Should I dither before using external devices?

[Kempo]

Hi there,

I'm becoming crazy trying to understand how Protools LE works.

The Protools LE audio engine work with 32 bit floating point precision.
I never red articles about how it really work (article like "48 bit dithered mixer" for Protools HD), so I can only do some hypothesis.

Imagine a 24 bit session:

1. I have to record my guitar with my 003 rack. So I connect the guitar to the DI input on the 003r.

Guitar sound(analog) ---> DI(analog) ---> A/D (digital 24 bit fixed) ---> PCM wave file (digital 24 bit fixed) on the hard disk (wich I see on the protools channel)

2. Now I want to use some RTAS processors (plugins) plus an external processor as inserts. For example only: Waves Q1 (wich use single precision processing) then Waves Renaissance Compressor (wich use double processing precision) then I want to use a digital outboard device as INSERT, the Cranesong HEDD (wich use 24 bit fixed point) for some distortion. Then the output of the channel strip go to the mix bus and then to my monitors.

PCM wave file (24 bit fixed) ---UPSAMPLED to 32 float---> PTLE mixing engine (32 bit float) ----> Waves Q1 (single precision=32 bit float) --UPSAMPLED to 64 bit float--> Waves Ren Comp (double precision=64 bit float) ---Dithered then DOWNSAMPLED to 32 bit float --> ---DOWNSAMPLED to 24 bit fixed (TRUNCATED) ----> 003r SPDIF out ----> HEDD (24 bit fixed) ---> 003r SPDIF in --upsampled to 32 bit float---> Channel fader + Pan control (32 bit float) ----> mix bus summing (32 bit float) --DOWNSAMPLED to 24 bit fixed (TRUNCATED) ---> 24 bit fixed 003r D/A (analog) ----> monitors


So, if this is correct (and I have no idea if it is correct):

I should Dither (24 bit) before external hardware.

I should Dither (24 bit) as the last process in the mix bus.

And last, to improve audio quality I shouldn't use channel fader or pan from PTLE mixer (32 bit floating point) but I should use one high quality plugin, for example waves S1, wich calcualte at 64 bit floating point then dither to 32 bit.
Leaving all the Pan and Fader at original position (0dBfor channel, extreme left and right for pan) though pan don't work for mono tracks.
The same for master fader.

Any IDEA?

P.S.

I know the processing chain have nosense, but it's only for example purposes.

If you use an analog external device, the process is the same, but you use the 003r D/A A/D but DOWNSAMPLE and UPSAMPLE process is the same

[albee1952]

I've read this thru twice and it makes my head hurt. I think you are over-thinking this, and maybe over-complicating your life. You only need dither when you are taking a 24 bit audio file and converting it to 16 bit. As this is all done "in the box", it really has nothing to do with hardware inserts or D>A conversion(apples and oranges I would call it). When you convert to analog, its a straight conversion so no dither is required. If you can be more specific about exactly WHAT you are trying to achieve and what hardware you are trying to use, someone may have better insight. With LE systems in general, I tend to talk folks out of using hardware on inserts(other than reverb) as the latency for the round trip of audio(D>A and then A>D back in) causes latency that needs to be compensated for manually. If the external hardware is a piece of "high-class" gear, AND you don't mind dealing with the latency, then go for it. BUT (big BUT) if you are planning on inserting a DBX or Alesis compressor(don't laugh, lots of folks want to do this), then you are making things a lot more complicated where a decent plugin will match or beat the quality(without any of the headache).

BTW, external reverbs would generally not be used as an insert, but would usually be used via sends and returns that are separate(sends from an AUX and returns to a new track-AUX track for simple listening, or AUDIO track so you could print the reverb). The latency in this scenario can be considered as a little "undocumented" pre-delay(which is a parameter usually set by ear anyway). As long as the reverb comes back(into PT) at 100% wet, there will be no phasing issue from latency. Is this helping, or did I miss your point completely?

[Bob Olhsson]

Originally Posted by Kempo

I should Dither (24 bit) before external hardware.

I should Dither (24 bit) as the last process in the mix bus.

And last, to improve audio quality I shouldn't use channel fader or pan from PTLE mixer (32 bit floating point) but I should use one high quality plugin, for example waves S1, which calculate at 64 bit floating point then dither to 32 bit.
...

Dither is for bit reduction and yes, 32 float dithered to 24 is the proper way. (It's too bad nobody makes a simple 24 bit dither plug with no latency.)

I haven't heard any particular problems with channel faders and pan although to be safe you can assign left and right to the mono L and R output busses/channels without a pan pot. I also think I read something from a developer saying that PTle calculations are only limited by the CPU hardware which in the case of any recent Intel is 80 bit float.

[Kempo]

Hi Albee,

I'm trying to unserstand how the PTLE audio engine work, so to understand how to optimize the audio quality of my works.

For Protools HD there is a "white paper": 48 bit dithered mixer, wich explain how the PT HD audio engine work.
I have never seen an article describing how the PTLE engine work.

1. Does the PTLE mixer dither before downsample, or just truncate?

2. Does PTLE upsample to 32 bit floating point in front of each insert and then downsample after each processing unit (like PTHD)?

3. Does PTLE upsample at the first processing unit and downsample only before the final DAC?

[Kempo]

Originally Posted by Bob Olhsson

Dither is for bit reduction and yes, 32 float dithered to 24 is the proper way. (It's too bad nobody makes a simple 24 bit dither plug with no latency.)

I haven't heard any particular problems with channel faders and pan although to be safe you can assign left and right to the mono L and R output busses/channels without a pan pot. I also think I read something from a developer saying that PTle calculations are only limited by the CPU hardware which in the case of any recent Intel is 80 bit float.

Hi Bob,

thank you for the reply, so the anwers to my LAST questions are:

1. Does the PTLE mixer dither before downsample, or just truncate?
Just TRUNCATE.

2. Does PTLE upsample to 32 bit floating point in front of each insert and then downsample after each processing unit (like PTHD)?
NO, it doesn't

3. Does PTLE upsample at the first processing unit and downsample only before the final DAC?
YES, it does

Is it right?

[Bob Olhsson]

Until you come out of the box or print to a file, everything's floating point.

[Kempo]

Originally Posted by Bob Olhsson

Until you come out of the box or print to a file, everything's floating point.

That's Speaking!

Only a doubt...

I did a test:
Generated a 1000 hZ sine waveform and put on a track

Duplicated this track and inversed the pahse of the sine
Sent this last track through the SPDIF OUT and IN (just straight 75 ohm coaxial cable) as insert (no dither)

When mixed these track null each other perfectly, (lifted the gain up more than 150dB and analizer show nothing on the bottom)

This means that there is no truncation and assuming the SPDIF is a 24 bit interface this means betwen inserts the bus is 24 bit (like PTHD).

I tried also to insert a trim plugin with -10 db gain as the first insert on both channel (to force PTLE to UPSAMPLE the signal), nothing changed.

Any Idea about it?

[Kempo]

I can do some assumption:

Protols LE reference manual state that there is no need to dither unless you want to bounce to a 16 or 20 bit file.

This could be because just after the the MIXBUS summing (wich is performed at 32bit float)the audio signal is truncated to 24bit (or dithered but I don't think so).

So this could make me think that PTLE engine work in the samme manner as PTHD engine (but with 32 bit float vs 48 bit fixed, and I believe without dithering).

If so there is no need to dither before external inserts, because the data is upsampled in front of each plugin and downsampled to 24bit just after. So the data in the "inserts bus" travel at 24 bit, then upsampled to 32 bit float for summing (faders, pan, channel summing).
And there is no need to dither before the last DAC, because the signal is downsampled juast after summing, as it is for PTHD.

I don't know what to think....

Any suggestion?

What do you think BOB, can you give some links to some technical documents?

[JMS40]

Originally Posted by Kempo

I can do some assumption:

Protols LE reference manual state that there is no need to dither unless you want to bounce to a 16 or 20 bit file.

Actually, that's not precisely what the Ref Guide says...
Basically, the recommendation is to use a dither plug any time you are bouncing to disc... with these exceptions:

"When bus recording a submix to a track in the
same session, do not use a dither plug-in on
the destination track."

"When mixing down to a 24-bit destination,
do not use a dither plug-in on the main
output."

"When mixing down to an analog destination
with any 24-bit capable interface, do not use a
dither plug-in on the main output."

"During normal recording and playback, bypass
any dither plug-in on the main output."

One of the often overlooked recommendations is to use dither even when bouncing to 16 bit from a 16 bit session.

"A dither plug-in is necessary even when mixing
down to 16-bit from a 16-bit session. Even
though 16-bit sessions use 16-bit files, they are
still processed at higher internal bit depths."

[Kempo]

Hey there,

after re-reading the reference manual I found that I believe is the answer:

• Pro Tools|HD systems use 24-bit audio input
and output signal paths, and internal 48-bit
mixing and processing
• Pro Tools LE and M-Powered systems use 24-
bit audio input and output signal paths, and
internal 32-bit floating point processing for
mixing and audio processing

This in my opinion means that protools UPSAMPLE in front of a processing block (Plugin, Fader, Pan, SummingBus) then DOWNSAMPLE to 24 bit just after.

So if I use an external device as insert I don't need to dither, and I don't need to dither before the final DAC (in a 24 bit session).

But I need to use 16 bit dither in front of each EXTERNAL insert, and before the final DAC (in a 16 bit session)

What do you think about?