to larger floor standing Meridians which really made a huge difference I did not expect.

[Phil Leigh]

I was listening to some mp3's on our common server at work and really liked one I had listened to a lot. So I bought the CD and ripped it to FLAC. It sounded aweful. Then I asked my coworker to bring in his CD. The one I bought was remastered as HDCD. Then I found out dbpoweramp can process the hdcd. After that it sounded great. Unprocessed HDCD is listenable but in no way sounds good. It's amazing how just a small shift of where the resolution is used can make significant difference. It's effect does vary per album and per song.

If you believe half the crap in this thread that you can attenuate 3 or 5 bits with no harm. HDCD would not work. But it most certainly does. You can think of HDCD as optimizing the gain dynamically. Don't confuse that with loudness.

(ahem)
I've got over 150 HDCD's - I used to make a point of buying them when I got an HDCD-capable CD player. Now they are all ripped as 24-bit flacs via DBP.

I have yet to hear an HDCD that sounds bad without decoding - on the other hand they all sound better (to differing degrees depending on which features of HDCD have actually been used) when decoded.

I can't believe anyone would say that Sailing To Philadelphia by Mark Knopfler sounds bad in non-HDCD replay...


There are various components of HDCD, one of which is the peak extend function that is a compander, making the peaks louder on playback than their 16-bit value would otherwise imply they should be. However, when this happens everything gets louder, it's exactly like momentarily lifting the volume control.

On normal playback, the peaks are not made louder.


Not every HDCD uses the peak extend or the filter switching.


HDCD does not optimise the gain - it simply boosts the level for loud bits making them louder than they otherwise would be, which gives an apparent increase in dynamic range.


ripped HDCD playback (which is 24-bit) is completely the same as normal 24-bit playback and thus the 5-bit limit holds.

Just to be clear on what you perceive to be "crap"... all I am saying is that in practice at best you can perceive 19 bits of DR and (for 24-bit) the only time you get to hear all 19 is when the music is near full scale - and the bottom 5 are inaudible at that point. As the music gets softer - ASSUMING YOU DO NOT TOUCH ANY VOLUME CONTROL - you'll be hearing the "middle bits" only, because there's now nothing in the MSB's and the LSB's are STILL too quiet to hear anyway.


When mixing, a 30-40dB cut in level of a track is effectively the same as muting it...

Try it for yourself in Audacity. Generate a full scale 90 second sine wave then apply a fade out effect to the whole of it
Then on a separate track generate a click track of say 128 measures at 80 BPM
Then try reducing the level of the click track - 10dB, -20dB and so on.

To what level does the click track have to be reduced to before you never hear it, even when the sine wave is nearly completely faded...?

(DO NOT ADJUST YOUR REPLAY VOLUME)

This is what real-world LSB audibility is about.

[Robin Bowes]

On 01/05/10 05:16, mswlogo wrote:
>
> I work with analog and digital signals at data rates that would make
> your head spin.
>
> Replies like yours are priceless.
>
> I'm sorry that the clipping I'm referring to is beyond your grasp.

Ah, here we go, the "my dick is bigger than your dick" reply.

Rising above that, I can see where carefully applied compression might
be beneficial to reduce the dynamic range of a track to enable the
overall level to be increased, albeit at the expense of SNR as any
increase in signal level would also increase noise level.

See, when I use the term "clipping", I mean this [1] or this [2],
depending on context. Is that what you mean by "clipping"? Or did you
mean "compression" ?

R.

[1] http://en.wikipedia.org/wiki/Clippin..._processing%29
[2] http://en.wikipedia.org/wiki/Clipping_%28audio%29

[Phil Leigh]

Final post on this thread for me:

take a 16/44.1 file into Audacity
make a copy, reduced by 6dB using Audacity's 24-bit level function (this mirrors an SB acting on redbook audio)

diff the two

result: <6dB of white noise - exactly what theory predicts due to reduction of SNR

keep going, lowering the level further each time...

12dB reduction = <12dB of noise
18dB reduction = <18dB of noise
...


So, no "loss of bits", "loss of resolution","loss of music" or any other effect - just the predicted decrease in the SNR.

So, 24-bit digital level control is transparent except for SNR.

This really is just as well because of course this is exactly how every mixing suite, mastering suite, DAW etc does level control (be it @ 24/32/48/56/64 or 72-bit!).

[mswlogo]

Here is an example of a well Mixed Recording. It is one of my standards I always use when evaluating systems.

http://softronix.com/pictures/levelp...u_Got_Full.jpg

But if you look close it's CLIPPED. Will you hear this tiny bit of clipping, never (it's too small and too short in TIME). The more I look at good recordings the more I see of this. It's a sign that the mixer pushed the dynamic range to the limit. Which you should do when dealing with 16bits.

http://softronix.com/pictures/levelp...at_You_Got.jpg

Here is an example of a poor mix. Way too much headroom over 6dB.

http://softronix.com/pictures/levelp...Lonely_Sea.jpg

[mswlogo]

Final post on this thread for me:

take a 16/44.1 file into Audacity
make a copy, reduced by 6dB using Audacity's 24-bit level function (this mirrors an SB acting on redbook audio)

diff the two

result: <6dB of white noise - exactly what theory predicts due to reduction of SNR

keep going, lowering the level further each time...

12dB reduction = <12dB of noise
18dB reduction = <18dB of noise
...


So, no "loss of bits", "loss of resolution","loss of music" or any other effect - just the predicted decrease in the SNR.

So, 24-bit digital level control is transparent except for SNR.

This really is just as well because of course this is exactly how every mixing suite, mastering suite, DAW etc does level control (be it @ 24/32/48/56/64 or 72-bit!).

What you are showing is why 24bit is good to EDIT in. Once you get into finite performance of your DAC and system then things are different.

If 24bit DACs were perfect then you are correct. The problem is they are far from perfect, they are not linear and don't cover 24bits. That's why I did the test. Far from perfect. If anything it shows a 24bit DAC is not 24bit no matter how much amplification you throw at it.

[darrenyeats]

Here is an example of a well Mixed Recording. It is one of my standards I always use when evaluating systems.

http://softronix.com/pictures/levelp...u_Got_Full.jpg

But if you look close it's CLIPPED. Will you hear this tiny bit of clipping, never (it's too small and too short in TIME). The more I look at good recordings the more I see of this. It's a sign that the mixer pushed the dynamic range to the limit. Which you should do when dealing with 16bits.

http://softronix.com/pictures/levelp...at_You_Got.jpg

Here is an example of a poor mix. Way too much headroom over 6dB.

http://softronix.com/pictures/levelp...Lonely_Sea.jpg

We have very different ideas about what looks like a good recording in Audacity. A good recording has very very few peaks at 0db, sometimes just one peak. That is because in real life music doesn't have a zillion peaks at exactly the same level. Neither does the recording, if you want it to sound real. This kind of dynamic compression is exactly what sound engineers should be criticised for.

Your recording might sound wonderful but it is not reference recording with respect to dynamic range. It's artificially restricted in respect of dynamics.

Nobody's reference recordings are going to satisfy in every respect, I understand that. But in a discussion specifically about dynamic range, as this is, to display that as a reference recording has made it more difficult to take your assertions seriously.
Darren

[darrenyeats]

Here is an example of a poor mix. Way too much headroom over 6dB.

http://softronix.com/pictures/levelp...Lonely_Sea.jpg

IMO a far far better looking waveform than your reference recording. The only problem is that it peaks too low. If it peaked at 0db it would better, of course I agree with that. But it looks to be a dynamic waveform.
Darren

[Robin Bowes]

On 02/05/10 16:35, mswlogo wrote:
>
> Here is an example of a well Mixed Recording. It is one of my standards
> I always use when evaluating systems.
>
> http://softronix.com/pictures/levelp...u_Got_Full.jpg

Hmm, we obviously have different ideas of "well-mixed"; it's certainly
clear, but the drums are a little too prominent for my liking, and it
sounds a little sterile overall. But that's a subjective thing.

Here are a few tracks I particularly like the sound of:

- "They're Red Hot" from "Me & Mr. Johnson" by Eric Clapton
- "Hurt" from "American IV" by Johnny Cash
- "Birdland" from "Heavy Weather" by Weather Report
- "Freddie Freeloader" from "Kind Of Blue" by Miles Davis

Interestingly, none of them are clipped.

> But if you look close it's CLIPPED. Will you hear this tiny bit of
> clipping never. The more I look at good recordings the more I see of
> this. It's a sign that the mixer pushed the dynamic range to the limit.
> Which you should do when dealing with 16bits.
>
> http://softronix.com/pictures/levelp...at_You_Got.jpg

IMHO, there is *NO* excuse to push levels into clipping. None. Nada.
It's just bad. By all means maximise the levels, but there really is no
need to push them into clipping.

>
> Here is an example of a poor mix. Way too much headroom over 6dB.
>
> http://softronix.com/pictures/levelp...Lonely_Sea.jpg

No, that's an example of a gentle track in the context of a whole album.
Sometimes musicians will want to play softly. Some things are more
important than SNR, you know?

R.

[Robin Bowes]

On 02/05/10 17:47, darrenyeats wrote:
>
> mswlogo;542627 Wrote:
>>
>> Here is an example of a poor mix. Way too much headroom over 6dB.
>>
>> http://softronix.com/pictures/levelp...Lonely_Sea.jpg
> A far far better looking waveform than your reference recording. The
> only problem is that it peaks too low. If it peaked at 0db that it
> would better, of course I agree with that. But it looks to be a dynamic
> waveform.

I agree with Darren; the Kenny Neal recording is typical of the things
being churned out by modern studios these days - compressed and sterile.
It is not a good recording, IMHO. Perhaps the guy is good in a live
environment, but those recordings don't do him any favours.

R.