celestial_sound Posted March 3, 2023 Author Share Posted March 3, 2023 I recently purchased the TEAC UD-701N and I am extremely pleased with its excellent qualities, particularly the transparency of vocals, resolution, and clarity. The unit is currently in the burn-in process, so I anticipate even further improvements in these aspects. However, as I experiment with the filters and options provided by the delta sigma modulator, I have a couple of questions that I would appreciate some clarification on. Firstly, there is an option for the DAC module to output the PCM signal in either multi-bit PCM or 1-bit DSD. As far as I understand, the delta sigma modulator oversamples the signal by default to x128, x246, or x512 (depending on the selected option), and then converts it into a 1-bit stream using interpolation. Therefore, how is it possible for the DAC to output multi-bit PCM? The only way I can think of is oversampling -> conversion to 1-bit -> downsampling -> conversion to multi-bit, but this seems nonsensical to me. Am I missing something? Secondly, my second question pertains to the "upconversion" of the signal. The options available are x2, x4, or x8. However, since this is a delta sigma modulator, doesn't it always upconvert or oversample the signal to x128, x256, or x512? If so, what is the purpose of having the option to upconvert the signal before it reaches the DAC section? I would be extremely grateful if someone with knowledge on these topics, such as @Miska@GoldenOne, could kindly provide me with some answers. Thank you very much. Link to comment
Miska Posted March 3, 2023 Share Posted March 3, 2023 7 hours ago, celestial_sound said: Secondly, my second question pertains to the "upconversion" of the signal. The options available are x2, x4, or x8. However, since this is a delta sigma modulator, doesn't it always upconvert or oversample the signal to x128, x256, or x512? If so, what is the purpose of having the option to upconvert the signal before it reaches the DAC section? I would be extremely grateful if someone with knowledge on these topics, such as @Miska@GoldenOne, could kindly provide me with some answers. DAC chips for example can run digital filters to max 8x. And then they go on from there by just copying same sample value N times. Or use something extremely simple and coarse such as linear interpolation. So above is in line with that. HQPlayer runs digital filters up to the final rate and never resorts to such simple and coarse approaches. But of course it needs much more processing power. Point of upconversion is that without proper digital filters, your signal keeps repeating itself in frequency domain. Meaning that it is not reconstructed properly back to the original smooth analog waveform. PCM needs to be "decoded" properly, which means combination of DSP processing, mixed domain electronics and analog electronics. But you should be able to run 701N with externally produced DSD256 or DSD512 and bypass it's built-in DSP. It may be an interesting device in such way. semente 1 Signalyst - Developer of HQPlayer Pulse & Fidelity - Software Defined Amplifiers Link to comment
mocenigo Posted March 4, 2023 Share Posted March 4, 2023 12 hours ago, Miska said: DAC chips for example can run digital filters to max 8x. And then they go on from there by just copying same sample value N times. Or use something extremely simple and coarse such as linear interpolation. So above is in line with that. This is not correct. Most modern digital chips run filters to upsample from the input frequency to 352.8 or 384khz (so up to 8x) or 705.6/768 (up to 16x but this is not very important) and then apply Delta Sigma modulation, not hold and repeat. bogi 1 Link to comment
Popular Post Miska Posted March 4, 2023 Popular Post Share Posted March 4, 2023 2 hours ago, mocenigo said: This is not correct. Most modern digital chips run filters to upsample from the input frequency to 352.8 or 384khz (so up to 8x) or 705.6/768 (up to 16x but this is not very important) and then apply Delta Sigma modulation, not hold and repeat. Yes it is correct. Because you cannot run proper delta-sigma modulation at such low rates. For example many AKM chips run the delta-sigma modulator at 5.6/6.1 MHz (128x) rates and so do Cirrus Logic. So it needs to make up for the difference. This is done by sample-and-hold (S/H) operation, aka zero-order-hold. Sometimes this is also shown in the chip's block diagram, like in case of AD1955: This is visible also when you look at output of the DAC from PCM inputs, here's one example from AKM chip based DAC I've measured, input is 44.1k PCM: This is peak-hold spectrum of 0 - 22.05 kHz sweep. You can see the images around multiples of 352.8 kHz which are due to this S/H operation. While in HQPlayer I run filters to upsample to 256x or 512x rates before the data goes to my delta-sigma modulator operating at 11.2/12.2/22.5/24.5 MHz before output to D/A conversion. So I never resort to such ugly methods myself. dknk and bogi 1 1 Signalyst - Developer of HQPlayer Pulse & Fidelity - Software Defined Amplifiers Link to comment
mocenigo Posted March 4, 2023 Share Posted March 4, 2023 33 minutes ago, Miska said: Yes it is correct. Because you cannot run proper delta-sigma modulation at such low rates. For example many AKM chips run the delta-sigma modulator at 5.6/6.1 MHz (128x) rates and so do Cirrus Logic. So it needs to make up for the difference. This is done by sample-and-hold (S/H) operation, aka zero-order-hold. Sometimes this is also shown in the chip's block diagram, like in case of AD1955: This is visible also when you look at output of the DAC from PCM inputs, here's one example from AKM chip based DAC I've measured, input is 44.1k PCM: This is peak-hold spectrum of 0 - 22.05 kHz sweep. You can see the images around multiples of 352.8 kHz which are due to this S/H operation. While in HQPlayer I run filters to upsample to 256x or 512x rates before the data goes to my delta-sigma modulator operating at 11.2/12.2/22.5/24.5 MHz before output to D/A conversion. So I never resort to such ugly methods myself. Oh please you know well that this is a different thing. The first upsampling (up to 8x or 16x) serves to ensure linear response, move the images at multiples of the nyquist frequency as far as possible from the audio band, give the desired shape to impulses. That’s the oversampling that counts and the only type that is needed. THEN you have the delta sigma modulation and this of course occurs at a further multiple. Holding the samples there is more of a technical aspect of implementation of the modulator, not oversampling of the signal, which is actually done by the modulator, and having anything else but holding or linear interpolation risks instability in corner cases (as you very well know). the images around multiples of 352.8khz may be of mathematical interest (and I am a mathematician), but they are so far from the audio band that they are irrelevant, and pretending that removing them will improve sound quality is snake oil. Link to comment
Popular Post Miska Posted March 4, 2023 Popular Post Share Posted March 4, 2023 28 minutes ago, mocenigo said: Oh please you know well that this is a different thing. The first upsampling (up to 8x or 16x) serves to ensure linear response, move the images at multiples of the nyquist frequency as far as possible from the audio band, give the desired shape to impulses. That’s the oversampling that counts and the only type that is needed. In my case, the "first upsampling" is typically 256x or 512x. 28 minutes ago, mocenigo said: the images around multiples of 352.8khz may be of mathematical interest (and I am a mathematician), but they are so far from the audio band that they are irrelevant, and pretending that removing them will improve sound quality is snake oil. Based on the sampling theory, for proper reconstruction, you need to remove the images entirely. Until that the original analog waveform is not reconstructed. Those are essentially just caused by the stair-stepping of output waveform. They can cause intermodulation within each other, especially at such high frequencies, because the frequencies between the images fall within audio band. They can also cause aliasing distortion with class-D amplifier (for example Hypex are susceptible to such). In addition, they are sign of incomplete reconstruction. For example in above case the first image is down just about -45 dB giving roughly 7 - 8 bits worth of reconstruction accuracy. 28 minutes ago, mocenigo said: having anything else but holding or linear interpolation risks instability Holding or linear interpolation can cause instability and in addition it causes signal dependent distortions inside the modulator. But since those chips don't have enough processing resources to do better, they need to cut corners on such things. I don't resort to such. dknk, mikicasellas, StreamFidelity and 1 other 4 Signalyst - Developer of HQPlayer Pulse & Fidelity - Software Defined Amplifiers Link to comment
mocenigo Posted March 5, 2023 Share Posted March 5, 2023 8 hours ago, Miska said: Based on the sampling theory, for proper reconstruction, you need to remove the images entirely. Until that the original analog waveform is not reconstructed. Those are essentially just caused by the stair-stepping of output waveform. Yes, but you are referring to an ideal reconstruction, and images will always be there, no matter how high you oversample. So one has to stop somewhere – you also do that, right? "stair-stepping" will always be there. So the aim is to push any artifacts outside the audible range. 8 hours ago, Miska said: They can cause intermodulation within each other, especially at such high frequencies, because the frequencies between the images fall within audio band. They can also cause aliasing distortion with class-D amplifier (for example Hypex are susceptible to such). True, and this will happen also with your upsampling. Re: the aliasing distorsion in some Hypex amplifiers: we all should keep in mind that modulation there is limited to about 500Khz, whereas in a modern DAC we are talking of 12 or 24 Mhz. 8 hours ago, Miska said: In addition, they are sign of incomplete reconstruction. For example in above case the first image is down just about -45 dB giving roughly 7 - 8 bits worth of reconstruction accuracy. Sorry, but this is both correct and completely misleading. First of all, the images you show are at multiples off 352.8 Khz, therefore they are per se outside audible range – if the original signal is rebook, the lowest frequencies in the negative image will be at 330Khzm roughly. In any DAC these are filtered out by the output filter. Therefore they can only intermodulate at that stage, and this only if the transfer function of the filter is not linear. A quite common LC, low-pass, elliptic filter will attenuate them, but also Butterworth will be more than enough. If you have such a filter, images are suppressed once they are far enough, and any residual artefact will be inaudible. Year, it is true that the unfiltered artefacts are at -45 Db. It is also true that they are completely inaudible and any intermodulation that they could cause in a downstream preamp or amp or transducer can be killed with an output filter. While it is also technically true that you get 7-8 bits worth of reconstruction accuracy if you do not have output filters and measure wide band, it is also a completely artificial measure: in the audible band you still have all your 20-21 bits of reconstruction accuracy, sometimes a bit less, and in fact also wide band if you include the filter, and nothing left that can audibly intermodulate. 8 hours ago, Miska said: Holding or linear interpolation can cause instability and in addition it causes signal dependent distortions inside the modulator. The question is the amount this distortion, inside the audible band, and you are clearly not addressing this. 8 hours ago, Miska said: But since those chips don't have enough processing resources to do better, they need to cut corners on such things. I don't resort to such. With HQPLayer you provide the most precise modulators money can buy, and your level of reconstruction is most likely unmatched. This is an achievement and it is nice to have. Actual audibility is only anecdotal, and please do not use red herrings, because this in fact detracts from your technical achievement. Link to comment
Miska Posted March 5, 2023 Share Posted March 5, 2023 41 minutes ago, mocenigo said: Yes, but you are referring to an ideal reconstruction, and images will always be there, no matter how high you oversample. So one has to stop somewhere – you also do that, right? "stair-stepping" will always be there. So the aim is to push any artifacts outside the audible range. Point is to oversample high enough that analog post-filter can remove the images without altering phase in audio band. Typical analog post-filter is 2nd order with fc=100 kHz. Already 44.1 kHz sampling rate has images "outside of audible range", since they start right above 22.05 kHz. 41 minutes ago, mocenigo said: True, and this will happen also with your upsampling. No it won't. 41 minutes ago, mocenigo said: Re: the aliasing distorsion in some Hypex amplifiers: we all should keep in mind that modulation there is limited to about 500Khz, whereas in a modern DAC we are talking of 12 or 24 Mhz. Except with your "modern DAC" leaks those images starting at 330.75 kHz (when source is 44.1k) and spreading to MHz range. This causes the low rate class-D modulator alias those high frequency signals. 41 minutes ago, mocenigo said: In any DAC these are filtered out by the output filter. No they are not, since the typical 2nd order filter positioned at around 100 kHz is not enough. The measurement result I shown IS measured from analog outputs of a typical DAC. I have tens of different DACs, with all the current notable and even esoteric DAC chips (RoHM for example) and I commonly measure things from the actual analog outputs. 41 minutes ago, mocenigo said: Year, it is true that the unfiltered artefacts are at -45 Db. It is also true that they are completely inaudible and any intermodulation that they could cause in a downstream preamp or amp or transducer can be killed with an output filter. Those measurements are from the DAC device outputs, after the output filter. And it is fully correlated distortion in itself, plain and simple. For fun, calculate what kind of analog filter you need to have no phase shift at 20 kHz, pass hires content, and that will attenuate to -96 dB (for 16-bit source) or to -144 dB (for 24-bit) for proper reconstruction, by 330.75 kHz. Your typical DAC doesn't have such... So instead of going for horribly complex (with extra noise and distortion) analog filter, it can be solved in digital domain by doing for example 256x digital filters. Then your first image for 44.1k source starts at 11.26755 MHz... So this, and number of other problems can be solved in a very straightforward way, if you anyway use computer for listening music. Tihon 1 Signalyst - Developer of HQPlayer Pulse & Fidelity - Software Defined Amplifiers Link to comment
mocenigo Posted March 5, 2023 Share Posted March 5, 2023 16 hours ago, Miska said: Point is to oversample high enough that analog post-filter can remove the images without altering phase in audio band. Typical analog post-filter is 2nd order with fc=100 kHz. Which is what I said 16 hours ago, Miska said: Except with your "modern DAC" leaks those images starting at 330.75 kHz (when source is 44.1k) and spreading to MHz range. This causes the low rate class-D modulator alias those high frequency signals. The point is not (only) whether there are such images and whether they are attenuated enough. The matter is whether they intermodulate, and this depends on the intermodulation distortion of the circuitry downstream. If the out of audio band frequencies are attenuated enough not to cause oscillations in downstream amps and they do not fry tweeters, then there is no problem. And in normal transducers these high frequencies cannot intermodulate because they cannot move at all at those frequencies. 16 hours ago, Miska said: No they are not, since the typical 2nd order filter positioned at around 100 kHz is not enough. The measurement result I shown IS measured from analog outputs of a typical DAC. Ok, I stand corrected on that. But the fact remains that there is no audible consequence. I do appreciate the effort and the mathematics behind it, and I stand on the objective side of audiophilia – but once artefacts remain below audibility even if processed by the downstream circuitry, any improvement becomes nice to have (and I am a nerd, so I do like that) but not indispensable. 16 hours ago, Miska said: So instead of going for horribly complex (with extra noise and distortion) analog filter, it can be solved in digital domain by doing for example 256x digital filters. Then your first image for 44.1k source starts at 11.26755 MHz... I said there are images. You said that with images will not happen with your upsampling (maybe you thought I was talking about instability?) and now you confirm what I said. They are pushed further out. 16 hours ago, Miska said: So this, and number of other problems can be solved in a very straightforward way, if you anyway use computer for listening music. Audibly, what would these problems cause? If they are not audible, they aren't problems. Link to comment
Miska Posted March 5, 2023 Share Posted March 5, 2023 2 hours ago, mocenigo said: I said there are images. You said that with images will not happen with your upsampling (maybe you thought I was talking about instability?) and now you confirm what I said. They are pushed further out. As stated earlier, point of upsampling is to move the images far enough that the analog post-filter will be able to eliminate them completely. And such analog post-filter that doesn't have adverse effects on audio band (meaning the corner needs to be at 100 kHz or higher). Only then you have properly reconstructed the analog waveform. Of course this is still not even touching intricacies of the filters themselves and modulator designs. 2 hours ago, mocenigo said: Audibly, what would these problems cause? I already covered this earlier. Only bad excuses for leaving any images. Signalyst - Developer of HQPlayer Pulse & Fidelity - Software Defined Amplifiers Link to comment
teac Posted March 21, 2023 Share Posted March 21, 2023 I for one find it hard to understand the current dislike campaign against Teac turntables (yes I know my user name looks like I work for them but trust me I do not) and I have a couple of viewpoints that turned my opinion of Teac totally. For years I have used a CS5000 and a top-range Aurex player, these players have served me well but I wanted a change. Ignoring the general consensus of these players I took a risk on a Teac TN 4D Direct Drive,, and upgraded the Oyster Stylus with LP Tunes and the results have been absolutely excellent. It out preforms the CS5000 in all ways, great controlled bass, trebles, and a rock-steady speed. Fast start-up and simple sweeet operation. It certainly looks good but I can vouch that it sounds good as well. Dont really understand the hate for this brand, they are as good as any in my opinion. Link to comment
teac Posted March 21, 2023 Share Posted March 21, 2023 Well, the more expensive ones are. I have no experience with the cheapest ones... Link to comment
DuckToller Posted March 21, 2023 Share Posted March 21, 2023 55 minutes ago, teac said: I for one find it hard to understand the current dislike campaign against Teac turntables (yes I know my user name looks like I work for them but trust me I do not) and I have a couple of viewpoints that turned my opinion of Teac totally. For years I have used a CS5000 and a top-range Aurex player, these players have served me well but I wanted a change. Ignoring the general consensus of these players I took a risk on a Teac TN 4D Direct Drive,, and upgraded the Oyster Stylus with LP Tunes and the results have been absolutely excellent. It out preforms the CS5000 in all ways, great controlled bass, trebles, and a rock-steady speed. Fast start-up and simple sweeet operation. It certainly looks good but I can vouch that it sounds good as well. Dont really understand the hate for this brand, they are as good as any in my opinion. Sounds not uninteresting but suggest we all know about the TEAC TT discussion, which we may not do ... Perhaps using the analog forum for an educated report on the subject could generate more interesting response and no OT discussion in a DAC themed thread. https://audiophilestyle.com/forums/forum/132-analog-components/ For myself, I used to find TEAC TTs quite interesting but have not followed up on the Reddit of 2017 concerning the subject you mentioned. But ready to learn about the subject and your positive experience ;-) Tom Link to comment
bodiebill Posted April 6, 2023 Share Posted April 6, 2023 Yesterday I received my UD-701N and I love it already. It was a "second deal" and indeed, it seems already broken in. Sound was great from the start: detailed, 3D, fluid, velvety and wonderful bass. I am still a bit confused regarding the Teac's upconvert / Fs process, as shown by my following question: My settings are - upconvert: x8 - DeltaSigma Fs: 256 Fs (or 512 Fs) - PCM Delta Sigma: DSD (1 Bit) - DSD Low Pass Filter: FIR1 When playing a 44.1 kHz wav file I would expect that the Teac's INFO button would show PCM 44.1kHz => DSD 11.2MHz as per the example in the manual. However it shows PCM 44.1kHz => PCM 352.8Hz Am I missing something? Jakenz 1 audio system Link to comment
stefano_mbp Posted April 6, 2023 Share Posted April 6, 2023 @bodiebill … while it can play up to DSD512 it seems that only PCM up to 384kHz upsampling is possible https://teac.jp/int/product/ud-701n/feature . check on pag 15 of the user guide Stefano My audio system Link to comment
bodiebill Posted April 6, 2023 Share Posted April 6, 2023 10 minutes ago, stefano_mbp said: @bodiebill … while it can play up to DSD512 it seems that only PCM up to 384kHz upsampling is possible https://teac.jp/int/product/ud-701n/feature . check on pag 15 of the user guide Thanks Stefano. That leaves me with the question what settings would achieve what is shown on page 17 of the manual. audio system Link to comment
stefano_mbp Posted April 6, 2023 Share Posted April 6, 2023 @bodiebill this is the same page of the manual I linked … something has changed … DuckToller 1 Stefano My audio system Link to comment
bodiebill Posted April 6, 2023 Share Posted April 6, 2023 24 minutes ago, stefano_mbp said: @bodiebill this is the same page of the manual I linked … something has changed … Ah, maybe I have an older version of the manual and they corrected a mistake. I will let go and enjoy the music :-) Thanks for sharing this info! audio system Link to comment
DuckToller Posted April 6, 2023 Share Posted April 6, 2023 28 minutes ago, bodiebill said: Thanks Stefano. That leaves me with the question what settings would achieve what is shown on page 17 of the manual. The actual website seems to be more precise: Upconversion The upconversion function uses RDOT-NEO (Refined Digital Output Technology NEO) to smoothly augment PCM digital audio signals and upconvert their sampling frequencies 2, 4 or 8 times (384kHz maximum). RDOT, which applies an analogous interpolation method using fluency logic, is a technology that was developed to enable reproduction and playback of frequencies higher than 20 kHz that are lost by 44.1kHz/48kHz digital signals. Based on the information read, analogous data is generated between the waveform samples, which also results in data above 20 kHz. If they were ever upsampling to DSD, they may have fixed that with a firmware upgrade. I guess it was rather a fault in information transfer from the 50x series ... (TEAC NT-505 X) Happy easter and enjouy your new TEAC toy ;-) bodiebill 1 Link to comment
celestial_sound Posted April 6, 2023 Author Share Posted April 6, 2023 6 hours ago, bodiebill said: When playing a 44.1 kHz wav file I would expect that the Teac's INFO button would show PCM 44.1kHz => DSD 11.2MHz as per the example in the manual. However it shows PCM 44.1kHz => PCM 352.8Hz As per my understanding the INFO button shows the up-conversation of the signal before it enters the delta sigma modulator, hence there are two up sampling stages with different algorithms of course. The first one is 2x, 4x, 8x fs, and the second during the 1 bit conversation - 128x, 256x, 512x fs bodiebill 1 Link to comment
Miska Posted April 6, 2023 Share Posted April 6, 2023 9 minutes ago, celestial_sound said: As per my understanding the INFO button shows the up-conversation of the signal before it enters the delta sigma modulator, hence there are two up sampling stages with different algorithms of course. The first one is 2x, 4x, 8x fs, and the second during the 1 bit conversation - 128x, 256x, 512x fs It is anyway a discrete 1-bit SDM DAC, so that's what the conversion stage needs. 352.8/384k is also what DAC chips do before further oversampling by just copying same sample enough many times to reach 128/256/512fs to the modulator. This is S/H (sample-and-hold) aka zero order hold. This will cause output to have images around multiples of the 352.8/384k digital filter output rate. Signalyst - Developer of HQPlayer Pulse & Fidelity - Software Defined Amplifiers Link to comment
celestial_sound Posted April 6, 2023 Author Share Posted April 6, 2023 Its really interesting that TEAC's algorithm for up-conversation is the first that makes any audible difference to me between off and 8x fs. I've tried several DAC's before with oversampling features and I couldn't hear any difference at all. Turning 8x fs on my UD-701n, always adds a bit of air and sparkle at the high frequencies. I think I can distinguish even in a ABX test. Link to comment
Miska Posted April 6, 2023 Share Posted April 6, 2023 1 hour ago, celestial_sound said: Its really interesting that TEAC's algorithm for up-conversation is the first that makes any audible difference to me between off and 8x fs. I've tried several DAC's before with oversampling features and I couldn't hear any difference at all. Yeah, it is pretty hard to find a DAC that wouldn't have oversampling features, since it is about 99% of the market. Much less DACs that wouldn't have. Difference to what? Since there are very very few NOS DACs. To me, different digital filters, delta-sigma modulators, D/A conversion structures and analog post-filters have always made differences. Signalyst - Developer of HQPlayer Pulse & Fidelity - Software Defined Amplifiers Link to comment
semente Posted April 7, 2023 Share Posted April 7, 2023 20 hours ago, bodiebill said: Yesterday I received my UD-701N and I love it already. It was a "second deal" and indeed, it seems already broken in. Sound was great from the start: detailed, 3D, fluid, velvety and wonderful bass. I am still a bit confused regarding the Teac's upconvert / Fs process, as shown by my following question: My settings are - upconvert: x8 - DeltaSigma Fs: 256 Fs (or 512 Fs) - PCM Delta Sigma: DSD (1 Bit) - DSD Low Pass Filter: FIR1 When playing a 44.1 kHz wav file I would expect that the Teac's INFO button would show PCM 44.1kHz => DSD 11.2MHz as per the example in the manual. However it shows PCM 44.1kHz => PCM 352.8Hz Am I missing something? HQPlayer? 😇 "Science draws the wave, poetry fills it with water" Teixeira de Pascoaes HQ Player Desktop/ Mac mini → HQ Player NAA/ CuBox-i → Intona 7054 → RME ADI-2 DAC FS Link to comment
bodiebill Posted April 12, 2023 Share Posted April 12, 2023 Can someone explain the logic behind the UD-701N's two clock sync settings, DIN ASYNC and DIN SYNC? I have the impression that here DIN ASYNC sounds slightly better, in spite of the fact that my reference clock (Mutec Ref10 SE120) also clocks a DDC (Audio-GD DI20HE) or -- when I use the network -- an EtherRegen. audio system Link to comment
Recommended Posts
Create an account or sign in to comment
You need to be a member in order to leave a comment
Create an account
Sign up for a new account in our community. It's easy!
Register a new accountSign in
Already have an account? Sign in here.
Sign In Now