Is Equalization in the Digital Domain Deleterious?

[wots]

Are the deleterious effects associated with digital control of level also present with digital equalization? Equalization is level control at various frequencies.

 

Thanks,

 

[ramagochi]

Like the volume control the equalization isn't a problem for bit deeps over of 16 bits, for today 24 bit systems isn't a problem, but for a very agressive cuts best with dither (Such as the volume control)

 

[bdiament]

Hi wots,

 

"Are the deleterious effects associated with digital control of level also present with digital equalization? Equalization is level control at various frequencies."

 

As ramagochi suggests, this will depend on the word length (i.e. "bit depth") of the source material.

 

A digital level control will "throw away" one bit of resolution for every 6 dB (actually ~6.02 dB) of attenuation below wide open, full level. That won't be as much an issue with 24-bit source material as it will with 16-bit material.

 

For example, with typical modern recordings, I find I need about 10-14 dB of attenuation just to get them down to "loud". That will turn a 16-bit source into a 14-bit source. (How much this ultimately matters with modern, dynamically squashed recordings is open to question but nonetheless, two of the bits are gone forever.)

If you want/need to turn them down to quieter levels, you could end up at 12-bits or less. (Things will start to sound quite coarse and rough-edged.)

 

With equalization, there are similarities and there are differences. Any change to a digital file, whether a .001 dB level change or the application of a process like EQ, will effectively lengthen the digital "word".

 

As a somewhat oversimplified but still appropriate example, imagine we have the value 5. It takes one place, or bit, to represent the this number. If we were to apply a process that cut this in half, we'd end up with 2.5, a number that will can no longer fit into the one place (or bit) used by the 5. We'd added "low level information". Since the new information won't fit into the "bucket" we have, the low level information will not be represented. It will be lost and the best we can do is store a 2 or a 3 but not the precise 2.5 value.

 

This is what happens when a process (any process, level, EQ, etc.) is applied to a digital file. The loss of the low order digits (i.e. low level information) is more noticeable with a 16-bit source (where low level information is under-represented to begin with). With the loss of that low level information goes the end of reverb tails, harmonic complexity and nuance, soundstaging information and overall "focus".

 

Processing a 24-bit source will result in the same lengthening of the digital word but this will be less noticeable than with 16-bit sources because the low order bits in a 24-bit file are already quite low in level.

 

That said, "less noticeable" loss is not the same as no loss. The ideal is to process at a word length that is longer than the target word length, then to use a good dither algorithm to decimate to the lower word length. In other words, if I want to end up with a 16-bit file, I'll do a Save As on the file in a good digital audio workstation and create a copy at a longer word length, say 24-bits. I'll apply the process (e.g. EQ) to the 24-bit file and then apply dither (as the very last step) to create a new 16-bit version.

 

The better processing algorithms and digital audio workstations will work internally at appreciably longer word lengths. Many consumer applications work internally at 32-bits. The better pro gear will be anywhere from 48-bits to 80-bits.

 

One thing to keep in mind though, is that many of these programs, especially the consumer ones, save their temporary files at the word length of the original source. That means, even if your program works at 32-bits internally, it is a good idea to first do a Save As to create a copy of the original - in this case, a 32-bit (float) copy. Then process the copy and lastly, apply dither to create a version with a shorter word length.

 

Hope this helps.

 

Best regards,

Barry

www.soundkeeperrecordings.com

www.barrydiamentaudio.com

 

[wots]

Thank you for your knowledge. I will refrain from digital processing of the 16-bit recordings I play back. So the trade-off for DRC (Digital Room Control) and other signal processing is having your collection of 16-bit Red Book recordings turned into 14-bit. I wonder if many are aware of this? Thanks again!

 

[bdiament]

Hi wots,

 

"So the trade-off for DRC (Digital Room Control) and other signal processing is having your collection of 16-bit Red Book recordings turned into 14-bit. I wonder if many are aware of this?"

 

I wouldn't say this is necessarily the case. DRC has other issues but from my perspective, this isn't one of them. A well designed algorithm will operate internally at a longer word length and dither the results back to the target. (If it doesn't, in my view it isn't a well designed algorithm.)

 

Whether one believes room problems (existing primarily in the time domain) can be successfully addressed by changing (i.e. distorting) the input to the loudspeakers (in the amplitude domain no less) is a separate issue. To me, this is like attempting to fix a broken arm by wearing a different hat.

 

Best regards,

Barry

www.soundkeeperrecordings.com

www.barrydiamentaudio.com

 

 

 

[Burak]

If I'm not hijacking the thread, is it possible to use digital volume control (and EQ) safely just

 

1. by using this (?) or that (?) software for ripping,

 

2. by using this (?) or that (?) software for "Saving As...",

 

3. by using this (?) or that (?) software for playing, and then

 

4. by using digital volume control on the keyboard or the one on the DAC?

 

(I have a feeling that I'm about to post a stupid post!)

 

[RayW]

Barry has explained well the word lengthening that occurs with digital signal processing. I just want to add one comment: As computer audiophiles, we do not need to convert the result back to 16 bits!

 

I've been doing some mild parametric EQ using Adobe Audition. I import audio from CD and temporarily save as 32-bit floating point. All processing is done in floating point. My final step is to convert the floating point to 24-bit integer (using dither). The 24-bit files go into my music library on disk and are played through a 24-bit DAC (Weiss DAC-2).

 

I believe the 16-bit CD remains the weakest link, that is, what quantization noise I've added is negligible compared to the quantization noise added in producing the CD.

 

Ray

 

[bdiament]

Hi Burak,

 

"...is it possible to use digital volume control (and EQ) safely just

1. by using this (?) or that (?) software for ripping,

2. by using this (?) or that (?) software for "Saving As...",

3. by using this (?) or that (?) software for playing, and then

4. by using digital volume control on the keyboard or the one on the DAC?...

 

First, I believe we must separate digital volume control from digital level adjustment.

 

In my experience, digital volume controls by their nature, throw away resolution when the volume is less than full up. It doesn't matter whether you use a keyboard or a knob. (This should not be confused with a digitally controlled analog volume control, as can be found on some current units.)

 

Digital level adjustment is something else and will fall under the same umbrella as digital EQ adjustment or any other digital process. By digital level adjustment I mean changing the inherent level of a file, regardless of the volume at which you choose to play it back.

 

Say, for example, you have several song files and one is in your view, too loud. Perhaps it was heavily compressed and now requires you to turn down your playback volume compared to the other songs. If you chose to, you could adjust the level of the file without suffering sonic degradation. (Note again, this is something completely different from the volume at which you choose to play the file.)

 

If you did want to adjust the level of the file, my suggestion would be to do a Save As to a longer word length, determined by the software you would preform the process in. Most consumer software will do its best with a 32-bit float input file. Some pro software just requires a 24-bit input file.

 

Keep in mind that audio applications all sound different and performing the same exact process in different applications will result in different sounding files. The differences are often subtle, certainly not "night and day" but they are audible and only the listener can decide if and when they are important.

 

Regarding your item #1, the above has nothing to do with ripping (though on that subject, I would rip only to raw, uncompressed PCM format like .aif or .wav).

 

Hope this helps.

 

Best regards,

Barry

www.soundkeeperrecordings.com

www.barrydiamentaudio.com

 

[Burak]

Thank you, Barry. It's been clarified.

Best...