16/44.1 vs Oversampling

[LAMF]

Since almost every cd is recording according to the standard Red Book, can you guys please explain to me which are the sonic gains using an oversampling DAC?

 

[Roseval]

The DA conversion creates multiple images of the audio.

To get rid of them the first generation DACs uses a very steep low pass filter (brick wall).

This filter causes all kind of artifacts.

As these images are dependent on the sample rate, upsampling pushes them far out of the sonic range, allowing a much ‘softer’ low pass filter.

It is much better explained here: http://www.thetadigital.com/upsampling.htm

 

 

[BobH]

Don't take this the wrong way, but the answer is whichever sounds best to you! It's another 'jitter' argument! There are technical reasons for and against, as with all things audio, it seems I cannot really tell a difference, on my system, between 44, 96 & 192. I can tell a very real difference between 16 & 24. I personally think non-oversampling dac's sound more musical, others will disagree. For me my choice is to go with 24/44.1 fed into a non-oversampling dac. Chris' reviews of various dac's, on this site, may give you a bit more insight as he does cover his own preferences, where a choice is available.

 

Short of the technical why's and wherefore's, the only way you can tell is to listen and see which you prefer. There is no right and wrong!

 

[PeterSt]

I don't think this is related to jitter of any sort. Although it is very true to state "whatever sounds best to you" I concur with Roseval for the theories.

It is not that simple though, because oversampling by itself causes anomalies (or fake hence wrong resolution if you want).

 

What is important though (and the fact what sounds best to you still counts !) is an oversampling DAC that does it for the reasons Roseval explained vs. a DAC that won't operate otherwise (like a sigma-delta DAC).

Both have very different merits. Note that the first category may use 8 x oversampling for the good sake of avoiding the filter, while the second category just *needs* 256 (etc.) times oversampling.

 

The latter is no virtue but just necessary. If you perceive it as a virtue ... OK.

 

Peter

 

[LAMF]

Thank you guys for the input.

I’d have a few DAC’s since the early 90’s – i still have my (heavily modded) Audio Alchemy V1.1 – and a few months ago I bought the Pro-Ject USB Box because I listen to a lot of music in my comp. I have compared it with a friend of mine’s DacMagic and to be honest I can’t tell any sonic differences, but i'm not one that believes in significant differences between competently-designed DACs, even though i expected a little more magic from that unit.

 

[audioengr]

I actually like the sound of good 24/96 upsampling, however not with hardware. I've modded more than a dozen different manufacturers DACs over the years, so I know what these hardware upsamplers sound like, even the custom-designed ones.

 

I much prefer S/W upsampling to 24/96 with SRC, or re-writing the files with R8Brain or Adobe Audition.

 

Upsampling to 24/96 with a good S/W algorithm like SRC does the following:

 

1) cleaner percussion, less edgy

2) smoother vocals and cymbols

3) improved dynamics

 

BTW, some interfaces also change the data from 16/44.1 into 24/44.1. This is not upsampling, but does improve the noise floor.

 

Oversampling for digital filtering is a bit different than upsampling BTW.

 

Steve N.

Empirical Audio

 

[Roseval]

I actually like the sound of good 24/96 upsampling, however not with hardware

Any queue why SW implementations sounds better than HW ones?

 

[PeterSt]

Hi Steve,

 

IMO there are so many things not right in your post because of lacking context, that I better stick to :

 

If you need S/W upsampling to begin with, then you have the wrong DAC.

Yeah, we heard that one before. :-)

 

But keep in mind what I said : because of lacking context.

On another side I'd leave it like that and don't eleborate because few people in here will gain from it I think.

 

Peter

 

[audioengr]

"Any queue why SW implementations sounds better than HW ones?"

 

I think you mean "clue". Sure, hardware implementations are usually limited with chip size and cost and the fact that they have to do the algorithm on-the-fly. Also, the chip designers dont always know the best algortihms.

 

Not all software upsamplers sound good. SSRC that comes with Foobar2000 is not good.

 

Steve N.

 

[audioengr]

"If you need S/W upsampling to begin with, then you have the wrong DAC.

Yeah, we heard that one before."

 

Huh? Who are you talking about?

 

[PeterSt]

I was joking Steve. You said similar to me over at AA a few weeks back. I thought you'd get that.

 

And not to start the subject afterall (where I wouldn't and won't), I was implying what you just said in the other post : First have a DAC that actually needs it because it's not in there (which would be explicit design) and then do it in software.

But there's more to it.

 

Peter

 

[ASB]

I have also converted 100+ albums ripped from CDs' as 16/44.1 using r8brain. I used the 24bit/88.2khz sample rate, as I found it better than the 24/96.

 

I liked the sound of the modified files much better than the original files.

 

You can test and see if you like it. It is for free.

 

http://www.voxengo.com/product/r8brain/

 

[Roseval]

Converting from 16 to 24 is trivial, padding 8 bits will do.

Upsampling form 44.1 to 88.2 is not difficult either, it is a factor of 2, an integer.

Upsampling to 96 can’t be done with integer arithmetic, you have to do some kind of interpolation. The results of these type of upsampling is more complex and therefore more prone to yield artifacts.

 

 

[The Computer Audiophile]

Hi Roseval - This is exactly why Reference Recordings did a D to A to D conversion for its 96k content on HDtracks if it originally was 176.4 digital.

 

[Rob]

One argument against a 16/44 DAC is the fact that it can't take advantage of 24/96 downloaded files from the likes of Linn etc. and the question is, if this is where audio is going, where does that leave the 16/44 DAC which some audiophile companies seem to tenaciously cling to (particularly on their USB input)? The pro-audio community has all but fully embraced 24/96+. The argument about non-upsampling, non filtered DACs vs the other becomes a moot point in part when you look at how the amp, speakers, and interconnects, deal with the signal coming from the DAC.

 

Rob

 

[PeterSt]

Hi Roseval - This is exactly why Reference Recordings did a D to A to D conversion for its 96k content on HDtracks if it originally was 176.4 digital.

 

Hi Chris. I assume you heard that from themselves, so I assume this is true. But if true indeed, this stupid. Says me.

If the original would have been 176.4, why not go to 88.2 ? Or must it go onto a DVD and a DVD doesn't dig 88.2 ? (I really don't know).

 

176.4 * 96 = 16934.4

16934.4 * 10 = 169344

88.2 * 10 = 882

169344 / 882 = 192

192 / 2 = 96

 

All integer for the divisions ...

(but AA is needed at downsampling; this is another matter)

 

 

 

[The Computer Audiophile]

Hi PeterSt. Yeah, I talked to RR directly about this a while ago. The requirements were 24 bit / 96 kHz. If RR could have use an 88.2 version I'm sure it would have been a no-brainer. To say Keith Johnson knows what he's doing is the understatement of the century.

 

[weiss2496]

Here are some facts about upsampling etc.:

 

- Whether an upsampled DAC sounds better than a non-upsampled depends on the particular implementation. There is nothing like "one is better than the other one".

- A software based upsampler is not necessarily better than a hardware based one. Again, depends on the implementation.

- upsampling and oversampling is the same in my book. Upsampling usually means an upsampling done separate to the DAC while oversampling means the same but within the DAC (chip).

- A 44.1/24 signal generated from a 44.1/16 one does not have better SNR. You would have to use a denoiser to get better SNR. But those claiming better SNR by just padding bits to a 16 bit signal are misleading the innocent.

- Whether additional upsampling done in front of an already upsampling DAC sounds better, again depends on the implementation of those upsamplers. No general rule.

 

Daniel

 

 

[LAMF]

Many tanx for the input Daniel.

 

[PeterSt]

Oversampling : a DAC which just needs that in order to operate, like a sigma-delta. It goes way up to 256fs or more, and then down to the output rate.

 

Upsampling : from the source to a higher rate output.

 

Both may mix, though this is usually useless if the first is able to output at the desired output rate of the second.

 

A multi bit DAC (meaning as much bits as the output implies) does not need any kind of oversampling, like the first (sigma-delta). It may upsample (and output at that upsampled rate) to soften or even avoid the filter.

 

An oversampling DAC makes sines of squares, and therewith loses precious data (harmonics);

Whether the NOS DAC is better for net result depends on the filtering, but with proper filtering it WILL.

 

Please point out where I am wrong in the above, and make it as technicle as you need.

Peter

 

 

 

 

[audioengr]

Peter said:

"I was joking Steve. You said similar to me over at AA a few weeks back."

 

I guess I get so much FLAC that I though this was more of the same.....

 

Steve N.

 

[audioengr]

"Upsampling to 96 can’t be done with integer arithmetic, you have to do some kind of interpolation. The results of these type of upsampling is more complex and therefore more prone to yield artifacts."

 

You would think so, but all of the S/W upsamplers I have tried sound poor with 88.2 and much better at 24/96.

 

Steve N.

 

[Max]

PeterSt said:

An oversampling DAC makes sines of squares, and therewith loses precious data (harmonics);

Whether the NOS DAC is better for net result depends on the filtering, but with proper filtering it WILL.

 

Please point out where I am wrong in the above, and make it as technicle as you need.

 

OK, here goes

 

Recreating the analogue signal from the PCM data takes two steps: the first is the digital to analogue conversion stage, the second is the sampled to continuous time stage. These correspond to the DAC and the reconstruction filter. An oversampling DAC includes (most) of the reconstruction filter within the DAC itself. Hence, if you compare the output of a NOS DAC chip, which doesn't do any reconstruction, with that of an OS DAC, which does, they will look different. The NOS DAC will indeed include 'harmonics' which are actually artefacts, to be filtered out in the next stage of conversion.

 

Nyquist tells us that a 44.1 ksps PCM signal cannot convey a 12kHz square wave. If you feed a 12kHz square wave and a 12kHz sine wave separately into the ADC of such a PCM system, and adjust the levels properly, you will get EXACTLY the same sequence of sample data.

 

Hope this helps!

 

Max

 

[weiss2496]

Upsampling to 96 can’t be done with integer arithmetic, you have to do some kind of interpolation. The results of these type of upsampling is more complex and therefore more prone to yield artifacts.

 

The necessary steps to be taken are the same in both cases (44.1 to 88.2 or 44.1 to 96). Both need interpolation. One way to implement it is with polyphase filters. In the case of 44.1 to 88.2 there is just one such filter required, i.e. the one which calculates the sample which is in between two 44.1 samples in order to get to the 88.2 grid of samples. In the case of 44.1 to 96 there are also intermediate samples to be calculated but not nicely positioned in the middle between two samples but rather on a grid of possible positions as implied by the 44.1 to 96 ratio. A bank of polyphase filters takes care of all those intermediate positions, i.e. for each intermediate sample to be calculated a different polyphase filter is engaged. The number of polyphase filters required for a given sampling rate ratio is exactly defined.

So in essence the quality of those filters determine the quality of the conversion. 44.1 to 96 can be designed as good as 44.1 to 88.2 or vice versa. It is not a matter of integer ratio conversions. The case 44.1 to 96 requires a larger filter to be designed at the design stage of the sampling rate converter (i.e. the whole bank of those polyphase filters), but once these filters are designed the program executing the sampling rate conversion requires the same amount of DSP power whether it is 44.1 to 88.2 or 44.1 to 96.

If an SRC doing 44.1 to 88.2 sounds poorer than one doing 44.1 to 96 it is that particular implementation, it does not have anything to do with the physics of SRCs.

See http://en.wikipedia.org/wiki/Sample_rate_conversion

 

An oversampling DAC makes sines of squares, and therewith loses precious data (harmonics);

 

An answer to this false statement would need a lengthy article on the basics of D/A conversion and reconstruction filters. I suggest to read the wikipedia article I give below. I try to give a summary: The staircase output of a DAC has a frequency content which goes way beyond the audio band. In fact the audio spectrum is repeated at multiplies of the sampling rate (with some attenuation). The usable content of audio signals goes from 0 Hz to half the sampling rate, i.e. harmonics can not be higher in frequency than half the sampling rate (called Nyquist frequency) and that is 22.05 kHz in the case of 44.1kHz sampling. It is a good idea to suppress frequencies above Nyquist in order not to upset subsequent stages in the audio chain. With non oversampling DACs that is done with an analog filter which ideally passes audio up to the Nyquist frequency. The purpose of an oversampling DAC is to make the job easier for the analog filter in that the Nyquist frequency is higher with oversampling (e.g. 48 kHz at a 96 kHz sampling rate). Thus the analog filter may start to cut off at e.g. 20kHz and full suppression is at e.g. 48kHz. This makes the analog filter much simpler and thus phase response is much better.

See http://en.wikipedia.org/wiki/Digital-to-analog_converter

 

Oversampling : a DAC which just needs that in order to operate, like a sigma-delta. It goes way up to 256fs or more, and then down to the output rate.

 

It goes up, yes, but there is no point to go down again. A multibit converter can also be sigma delta, actually the modern sigma delta converters are multibit, e.g. 5 bits.

 

Daniel

 

 

 

[PeterSt]

Hi Max, thank you !

 

(btw, if you can find some lacking /cite in your last (and first hehe) post, maybe you can edit it in to avoid ever lasting italics in this thread)

 

"The NOS DAC will indeed include 'harmonics' which are actually artefacts, to be filtered out in the next stage of conversion."

 

This would be true for the "too square" output above 14700 Hz I think (not 12000) which indeed will create artificial "harmonics'. However, these are not the harmonics I referred to. I meant the harmonics from squares (squarish of course) from the source, those harmonics being squares themselves (in the end (n'th harmonic) indeed crossing the 14700 border and aliasing will start to emerge).

 

What I implied is :

 

a. the base squar(ish) sound will turn into a sine by heavy oversampling, so it can't recreate the harmonics in mid air anymore;

b. the harmonics present in the source itself also turn into sines and I can't tell what actually comes from that (apart from it not being reality anymore).

 

Furthermore, what I said is that an NOS DAC retains squares where they were there in the source, while the OS DAC makes sines of it. This is unrelated to the "reconstruction filter" because not any filter will make squares out of sines.

 

Peter