my recent epiphany concerning low NFB linear amps got me thinking.
I keep hearing about people being impressed by vintage CD players, even compared to relatively expensive recent production DACs. my CD600 died recently and I have no intention to repair it but intuition tells me that said people are not crazy.
since I know what is usually found inside such players, it's really weird. discrete regulators and output stages or fancy oscillators are never to be found. but they're all based on R2R DACs. apart from that, the infamous oversampling filters (no dithering or subtleties of the kind), opamp-based I/V stages etc.
my personal experience, mirrored by that of other people tells me that all cheap delta-sigma based implementations sound almost identical.
do you have similar experiences? if so, what do you think is the cause? is it USB-related? is it the R2R chips? what do you think?
I keep hearing about people being impressed by vintage CD players, even compared to relatively expensive recent production DACs. my CD600 died recently and I have no intention to repair it but intuition tells me that said people are not crazy.
since I know what is usually found inside such players, it's really weird. discrete regulators and output stages or fancy oscillators are never to be found. but they're all based on R2R DACs. apart from that, the infamous oversampling filters (no dithering or subtleties of the kind), opamp-based I/V stages etc.
my personal experience, mirrored by that of other people tells me that all cheap delta-sigma based implementations sound almost identical.
do you have similar experiences? if so, what do you think is the cause? is it USB-related? is it the R2R chips? what do you think?
Here is a related thread:
http://www.diyaudio.com/forums/digital-source/232823-attacking-sound-quality-denon-dcd-680-a.html
http://www.diyaudio.com/forums/digital-source/232823-attacking-sound-quality-denon-dcd-680-a.html
having read a bit of psychoacoustics over the years, looking at the numbers involved, I really don't see the justification for the worry about better delta-sigma converter audio band noise floor modulation
for the best chips we're talking below -120 dB with the -60 dB fs reference signal - the datasheet standard number
which as far as I can tell was chosen to emphasize quantization noise, not to hide noise modulation
from the few plots showing this "problem" we see <20 dB noise floor rise at full scale output - which gives noise -100 dB relative to signal
there is absolutely no expectation that this can be audible when comparing to any previous analog recording/playback media
even less when comparing with psychoacoustic lossy codecs - which are extensively DBT tested - the lesson from them is that <50 dB is S/N needed in each critical band relative to its peak amplitude
likewise other demos, tests out there show that a Sousa March is audibly undetectable at 60 dB below a Brahms lullaby, the Noise Power Ratio/Belcher Noise fill test only requires <40dB resolution before all correlation with listening results goes away (good thing they were testing speaker, not amps)
for the best chips we're talking below -120 dB with the -60 dB fs reference signal - the datasheet standard number
which as far as I can tell was chosen to emphasize quantization noise, not to hide noise modulation
from the few plots showing this "problem" we see <20 dB noise floor rise at full scale output - which gives noise -100 dB relative to signal
there is absolutely no expectation that this can be audible when comparing to any previous analog recording/playback media
even less when comparing with psychoacoustic lossy codecs - which are extensively DBT tested - the lesson from them is that <50 dB is S/N needed in each critical band relative to its peak amplitude
likewise other demos, tests out there show that a Sousa March is audibly undetectable at 60 dB below a Brahms lullaby, the Noise Power Ratio/Belcher Noise fill test only requires <40dB resolution before all correlation with listening results goes away (good thing they were testing speaker, not amps)
There's no worry, just listening observations - noise modulation is the best hypothesis I have at present which fits the observations.
The 20dB difference is only in the averaged noise floor - so short term noise bursts are potentially higher. Added to this that the stimulus signal is a simple sine, low crest factor whereas the noise being added is a function of the instantaneous level in the modulator loop.
The Sousa march is the usual red herring because its uncorrelated with the music.
The 20dB difference is only in the averaged noise floor - so short term noise bursts are potentially higher. Added to this that the stimulus signal is a simple sine, low crest factor whereas the noise being added is a function of the instantaneous level in the modulator loop.
The Sousa march is the usual red herring because its uncorrelated with the music.
the correlation argument doesn't help your case
again studies of timbre perception show that -40 dB, ~ 1% resolution of harmonics, wavefrom envelope is the perceptual limit
according to frequency masking theory a broad audio band noise's amplitude that is porportional to a signal amplitude envelope will be heavily masked, ~ 4th order slope at low level, at higher >80 dB spl the masking of higher freuquency content is much stonger
temporal masking - another fail for correlated errors - anything within ~ms but at <40 dB down is fused with the larger amplitude signal - whether before or for even longer time after
again studies of timbre perception show that -40 dB, ~ 1% resolution of harmonics, wavefrom envelope is the perceptual limit
according to frequency masking theory a broad audio band noise's amplitude that is porportional to a signal amplitude envelope will be heavily masked, ~ 4th order slope at low level, at higher >80 dB spl the masking of higher freuquency content is much stonger
temporal masking - another fail for correlated errors - anything within ~ms but at <40 dB down is fused with the larger amplitude signal - whether before or for even longer time after
just the check of whether can you hear whatever special proposed component's difference between the source with the desired property and the same output passed thru a high resolution ADC/DAC in series
in the case at hand you can start with components you assert can be clearly heard such as R2R DAC
then pass thru a delta-sigma ADC/DAC combo, a strong hypothesis is that some pro quality 24/192 ADC/DAC will be audibly transparent
if so, then an additional test of 2 source that you can audibly differentiate (must also demonstrate controlled, blind) passed thru the same high res chain - can you still differentiate by ear?
it seems if both can be achieved then you have to conclude the the high res ADC/DAC was in fact transparent for whatever hypothetical difference
only apples to the specific combo of equipment, still many possible confounding factors like system IC, gnd, noise, antialias filters – but a chance of learning something exists
in the case at hand you can start with components you assert can be clearly heard such as R2R DAC
then pass thru a delta-sigma ADC/DAC combo, a strong hypothesis is that some pro quality 24/192 ADC/DAC will be audibly transparent
if so, then an additional test of 2 source that you can audibly differentiate (must also demonstrate controlled, blind) passed thru the same high res chain - can you still differentiate by ear?
it seems if both can be achieved then you have to conclude the the high res ADC/DAC was in fact transparent for whatever hypothetical difference
only apples to the specific combo of equipment, still many possible confounding factors like system IC, gnd, noise, antialias filters – but a chance of learning something exists
Seems we have rather a different focus - I'm not at all interested in 'can I hear a difference?' - to me its purely an academic question. I'm far more interested in describing and then understanding the characteristics of the sound of each. My customers are going to be interested in long-term listening satisfaction, not in whether an audible difference exists between two arbitrarily chosen contenders.
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.