21KHz ripple
Hi poobah,
By the way the upper trace is a quad DI-DAC
Jes, a ripple (amplitude modulation) occurs on the signal, just like like with the reference NOS-DAC (as shown in the oscillograph).
The sinewaves are triangular shape (the NOS-DAC just outputs a squarewave), as I have mentioned in a previous post. I just ran the same test on a commercial CD player, there is still a ripple on the sinewave signal but it is significantly smaller than both the reference NOS-DAC and the D-I DAC (the sound however isn't nearly as good as both the NOS-DAC or the D-I DAC). So what makes the NOS-DAC sound so good? is it phase linearity?
So, is this ripple seriously affecting the audible sound quality? If it does perhaps we can reduce the ripple to an acceptible level without compromizing the audio quality.
Suggestions?
Hi poobah,
By the way the upper trace is a quad DI-DAC
Jes, a ripple (amplitude modulation) occurs on the signal, just like like with the reference NOS-DAC (as shown in the oscillograph).
The sinewaves are triangular shape (the NOS-DAC just outputs a squarewave), as I have mentioned in a previous post. I just ran the same test on a commercial CD player, there is still a ripple on the sinewave signal but it is significantly smaller than both the reference NOS-DAC and the D-I DAC (the sound however isn't nearly as good as both the NOS-DAC or the D-I DAC). So what makes the NOS-DAC sound so good? is it phase linearity?
So, is this ripple seriously affecting the audible sound quality? If it does perhaps we can reduce the ripple to an acceptible level without compromizing the audio quality.
Suggestions?
The upper hearing limit is not the limiting factor of someones hearing abilities. I have a large hole at around 4k (from loud music/popconcerts?) And my top is only 15.000 Hz with normal music levels, tried this with the this months test CD included in german magazine "Audio".
I can detect faults in the music very easy now, better then when about 20 years ago. Its the same story with supertweeters afaik, nobody can hear 40.000 Hz, (the dog maybe?) but soundstage can improve a lot with a supertweeter.
I can detect faults in the music very easy now, better then when about 20 years ago. Its the same story with supertweeters afaik, nobody can hear 40.000 Hz, (the dog maybe?) but soundstage can improve a lot with a supertweeter.
Whoa boy...
The ripple is induced because when you linear interpolate you are injected false (wrong) numbers, voltages into the stream. Your CD player will not exhibit this behavior because the oversampling is done with a filter.
If you could put both waveworms, DI & REF, through an analog brick wall you would see that the reference DAC would have the best fidelity.
I suspect that the improvement you are hearing might be two things at play. First, there might be some sonic "bonus" (extra harmonics) you are hearing... not all distortions sound bad... and this is where "listen-only" people are detrimental to the craft. That's all about perception and nobody wants to be told they're nuts.
But most importantly... a 16 bit DAC is workin' its butt off to achieve 15 bit accuracy... and it sometimes runs with 14 bit accuracy. The linearity erros in a DAC are complex...some are random... some have patterns relating to which bit is being toggled. the biggest error is usually at the transistion from:
0111 1111 1111 1111
to
1000 0000 0000 0000
When you sum (average) 4 DACs together I think are taking a big step toward reducing the errors... certainly more the random errors than the pattern type.
I think you are seeing an improvement in linearity... at least below 10 kHz. All we hear above 10 Khz is really just pop, sizzle, and zing anyway... hiss.
Now try this... rewire your DACs to fire all at once... kill the linear interpolation. I'll bet you hear same improvement with less grit or mush in the real high end.
I'd be curious to learn of your results.
The ripple is induced because when you linear interpolate you are injected false (wrong) numbers, voltages into the stream. Your CD player will not exhibit this behavior because the oversampling is done with a filter.
If you could put both waveworms, DI & REF, through an analog brick wall you would see that the reference DAC would have the best fidelity.
I suspect that the improvement you are hearing might be two things at play. First, there might be some sonic "bonus" (extra harmonics) you are hearing... not all distortions sound bad... and this is where "listen-only" people are detrimental to the craft. That's all about perception and nobody wants to be told they're nuts.
But most importantly... a 16 bit DAC is workin' its butt off to achieve 15 bit accuracy... and it sometimes runs with 14 bit accuracy. The linearity erros in a DAC are complex...some are random... some have patterns relating to which bit is being toggled. the biggest error is usually at the transistion from:
0111 1111 1111 1111
to
1000 0000 0000 0000
When you sum (average) 4 DACs together I think are taking a big step toward reducing the errors... certainly more the random errors than the pattern type.
I think you are seeing an improvement in linearity... at least below 10 kHz. All we hear above 10 Khz is really just pop, sizzle, and zing anyway... hiss.
Now try this... rewire your DACs to fire all at once... kill the linear interpolation. I'll bet you hear same improvement with less grit or mush in the real high end.
I'd be curious to learn of your results.
High frequencies
Hi, tubee
If persons can't hear a single 18KHz sinewave, they sure can hear minute phase differences in the microsecond range, direct and reflected sound or two different frequencies at the same time. We use this ability to locate sounds around us. The perceived phase difference between L and R helps us to pinpoint the sound source. Perhaps this "phase sensitivity" is playing a large part in how we perceive sound quality. The slightest phase distortion (non linearity) could make music sound unrealistic or distorted as it conflicts with the way we interpret "live" sounds.
Hi, tubee
If persons can't hear a single 18KHz sinewave, they sure can hear minute phase differences in the microsecond range, direct and reflected sound or two different frequencies at the same time. We use this ability to locate sounds around us. The perceived phase difference between L and R helps us to pinpoint the sound source. Perhaps this "phase sensitivity" is playing a large part in how we perceive sound quality. The slightest phase distortion (non linearity) could make music sound unrealistic or distorted as it conflicts with the way we interpret "live" sounds.
)DAC rewiring
Hi poobah,
Thanks for your reply,
I already tried that during the D-I design. The details dissapeared compared to the twin-DAC with 8th order filter I build earlier. Now I get significantly more detail with the DI-DAC without that 8th order filter.
DI-DAC outputs are also summed, so this should improve linearity.
When you look at the oscillogram the DI-DAC has a triangle wave (EVEN harmonics). The NOS-DAC has a squarewave with a very high slew rate (more ODD high frequency harmonics). So the D-I DAC must have far less (very) high frequency harmonics compared to the NOS-DAC. This could explain the significantly clearer sound of the D-I DAC as those ODD high frequency harmonics are removed in the D-I process.
Hi poobah,
Thanks for your reply,
I already tried that during the D-I design. The details dissapeared compared to the twin-DAC with 8th order filter I build earlier. Now I get significantly more detail with the DI-DAC without that 8th order filter.
DI-DAC outputs are also summed, so this should improve linearity.
When you look at the oscillogram the DI-DAC has a triangle wave (EVEN harmonics). The NOS-DAC has a squarewave with a very high slew rate (more ODD high frequency harmonics). So the D-I DAC must have far less (very) high frequency harmonics compared to the NOS-DAC. This could explain the significantly clearer sound of the D-I DAC as those ODD high frequency harmonics are removed in the D-I process.
Hmmmm,
Hate to say it, but the detail you are referring to might be sonic enrichment... pleasant errors. Might be some other effect.
You need to do a spectral analysis of both types to see what's going on... an FFT on dry data would do as well.
If you can, try your 8th order filter on both the DI and REF dac and compare your results. Might find something on the scope
And also, have you checked and plotted the response of your 8th order filter?
Hate to say it, but the detail you are referring to might be sonic enrichment... pleasant errors. Might be some other effect.
You need to do a spectral analysis of both types to see what's going on... an FFT on dry data would do as well.
If you can, try your 8th order filter on both the DI and REF dac and compare your results. Might find something on the scope
And also, have you checked and plotted the response of your 8th order filter?
8th order filter
Hi, poobah
Thanks for your reply,
We must have had the same thoughts.. I just quickly connected the 8th order filter after the D-I DAC to see what would happen, guess what, hardly any change. Only the triangle waves look much more like sinewaves. I scanned the 8th order filter with a sweep generator during design. It starts rolling off at 20KHz, the pass band has constant amplitude (butterworth characteristic)
Hi, poobah
Thanks for your reply,
We must have had the same thoughts.. I just quickly connected the 8th order filter after the D-I DAC to see what would happen, guess what, hardly any change. Only the triangle waves look much more like sinewaves. I scanned the 8th order filter with a sweep generator during design. It starts rolling off at 20KHz, the pass band has constant amplitude (butterworth characteristic)
poynton said:
Standard digital techniques will recover the 20Khz sinewave from a 44k1 sample rate through the very ringing that seems so objectionable to some. Combined with the correct analogue filter, pre and post ringing recreates the sinewave.
Linear interpolation simply joins the sample points with straight line. At 2K you have 20+ samples, enough for a fair representation of the original waveform. At 20k, with only two samples, simply joining the sample points misrepresents the original waveform.
