Oversampling DAC vs NON Oversampling DAC

[TKYR1967]

I am confused here and might be FAR off but I'm going to ask anyway. If Audiophiles are in pursuit of recreating what was recorded exactly, note for note and bit for bit, and have it sound as close to the live experience as possible, then why use oversampling DACs? Isn't that altering what is on the disc much like an equalizer alters what was originally recorded?

 

[PeterSt]

Correct.

 

But it is not easy to create an NOS DAC with low harmonic distortion so YMMV at listening to random NOS DACs and conclusions.

But when the DAC is implemented with the utmost performance on THD+N you can't imagine what you will be hearing.

 

The main problem is that 24 bit NOS DAC chips are not produced anymore ...

 

Peter (who wants to keep it brief this time :-)

 

[Audio_ELF]

Oversampling changes bit in your data, but it doesn't change the notes in the music.

 

There are good arguments both ways for NOS vs OS DACs, however both have their advantages and limitations. Some people will say NOS DACs are nicer sounding, yet very few companies use NOS technology. Yes most manufacturers rely on a small number of chips for their DACs which limit what is available ... but dCS and Chord both create their discrete solutions and both use over sampling in their DACs. Also I'm sure if Denon, Marantz, Yamaha, et al felt that they needed to use NOS technology, they would be putting pressure on Texas Instruments, etc to produce NOS DACS.

 

Yes there is always the mainstream way and the niche market. No one would start comparing a Morgan car with it's wooden chassis to a Ferrari and say that Morgan should stop using wood. However Ferrari, Porsche, Aston Martin, etc use metal chassises for a reason.

 

There are reasons to use NOS DAC, but I think not changing notes in the music is something that can NEVER happen, and sometimes worrying about bit-perfection within the DAC is superfluous. If it sounds good, then it probably is whatever is happening.

 

Eloise

 

[Lizard_King]

I use a NOS Dac with USB and I feel is sound a lot better playing Red book Cds ripped as FLAC files over the USB .

 

[emmodad]

@PeterST: were 24-bit NOS DAC chips ever produced? can you identify any?

 

[Suteetat]

At the end of the day, I don't think much about upsampling, oversampling, 16 bits 24 bits much anyhow.

The most important thing is how it sounds. I think each company choose whatever they feel is the best

for their products at a given price point. If one product sounds better than the other (to my ear and for my personal taste and preference) I don't really care what chipset it uses or what other specs are in there.

There are companies that make great sounding DAC with no oversampling such as Zanden and there are plenty of companies that make great sounding DACs with oversampling so I think it is more of an implementation than

anything else.

 

 

[ggking7]

For a long time my USB DAC of choice was the DIY Paradise USB Monica. I loved it and I thought the excellent sound was due in large part to the fact that it's a NOS DAC. Then I heard the Wavelength Proton in my system and I don't care about NOS/OS anymore.

 

[TKYR1967]

well I am not sure if it is NOS vs OS or the tube pre-amp stage that is making the difference but I can say EASILY that I prefer the sound of the Peachtree Decco to the PS Audio DL III and the Bel Canto DAC 3 that I owned previously.

 

[KMASCII]

Don't confuse bit depth (amount of data, 16bit vs 24bit, etc) with the sample rate (number of times the data is presented, 44.1khz vs 96khz, etc). I believe if your changing the bit depth such as 16 bit to 24 bit, this involves converting the data. But if you're adjusting the sample rate, such as 44.1khz to 96khz, this is resampling. From my understanding, resampling is an attempt to better represent a natural anolog sine. Which would be represented by an infinite/continuous sample rate.

 

Although I can understand the benefits of resampling, I can't see any benefit to increasing the bit depth of the source.

 

[sshd]

Resampling has no benefits in the digital domain - that is representing the analog signal.

 

Visualise two dots on a piece of paper and draw a straight line between them. The two dots represent the line and every possible dot on that line.

 

Now visualise three dots on a piece of paper. One dot is the starting point, one the middle and one the end point. Now draw a perfect soft curve through the three dots (there is only one perfect curve through the three dots). These three dots represent the curve and every possible dot on that curve.

 

Now think of the curve as part of an analog signal. Adding more dots (upsampling) only increases the number of calculations. It does not add any accuracy or represent the signal better.

 

The bit-depth represents the signal leve. Increasing to 24 bit makes sense when modifying the source signal (performing DSP operations like volume or digital room correction).

 

[KMASCII]

Thanks sshd. I do understand that upsampling is inserting dots. lol.

 

[Fyper]

Resampling or upsampling also has consequences on jitter.

Upsampling will reduce the "noise" created by jitter to level which are much less perceptible by human ear (4X upsampling will push the noise back by -6db and we are already in the -100db and less levels).

It will however spread this "lowered noise" over a larger frequency spectrum.

In other words it will level those noise peaks created by jitter off, but spread them over a larger frequency.

 

Therefore, with a high jitter source, it may interesting to use upsampling to reduce perceptible jitter and have consistant sound over time.

With a low jitter source, it may be interesting not to spread too much noise (even if at very low level) and have a "cleaner" sound (less "noise") by keeping the original sampling rate.

If you've have very low jitter, upsampling will have almost no effect on it (doesn't mean that it may not have other effects such as those written above though).

 

My current DAC proposes different sample rates from "original" to "X4". I must admit that it is honestly quite difficult to hear significant differences. Would that be because I have a low jitter system?

Maybe the upsampling brings a more liquid, more flowing sound whereas keeping the original rate seems to be slightly more precise on details. I never really sat down and compared though.

 

[bdiament]

The suggestion that oversampling reduces jitter doesn't make sense to me. As the sample rate goes up, so do the demands on clocking in the DAC. This in fact, results in *more* jitter on those DACs with less than wonderful clocking. I believe this is one reason why some of these DACs actually sound better when used at lower sample rates.

 

Aside from this, I have compared a few dozen sample rate algorithms for use in my work. (I record at 24/192 and must convert to lower rates for 24/96 DVD and 16/44 CD release.) What I've found is that even among the off-line (i.e. outside of real time) SRC algorithms, the majority will among other things, generate spurious harmonics, not present in the original signal. The result is a subtle to not-so-subtle brightening and hardening of the sound when comparing the result with the unconverted original.

 

Many of these do a slightly better job at integer conversion (e.g. 88.2 to 44.1 vs. 96 to 44.1), which is why I believe many folks use 88.2 or 176.4 instead of 96 and 192 for their recordings. The most transparent of these, does an equally good job (in my experience) whether performing integer conversion or non-integer conversion.

 

What the best have in common is they tend to take a while to perform their magic. This is to say, with most off-line SRC not doing a particularly great job, even when given lots of time, I would not expect real-time SRC to provide transparent results. And in fact, the converters I've found most transparent do not perform any sort of sample rate manipulation.

 

To be clear, 192, when done well, provides the best sound I've ever experienced. But I want my DAC to play sources at their native rates: 192 if the source is 192 but not when it isn't.

 

By the way, with my Metric Halo ULN-8, 24/192 sounds like I'm listening directly to the mic feed!

 

Best regards,

Barry

www.soundkeeperrecordings.com

www.barrydiamentaudio.com

 

 

[cfmsp]

 

Barry,

 

Glad to read your posts here!

 

As another Metric Halo fan, I run across your posts from time to time, and almost always am surprised at how much I am in agreement with your words.

 

Clay

 

PS, Ditto for Brian Willoughby, wherever you are!

 

[Fyper]

Hi, I didn't say that oversammpling reduced jitter. Jitter is there or not and sampling rate won't make it disappear. However this theory says that upsampling reduces the magnitude of jitter and spread it over more frequencies than the original jittery one.

Here is the link of the theory, but unfortunately it is in French so won't help you much... (that, by the way, may be the reason why my English may look strange or wrong sometimes....)

here

 

I'm with you when you talk about the brightening and hardening of the sound when Hirez original is converted to lower sample rate.

It then makes sense that this same downsampled sound if upsampled again will actually also include and multiply these "spurious harmonics" and possibly make it sound less good than at original sample rate.

 

I'm not sure I understand the comment about the fact that good downsampling algorithms take time to operate their magic. I mean I understand this, but the upsampling operation doesn't have to be as sophisticated as downsampling algorithms (no?).

To use the same image as above: it is more complicated to reduce the number of points in a curve and still make it look like the original curve than to add points between already existing points.

Upsampling can't recreate an original hirez signal but, a part possibly from these harmonic distortions added to the signal, no information is lost compared to the original and I still think that it can be beneficial in some cases, less in others.

 

Besides, I'm not hundred percent convinced by this but a number of ABX double blind tests showed that most human beings can't make a difference between 16/44 and 24/96 and I'm not talking about 192...

This may be linked to the quality of systems used for these tests and/or the limits of human hearing...

 

[CG]

"What I've found is that even among the off-line (i.e. outside of real time) SRC algorithms, the majority will among other things, generate spurious harmonics, not present in the original signal. The result is a subtle to not-so-subtle brightening and hardening of the sound when comparing the result with the unconverted original."

 

If it's not a trade secret, could you suggest some that you find to be good? Thanks in advance...

 

[bdiament]

Hi Fyper,

 

You said:

"However this theory says that upsampling reduces the magnitude of jitter and spread it over more frequencies than the original jittery one."

 

To me, his is suggesting oversampling reduces jitter or at least it suggests oversampling reduces the audibility of jitter. I'm sorry my understanding of French is limited to what I learned in high school, so I cannot address the writings in the link you provided.

 

Perhaps the author is confusing jitter with other things the a high sample rate

[bdiament]

Hi CG,

 

I've heard some positive things about Sox (used in its Linear Phase setting) but by far my favorite -the one I find to produce results that sound most like the unconverted original- is Alexey Lukin's algorithm, sold as iZotope 64-bit SRC. (I prefer its Steep, No Alias setting.)

 

Mac users can get it in Audiofile Engineering's Wave Editor and in their Sample Manager (but the former currently offers more control).

 

It is also available in iZotope's RX Advanced (both Mac and PC).

 

An interesting web site with graphic comparisons of many SRC algorithms can be found at http://src.infinitewave.ca

I find the Sweep and 1 kHz tests to be most informative and to correlate extremely well with what I hear in the many I've tested.

(Look at the 1k test for iZotope, Steep, No Alias and compare it with any of the others.)

 

Best regards,

Barry

www.soundkeeperrecordings.com

www.barrydiamentaudio.com

 

 

[bdiament]

Hi Clay,

 

Glad to find another MH user here.

Brian's here too? Cool.

 

I can't stop raving about my ULN-8. So many things about it: the conversion (A-D and D-A, the mic pres, the software console and record panel, the last being the best recorder I've ever used regardless of price or format, etc.)

 

Hope Chris gets to review one here.

 

Best regards,

Barry

www.soundkeeperrecordings.com

www.barrydiamentaudio.com

 

 

[Olive Empithrie]

Very nice to see you posting here.

 

I'm one of the many idiots who thought it was a good idea at the time to sell all my original Led Zep CDs :-(

 

All the best

 

Olive

 

[PeterSt]

Through the posts it looks like you are talking about inboard upsampling, offline upsampling and filtering and everything, while the DACs you use are sigma-delta types, hence massively oversample to begin with *and* filter to remove the HF noise from the oversampling.

 

Although applying something like SoX and the like will change the sound, IMO there is no way you can make it better because all starts with the result product of the filtering in the DAC chip.

 

Also, while these DACs will oversample 64 times to begin with (can be 256 times as well), I don't see the point in starting with e.g. 4 x upsampling yourself. 64 would change to 68 ...

 

Just 2c.

Peter

 

[Audio_ELF]

Surely if you upsampled from 44.1 to 176.4 (4x) then applied 64 times oversampling ... you would be "upsampling" 256x (4x64) not 68 times?

 

Nothing to do with debate if upsampling and oversampling is good or not, just the maths.

 

Eloise

 

[PeterSt]

That seems the only thing I am capable of today !

 

I stand corrected. Djeezz ...

 

But this makes me think now;

Suppose you have a DSD compatible chip (fs 512) and you feed that with 4 x upsampled material ... it should take that in consideration and multiply with 64 in that case.

So would an fs64 chip I guess (multiply with 16 instead of 64).

 

This would mean that your own upsampling doesn't make a difference, apart from the possible (no, assumeable) different used algorithm.

 

1c ?

:-)

 

[idiot_savant]

I believe the way it works is that the DAC runs internally at the same frequency, so if you have a native 16x oversampling DAC (say), then the DAC oversamples/upsamples 16x at 44.1, 8x at 88.2, 4x at 176.4 etc.

Also, the filtering in an oversampling DAC is nothing to do with noise from the sigma-delta - the digital filtering is done before this stage, and is done to remove the images above 22.05kHz. This is why NOS DACs tend to measure so badly - the (large) images above 22.05 can cause IM products in the audio band, depending on the equipment connected to the DAC ( amp, speakers etc ) - to verify this, look at a scope showing anything above an 11kHz sine on a NOS, and the same signal on a properly designed oversampling DAC...

 

[cfmsp]

 

"Brian's here too? Cool."

 

Haven't seen him post recently, I believe he posted a couple of times some months back.

 

cheers