If I dare...
Greetings Quantran,
If I'm worthy enough of the huge pedagogical efforts made by the knowledged heads of this thread to explain to us, I would try to drop a basic answer... that might be rectified if needed by them. I had more or less your same question...
1. It's advisable but for most of people ears, the quality gain is not about curing the treble roll off but more of what you pointed out in your last sentence. My bet is the roll off some hear or not has more to see with the mixing of the reccording. I have a NOS dac and it doesn't suffer from high treble limitations... perhaps I'm deaf, biased, etc... Upsampling playback for Treble is not really the point for 99% of hifi we have. If one always suffers from missing treble with his NOS DAC, the culpritt is imho the post dac chip process : I/V, buffer, parts, layout... let's call that the output analog stage to be simple... Maybe also the quality of the clock if we stay in the digital side of your topic.
2.No it doesnt. In fact it's the opposit : upsampling software are proceeded by more powerfull processor that the onboard upsampling/filter board. SO they can provide more sophistacated filters in association with the upsampling process. Also on board upsampling/filtering chips suffer from some other limitations you also pointed out in your last sentence but non only. You can add jitter, on chip crosstalk, active and passive parts have a noticeable influence as well. So better a renderer or a pc as far you process to noise & groundloop isolation between them and your DAC.
Upsampling and filtering, which is the same process, will smooth the fragmented (stepped) sampled curve by linking up (so smooth in the geometrical sense of this word) the top of the squarred steps that the samples are with a more smooth AND continuous (unstepped) curve... at least for our ears as a sample stays a sample.
Samples are a quantity expressed by a height measured at same windows intervals - the more the samples the more the precision) : This last paragraph is my translation for myself of what I believe I understood... so can be false, lol ! Also upsampling are removing the artifacts you wrote in your last sentence : ringing, phantom images..
Hope that helps and I understand well what have been said above, they certainly add some pedagogical if not.
Greetings Quantran,
If I'm worthy enough of the huge pedagogical efforts made by the knowledged heads of this thread to explain to us, I would try to drop a basic answer... that might be rectified if needed by them. I had more or less your same question...
1. It's advisable but for most of people ears, the quality gain is not about curing the treble roll off but more of what you pointed out in your last sentence. My bet is the roll off some hear or not has more to see with the mixing of the reccording. I have a NOS dac and it doesn't suffer from high treble limitations... perhaps I'm deaf, biased, etc... Upsampling playback for Treble is not really the point for 99% of hifi we have. If one always suffers from missing treble with his NOS DAC, the culpritt is imho the post dac chip process : I/V, buffer, parts, layout... let's call that the output analog stage to be simple... Maybe also the quality of the clock if we stay in the digital side of your topic.
2.No it doesnt. In fact it's the opposit : upsampling software are proceeded by more powerfull processor that the onboard upsampling/filter board. SO they can provide more sophistacated filters in association with the upsampling process. Also on board upsampling/filtering chips suffer from some other limitations you also pointed out in your last sentence but non only. You can add jitter, on chip crosstalk, active and passive parts have a noticeable influence as well. So better a renderer or a pc as far you process to noise & groundloop isolation between them and your DAC.
Upsampling and filtering, which is the same process, will smooth the fragmented (stepped) sampled curve by linking up (so smooth in the geometrical sense of this word) the top of the squarred steps that the samples are with a more smooth AND continuous (unstepped) curve... at least for our ears as a sample stays a sample.
Samples are a quantity expressed by a height measured at same windows intervals - the more the samples the more the precision) : This last paragraph is my translation for myself of what I believe I understood... so can be false, lol ! Also upsampling are removing the artifacts you wrote in your last sentence : ringing, phantom images..
Hope that helps and I understand well what have been said above, they certainly add some pedagogical if not.
Last edited:
Hi diyiggy
Thank you for answering my question. I tend to agree that the treble roll off is hardly noticeable. The fact is that when I tested my hearing using a frequency generator, I can’t hear single tone higher than 15khz. However, there is also a theory that human can hear or feel complex tone higher than the limit for single tone sound.
With regards to the ringing effect, I understand that this is the side effect of oversampling. Different types of digital filters have different effect. This is why various dac have different DF options to choose from. The attached picture shows the ringing effect.
However, I am still not clear if the upsampling function by software also has such effects or not.
Thank you for answering my question. I tend to agree that the treble roll off is hardly noticeable. The fact is that when I tested my hearing using a frequency generator, I can’t hear single tone higher than 15khz. However, there is also a theory that human can hear or feel complex tone higher than the limit for single tone sound.
With regards to the ringing effect, I understand that this is the side effect of oversampling. Different types of digital filters have different effect. This is why various dac have different DF options to choose from. The attached picture shows the ringing effect.
However, I am still not clear if the upsampling function by software also has such effects or not.
Attachments
Well, fundamental of a triangle is 7 K Hz iirc, so...
This theory you're talking about is maybe to sell Super tweeter ??? It is a known fact there are some rare people hearing till 22/23 K Hz... Rare enough not to bother imo.
For the few I understand the ringing is not due to the oversampling but by the sample process, so as soon than the analog to digital process. Over/up-sampling is the counterpart needed to reject or to move it higher in frequency in the non hearable frequency area AND reconstruct a more accurate signal by this linking up/smoothing upsampling/filtering process during the playback !
At least that's what I understood. More complex up/oversampling filters from the software you're talking about are known to "sound" better than the simpliest one of the upsampling/filter on board chips. Pcs and renderers have more processor power to make possible more complex filters. Indeed Audirvana for Mac or Signalyst for W. are saying they have such filters they developp themselves and/or the choice to choose your filter if i don't mistake. SoX in the Linux world are a part of many playback software for renders as Raspberry Pi.
NOS is said to have better micro dynamic but maybe it's just a myth, I don't know. I'm going to test upsampling from a Rpi that feeds my NOS DAC to discover what I subjectivly prefer. No doubt for me todays one should avoid in board upsampling/filter chips... but if already embeded in the modern dac chips (AK, ESS, etc)
This theory you're talking about is maybe to sell Super tweeter ??? It is a known fact there are some rare people hearing till 22/23 K Hz... Rare enough not to bother imo.
For the few I understand the ringing is not due to the oversampling but by the sample process, so as soon than the analog to digital process. Over/up-sampling is the counterpart needed to reject or to move it higher in frequency in the non hearable frequency area AND reconstruct a more accurate signal by this linking up/smoothing upsampling/filtering process during the playback !
At least that's what I understood. More complex up/oversampling filters from the software you're talking about are known to "sound" better than the simpliest one of the upsampling/filter on board chips. Pcs and renderers have more processor power to make possible more complex filters. Indeed Audirvana for Mac or Signalyst for W. are saying they have such filters they developp themselves and/or the choice to choose your filter if i don't mistake. SoX in the Linux world are a part of many playback software for renders as Raspberry Pi.
NOS is said to have better micro dynamic but maybe it's just a myth, I don't know. I'm going to test upsampling from a Rpi that feeds my NOS DAC to discover what I subjectivly prefer. No doubt for me todays one should avoid in board upsampling/filter chips... but if already embeded in the modern dac chips (AK, ESS, etc)
Last edited:
Abraxalito's was probably the most succinct & clearest explanation so far in the thread - it would benefit you to study it & try to understand rather than knee-jerk
I have done some reading about an upsampling software Izotope and found that its behaviour is similar to the digital filter on a chip with respect to pre/post-ringing in time domain
resample
However depending on the sophistication of the software, the filter phase may be adjustable according to user preference. At the moment the common opinion seems to prefer minimum phase filter which has no pre-ringing and max post-ringing.
resample
However depending on the sophistication of the software, the filter phase may be adjustable according to user preference. At the moment the common opinion seems to prefer minimum phase filter which has no pre-ringing and max post-ringing.
A look at Cirrus/Wolfson, AKM, and ESS suggests their higher-spec chips are quite capable, and that dedicated hardware isn't buying you much. But people enjoy a wide range of solutions, and I'm sympathetic to the argument of doing whatever makes you happy. I'd argue people pursue the simplest/most straightforward (i.e. using the hardware given to us) first, and then diverging from there, but NOS DACs are inherently not as integrated.
Digital filtering and oversampling are EXACTLY THE SAME THING when it comes to D/A.
Philips isn't mincing words when they label in front of their players "Oversampling Digital Filter".
"Image spectra in nonoversampled and oversampled reconstructions. A. A brick-wall filter must sharply bandlimit the output spectra. B. With four-times oversampling, images appear only at the oversampling frequency. C. The output S/H circuit can be used to further suppress the oversampling spectra." From: Principles Of Digital Audio (Pohlmann, 2005)
So, by "pushing" the images well away from the orig. sampling (by oversampling) you have filtered the signal. All in the digital domain.
Now, class, repeat after me:
Digital filtering and oversampling are EXACTLY THE SAME THING
Digital filtering and oversampling are EXACTLY THE SAME THING
Digital filtering and oversampling are EXACTLY THE SAME THING
Digital filtering and oversampling are EXACTLY THE SAME THING
Digital filtering and oversampling are EXACTLY THE SAME THING
....
Philips isn't mincing words when they label in front of their players "Oversampling Digital Filter".
"Image spectra in nonoversampled and oversampled reconstructions. A. A brick-wall filter must sharply bandlimit the output spectra. B. With four-times oversampling, images appear only at the oversampling frequency. C. The output S/H circuit can be used to further suppress the oversampling spectra." From: Principles Of Digital Audio (Pohlmann, 2005)
So, by "pushing" the images well away from the orig. sampling (by oversampling) you have filtered the signal. All in the digital domain.
Now, class, repeat after me:
Digital filtering and oversampling are EXACTLY THE SAME THING
Digital filtering and oversampling are EXACTLY THE SAME THING
Digital filtering and oversampling are EXACTLY THE SAME THING
Digital filtering and oversampling are EXACTLY THE SAME THING
Digital filtering and oversampling are EXACTLY THE SAME THING
....
Digital filtering and oversampling are not exactly the same thing.
For example, upsampling can be accomplished by zero-stuffing. The next step is where digital LP filtering would be used, to complete the oversampling process.
That being said, digital filtering can be used for many other applications.
For example, upsampling can be accomplished by zero-stuffing. The next step is where digital LP filtering would be used, to complete the oversampling process.
That being said, digital filtering can be used for many other applications.
The quickest way to test it is just playing back 192K source with both type of DACs, because the ringing is most likely inaudible at that cutoff frequency. I doubt that there is any difference between them, or OS DAC should sound better due to newer architecture.
If you are OK with one second latency, a few years old Intel i5 can handle FIR at 260000 steps, I believe.
Is it ok to ressample off line, with FIR software like Izotope, some 16/44.1 with our filter choice and upsampling choice then playback this new material to a NOS DAC with the same result (phase, sampling rate,etc) instead the operation was made online (as per the discussion above) ?
... then of course playback through the Rpi if the clock of the NOS DAC permitt it ? Or these new upsampled materials will be just filled with zeros between the original sample rate
Not sure I understood difference between FIR filter and IIR filter in relattion to upsampling process,I do think I have to re read the thread
... then of course playback through the Rpi if the clock of the NOS DAC permitt it ? Or these new upsampled materials will be just filled with zeros between the original sample rate
Not sure I understood difference between FIR filter and IIR filter in relattion to upsampling process,I do think I have to re read the thread
Last edited:
diggy,
When you ask if something is okay, sure its okay. Don't expect it to sound better than a Chord Dave, of course. If you like the resulting sound then its good, that's all.
FIR and IIR are just two basic types of digital filters. There are different engineering design trade-offs with different filter types. Sometimes, both types may be used in some particular design.
Regarding zero stuffing, it simply involves something like, say, upping the sample rate 8 times and stuffing 7 samples with value zero between each of the original samples. Now the sample rate is 8 times higher and there are 8 times more samples. Only issue is that the original signal we are interested in was recorded in the frequency range of, say, 20Hz - 20kHz. However, the upsampling we did also created a bunch of images at higher frequencies that we don't want to run through our whole reproduction system as they can cause various problems and make for poor sound quality. So, we need to digital LP filter our upsampled digital audio to help separate out the 20Hz-20kHz audio frequencies we really care about reproducing faithfully, from all the other, more HF stuff we don't want. After passing through the digital LP filter, the resulting samples visually would look like a much smoother representation of the 20Hz-20kHz audio we do want. The better the filter we use, the more we will have attenuated the HF image effects of the zero samples we added (stuffed) and replaced them with more accurate (for our purposes) 20Hz-20kHz band-limited sample values.
EDIT: The above somewhat simplified with the hope that clarity can be arrived at in steps.
When you ask if something is okay, sure its okay. Don't expect it to sound better than a Chord Dave, of course. If you like the resulting sound then its good, that's all.
FIR and IIR are just two basic types of digital filters. There are different engineering design trade-offs with different filter types. Sometimes, both types may be used in some particular design.
Regarding zero stuffing, it simply involves something like, say, upping the sample rate 8 times and stuffing 7 samples with value zero between each of the original samples. Now the sample rate is 8 times higher and there are 8 times more samples. Only issue is that the original signal we are interested in was recorded in the frequency range of, say, 20Hz - 20kHz. However, the upsampling we did also created a bunch of images at higher frequencies that we don't want to run through our whole reproduction system as they can cause various problems and make for poor sound quality. So, we need to digital LP filter our upsampled digital audio to help separate out the 20Hz-20kHz audio frequencies we really care about reproducing faithfully, from all the other, more HF stuff we don't want. After passing through the digital LP filter, the resulting samples visually would look like a much smoother representation of the 20Hz-20kHz audio we do want. The better the filter we use, the more we will have attenuated the HF image effects of the zero samples we added (stuffed) and replaced them with more accurate (for our purposes) 20Hz-20kHz band-limited sample values.
EDIT: The above somewhat simplified with the hope that clarity can be arrived at in steps.
Last edited:
I'm not sure how you arrived at "can't be sufficient", but I suspect we have very different versions of good enough, whether real or not (and all kinds of implementation details to muddy this distinction).
Please note these aren't even the top of the line chips:
https://www.akm.com/akm/en/file/datasheet/AK4490EQ.pdf
pg.12, 14
https://statics.cirrus.com/pubs/proDatasheet/WM8741_v4.3.pdf
pg. 54+
http://www.ti.com/lit/ds/symlink/pcm1798.pdf
pg. 10
ESS not represented due to their silliness around datasheets
Additionally, as was hashed again and again within this thread, one can separate up/oversampling (whether integer multiples or not) into blocks of decimation and interpolation/filtration or integrate them, but the aliases are there whether we like them or not, and need to be addressed regardless.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.