Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Resampling and Fidelity (Read 24877 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Resampling and Fidelity

Reply #50 –
Quote from: pbelkner on
What's wrong with my argument that in general up-sampling a discrete (i.e. digital, stair case function) signal results into a more smooth (please allow me to speak of a "more smooth" discrete function in time) and more dynamic discrete signal?


The reconstruction filter determines the smoothness of the output wqveform, not the sample rate, If the reconstruction filter is doing its job, waveform has the same degree of smoothness  (freedom from short-term irregularities) regardless of the sample rate.

If we sample a 1 KHz tone at 2.205, 4.41, 11.025, 22.05 or 44.1 KHz, the reconstructed wave is identically the same if the reconstruction filter is identically the same, only scaled to suit the sampling rate.

Since the reconstruction filter is usually a digital filter, this degree of matching across various sample rates is not only possible but  guaranteed,

Resampling and Fidelity

Reply #51 –
Quote from: saratoga on
Quote from: WernerO on
Where is stated that the Hifiman is not oversampled? There happens to be a DF1704 digital
filter on its board.


So they used a non oversampling DAC, but oversampled using a second chip?  I wonder why it has such bad high frequency roll off then.



A little more research shows that the DF 704  digital reconstruction filter chip they used has an optional "slow roll off" feature whose frequency response pretty well predicts the relatively poor high frequency response of the Hifiman portable digital player. The chip also has a "sharp roll off" feature that provides frequency response that is about as flat and extended as you could reasonably hope for.

My opinion is that the slow roll off mode was originally intended for use at far higher sample rates than 44.1 KHz, and thus intended to not cause audible effects. There is probably some TI engineer who pulled on his hair pretty hard when he discovered how the Hifiman is abusing it.

The chip can be programmed for either mode by connecting pins on the chip to power or ground,  or by loading data into its internal registers. All pretty usual stuff.

Resampling and Fidelity

Reply #52 –
Quote from: Arnold B. Krueger on
Since the reconstruction filter is usually a digital filter

Do I misunderstand Wikipedia?
[blockquote]A DAC converts an abstract finite-precision number (usually a fixed-point binary number) into a concrete physical quantity (e.g., a voltage or a pressure). In particular, DACs are often used to convert finite-precision time series data to a continually varying physical signal.

http://en.wikipedia.org/wiki/Digital-to-analog_converter[/blockquote]That's exactly what I expect from a DAC. As far as I understand this statement the result of this step should already be in the analogue domain.
[blockquote]A typical DAC converts the abstract numbers into a concrete sequence of impulses that are then processed by a reconstruction filter using some form of interpolation to fill in data between the impulses.

http://en.wikipedia.org/wiki/Digital-to-analog_converter
[/blockquote]This is a bit open to interpretation. But to read it consistently with the above qoute I have to interpret the term "impuls" as an anlogue impuls.
[blockquote]In systems that have both, the anti-aliasing filter  and a reconstruction filter may be of identical design. For example, both the input and the output for audio equipment is sampled at 44.1 kHz. Both audio filters  block as much as possible above 22 kHz and pass as much as possible below 20 kHz. Typically both filters are active op-amp filters, with exactly the same selection of resistors and capacitors.

http://en.wikipedia.org/wiki/Reconstruction_filter
[/blockquote]Oops, "op-amp" and "resistors and capacitors" finally sound like analogue domain.

Resampling and Fidelity

Reply #53 –
Quote from: pbelkner on
Quote from: Arnold B. Krueger on
Since the reconstruction filter is usually a digital filter

Do I misunderstand Wikipedia?
[blockquote]A DAC converts an abstract finite-precision number (usually a fixed-point binary number) into a concrete physical quantity (e.g., a voltage or a pressure). In particular, DACs are often used to convert finite-precision time series data to a continually varying physical signal.

http://en.wikipedia.org/wiki/Digital-to-analog_converter[/blockquote]That's exactly what I expect from a DAC. As far as I understand this statement the result of this step should already be in the analogue domain.
[blockquote]A typical DAC converts the abstract numbers into a concrete sequence of impulses that are then processed by a reconstruction filter using some form of interpolation to fill in data between the impulses.

http://en.wikipedia.org/wiki/Digital-to-analog_converter
[/blockquote]This is a bit open to interpretation. But to read it consistently with the above qoute I have to interpret the term "impuls" as an anlogue impuls.
[blockquote]In systems that have both, the anti-aliasing filter  and a reconstruction filter may be of identical design. For example, both the input and the output for audio equipment is sampled at 44.1 kHz. Both audio filters  block as much as possible above 22 kHz and pass as much as possible below 20 kHz. Typically both filters are active op-amp filters, with exactly the same selection of resistors and capacitors.

http://en.wikipedia.org/wiki/Reconstruction_filter
[/blockquote]Oops, "resistors and capacitors" finally sounds like analogue domain.


The wikipedia article is sort of obsolete.  Digital reconstruction filters were used in the origional first generation Philips CD-7 CD player (1982-1983). The contemporaneous Sony CDP 101 used passive analog reconstruction filters (but there were no op amps in the filters). In the second and suceeding generation CD players, the reconstruction filters used by all manfuacturers were almost always digital.  This trend continues to this day.

Now for the giant but...  Digital domain sharp cutoff reconstruction filters are generally followed by additional filters that deal with artifacts related to the oversampling frequency (typically 8x Fs). These filters are implemented in the analog domain, and may be active filters composed of resistors, capacitors and op amps.

Resampling and Fidelity

Reply #54 –
Thanks a lot, Arnold.

Quote from: Arnold B. Krueger on
In the second and suceeding generation CD players, the reconstruction filters used by all manfuacturers were almost always digital.  This trend continues to this day.

Is this architecture ment by the term "up-sampling DAC"?

Do you know whether the E-MU 0404 USB is based on this design?

Resampling and Fidelity

Reply #55 –
Quote from: pbelkner on
Do you know whether the E-MU 0404 USB is based on this design?


Upsampling and oversampling are mathematically identical, although depending on the precise
context sometimes one term is preferred over the other.
In audio marketing it is something else. After a decade of 'oversampling' someone started using 'upsampling' to distinguish, and sell, his products. Ironically, what most don't know is that the technical documentation of Philips' 1981-era digital filter used the term 'upsampling'. Really.

And yes, 99.99% of audio DACs and ADCs since 1985, indicating chips as well as complete components, use oversampling and digital anti-imaging respectively anti-aliasing filters.


BTW Wikipedia and many many internet sites are notoriously wrong, inaccurate, or vague when it comes to Shannon's theorems and it implications and implementations in the audio world.

In this respect the original sin was commited by Philips, whose marketing talk on oversampling also was utter bollocks.


Resampling and Fidelity

Reply #56 –
Quote from: pbelkner on
Is this architecture ment by the term "up-sampling DAC"?

Do you know whether the E-MU 0404 USB is based on this design?


That's like asking whether your new Volkswagen's petrol engine is also based on this four-stroke cycle design instead of a Wankel engine.

Resampling and Fidelity

Reply #57 –
Great, very informative thread, but I still have a question regarding the internal workings of an oversampling DAC.

Is it accurate to say that once a (for example) 44.1 kHz PCM bitstream "gets in the front door", so to speak, of an oversampling DAC, we never "see" a 44.1 kHz bitstream again?

That is, a DAC that is converting a 44.1 kHz stream is actually "running" at 352.8 kHz (or much higher), so it's really just a matter of what clocks and multipliers it's designed with?

I guess what I'm trying to get at is just how ridiculous this whole obsession with "upsampling DACs" is...that, as WernerO said, oversampling DACs *are* upsampling DACs, and therefore virtually every DAC made since the mid-80's is already an "upsampling" DAC...it just doesn't have that magic word printed anywhere on the box.
"Not sure what the question is, but the answer is probably no."

Resampling and Fidelity

Reply #58 –
Quote from: mixminus1 on
Great, very informative thread, but I still have a question regarding the internal workings of an oversampling DAC.

Is it accurate to say that once a (for example) 44.1 kHz PCM bitstream "gets in the front door", so to speak, of an oversampling DAC, we never "see" a 44.1 kHz bitstream again?

That is, a DAC that is converting a 44.1 kHz stream is actually "running" at 352.8 kHz (or much higher), so it's really just a matter of what clocks and multipliers it's designed with?


Yes. A digital lowpass is applied at the oversampled rate with a cutoff frequency of about half the input sample rate. The artifacts caused by oversampling (far above the base band) can be removed with a simple analog lowpass afterwards.

Quote from: mixminus1 on
I guess what I'm trying to get at is just how ridiculous this whole obsession with "upsampling DACs" is...that, as WernerO said, oversampling DACs *are* upsampling DACs, and therefore virtually every DAC made since the mid-80's is already an "upsampling" DAC...it just doesn't have that magic word printed anywhere on the box.


It's really mainly marketing blur, at least regarding theory and actual performance differences. In practice, those DACs, which are sold with the label 'upsampling', often have a discrete resampler before a regular oversampling DAC. The resampler does nothing conceptually different, oversample and lowpass filter, even sometimes at non integer ratios (which doesn't have to be worse but usually not better). So technically it is not superior in any way over a regular, oversampling DAC, that houses the complete pipeline in a single chip enclosure. Sometimes the resampler chips have additional features, as asynchronous sample rate conversion, which is great to eliminate interface jitter. They can indeed be found in excellent products. On the other hand, there are also single-package (regular, "oversampling") solutions with excellent jitter tolerance, which don't need a separate IC to accomplish that task. To cut a long story short: look at the specs and not the marketing bla. Even shorter: don't care about it, nowadays the $1.50 audio codec on your budget motherboard does all this already better than your ears could discern outside an anechoic chamber. *

* Or with synthetic test samples over very sensitive, noise-blocking headphones at very high volume levels. But already a $2.50-$5.00 part would also take care of those...

Resampling and Fidelity

Reply #59 –
Fascinating to see how much interest my short comment sparked.

One topic that was not covered by the excellent technical and empirical posts:
What happens if someone resample their 44.1kHz CD audio to some high intermediate rate using software, then transmit the data across a really, really bad spdif connection to a really, really bad ("boutique"?) external DAC with little jitter suppression? Let us assume perfect sinc lowpassfilters (or as perfect as time permits) all over the place. Would she be better of leaving the signal at 44.1kHz then?

-k

Resampling and Fidelity

Reply #60 –
Quote from: pbelkner on
Thanks a lot, Arnold.

Quote from: Arnold B. Krueger on
In the second and suceeding generation CD players, the reconstruction filters used by all manfuacturers were almost always digital.  This trend continues to this day.

Is this architecture ment by the term "up-sampling DAC"?


Yes.

Quote
Do you know whether the E-MU 0404 USB is based on this design?


Yes.

Article about eMu 0404 USB intersnals

"•24-bit 192 kHz AKM AK5385A ADC and AK4396 DAC "

Looking up the AK4396 on the AKM web site:

AKM 4396 spec sheet

"24 Bit 8x Digital Filter (Slow-roll-off option)"


Resampling and Fidelity

Reply #61 –
Quote from: knutinh on
Fascinating to see how much interest my short comment sparked.

One topic that was not covered by the excellent technical and empirical posts:
What happens if someone resample their 44.1kHz CD audio to some high intermediate rate using software, then transmit the data across a really, really bad spdif connection to a really, really bad ("boutique"?) external DAC with little jitter suppression? Let us assume perfect sinc lowpassfilters (or as perfect as time permits) all over the place. Would she be better of leaving the signal at 44.1kHz then?


That's a rhetorical question, right? ;-)

Resampling and Fidelity

Reply #62 –
Thanks much, googlebot, and everyone else for your comments and responses in a very informative, yet eminently understandable thread.
"Not sure what the question is, but the answer is probably no."

Resampling and Fidelity

Reply #63 –
Quote from: googlebot on
Quote from: mixminus1 on
I guess what I'm trying to get at is just how ridiculous this whole obsession with "upsampling DACs" is...that, as WernerO said, oversampling DACs *are* upsampling DACs, and therefore virtually every DAC made since the mid-80's is already an "upsampling" DAC...it just doesn't have that magic word printed anywhere on the box.

It's really mainly marketing blur, at least regarding theory and actual performance differences. In practice, those DACs, which are sold with the label 'upsampling', often have a discrete resampler before a regular oversampling DAC.
The original intent is that the separate "upsampling" resampler is higher quality (e.g. includes more accurate, and hence more computationally expensive digital filters) than the "oversampling" resampler in the DAC. Alternatively the "upsampling" resampler uses some "favourite" filtering algorithm of the designer, rather than whatever is available in the "oversampling" resampler. The oversampling is a commodity - off the shelf - unavoidable. The separate upsampling is supposed to be better.

If there is any intended difference in meaning, this is it. However, don't assume everyone understands this difference. Also, the meanings of words don't tend to survive their use by a marketing department.

Of course the upsampler and resampler perform the same operation (resampling), and could be identical. One DAC's oversampler could even be better than a specific resampler.

What matters more is the overall response.

What matters even more is that people can't agree on what the overall response should be! Plenty of evidence that the "ideal" response isn't the best choice for audio.

What matters most is whether you can hear any difference!

Cheers,
David.

Resampling and Fidelity

Reply #64 –
Quote from: 2Bdecided on
What matters even more is that people can't agree on what the overall response should be! Plenty of evidence that the "ideal" response isn't the best choice for audio.


While the ideal reconstruction filter is a given, no such best approach exists for
the recording-side anti-aliasing filter, and the total system response
is of course the concatenation of the two filters.

There are two dominant AA filter architectures found back in commercial CDs:

-the historical Sony PCM1610/1630 analogue AA filters, with quite significant phase
distortion but apparently without aliasing. (The true 80s CDs in my collection I investigated
show remarkably little content above 21 kHz. However, I have never seen
the actual response graphs of these ADCs.)

-half-band digital FIR filters, as found in delta-sigma ADCs. These are -6dB down at
Fs/2 (or a multiple of 6dB) and thus exhibit some aliasing, violating the sampling
theorem.

Oh, there is a third category, that of all half-assed software resamplers spread
through the recording industry, with the more expensive pro tools often
the most compromised...