Digital Explained – “Too Many Notes”

There is a contingent of digital audio convertor manufacturers who do not use – or even like – oversampling in their products.  These designs are frequently combined with the intentional omission of a vital part of the digital conversion process: The reconstruction or “anti-image” filter.

Without this filter, the sampling process is not complete, and the side effects range from actual stair-steps remaining in the audio, to massive amounts of interference being output that causes unpredictable results in preamps and power amplifiers.  

While some may think that digital audio contains steps, it’s not really the case at all!  Or that digital somehow needs to be “re-constructed,” as if broken, this filter is part of the system.

Yet here we can see actual steps in a USD$12,000 non-oversampled CD player that also uses a R/2R type of conversion, but no output filter.

And, as before, all sorts of new notes are created by the conversion:

Of particular concern are all the extra notes around 5kHz as this is a highly audible range. They are not on the sheet music; the CD player is making them up!

As always, a person’s preference for NOS, R/2R, or whatever it may be is inviolate, but I think it’s easy to see 

the difference in accuracy of conversion in these examples.

Steven asked me to address two questions for this next article:

“Has digital processing gotten to the point that PCM has come to a technological conclusion?”

Emphatically no! 

I think it’s clear that digital conversion and processing has made tremendous strides, especially as we covered in the last article with the introduction of oversampling A/D converters in the early 1990’s.  After all, analog has had a hundred years to evolve and certainly digital will do the same.  In doing a bit (Har!) of research for the previous article, I came upon this review, from 1982, of a consumer digital recorder called the Sony F1.  When I was working in audio in the early days of Compact Disc, this was a popular format to record master mixes, and was used on many records that still sound good today.  It did not use oversampling, and had the complex and error-prone circuitry we talked about in the first article.

Nevertheless, the reviewer George Chkiantz (who has recorded more classic records than I’ve had hot meals) had this to say about the quality of the conversion in Studio Sound: “The extraordinary quality of the PCM system, (in one test it was found that the feedthrough signal was closer to the original than the line-in signal on a certain well-respected tape recorder!) makes it extremely difficult to find anything to say about the positive aspects of the quality of the machine. A very brief listen convinces everyone who has heard it that it ‘sounds like it isn’t there’ which, of course is as it should be. Extended listening only confirms this impression, and should make the dedicated listner grateful that such a large area of compromised performance can now be forgotten about or would you really want to be listening to a mechanica; gramaphone, however refined? In essence one immediately looks on PCM-F1 as if it were some in-line bit of processing circuitry, like a line or mix amp, and judges it accordingly, and even by these criteria it is winning all the way…”

So, what is going on here?  Is this the Perfect Sound Forever crew again, claiming that all technological hurdles were solved in 1982?  After all, true transparency is a remarkable achievement.  Or was it that digital had solved many problems (noise, pitch accuracy, peak overload) while creating others that were not immediately obvious?

Certainly an experienced engineer would have heard other aspects, like the unspeakably bad filter ringing etc.? This reviewer even took time to record music at very low levels (-45dBFS) on the F1 and then amplify the output to see if the low-level distortion was audible, as it had been in other systems.  I don’t know the answer, but can definitely say that some CD’s recorded with the F1 system still sound good today, in spite of the imperfections of the 1980’s technology. 

“Or can further up-sampling or bit density improve fidelity?”

 Much like when they asked Bob Dole “boxers or briefs?” the answer is “Depends.”  

 Up-sampling works like this:

 Here are four original samples, say at 44.1 kHz.

 Sample #1| Sample #2| Sample #3| Sample #4|…

 Now let’s up-sample them to 88.2 kHz:

Sample #1| Zero| Sample #2| Zero| Sample #3| Zero| Sample #4| Zero|…

No fair! All you did was put zeros in-between the original samples.

Zero is actually a great number to put in there, as it is guaranteed not to affect the values of the original four samples.  Any number of zeros may be added to achieve a new sampling rate.  16 times?  No problem, zeros are cheap.

While no company is putting the phrase: “Now, with more zero-padding!” in their ad brochures, it’s a common and legitimate way to up-sample audio.

Since we have this new sampling rate we must, as always, apply a Nyquist filter to it, and this is where it gets tricky.  Adding zeros won’t affect the sound, but this filter certainly can, so this part is key to any sampling frequency conversion.

The important point is that no new information is ever added in audio up-sampling. You might like the sound better, or not, but you are never adding anything new to the music…