Author Topic: Looks like an interesting read  (Read 588 times)

Offline jimbones

  • Audio Neurotic
  • *****
  • Posts: 1170
  • Two plus Two Speakers
Modwright LS100 Pre, Art Audio Vinyl Reference, Wells Audio Inamorata Pwr, VPI Classic II/Dynavector 20X2L, Two plus Two Speakers,  Roon Rock, Auralic Vega DAC, Emotiva ERC-3, MIT, WireWorld, Wywires, Shunyata, TWL

Offline steve

  • Audiologist
  • *
  • Posts: 804
    • Reference components
Re: Looks like an interesting read
« Reply #1 on: September 21, 2019, 09:40:06 PM »

Just a few general comments concerning the article.

1. It takes two samples to reproduce a portion of a signal. One sample won't produce anything.
With that in mind, Fig. 1 (bottom), if a signal starts at A', just immediately after sample A, the red portion of the initial signal will not be reproduced. The signal will start point B since pt B and pt C are the first two samples. So the musical waveform timing and slope will be altered.

This occurs with the start of every note in a musical selection, from every instrument. (This does not occur with all analog LPs.) The lower the frequency, the less information is lost.

The more samples taken, the less initial musical signal (in red) is missing. However, the waveform is still
distorted vs a pure analog reproduction.

2. In figure 2, we see a signal wave form being sampled. Notice during sampling, the wave form does not cross
at a particular level (short lines on the Y axis), but in between the lines. So which level/line does the dac "use"? Whichever is closer to the signal. This means the wave form slope is altered. Anytime we alter the slope, we
alter the frequency of the signal.

The more levels/lines/bits, the less alteration of the musical signal.

3. Another major problem with both analog to digital and digital to analog converters is that there are analog stages present. I think we all know each analog designer's/manufacturer's components "sound" different. The same holds true with all those analog to digital and digital to analog converters because there are analog stages in both.

Below is a quote from an expert I found, general public information.

  .... discussion of the stepless output is based on pure sampling theory, where each sample is a zero-width Dirac delta function (his 'lollipop diagram').

That's correct in theory, but in the real world, a DAC does implement the zero-order hold he mentions, and the output of the DAC chip DOES have steps; it does not produce zero-width delta function outputs. A DAC chip is therefore followed by a Nyquist reconstruction filter, to remove the steps.

A real-world ADC will also generate a sampling (or quantisation) error, as, unless you have infinite bits, there will be an error between the real sample, and the nearest ADC quantising value. His lollipops are assumed to be perfect samples, with zero quantisation error.

Sampling theory assumes perfect, 'brick-wall' Nyquist filters. In the real world, these do not exist. Real filters have problems like roll-off rates, and ripple in passband amplitude & phase.

I was once asked to look at the design of a digital radio transceiver. They were having trouble with harmonics in the transmitted RF, violating the spectrum mask. It turned out to be the the use of a 4-bit, 12 sample per cycle sine wave generator used in the digital IF mixer (implemented in an FPGA). The quantisation error caused by the 4 bit samples was causing the harmonic distortion. I increased the size of the samples, which allowed us to reduce the harmonics to an acceptable level. The analysis of the problem needed nothing more than a look at the VHDL code to understand the sample scheme, and creating an excel spreadsheet to create a set of repeated cycles, on which I got excel to compute an FFT, which showed the harmonics.

It would have been illuminating for him to have changed the precision of his samples, from the 16 bits he used, to 8 bits, or even 4 bits. Using 16 bits, the quantisation error will be below the noise floor of the analyser he was using. It's not a good idea to try to claim an effect doesn't exist because you can't measure it.

Much of the effort expended in the design of DACs has been to try to eliminate that zero-order hold step problem; to smooth the transitions in some way. A first order interpolator would draw a straight line between the samples, and higher order interpolators would try to draw a smooth curve. The most common approach is a digital oversampling filter, which generates additional samples to fill in the gaps, with a higher precision DAC; e.g. a 4x oversampling DAC, generating 176.4kSa/s, 18-bit samples. Then there are the 'noise-shaping' DACs, such as Bitstream and MASH, that use an entirely different approach to reconstruction, with a stream of very high rate pulses; they push the noise way above the Nyquist frequency, allowing simpler, more linear filters to be used.

I started my career working on the development of the GSM standard, and the first network and handsets. In particular, the frequency synthesis and modulation. We used a technique called Digiphase, a type of fractional-N synthesiser. It did direct digital modulation by constantly changing the synthesiser frequency. It used a third-order interpolator, combined with digital predistortion to meet the modulation and spectral mask requirements. Essentially, a noise-shaping DAC.


« Last Edit: September 21, 2019, 10:23:36 PM by steve »
Steve Sammet (retired, owner, SAS Audio Labs)
"V" Very Low Capacitance ICs, 40pf 1 meter
SAS Audio Labs Test Phono Stage
SAS Audio 11A Tube Reference Preamplifier
SAS Audio Labs 25 W Triode Reference Monoblocks
2 way test Spkrs, 28 - 20khz  -3db (28hz)

Offline HAL

  • Obsessively Audiophilic
  • ****
  • Posts: 835
  • Audio Mad Scientist
    • Hollis Audio Labs
Re: Looks like an interesting read
« Reply #2 on: September 22, 2019, 04:14:41 AM »
Very nice review of the book. 

The reviewer did a good job of showing the incorrect information on digital presented.  At this time, there really is no need of any type of lossy data compression when storage size and data bandwidth are well beyond what is needed for lossless data transmission or file storage.


Offline Folsom

  • Obsessively Audiophilic
  • ****
  • Posts: 597
Re: Looks like an interesting read
« Reply #3 on: September 22, 2019, 06:46:18 PM »
I wish I could get ahold of Neil Young and propose some stuff to do in a studio to modify them to make the sound better. I've got some ideas that don't exist in the studio yet... and some of them actually reflect back to some of the earliest studio equipment where things were magical in the 50's and no one knows why other than "tubes" (maybe).