Digital Drive Upsamplers, DACs, jitter, shakes and analogue withdrawals, this is it.

---

[ Follow Ups ] Thread: [ Display  All  Email ] [ Digital Drive ]

---

[ Alert Moderator ] [ Edit ] [ Delete ]

---

Fluff

Posted by JimF on January 1, 2003 at 07:47:31

In Reply to: The Perpetual Website is Gone... posted by Todd Krieger on December 31, 2002 at 19:52:51:

Todd Krieger wrote:

> [T]he infamous "Level III" white paper claiming
> Perpetual's innovation of "interpolation" can be
> found here (link). The white paper erroneously implies
> that "interpolation" is both an alternative to
> oversampling/upsampling and an improvement over
> oversampling/upsampling. But in reality,
> oversampling/upsampling and interpolation are inseparable,
> and have been the standard method of digital filtering for
> 15 years.

Sigh. Yes, this is an embarrassing bit of fluff. I
can guess who wrote it (I doubt if it was Peter Madnick
and/or Keith Allsop).

At the risk of annoying the folks at _Stereophile_, I'm
posting a graph and an excerpt from the review of the
original DTI Pro in November, 1994 (Vol. 17, No. 11).

Robert Harley wrote (p. 112):

"The DTI Pro's 'resolution enhancement' algorithm is run
on a Star Semiconductor quad DSP chip driven at 40 MHz.
A socketed Erasable Programmable Read-Only Memory (EPROM)
chip contains the software that tells the DSP chip what
to do. Specifically, the algorithm looks at the 16-bit
data over a sequence of samples and interpolates (estimates)
what data would have been present had the data been
encoded with greater than 16-bit quantization. . .

I was curious about how resolution enhancement differs
from standard digital filtering found in off-the-shelf
oversampling digital filter chips. Those devices take
in 16-bit words and output 20-bit words. The extra bits
are created from the intermediate calculations performed
in the filter. The digital filter may internally process
(in the filter's accumulator) the audio samples with very
long words, then round off those long words to 20 bits
for output to the DACs. Note that no new information is
created by this process. Instead, the technique [of
sending more than 16 bits to the DAC] merely reduces
the requantization error added by the filter.

Audio Alchemy claims that their resolution enhancement
is very different. The additional bits created by
resolution enhancement **do** contain new information;
the bits are calculated by looking at the sample values
over a time window, and adding the bits the algorithm
guesses would have been there had the signal been of
higher resolution. . .

Audio Alchemy provided _Stereophile_ with a three-dimensional
plot that AA claims shows the measured resolution improvement
afforded by their resolution enhancement technology (Fig. 1).
Note that this graph is not a theoretical model, but the
result of an actual measurement made on a DTI Pro with a
Sun workstation computer. The plot appears to show about
a 2-bit overall improvement in resolution up to 7 kHz for
high-level signals, and increased resolution up to 11 kHz
with low-level signals. The sharp spikes of very high
resolution occur at frequencies at which the algorithm
happens to be most efficient. I'd be interested in hearing
more about how this measurement was made (perhaps in Audio
Alchemy's 'Manufacturer's Comment'?)."

Audio Alchemy replied in a letter signed by Peter Madnick
and Keith Allsop (pp. 241 - 243):

"Bob asked us to explain how the Resolution Enhancement vs.
Frequency vs. Amplitude plot was obtained. It was created
from actual digital data collected from the DSP in a
production DTI Pro. A 16-bit-accurate (undithered) digital
sinewave source was used, and the frequency and amplitude
varied in discrete steps while collecting the digital
output data from the DTI Pro. The frequency was varied
from 172 Hz (44,100 Hz / 256) to 11 kHz in steps of 172 Hz.
At each frequency the amplitude of the sinewave was varied
from -6 dB to -84 dB in steps of 6 dB. Each data set at
this stage is in the time domain. Next an FFT is performed
to transform the data into the frequency domain. Actually,
two FFTs are performed, one for the input data and the
other for the output data to/from the DTI Pro. The noise
energy in the FFTs is summed (sum of all components
except for the fundamental), and the difference between
the output and input is the improvement due to the
resolution-enhancement software in the DTI Pro. For the
plot, the noise energy is converted into bits (each bit
is an improvement of 6 dB). This whole process has been
automated so that a plot can be created automatically,
enabling rapid evaluation of DTI Pro software algorithms.
'Rapid' is a relative term in this context, as it still
takes approximately six hours to collect and process
the data to create a single 3D plot!

As DTI Pro's software matures, the resolution-enhancement
peaks (as much as 12 additional bits now!) will be reduced,
while the overall improvement will increase and extend
further into the higher frequencies and amplitudes. . .

Regarding how DTI Pro's resolution enhancement differs
from that of an upsampling [read: oversampling -- this is
pre-DVD-Audio 1994] digital filter, these devices, after
interpolation (output filtering), **can** have more bits
at the end than at the beginning (for example, 16 bits
in, 20 bits out). However, this process **does not
increase** the resolution of the signal. . . In fact,
the resolution-enhancement algorithm does not use the
upsampling/interpolate principle."

And that, together with the Asylum post on the Perpetual
P-1A from September, 2001 signed "misterdsp" (who I guess
is Keith Allsop himself):
http://www.AudioAsylum.com/audio/digital/messages/28585.html
is likely to be all anybody is going to find out from
the manufacturers about the "secret" of the Audio Alchemy/
Perpetual Technologies "resolution enhancement" algorithm.
They consider the details proprietary information.

That's hardly surprising -- you wouldn't have expected
Yoshiaki Sugano, when he was alive (or his sons, now)
to talk about exactly what they do to make Koetsu cartridges
sound like Koetsu cartridges. The only thing that you,
as a consumer, have to know is whether or not you like
the sound.

The bottom line on the Perpetual P-1A is whether or not
you like what it does to the sound of CD playback.
I think it sounds great. But if someone is philosophically
uncomfortable with the notion of "sweetening" the sound
of digital playback with mysterious and proprietary DSP processing,
well, that's his prerogative.

Jim F.

• http://www.goldenagemusic.se/swe/proaudio/perpetualtech/P1Alev3.html   (Open in New Window)

This post is made possible by the generous support of people like you and our sponsors:

VH Audio

---

[ Top of Page ] [ Contact Us ] [ Support/Wish List ] [ Copyright © 2026, Audio Asylum, all rights reserved ]

---

[ General ] [ Speakers ] [ Tubes ] [ Vinyl ] [ Digital ] [ Hi-Rez ] [ Video Asylum ] [ Cables ] [ Tweaks/DIY ] [ Music ] [ Films ]

---