[macaque]

Quote:

Originally Posted by leeperry /img/forum/go_quote.gif simple link: Upsampling vs. Oversampling for Digital Audio oversampling in the DAC: you want it as high as possible to decrease aliasing and increase the conversion resolution. the AK4396 does it at 128X rate for all sampling rates. upsampling in the source: terrible idea IMO(and that link conclusion agrees), all it does is feeding worthless interpolated data and increase THD+N dramatically(the sound will appear brighter and more distorted):

I don't think you understood my post or are relating it to the quoted paper correctly...also please explain your graphs and what they are actually from...

The ADC oversampling/aliasing portion is irrelevant to what I'm talking about as it is done before you get the music, the point is you obtain the music encoded at 16/44.1 and the assumption with good recordings is they have done the work to to ensure a good noise shaped data set.

Now, you have to play back the 16/44.1 data. You need a reconstruction filter with a sharp cutoff in the 20-22.05 KHz range. We have already concluded that doing this with analog components is impossible and as good as you can make it is still impossible to match 2 channels (or more) exactly with physical components. So by oversampling or upsampling you get to make a digital filter to do this part and use an analog reconstruction filter for the new sample rate at a cutoff so far away from any actual signal that it doesn't matter how bad it behaves near the cutoff or if the filters for each channel even match closely. (and the digital filters match perfectly for multiple channels)

So oversampling DACs have such a digital filter built in. How good is it? It depends on the DAC, how much of a data window they work with, the precision of their calculations, etc. Once all this is set it is stuck in the DAC forever. Look at any oversampling DAC data sheet and they will describe this digital filter (often they have more than one selectable with sharp or slow cutoffs).

There are really four choices for doing the 20-22.05 Khz filter:
1) non-oversampling DAC and analog filter (with accuracy and matching problems)
2) non-oversampling DAC and no filter (the paper describes potential problems) - this can be simulated by upsampling with no filter and playing back with an oversampling DAC. I've tried it - sounds pleasant until you compare it to doing proper filtering (on my system). I suspect this is what you are showing in your graphs.
3) oversampling DAC with its built in filter
4) upsampling before oversampling DAC. You get to make the filter and and the DAC filter is still present but has a cutoff far-away from the audio band so you are essentialy moving the 20-22.05 filter from it to you.

#4 is what I am doing because I believe I can make a better offline filter than the one built into any DAC - mainly because I have the benefit of much more information than the DAC could possibly have in real-time - and I can use higher precision math (64 bit floating point, 80 bit internal in the FPU).

This is not about finding "worthless interpolated data" it is about using the math to calculate the correct values of this data much the same way the DAC will be trying to do it only _better_. We know what the filter should be to get the right values for the new samples we can just implement a more accurate version of it. It is not curve fitting or any other kind of "connect the dots" interpolation, it is "applying the correct filter" interpolation. This is why you likely will get peaks above those in your original data set.

This whole exercise has nothing to do with adding or manipulating information - this is 100% about extracting the 16/44.1 to an analog waveform as accurately as possible and one of the biggest impediments to doing that it building the 20-22.05 filter.

 

[leeperry]

well, I fully agree w/ everything that's said on this page: Upsampling vs. Oversampling for Digital Audio
Quote:

Upsampling would give us the same benefits in frequency response that we have gone over, however we can achieve the same effects by sufficiently oversampling our signal both at the DAC and ADC. Upsampling has no effect on our digital filter design problem since all our digital filters are FIR (finite impulse response) and all have linear phase. By sufficiently oversampling at the ADC, we can design a very simple, linear phase, digital filter that has no problems with our audio signal. There has been much misinformation surrounding upsampling and many claims have been made that state that upsampling is necessary to allow for such a desirable digital filter. However, it is the oversampling at the ADC takes care of this, not upsampling.

when you upsample, you're resampling and adding useless interpolated data that will mislead the DAC oversampling algorithm and post-filtering.

you're not going to create any useful data, and a recent powerful DAC is fully able to process 128X oversampling on 44.1 audio...something you'll NEVER be able to achieve by software upsampling, and I really don't think that you want to feed bogus interpolated data to your DAC...as it'll do a far better job than any resampler on your PC.

the graphs are from WaveSpectra...you can have fun generating test tones in SineGen, upsample them in r8brain Pro, libsamplerate or other resamplers...and measure the THD/THD+N/SNR differences, the only thing upsampling does is increasing the harmonic distortion IMHO, and make the sound brighter due to distortion...and you're feeding a distorted signal to your DAC, that will process its 128X internal oversampling/post-filtering on bogus interpolated data..

the WM8740/AK4396 have a very thorough and highly accurate 128X oversampling stage..we're not in the 80's anymore

Quote:

Richard Kulavik of AKM Semiconductors explained it this way : "The AK4396 DAC is a large departure from other delta-sigma DACs designed by us and others like BurrBrown, Analog Devices and Cirrus Logic. The AK4396 is an entirely new modulator, pioneered and patented by AKM. It achieves something unique. In the past, many of the old Phillips and BurrBrown parts were R-2R* based products. These older products were looked upon as some of the best. One of the reasons was high frequency noise. In older R-2R parts, HF noise was not present. In all delta-sigma parts prior to the AK4396, everyone has fought HF noise caused from the delta-sigma modulator with the insertion of large filters and other parts to attempt to solve a problem created by the delta-sigma design. The AK4396 today effectively does not suffer any modulator-induced HF noise and is over 60dB better than the nearest Cirrus and BB devices. All of this HF noise can cause many audible artifacts downstream. That is the 'miracle' we believe is making the difference today. This part gives you the performance and linearity of a delta-sigma device with the noise performance of an R-2R part, something that was never previously available."

upsampling your CD's to 176.4 kHz FLAC doesn't offer a single advantage and only worsens the SQ IMHO...well, except if you got one of those NOS DAC's from the 80's that is....but I'll take 128X hardware oversampling over 4X software upsampling anytime of the day

PS: but I don't think that there is much need to get too hung up about all this...because the SQ will be far more colored by the quality of your PSU and the opamps you'll use as DAC LPF: http://www.head-fi.org/forums/6490323-post3.html

as usual, you're only as strong as your weakest link...and the jitter of the signal fed to the DAC is also very prone to change its subjective SQ

 

[SP Wild]

Off-line upsampling sounds interesting. Trying to use the secret rabbit code at it's highest level (to upsample to 48khz to bypass cheap soundcard upsampling before optical into dac) is not possible on my net book because it's not powerful enough. No problems with the desktop, but the desktop is noisy. Or I could invest in the M2tech Hi-face - because adaptive USB is crap. But still...is there free software to do high quality off-line software upsampling?

 

[macaque]

Quote:

Originally Posted by leeperry /img/forum/go_quote.gif well, I fully agree w/ everything that's said on this page: Upsampling vs. Oversampling for Digital Audio the WM8740/AK4396 have a very thorough and highly accurate 128X oversampling stage..we're not in the 80's anymore upsampling your CD's to 176.4 kHz FLAC doesn't offer a single advantage and only worsens the SQ IMHO...well, except if you got one of those NOS DAC's from the 80's that is....but I'll take 128X hardware oversampling over 4X software upsampling anytime of the day PS: but I don't think that there is much need to get too hung up about all this...because the SQ will be far more colored by the quality of your PSU and the opamps you'll use as DAC LPF: http://www.head-fi.org/forums/6490323-post3.html as usual, you're only as strong as your weakest link...and the jitter of the signal fed to the DAC is also very prone to change its subjective SQ

I also agree with the paper but you are still mising the point. The part you quoted:

Quote:

By sufficiently oversampling at the ADC, we can design a very simple, linear phase, digital filter that has no problems with our audio signal. There has been much misinformation surrounding upsampling and many claims have been made that state that upsampling is necessary to allow for such a desirable digital filter

.

The point is that "we can design a very simple, linear phase, digital filter that has no problems with our audio signal." but the oversampling DAC has to IMPLEMENT it with a limited scope of source data. What I am talking about here is oversampling offline (changing the sample rate to an exact multiple of the original, just like the DAC). But the reason for doing it is I get to IMPLEMENT the SAME (or better) digital filter the DAC is implementing. I think I can do it better offline than the DAC because I have more data (not constrained by a real-time window) and a killer FPU with more precision. It also means the DAC's version of this filter becomes much less important since the new signal will be in a small portion of the passband of the DAC's version of the digital filter. The very large ratio of oversampling in most oversampling DACs (delta sigma types) type mainly use the larger ratios because they are internally using less bits at the higher rates to create the same effect as a larger number of bits at a lower sample rate (using noise shaping). Either way the DAC still must implement the digital filter we are talking about.

Again, the interpolated data is not worthless - it is the SAME data the oversampling DAC would be trying to get from its own filter except you can make a more accurate filter - in fact you missed one very important graph in you series (and you seem to have used the same sample window in terms of number of samples in all your graphs instead of the same amount of signal time). The oversampled versions (which actually do not look to be exact integral multiples of the sample rate) all have had the 20-22.05 Khz filter applied (well, actually the equivalent for the 48K sample rate you seem to use) - that is part of the oversampling process. The graph you need to show is the oversampled signal INSIDE the oversampling DAC after it applies it's 20-22.05 digital filter and compare that to the rest (of course you can't see that data). My point is all these techniques are doing the same thing (including the oversampling in the DAC) but my bet is if you compared the result you would find the upsampling algorithms produced a better filter. Also, what you really want to compare is with a fundamental and a lot of harmonics that go all the way up to Nyquist and take a look at the post filter results in terms of how the harmonic relations have been distorted - because that is what makes music. This was the beauty of analog, timing distortions changed pitch while leveling harmonic structure completely in tact.

With regard to the DAC being a more powerful filter engine, that is nonsense. Your PC and mine are far more powerful tools for this -if you had the digital filter algorithm the DAC was using you could easily implement it on a PC (very liklely faster and undoubtedly more accurate). You could do 1,000,000 times oversampling if you wanted to but you have no device to take the data.

There is one very important difference between doing this offline and in a device. It is a _platform_ vs. a _component_. Digital audio was for a long time a bunch of components strung together - some to get the data (which also usually controlled the timing), somtimes transmission, conversion, filtering etc... From the point of view of an oversampling DAC you need include a digital filter, but the amount of signal window you can use is constrained by how much delay you can live with in real-time - since the component will be used in real-time - often from a source transmitted data and timing information. To have a data window you have to have at least that many samples delivered to the DAC before you can start. Needless to say this places an immediate constraint on your digital filter design - that is you cannot _implement_ the best filter you can _design_. The constraint of precision in calculations is based on how complex you are willing to make the DAC. The fact that a typical DAC contains 2 different digital filter algorithms is because each one has its trade-offs (better phase response but slower cutoff or sharper cutoff with less than ideal phase response).

Computers have made digital audio a platform. Timing can be completely independent of retrieval of source data - and digital filtering can be done in a highly accurate fashion by the computer. In fact the number of filter designs you can use are limitless. When you oversample offline you get rid of the sample window constraint and you can select from devices of incredible precision (your PC FPU can do 80 bit internally). In other words you can likely _implement_ ANY filter you choose to design and implement it very accurately - you are not constrained by real-time concerns and you can have as much signal data as you need for your design.

 

[macaque]

Quote:

Originally Posted by leeperry /img/forum/go_quote.gif well, I fully agree w/ everything that's said on this page: Upsampling vs. Oversampling for Digital Audio the WM8740/AK4396 have a very thorough and highly accurate 128X oversampling stage..we're not in the 80's anymore upsampling your CD's to 176.4 kHz FLAC doesn't offer a single advantage and only worsens the SQ IMHO...well, except if you got one of those NOS DAC's from the 80's that is....but I'll take 128X hardware oversampling over 4X software upsampling anytime of the day PS: but I don't think that there is much need to get too hung up about all this...because the SQ will be far more colored by the quality of your PSU and the opamps you'll use as DAC LPF: http://www.head-fi.org/forums/6490323-post3.html as usual, you're only as strong as your weakest link...and the jitter of the signal fed to the DAC is also very prone to change its subjective SQ

I also agree with the paper but you are still mising the point. The part you quoted:

Quote:

By sufficiently oversampling at the ADC, we can design a very simple, linear phase, digital filter that has no problems with our audio signal. There has been much misinformation surrounding upsampling and many claims have been made that state that upsampling is necessary to allow for such a desirable digital filter

.

The point is that "we can design a very simple, linear phase, digital filter that has no problems with our audio signal." but the oversampling DAC has to IMPLEMENT it with a limited scope of source data. What I am talking about here is oversampling offline (changing the sample rate to an exact multiple of the original, just like the DAC). But the reason for doing it is I get to IMPLEMENT the SAME (or better) digital filter the DAC is implementing. I think I can do it better offline than the DAC because I have more data (not constrained by a real-time window) and a killer FPU with more precision. It also means the DAC's version of this filter becomes much less important since the new signal will be in a small portion of the passband of the DAC's version of the digital filter. The very large ratio of oversampling in most oversampling DACs (delta sigma types) type mainly use the larger ratios because they are internally using less bits at the higher rates to create the same effect as a larger number of bits at a lower sample rate (using noise shaping). Either way the DAC still must implement the digital filter we are talking about.

Again, the interpolated data is not worthless - it is the SAME data the oversampling DAC would be trying to get from its own filter except you can make a more accurate filter - in fact you missed one very important graph in you series (and you seem to have used the same sample window in terms of number of samples in all your graphs instead of the same amount of signal time). The oversampled versions (which actually do not look to be exact integral multiples of the sample rate) all have had the 20-22.05 Khz filter applied (well, actually the equivalent for the 48K sample rate you seem to use) - that is part of the oversampling process. The graph you need to show is the oversampled signal INSIDE the oversampling DAC after it applies it's 20-22.05 digital filter and compare that to the rest (of course you can't see that data). My point is all these techniques are doing the same thing (including the oversampling in the DAC) but my bet is if you compared the result you would find the upsampling algorithms produced a better filter. Also, what you really want to compare is with a fundamental and a lot of harmonics that go all the way up to Nyquist and take a look at the post filter results in terms of how the harmonic relations have been distorted - because that is what makes music. This was the beauty of analog, timing distortions changed pitch while leveling harmonic structure completely in tact.

With regard to the DAC being a more powerful filter engine, that is nonsense. Your PC and mine are far more powerful tools for this -if you had the digital filter algorithm the DAC was using you could easily implement it on a PC (very liklely faster and undoubtedly more accurate). You could do 1,000,000 times oversampling if you wanted to but you have no device to take the data.

There is one very important difference between doing this offline and in a device. It is a _platform_ vs. a _component_. Digital audio was for a long time a bunch of components strung together - some to get the data (which also usually controlled the timing), somtimes transmission, conversion, filtering etc... From the point of view of an oversampling DAC you need include a digital filter, but the amount of signal window you can use is constrained by how much delay you can live with in real-time - since the component will be used in real-time - often from a source transmitted data and timing information. To have a data window you have to have at least that many samples delivered to the DAC before you can start. Needless to say this places an immediate constraint on your digital filter design - that is you cannot _implement_ the best filter you can _design_. The constraint of precision in calculations is based on how complex you are willing to make the DAC. The fact that a typical DAC contains 2 different digital filter algorithms is because each one has its trade-offs (better phase response but slower cutoff or sharper cutoff with less than ideal phase response).

Computers have made digital audio a platform. Timing can be completely independent of retrieval of source data - and digital filtering can be done in a highly accurate fashion by the computer. In fact the number of filter designs you can use are limitless. When you oversample offline you get rid of the sample window constraint and you can select from devices of incredible precision (your PC FPU can do 80 bit internally). In other words you can likely _implement_ ANY filter you choose to design and implement it very accurately - you are not constrained by real-time concerns and you can have as much signal data as you need for your design.

And you are not taking 4x oversampling over 128x oversampling - you are taking 512x oversampling over 128x (actually at the end of the day it is likely the total is still 128x because it'll be based on whatever actual clock frequency in on the crystal driving the DAC), and the DACs filter becomes relatively beign.

 

[b0dhi]

Interesting stuff. I wonder if it's correct about the DAC filter being benign in that scenario, though. If you have two filters in series, won't the inaccuracies of the the poorer filter still pass through the filter chain, even if the other filter is much higher fidelity?

 

[leeperry]

hehe, I completely disagree w/ everything macaque said...but the new forum is just too painful to use....if he wants to feed his DAC w/ a distorted signal, fine.
 
you can't disable the built-in DAC oversampling stage, and doing a double SRC is a terrible idea IMHO.
 
you can run WaveSpectra as much as you want, at whatever rate, upsampling will aways increase the harmonic distortion....and many DAC's have a much lower internal oversampling >96kHz.

 

[Audiophiliac]

The best thing about being a tech type of person, is that once you learn to toggle up the ADC to the max sample rate, you can re-release all the old stuff from tape again at higher bitrates than anyone else is using.
Better than bitcoin mining, if you ask me.

 

[Ghoostknight]

while the nyquist theory is right imo resampling to 176,4/192khz actually sounds better, specially for 44,1khz files (if you use a high quality resampling algorithm or it doesnt sound better)
i think the reason is that the reconstruction filter effects are higher in frequency (or better quality filters from resampling algorithms)

imo everyone should have tried resampling, there definitely is a difference, higher samplerate usually sound smoother (atleast for me)

 

[bigshot]

Your opinion is 100% unadulterated expectation bias.

 

[gregorio]

while the nyquist theory is right imo

Ah good, you have an opinion that actually agrees with the proven facts. Although: 1. The proven facts are the proven facts, regardless of your opinion and 2. Unfortunately, that’s just about the only assertion in your whole post which does agree with the facts!

resampling to 176,4/192khz actually sounds better

Technically, resampling adds distortion and therefore is actually worse! Although it’s such a tiny amount of distortion it’s way beyond inaudible, even if you resample numerous times. So you’re hearing a difference that: A. You think is better when it’s actually worse! And B. Is way below audibility anyway.

imo everyone should have tried resampling,

Who hasn’t tried resampling? All ADCs resample, pretty much all DACs resample and most recordings made in the last 20 years have multiple additional resampling processes during mixing/mastering, so how do you avoid listening to a digital audio recording without any resampling?

there definitely is a difference,

Yes, there definitely is but you need specialist equipment/tests to detect it.

higher samplerate usually sound smoother (atleast for me)

The difference is so small it probably doesn’t even affect the sound, in which case there can be no question of audibility of course and even if it did affect the sound, it would not make it “smoother”! Unless something in your system is seriously broken/faulty, then what you’re hearing has nothing to do with the actual sound and everything to do with your personal perception biases (as already mentioned)!

G

 

[71 dB]

There are benefits to higher sample-rates when applying non-linear processing onto the signal such as saturator and limiter. Non-linear processing adds harmonics that can easily go beyond the original Nyquist frequency. Without using higher sample-rate and lowpass filtering the higher harmonics away before downsampling back original lower sample-rate there will be aliasing.

Higher samplerates are not without problems. More computational power is needed and resampling the signal back and forth cumulatively degrades the sound quality, at least in theory if not audibly. A few years ago an advantaged and smart way to do non-linear processing was develop: Anti-Derivative Anti-Aliasing (ADAA). The idea is to do the processing with the anti-derivative of the non-linear function to make it "smoother" and then taking the derivative of the operation resulting less aliasing. If we for example have a non-linear saturation function

f(x) = x - x³/3,​

where f(x) is the output and x the input, -1 ≤ x ≤ 1. The anti-derivative F(x) of f(x) is

F(x) = ∫ f(x) dx = x²/2 - x⁴/12 + C.​

We set F(0) = 0 and have C = 0 for simplicity, but the value of C doesn't matter, because the output of the saturator is calculated by taking the "discrete derivative" of F(x[n]):

y[n] = (F(x[n]) - F(x[n-1])) / (x[n] - x[n-1])),​

where x[n] is the nth input sample value and y[n] is the nth output sample value. Those into math will immediately see, that the denominator x[n] - x[n-1] is a troublemaker, because we are dividing by zero, if there are two identical consecutive sample values. Those cases have to be separately calculated:

y[n] = f(x[n]), if x[n] = x[n-1]​

To make the aliasing reduction more substantial, this idea can be extended into 2nd, 3rd etc. order antiderivatives. Combining for example doubled sample-rate with 3rd order anti-derivative anti-aliasing method can reach anti-aliasing reduction at levels beyond the dynamic range of 16 bit digital audio.

 

[bigshot]

You think that’s going to help him understand?

 

[Ghoostknight]

Your opinion is 100% unadulterated expectation bias.

well, its what i hear and there may be different reasons for that but im not going to argue with people that say every dac sounds the same

Technically, resampling adds distortion and therefore is actually worse! Although it’s such a tiny amount of distortion it’s way beyond inaudible, even if you resample numerous times. So you’re hearing a difference that: A. You think is better when it’s actually worse! And B. Is way below audibility anyway.

or there are different reasons for it and not just distortion which sits at -170db for good resampling algorithms, way below 24bit noisefloor

Who hasn’t tried resampling? All ADCs resample, pretty much all DACs resample and most recordings made in the last 20 years have multiple additional resampling processes during mixing/mastering, so how do you avoid listening to a digital audio recording without any resampling?

thats not the same as applying resampling to the whole song tho you only know the different resampled parts of song (or dacs) already resampled and never heared it differently
so A-B resampling vs non resampling yourself!

The difference is so small it probably doesn’t even affect the sound, in which case there can be no question of audibility of course and even if it did affect the sound, it would not make it “smoother”! Unless something in your system is seriously broken/faulty, then what you’re hearing has nothing to do with the actual sound and everything to do with your personal perception biases (as already mentioned)!

"probably", well everyone should try for themself, this is just my opinion

--

honestly after reading alot the only logical reason for resampling to sound better is a better antialising filter (at 44,1khz, and the worse dac filter gets shifted to higher frequencys) and maybe "helping the dac out in reconstruction" in throwing higher samplerates at it, then it also makes sense why it may sound "smoother"
dont bother throwing objective reasons out why this can not be etc.. atleast i dont care if it still sounds different/better

 

[bigshot]

You make up what you want to believe. You think up something and then defend it. You don’t look at things, test carefully and let the truth reveal itself. I’m not telling you to believe me. I’m just telling you that you need carefully controlled observation and analysis, not sloppy self serving delusion. Get smarter.