Print

Print


"and if the codes are made for first-order filters, they will be
as much minimum-phase as analogue first-order networks--minimum-phase being
the requirement for phase compensation (this was Ted Kendall's bid for the
problems)."



Quite so - the operative word being "if" - they all too often aren't. There 
are some ghastly algorithms hiding behind pretty interfaces...

----- Original Message ----- 
From: "George Brock-Nannestad" <[log in to unmask]>
To: <[log in to unmask]>
Sent: Thursday, March 11, 2010 12:38 AM
Subject: Re: [ARSCLIST] Disc EQ in the digital domain


From: Patent Tactics, George Brock-Nannestad

Hello,

I am a bit puzzled over this debate. I could just pretend that I understood
and say nothing (after all, Hans Christian Andersen's Emperor's New Clothes
was from Denmark), but I think that at least I merit a better explanation.

I can accept the following in the "flat" case, meaning that we have a
velocity reproducer of the groove. As the amplitude of the groove does not 
go
above the maximum excursion at, say 1000 Hz, the output level from a 20 Hz
signal is about 20 dB down, or about 1/10 of the level, although, nom�nally,
the input to the recording system was at the same level when that frequency
was recorded. So, we have to amplify, boost, it, either in the pre-amp or
digitally, by shifting the digital words representing this low frequency "to
the right". The least significant bit now represents a higher step, so our
resolution in the digital representation has suffered. The rules that govern
this shifting are the codes for the digital filter that actually performs 
the
bass boost, and if the codes are made for first-order filters, they will be
as much minimum-phase as analogue first-order networks--minimum-phase being
the requirement for phase compensation (this was Ted Kendall's bid for the
problems).

I do not really see in which way this loss of resolution should have any
practical importance if you use 24 bits, unless it is because the 24 bits 
are
not accurate as to resolution when you go below 14-16 bit at the sample 
rates
used. In other words, if the steps that are not at all equidistant when you
get into the less significant bits suddenly mean more, then you have non-
equidistant steps at levels where this becomes audible distortion. So my
question is: is this the most plausible explanation for digital EQ sounding
worse?

The noises from pre-vinyl records have a very broad spectrum, and if you
boost the bass before using programs to remove some of the noise, these
programs will behave differently. Similarly if you reduce the treble as in
the vinyl RIAA case. I am quite convinced that those professional and semi-
professional programs that are available for noise reduction have taken this
into account, but I think that a skewed spectral distribution is just one
more problem that such a program would have to deal with. For this reason it
would appear that going flat to your A-D, going into noise reduction (not
here discussing the various characters of noises) and then perform the EQ
digitally, while you are at it, would give the best effect. However, as more
and more noise reduction is using physiological properties of the ear for
noise masking, possibly this approach will be thwarted, as the ear does not
like to listen to a non-boosted bass from a pre-vinyl record, and so the
wrong masking procedure would be applied if you had not already corrected 
the
response in the pre-amp. It is probably a compromise situation.

Kind regards,


George



> On Wednesday, March 10, 2010 8:44 AM, Jim Sam wrote:
>
> > On Tue, Mar 9, 2010 at 6:42 PM, Eric Jacobs wrote:
> >
> >> Transfering FLAT will cost you about 6-7 bits of dynamic
> >> range - an audible loss.
> >>
> >
> > 36-42 dB*?  ~10-12 is more like it, which is two bits max.
> > That leaves 22 bits, which is more than adequate.
>
> Hi Jim,
>
> Just a thumbnail calculation on word length loss for two common EQs:
>
> RIAA boost = 19.27 dB @ 20 Hz
> RIAA cut = 19.62 dB @ 20,000 Hz
> RIAA dynamic range lost = 38.89 dB
> RIAA approximate word length loss with FLAT EQ transfer = 6.5 bits
>
> NAB boost = 19.85 dB @ 20 Hz
> NAB cut = 21.52 dB @ 20,000 Hz
> NAB dynamic range lost = 41.37 dB
> NAB approximate word length loss with FLAT EQ transfer = 6.8 bits
>
> To the extent that fractional bits exist.  <smile>
>
> Keep in mind that this is word length loss due to a FLAT transfer,
> where the bass boost is lost, and extra headroom is needed to
> accomodate the lack of treble cut.
>
> Eric Jacobs
>
> The Audio Archive, Inc.
> tel: 408.221.2128
> fax: 408.549.9867
> mailto:[log in to unmask]
> http://www.TheAudioArchive.com
> Disc and Tape Audio Transfer Services and Preservation Consulting