Hi Bernhard,
If you plan to detect / measure subtle differences in sound quality between DAC chips, it takes much more than a spectrum scan of a static sine wave test signal as this will not reveal THE most important property of a DAC: Dynamic response to complex signals like a recording of a Symphony orchestra with many different musical instruments being played at the same time, producing highly complex harmonics patterns.
I understand your scientific approach, and respect this, don't get me wrong. It would be perfectly fine if the DAC would only need to produce single sine wave of a certain frequency, producing as low THD as possible, similar to a function generator.
When performing measurements on dynamic DAC parameters, the measuring instrument must be able to pick-up subtle differences in complex dynamic signals, like our auditory system can do quite easily. This means that the measuring instrument then needs to have better specs then the device being tested. Then even a galvanic connection between DAC and sound card could be quite problematic as it pollutes the device under test with interference. The effects of this interference on complex dynamic signals doesn't show up in your sine wave spectrum scan because (time) resolution is too low, and you are using only one fundamental instead of a few thousand simultaneously.
Other parameter is transient response. How does a DAC respond to a sudden transient from a bell, triangle or drum. brickwall filtering limits max. frequency to say 22KHz. This means that minimum transient time (from bipolar zero to maximum) equals t(min) = (1/22,000)/4 = 11.364us, for realistic transient reproduction of live sound this is too slow. Similar, digital brickwall filtering is based on curve fitting, assuming highest possible frequency never exceeds fs / 2. This reconstructs sine wave shape perfectly, but now time resolution suffers (time resolution between L and R channel for example). This in turn leads to synthetic sound, as we are used to listen to sounds that have infinite time resolution between (L and R channels). With this reduced time resolution, we have difficulties with exact location of sound sources and their characteristic sound. Our brain has to do much more effort in order to interpret this artificial sound, this leads to listening fatigue. The interpretation of this band limited sound reproduction is "flat" and dull (analytical), not natural.
Digital sound reproduction systems that closely match live sound won't cause listening fatigue, it must be possible to listen a whole day to these systems without developing the slightest listening fatigue.
Proof of the importance of time resolution is with hi-res playback where time resolution has been slightly increased (96 / 192 KHz sample rate). Even if these hi res formats would be 16 bits, the higher time resolution would introduce similar advantages.
I use NOS without any filtering up to the speakers. I use dedicated large bandwidth audio components to make sure these remain unconditionally stable over required larger bandwidth. The system has very fast transient response that closely matches that of live music. I use no digital filtering, so time resolution that survived the anti-aliasing and A/D process will be preserved, not perfect, but as good as it gets.
Now I can extract subtle timing differences between both L and R channels, and can produce faster attack (transient response). this enables more natural sound reproduction.
About DLE / bit errors, I agree that these could have minor effect on perceived sound quality. But my personal opinion is that other errors like timing errors and time resolution are most important right now.
TDA1541A DLE error is specified at 0.5 LSB, glitch (MSB change) is specified at 0.25 LSB:
RapidShare: 1-CLICK Web hosting - Easy Filehosting
Page 427, 428
I don't want to start arguments here, just share my personal experiences and opinions, and yes I do make mistakes so now and then, I am only human. Let's be constructive and combine our efforts, and enjoy this hobby. You built an impressive master clock, and the PCB design looks very professional too. This is exactly what this forum is about.
If you plan to detect / measure subtle differences in sound quality between DAC chips, it takes much more than a spectrum scan of a static sine wave test signal as this will not reveal THE most important property of a DAC: Dynamic response to complex signals like a recording of a Symphony orchestra with many different musical instruments being played at the same time, producing highly complex harmonics patterns.
I understand your scientific approach, and respect this, don't get me wrong. It would be perfectly fine if the DAC would only need to produce single sine wave of a certain frequency, producing as low THD as possible, similar to a function generator.
When performing measurements on dynamic DAC parameters, the measuring instrument must be able to pick-up subtle differences in complex dynamic signals, like our auditory system can do quite easily. This means that the measuring instrument then needs to have better specs then the device being tested. Then even a galvanic connection between DAC and sound card could be quite problematic as it pollutes the device under test with interference. The effects of this interference on complex dynamic signals doesn't show up in your sine wave spectrum scan because (time) resolution is too low, and you are using only one fundamental instead of a few thousand simultaneously.
Other parameter is transient response. How does a DAC respond to a sudden transient from a bell, triangle or drum. brickwall filtering limits max. frequency to say 22KHz. This means that minimum transient time (from bipolar zero to maximum) equals t(min) = (1/22,000)/4 = 11.364us, for realistic transient reproduction of live sound this is too slow. Similar, digital brickwall filtering is based on curve fitting, assuming highest possible frequency never exceeds fs / 2. This reconstructs sine wave shape perfectly, but now time resolution suffers (time resolution between L and R channel for example). This in turn leads to synthetic sound, as we are used to listen to sounds that have infinite time resolution between (L and R channels). With this reduced time resolution, we have difficulties with exact location of sound sources and their characteristic sound. Our brain has to do much more effort in order to interpret this artificial sound, this leads to listening fatigue. The interpretation of this band limited sound reproduction is "flat" and dull (analytical), not natural.
Digital sound reproduction systems that closely match live sound won't cause listening fatigue, it must be possible to listen a whole day to these systems without developing the slightest listening fatigue.
Proof of the importance of time resolution is with hi-res playback where time resolution has been slightly increased (96 / 192 KHz sample rate). Even if these hi res formats would be 16 bits, the higher time resolution would introduce similar advantages.
I use NOS without any filtering up to the speakers. I use dedicated large bandwidth audio components to make sure these remain unconditionally stable over required larger bandwidth. The system has very fast transient response that closely matches that of live music. I use no digital filtering, so time resolution that survived the anti-aliasing and A/D process will be preserved, not perfect, but as good as it gets.
Now I can extract subtle timing differences between both L and R channels, and can produce faster attack (transient response). this enables more natural sound reproduction.
About DLE / bit errors, I agree that these could have minor effect on perceived sound quality. But my personal opinion is that other errors like timing errors and time resolution are most important right now.
TDA1541A DLE error is specified at 0.5 LSB, glitch (MSB change) is specified at 0.25 LSB:
RapidShare: 1-CLICK Web hosting - Easy Filehosting
Page 427, 428
I don't want to start arguments here, just share my personal experiences and opinions, and yes I do make mistakes so now and then, I am only human. Let's be constructive and combine our efforts, and enjoy this hobby. You built an impressive master clock, and the PCB design looks very professional too. This is exactly what this forum is about.
While I applaud the idea, in practice it's a lot harder than than. EC has already spoken to this. The ADC has to be better than the DAC for meaningful measurements. Well, at least for anything beyond the basics. And I also wonder if the DAC and ADC need to be locked to the same clock.
It's easy enough to look at a wave file sample by sample. If you can re-record the file via the DAC/ADC loop and show that it's sample perfect, then no worries. So far, I've not been able to do that. Close, but not quite.
So yeah, the measurements would be great. Getting good ones is not going to be easy, IMO. I'd love to see someone prove me wrong!
Dear Bernhard wrote:
I would say, why don't you go and listen to the damn thing!
Maybe he is right.
This hobby is about satisfaction in audio reproduction, at least IMHO. We seek different things in audio that are important for us: some like extreme detail and black backgrounds; some prefer huge dynamics and natural timbre...etc.
It is hard to have all...and expensive. I, for a couple of rupies, can build a DAC that is satisfactory to me because it has most of the things that I find are sine qua non to audio experience.
An example:
You introduce me to your fiancée. You love her because she is very pretty and inteligent and refined and also she is a very good cook...
I don't know how to tell you (I prefer to remain silent with a polite smile ) because to me she is too skinny and to pale and blonde, and she has no smell...briefly speaking, she is ugly We, latins, prefer the "brunettes" which have a beautifull skin, with a very nice aroma and are in general warm persons, but hey! it's your choice. If you are satisfied, who am I too make you fall from your skies!
Yours,
M.
I would say, why don't you go and listen to the damn thing!
Maybe he is right.
This hobby is about satisfaction in audio reproduction, at least IMHO. We seek different things in audio that are important for us: some like extreme detail and black backgrounds; some prefer huge dynamics and natural timbre...etc.
It is hard to have all...and expensive. I, for a couple of rupies, can build a DAC that is satisfactory to me because it has most of the things that I find are sine qua non to audio experience.
An example:
You introduce me to your fiancée. You love her because she is very pretty and inteligent and refined and also she is a very good cook...
I don't know how to tell you (I prefer to remain silent with a polite smile
) because to me she is too skinny and to pale and blonde, and she has no smell...briefly speaking, she is ugly We, latins, prefer the "brunettes" which have a beautifull skin, with a very nice aroma and are in general warm persons, but hey! it's your choice. If you are satisfied, who am I too make you fall from your skies! Yours,
M.
Dear Bernhard wrote:
I would say, why don't you go and listen to the damn thing!
Maybe he is right.
This hobby is about satisfaction in audio reproduction, at least IMHO. We seek different things in audio that are important for us: some like extreme detail and black backgrounds; some prefer huge dynamics and natural timbre...etc.
It is hard to have all...and expensive. I, for a couple of rupies, can build a DAC that is satisfactory to me because it has most of the things that I find are sine qua non to audio experience.
An example:
You introduce me to your fiancée. You love her because she is very pretty and inteligent and refined and also she is a very good cook...
I don't know how to tell you (I prefer to remain silent with a polite smile
Yours,
M.
I would say audio is less subjective.
You are right, the ADC has to be better than the DAC or at least equal it.
So for CD-audio you need 16 bit and the ADC needs to be very high quality which means it must have a spurious free dynamic range.
It also means that I can not measure a PCM1704 over its entire range.
But:
All what is written above is only relevant if you want to measure the DAC at 0 dB / full scale.
At -60 dB the inherent distortion of a -60 dB signal on a CD is already 1%.
This is not changed even when a 20 bit or 24 bit DAC is used.
The problem is that the TDA1543 transforms the 1% to 8%.
On the graph below you can see that for the PCM1704, for example if you play a -30 dB test signal, the THD+N is -72 dB which means the 16 bit ADC still has an unused 20 dB dynamic range safety.
The 16 bit ADC can test accurately all signals above the blue line.
From PCM1704 datasheet:
@ maxlorenz
sorry for another graph
Last edited:
with no decoupling caps? very badly, specific sound for specific people
for first look this seems to be true, but only for first...
+1
if you like dirty sound then yes, it is religion...
if 1khz 0db test then yes, but
for know almost all about the sound level of desired DAC or ADC you simply need THD of 100Hz, 1Khz, 10khz with output levels 0,-20,-40,-60,-80,-90db, glith energy and outband noise, thats all. Implementation of desired dac/adc you can "get on output" 20-99% sound abilities of DAC/ADC depending on you engeneering level and experience
our ears is perfect brickwall filter
dont use bad DFs and you get more natural and clean sound than NOS implementation
only if we dont use brickwall filters, only monotonic
for measure on -60db level all modern DS ADC have almost zero distortion (good 1bit ADC with -60db input have no distortion at all(<-160db))
Last edited:
not distortion! THD+N. quantity of bits only determine s/n ratio, distortion determines DNL/INL of ADC/DAC.
for CDDA s/n will be(without NS and dithering) 98db, 98-60=38db that is ~1% THD+N.
if you use ADC and DAC with zero DNL, INL than you has zero distortion, even if you ADC/DAC is only 1-5bit,
Last edited:
I must disagree.
If you have 16 bit ADC, it means max. 65536 steps that is an uncertainty of 0,000015 = 0,0015 %
when you sample a low level signal that uses only 7 bit of the ADC, you have max. 128 steps and uncertainty of 0,0078 = 0,78 %
no uncertainty, all definitely
with zero DNL, INL you steps would be ideally the same, transfer function linear.
Look at 1bit prom. ADC/DAC, real 1bit dac/adc have almost zero DNL/INL, with input signals <-10db he has distortions <-150db not depending of the output bit (simply DF output bit lenght), the noise is taken out of band by modulator. (low bit ADC for radiolocation with very big DR etc)
in multilevel DS DACs use DEM, DWA, partially DWA, DDS etc to reduce DNL/INL and lowering distortions
Look at you measurements http://www.beautyphoto.de/bf-analyze/PCM56@DCD1500mkII.jpg you have -122db 3rd but from you words it is impossible (not smaller than -95db)
Last edited:
But this is a very good dithered & noiseshaped signal, 4 x oversampling and the analyzer uses lots of averaging to reduce the noise floor.
http://www.beautyphoto.de/PCM56@DCD1500mkII.jpg
bf-analyze/ ???
http://www.beautyphoto.de/PCM56@DCD1500mkII.jpg
bf-analyze/ ???
Last edited:
No, you misunderstand.
I was talking about numbers without tricks applied.
And I assume that the datasheets do the same.
Look at PCM56 datasheet. Figure 11.
THD test is raw data from Eprom, no tweaked wave file from CD like I use.
Anyway, the posting was about what can be measured accurately with my Analyzer.
For 16 bit DAC @ -60 dB with dithering, noiseshaping, oversampling, 16 x averaging in the analyzer, the noise floor is -66 dB referenced to analyzer input level.
Makes 30 dB safety margin for my analyzer.
Too much discussion where those curves came from only spreads confusion.
I was talking about numbers without tricks applied.
And I assume that the datasheets do the same.
Look at PCM56 datasheet. Figure 11.
THD test is raw data from Eprom, no tweaked wave file from CD like I use.
Anyway, the posting was about what can be measured accurately with my Analyzer.
For 16 bit DAC @ -60 dB with dithering, noiseshaping, oversampling, 16 x averaging in the analyzer, the noise floor is -66 dB referenced to analyzer input level.
Makes 30 dB safety margin for my analyzer.
Too much discussion where those curves came from only spreads confusion.
Last edited: