DSD comparison to PCM.
- Thread starter U47
- Start date
Acknowledged; and I agree. Magazine reviewers are in a different position than I am; this technology is a magic box full of sparks to most reviewers, so they have the manufacturer's word (and hype) at face value. I don't. I want to know what's inside. That's what determines the sound, not the price-tag, ad-budget, or reviewer-schmoozing.
That said, I only have a bare minimum of technical understanding of these things. I've designed direct-radiator speakers for nearly thirty years, so I know them fairly well, vacuum-tube amps for fifteen years, so a little less, and horns for three years (but fortunate enough to have Bjorn Kolbrek, Jean-Michel LeCleac'h, and Martin Seddon of Azurahorn as mentors and collaborators).
What really baffles me about digital conversion is noise-shaping. To the best of my knowledge, it uses digital feedback around a single or 5 to 6-bit switch array to linearize the array. By adding just the right amount of dither-noise, the switch is PWM-modulated to achieve intermediate analog values that fall between what the switch can do on its own. A single-bit switch only has ON and OFF and is obviously grossly nonlinear, while a 6-bit switch is only somewhat linear, with 64 possible levels. Obviously far, far short of a 20-bit transmission system with a million possible levels. This is where the PWM-modulation and digital-feedback comes in, synthesizing the intermediate levels.
But stability problems can - and will - arise in complex feedback systems. In the analog world, we are limited by component parts variation and transit speeds through the forward path; get too complex and rely too much on the SPICE simulation, and the real world will bite you - hard. No fun to have a brand-new transistor amplifier destroy itself in a few milliseconds, and possibly take the speaker along, too.
In the digital world of noise-shaping, we get so-called "idle tones" and dynamic noise-floor modulation. The 20 dB jumps in noise-floor levels I saw at the 2011 RMAF presentation by ESS really got my attention; it was obvious that artifact-free noise-shaping design is extremely difficult.
I am also not comfortable with how the digital guys call the rising wall-of-noise (above 20 kHz) that's created by noise-shaping "noise". It's not noise as analog guys know it; it's the chopped-up debris of switch-errors, shoved to the top of the audio band, but as far as I can tell, is very much correlated with the audio signal. The beautiful thing about most sources of analog noise is that it is fully uncorrelated with the audio signal. If the transmission path has good low-level linearity, it'll stay that way, too, all the way to the loudspeaker. As long as it is uncorrelated, the ear/brain/mind system will easily ignore it, just as we ignore audience noise at a live concert. Once correlation starts, though, all bets are off.
After living with delta-sigma converters, I now feel that the noise-shaping algorithm is what we're hearing. What makes this problematic are the artifacts of noise-shaping are unfamiliar to most audiophiles and reviewers; once again, we're back in the "perfect sound forever" world, before we realized what lack of dither, jitter, and sample-rate conversion artifacts sounded like. I'm convinced that over time, we'll all start to hear the artifacts and recognize them.
That said, I only have a bare minimum of technical understanding of these things. I've designed direct-radiator speakers for nearly thirty years, so I know them fairly well, vacuum-tube amps for fifteen years, so a little less, and horns for three years (but fortunate enough to have Bjorn Kolbrek, Jean-Michel LeCleac'h, and Martin Seddon of Azurahorn as mentors and collaborators).
What really baffles me about digital conversion is noise-shaping. To the best of my knowledge, it uses digital feedback around a single or 5 to 6-bit switch array to linearize the array. By adding just the right amount of dither-noise, the switch is PWM-modulated to achieve intermediate analog values that fall between what the switch can do on its own. A single-bit switch only has ON and OFF and is obviously grossly nonlinear, while a 6-bit switch is only somewhat linear, with 64 possible levels. Obviously far, far short of a 20-bit transmission system with a million possible levels. This is where the PWM-modulation and digital-feedback comes in, synthesizing the intermediate levels.
But stability problems can - and will - arise in complex feedback systems. In the analog world, we are limited by component parts variation and transit speeds through the forward path; get too complex and rely too much on the SPICE simulation, and the real world will bite you - hard. No fun to have a brand-new transistor amplifier destroy itself in a few milliseconds, and possibly take the speaker along, too.
In the digital world of noise-shaping, we get so-called "idle tones" and dynamic noise-floor modulation. The 20 dB jumps in noise-floor levels I saw at the 2011 RMAF presentation by ESS really got my attention; it was obvious that artifact-free noise-shaping design is extremely difficult.
I am also not comfortable with how the digital guys call the rising wall-of-noise (above 20 kHz) that's created by noise-shaping "noise". It's not noise as analog guys know it; it's the chopped-up debris of switch-errors, shoved to the top of the audio band, but as far as I can tell, is very much correlated with the audio signal. The beautiful thing about most sources of analog noise is that it is fully uncorrelated with the audio signal. If the transmission path has good low-level linearity, it'll stay that way, too, all the way to the loudspeaker. As long as it is uncorrelated, the ear/brain/mind system will easily ignore it, just as we ignore audience noise at a live concert. Once correlation starts, though, all bets are off.
After living with delta-sigma converters, I now feel that the noise-shaping algorithm is what we're hearing. What makes this problematic are the artifacts of noise-shaping are unfamiliar to most audiophiles and reviewers; once again, we're back in the "perfect sound forever" world, before we realized what lack of dither, jitter, and sample-rate conversion artifacts sounded like. I'm convinced that over time, we'll all start to hear the artifacts and recognize them.
Last edited:
Re: the Invicta. It has no less than *seven* front-panel-selectable filters, two that are integral to the 9018, and five Resonessence-designed custom filters.
Anyway, the default Resonessence-designed apodizing filter sounded the best, followed by the slow-rolloff IIR filter (which mimics a 4th-order analog lowpass filter).
I am curious how a 9018 would sound with a passive I/V converter; the ESS engineers I spoke to at the 2012 RMAF said anything lower than 10 ohms would gently transition the 9018 into current mode, and 2~5 ohms would be a good overall compromise if you wanted to build a passive I/V converter. Once you passively low-pass with a cap in parallel to the resistor, the analog design is easy - basically, a microphone preamp, not the hardest thing to do. Much, much easier than a RIAA preamp, which is not at all easy to design.
I have read elsewhere that the 9018 requires 50~100x quieter clocks and power supplies than other delta-sigma chips, which is why the sound of 9018 DACs is all over the place. This is something I know nothing about; I'd probably call John Atwood or Gary Pimm for info on how to do this.
Anyway, the default Resonessence-designed apodizing filter sounded the best, followed by the slow-rolloff IIR filter (which mimics a 4th-order analog lowpass filter).
I am curious how a 9018 would sound with a passive I/V converter; the ESS engineers I spoke to at the 2012 RMAF said anything lower than 10 ohms would gently transition the 9018 into current mode, and 2~5 ohms would be a good overall compromise if you wanted to build a passive I/V converter. Once you passively low-pass with a cap in parallel to the resistor, the analog design is easy - basically, a microphone preamp, not the hardest thing to do. Much, much easier than a RIAA preamp, which is not at all easy to design.
I have read elsewhere that the 9018 requires 50~100x quieter clocks and power supplies than other delta-sigma chips, which is why the sound of 9018 DACs is all over the place. This is something I know nothing about; I'd probably call John Atwood or Gary Pimm for info on how to do this.
welcome to WBF.
your mention of 'a good friend' compells me to mention that that same gentleman (also my friend) also built the 'Found Music' amps that i had in my room for a time. i assume that possibly that 'flea-power 1 watt SET' amp (not of your design) might be one i had in my room....or maybe the 'other one' he mentioned he was working on.
it's a small audio world we live in.
Our 'special audio/music' community is indeed concentrated related.
...The discoveries are ahead, yet undiscovered.
The amplifier built by our mutual friend actually sounded really good; but then again, I've never heard a bad 45 SET amplifier, and I've heard plenty of funky 2A3 amplifiers. Go figure.
But I did tell him, when I was in Dallas, that 1 watt just ain't enough for any loudspeaker, no matter how efficient, and recounted the story of Paul Klipsch carrying a Brook 10-watt amplifier around the country when he was demonstrating the new Klipschorn in the late Forties. If PWK thought the 103~107 dB/meter/watt Khorn needed 10 watts, what does that say about a 1-watt amplifier with a 99 dB/meter/watt loudspeaker? To me, it says Not Enough Power, a problem easily solved with a PP 2A3 amplifier, which is simple as pie to build, and sounds really, really good, with super-low distortion and the gentlest overload imaginable.
Regarding the Keith Howard review, that's kind of an odd conclusion, but for an interesting reason: the Invicta actually uses a completely separate DAC, with one of the voltage-mode ESS converters (not the 9018) for the headphone output. The headphone section is completely separate from the RCA and XLR outputs, although controlled by the same volume control.
My reservations about most high-end DACs might come down to the op-amps; in my system, even the best ones just sound so-so, and far short of even a mediocre tube stage. Discrete transistor circuits tend to sound pallid and undynamic, and the not-so-good ones sound grainy and harsh. The only decent-sounding transistor line amps I've heard (that I was tempted to buy) came from Rowland Research. There are probably lots of transistor linestages out there in DIY-land that are good, but the sound I like comes from vacuum-tube circuits optimized for low distortion in the midrange.
But I did tell him, when I was in Dallas, that 1 watt just ain't enough for any loudspeaker, no matter how efficient, and recounted the story of Paul Klipsch carrying a Brook 10-watt amplifier around the country when he was demonstrating the new Klipschorn in the late Forties. If PWK thought the 103~107 dB/meter/watt Khorn needed 10 watts, what does that say about a 1-watt amplifier with a 99 dB/meter/watt loudspeaker? To me, it says Not Enough Power, a problem easily solved with a PP 2A3 amplifier, which is simple as pie to build, and sounds really, really good, with super-low distortion and the gentlest overload imaginable.
Regarding the Keith Howard review, that's kind of an odd conclusion, but for an interesting reason: the Invicta actually uses a completely separate DAC, with one of the voltage-mode ESS converters (not the 9018) for the headphone output. The headphone section is completely separate from the RCA and XLR outputs, although controlled by the same volume control.
My reservations about most high-end DACs might come down to the op-amps; in my system, even the best ones just sound so-so, and far short of even a mediocre tube stage. Discrete transistor circuits tend to sound pallid and undynamic, and the not-so-good ones sound grainy and harsh. The only decent-sounding transistor line amps I've heard (that I was tempted to buy) came from Rowland Research. There are probably lots of transistor linestages out there in DIY-land that are good, but the sound I like comes from vacuum-tube circuits optimized for low distortion in the midrange.
I have no engineering background, nor am I an employed audio reviewer. But, I do have 40 years listening to many, many audio products. I only know what gives me goosebumps. To me, in the audio chain, DSD vs PCM is not on the top of my list. The most important factor IMHO is system integration to what the buyer/listener loves. Do you love Porches, or Ferraris, or Aston Martins? There is no right in what people perceive to be superior. That is why people come in different sizes, sexes, color and culture. FWTW, I just re-discovered the joy of a tubed source, the sound of glass is so romantic. To hear Diana Krall's lips part on a recording, makes me want to get a divorce.
That is the great luxury of designing your own loudspeakers and amplifiers. I design for what Karna and I like. We hear similar things, she just uses different language than I do. She often points out things I don't hear right away - and then I focus in on what she's describing (it sounds constipated!) and realize she's right, it DOES sound constipated! I'd call it compressed, or flat, or some other phrase, and twiddle around until it sounds better (or worse).
Measurements are almost useless for a reviewer or listener, except as a way to discover outright design errors (these very commonly appear in Stereophile measurements, by the way). Measurements are extremely useful for the designer: as a way of discovering which parts of the design need improvements, or revealing subtle or gross errors. They are certainly more useful than any amount of SPICE simulation.
Putting it another way, measuring a "black box" only reveals gross design defects or manufacturing QC problems (which are plenty common in the high-end biz). Once the box is opened, though, and you look at subsystems or individual components, problems are quickly revealed, and can usually be quickly resolved.
Measurements are almost useless for a reviewer or listener, except as a way to discover outright design errors (these very commonly appear in Stereophile measurements, by the way). Measurements are extremely useful for the designer: as a way of discovering which parts of the design need improvements, or revealing subtle or gross errors. They are certainly more useful than any amount of SPICE simulation.
Putting it another way, measuring a "black box" only reveals gross design defects or manufacturing QC problems (which are plenty common in the high-end biz). Once the box is opened, though, and you look at subsystems or individual components, problems are quickly revealed, and can usually be quickly resolved.
I'm still puzzled why DSD sounds as good as it does (at the Invicta-and-above level). There is far more noise-shaping than a delta-sigma converter (the system would only have 6-bit resolution without it), there were many many problems stabilizing single-bit converters (so many that single-bit converters were replaced by delta-sigma converters), so on and so forth.
Operating the converter at 2.8 MHz (or a multiple of that) might be part of it, which makes me wonder what a delta-sigma converter operating at native rate, with the 6-bits of resolution preserved, might sound like. Huge data files, of course: 2.8 MHz at 6-bit resolution is honking big, although not as large than 1080p HDTV, I guess.
Operating the converter at 2.8 MHz (or a multiple of that) might be part of it, which makes me wonder what a delta-sigma converter operating at native rate, with the 6-bits of resolution preserved, might sound like. Huge data files, of course: 2.8 MHz at 6-bit resolution is honking big, although not as large than 1080p HDTV, I guess.
i'm a Playback Designs user, and big vinyl and tape guy. and i do enjoy dsd (both SACD disc and streaming) over PCM most times. and i like PCM too. i just prefer dsd.
i have a (totally un-scientific) theory about why 'most' analog lovers prefer (sufficiently high level) dsd regardless of it's theoretical problems.
it's not that dsd is perfect. in reality or in theory.
it's that PCM is simply more harmful to an analog signal. more math, more phase shifts, less ambient info. more distance from analog. the idea that the PCM conversion is not harmful to the analog signal is relatively wrong.
the lesser of two (evils) approaches.
can't support my perspective other than years and years of listening experiences.
i've sat in Bruce's studio and listened to Bruce switch an analog signal thru many different digital formats and dacs, and then back to analog. easy to hear how the dsd comes closest to analog....especially easy to hear is how the soundstage and ambient information changes. how your body reacts to the dsd after the pcm.
it's maybe a puzzle why this happens; but no doubt to my ears that it does happen.
I concur with the conclusions that Mike drew-meaning that DSD sounds the most like analog. And maybe that is part of the problem for people who listen exclusively to digital. Good analog doesn't serve as a reference point for them in their systems.
Well, one difference, discussed in more detail in Part Two, between PCM and DSD is a very different ultrasonic spectra. DSD is noiselike (but not quite noise in the analog sense), while PCM is a collection of narrow spectral spikes that are tightly correlated with the audio signal.
The reason I mention this is that PCM seems to greatly suffer in DACs that use opamps or discrete-transistor circuits with high feedback ratios combined with low slew rates. By "low" I mean a slew rate appreciably slower than 1000V/uSec. I've measured comb spectra coming out of a Burr-Brown PCM-63 that was flat to 20 MHz, and finally disappeared in the noise at 50 MHz. To keep up with that kind of rate-of-change requires a slew rate 1000V/uSec, or more.
This is why I favor passive I/V conversion in parallel with a lowpass capacitor. It is nearly impossible to build an analog stage that has low distortion in the 1~20 MHz region; RF circuits avoid this necessity by extensive use of filtering, as in superheterodyne conversion and amplification.
Passive I/V conversion distortion is set by the distortion of the resistor, which if well-chosen, is less than -150 dB over a wide bandwidth extending into the GHz region. A 70 kHz 1st-order lowpass (parallel cap) shunts RF energy away from the following analog section, which still needs to be reasonably fast (5532/5534 and 797 need not apply).
The 6DJ8, although not my favorite audio tube (more 3rd-harmonic than I'd like), has exceptionally good RF performance into the hundred-MHz range (the 6DJ8 was originally designed as a RF preamp tube for color-TV tuners). Although the 2nd and 3rd-harmonic are probably in the 0.1% range for 2V rms out, that can be lowered (if desired) by choke or current-source plate loading. It has plenty of headroom with the plate at 100~120V; the available voltage swing is about 60V.
The most important aspect of the I/V conversion stage is avoidance of RF crossmodulation from ultrasonic components from the converter. It's basically akin to designing a MC preamp that has to operate in close proximity to a flourescent light. The primary requirement is noise resistance. Compared to that, everything else is secondary; it is more important to reduce gross distortion over the working bandwidth of the I/V converter than chase the last decimal point in the audio band.
The second stage, if necessary, can be perfectly conventional, since by then the signal has been effectively lowpassed to the 70 kHz or lower range.
Returning to the original point, ladder/R-2R DACs with correctly designed I/V conversion sound remarkably "analog" and DSD-like. On the other hand, if you simply follow the manufacturer's app-note and use a 797 or 5532 active I/V converter followed by a 797 or 5532 Sallen & Key active lowpass filter, all of the grit-n-grain of PCM returns.
Protip: Any time you hear flattening of the soundstage, or outright grittiness, do a thorough analysis for any possible RFI intrusion into the analog circuits. Solid-state audio-analog circuits, in particular, do not behave well in the RF region, so content above 100 kHz needs to be well-filtered before it touches the first transistor base or FET gate.
The Karna amplifier is very overdesigned from this viewpoint; the input transformer is a 2nd-order lowpass around 50 kHz, the secondary is electrostatically screened to filter RFI, the grounds of the amplifier are electrically isolated from the DAC (no ground loops are possible), and the input tubes retain their linearity in the MHz region. In addition, there's no output-to-input feedback loop, so RFI picked up by the speaker cables (which are good antennas for AM radio) does not crossmodulate with audio in the input stage.
The reason I mention this is that PCM seems to greatly suffer in DACs that use opamps or discrete-transistor circuits with high feedback ratios combined with low slew rates. By "low" I mean a slew rate appreciably slower than 1000V/uSec. I've measured comb spectra coming out of a Burr-Brown PCM-63 that was flat to 20 MHz, and finally disappeared in the noise at 50 MHz. To keep up with that kind of rate-of-change requires a slew rate 1000V/uSec, or more.
This is why I favor passive I/V conversion in parallel with a lowpass capacitor. It is nearly impossible to build an analog stage that has low distortion in the 1~20 MHz region; RF circuits avoid this necessity by extensive use of filtering, as in superheterodyne conversion and amplification.
Passive I/V conversion distortion is set by the distortion of the resistor, which if well-chosen, is less than -150 dB over a wide bandwidth extending into the GHz region. A 70 kHz 1st-order lowpass (parallel cap) shunts RF energy away from the following analog section, which still needs to be reasonably fast (5532/5534 and 797 need not apply).
The 6DJ8, although not my favorite audio tube (more 3rd-harmonic than I'd like), has exceptionally good RF performance into the hundred-MHz range (the 6DJ8 was originally designed as a RF preamp tube for color-TV tuners). Although the 2nd and 3rd-harmonic are probably in the 0.1% range for 2V rms out, that can be lowered (if desired) by choke or current-source plate loading. It has plenty of headroom with the plate at 100~120V; the available voltage swing is about 60V.
The most important aspect of the I/V conversion stage is avoidance of RF crossmodulation from ultrasonic components from the converter. It's basically akin to designing a MC preamp that has to operate in close proximity to a flourescent light. The primary requirement is noise resistance. Compared to that, everything else is secondary; it is more important to reduce gross distortion over the working bandwidth of the I/V converter than chase the last decimal point in the audio band.
The second stage, if necessary, can be perfectly conventional, since by then the signal has been effectively lowpassed to the 70 kHz or lower range.
Returning to the original point, ladder/R-2R DACs with correctly designed I/V conversion sound remarkably "analog" and DSD-like. On the other hand, if you simply follow the manufacturer's app-note and use a 797 or 5532 active I/V converter followed by a 797 or 5532 Sallen & Key active lowpass filter, all of the grit-n-grain of PCM returns.
Protip: Any time you hear flattening of the soundstage, or outright grittiness, do a thorough analysis for any possible RFI intrusion into the analog circuits. Solid-state audio-analog circuits, in particular, do not behave well in the RF region, so content above 100 kHz needs to be well-filtered before it touches the first transistor base or FET gate.
The Karna amplifier is very overdesigned from this viewpoint; the input transformer is a 2nd-order lowpass around 50 kHz, the secondary is electrostatically screened to filter RFI, the grounds of the amplifier are electrically isolated from the DAC (no ground loops are possible), and the input tubes retain their linearity in the MHz region. In addition, there's no output-to-input feedback loop, so RFI picked up by the speaker cables (which are good antennas for AM radio) does not crossmodulate with audio in the input stage.
Last edited:
Yeah I can't find fault with that analysis - DSD sounds most like analog, meaning the sound of analog tape.
Yep, we're forced by absence of tape modulation noise to use real-life sounds as our reference. Real, live sounds do indeed sound... now what was that phrase.... ah yes, 'wowie zowie'.
Firstly a warm welcome to WBF Lynn - I've been a fan of your writings on DIYA for some time now. I'll just pick up on one point you've raised so as not to write a huge long post...
I agree here and this is why your proposed single capacitor filtering for an analog stage for the ESS won't cut the mustard.
Consider for a moment that the ESS is made from a much smaller geometry than your PCM63. I think the PCM63 hails from the early 1990s at which stage we were going below 1um feature sizes for digital but analog tends to run considerably behind digital. I'd hazard that the ESS is made below 180nm feature size, its a digital process. So a conservative estimate would be that there's at least a 10X difference in the RFI spectrum out of the two parts - meaning there's output up to 500MHz from this chip based on your experiments with the PCM63. Add to this the fact that its S-D meaning that there's a much higher noise output even in the presence of very low level signals. Its not as bad as DSD because its 6bits, but for the lowest level signals there's still going to be 3 or 4 levels of activity out of the 6bit DAC.
A single pole RC at 70kHz is going to have a capacitor value of roughly 10nF, given the output impedance of the ESS is of the order of 200ohms. Say we adopt your earlier hand-waving suggestion of using it in current mode with 10ohms as the I/V resistance - this increases the capacitor value to 200nF. The best capacitor here (lowest inductance, lowest losses, lowest ESR) is undoubtedly an NP0 ceramic, if we were to parallel two 100nFs (the largest generally available value) the SRF is 12.5MHz (using Kemet spice). Above that the capacitor is inductive meaning no further attenuation. But remember we need to slug to 500MHz by my handwaving arguments.
It seems obvious to me that some series inductor(s) are going to be required - my suggested output stage for an ESS DAC is 4th order LC. Looks like this : http://www.diyaudio.com/forums/digi...ad-diyinhk-es9018-dac-ebay-3.html#post3332420
I agree here and this is why your proposed single capacitor filtering for an analog stage for the ESS won't cut the mustard.
Consider for a moment that the ESS is made from a much smaller geometry than your PCM63. I think the PCM63 hails from the early 1990s at which stage we were going below 1um feature sizes for digital but analog tends to run considerably behind digital. I'd hazard that the ESS is made below 180nm feature size, its a digital process. So a conservative estimate would be that there's at least a 10X difference in the RFI spectrum out of the two parts - meaning there's output up to 500MHz from this chip based on your experiments with the PCM63. Add to this the fact that its S-D meaning that there's a much higher noise output even in the presence of very low level signals. Its not as bad as DSD because its 6bits, but for the lowest level signals there's still going to be 3 or 4 levels of activity out of the 6bit DAC.
A single pole RC at 70kHz is going to have a capacitor value of roughly 10nF, given the output impedance of the ESS is of the order of 200ohms. Say we adopt your earlier hand-waving suggestion of using it in current mode with 10ohms as the I/V resistance - this increases the capacitor value to 200nF. The best capacitor here (lowest inductance, lowest losses, lowest ESR) is undoubtedly an NP0 ceramic, if we were to parallel two 100nFs (the largest generally available value) the SRF is 12.5MHz (using Kemet spice). Above that the capacitor is inductive meaning no further attenuation. But remember we need to slug to 500MHz by my handwaving arguments.
It seems obvious to me that some series inductor(s) are going to be required - my suggested output stage for an ESS DAC is 4th order LC. Looks like this : http://www.diyaudio.com/forums/digi...ad-diyinhk-es9018-dac-ebay-3.html#post3332420
Last edited:
Lynn, did you miss my previous post (quote above)?
* What do you think of the TI Burr-Brown DSD-1792 (best/latest TI BB) DAC chip?
Last edited:
They make K grade in the DS chips now do they? 
They make 'highest grade' ? I wasn't aware there were any grades in DS chips. That to me is one of the whole points of designing DS chips, so they're ready to rock straight off the line without any kind of characterization or laser trimming.
Although this might not seem relevant at first blush, I believe its the 'Holy Grail' of the designers - to make a part that beats an analog one but is purely digital. DSD seems to me to be in the same vein - the 'digital dream' format that's totally digital, no intermediate steps into analog whatsoever.
They make 'highest grade' ? I wasn't aware there were any grades in DS chips. That to me is one of the whole points of designing DS chips, so they're ready to rock straight off the line without any kind of characterization or laser trimming.Although this might not seem relevant at first blush, I believe its the 'Holy Grail' of the designers - to make a part that beats an analog one but is purely digital. DSD seems to me to be in the same vein - the 'digital dream' format that's totally digital, no intermediate steps into analog whatsoever.
Last edited:
The 1792 is delta-sigma, isn't it? We're right back into the noise-shaping soup once you move to delta-sigma. Remember, to avoid noise-shaping (digital feedback loops that are prone to instability), the converter would have to operate at GHz speeds - and none of them do that just yet.
As for the effectiveness of a 1st-order lowpass filter, the most important function is to protect the first analog stage from slewing - additional filtering can be deferred to a later stage, if desired. The raw output of the converter is undoubtedly harmful and most certainly will induce slewing in most analog stages. But a filter as simple as a 100 kHz 1st-order lowpass filter reduces the magnitude of 1 MHz by 20 dB, and 10 MHz by 40 dB. That makes all the difference, especially if the first analog stage can be arranged so it is reasonably linear up to 1 MHz. That's a tall order for a high-feedback opamp, but I understand that current-mode devices handle it well, and for vacuum-tubes, no problem in the slightest (they have state-of-the-art RFI immunity).
As for the physical capacitor, we use standard RF isolation techniques, bypassing in 1 to 20 to 400 ratio caps as necessary, measuring as we go. Nothing that can't be figured out on a RF bridge, or measured with an RF network analyzer. Broadband filtering that is effective out to several GHz has been standard practice in the RF and microwave world since the 1950's.
Another technique to block RFI (immediately following a 5 to 50-ohm resistive I/V converter) is a moderate-ratio stepup transformer (1:2, 1:3, 1:4) with a 80% nickel core and a screened secondary - a high-quality mike transformer, in other words. High stepup ratio transformers have ripples in the HF response that are problematic in this application - we want a nice smooth Bessel characteristic with the intended source and load impedances. Another not-so-subtle benefit is fully isolating the digital and analog grounds - there's no galvanic connection between primary and secondary, which really simplifies board layout.
As for the audibility of different digital filters ahead of the Philips or Burr-Brown ladder converters, I dunno. I've heard the no-longer-in-production (thanks, Microsoft) Pacific Microsonics filters, but in a solid-state DAC with solid-state I/V converters, so that scrambles the comparison. Never heard the PMD-100 or PMD-200 in a DAC with the topology similar to the one I have now.
I've tempted one of my audio-friends to build a tube-based DAC based on some of these thoughts, so I'm curious how that turns out. For one thing, the Philips or Burr-Brown converters can be run in NOS mode - but at a very high speed - and the computer can do the upsampling with modern algorithms.
As for the effectiveness of a 1st-order lowpass filter, the most important function is to protect the first analog stage from slewing - additional filtering can be deferred to a later stage, if desired. The raw output of the converter is undoubtedly harmful and most certainly will induce slewing in most analog stages. But a filter as simple as a 100 kHz 1st-order lowpass filter reduces the magnitude of 1 MHz by 20 dB, and 10 MHz by 40 dB. That makes all the difference, especially if the first analog stage can be arranged so it is reasonably linear up to 1 MHz. That's a tall order for a high-feedback opamp, but I understand that current-mode devices handle it well, and for vacuum-tubes, no problem in the slightest (they have state-of-the-art RFI immunity).
As for the physical capacitor, we use standard RF isolation techniques, bypassing in 1 to 20 to 400 ratio caps as necessary, measuring as we go. Nothing that can't be figured out on a RF bridge, or measured with an RF network analyzer. Broadband filtering that is effective out to several GHz has been standard practice in the RF and microwave world since the 1950's.
Another technique to block RFI (immediately following a 5 to 50-ohm resistive I/V converter) is a moderate-ratio stepup transformer (1:2, 1:3, 1:4) with a 80% nickel core and a screened secondary - a high-quality mike transformer, in other words. High stepup ratio transformers have ripples in the HF response that are problematic in this application - we want a nice smooth Bessel characteristic with the intended source and load impedances. Another not-so-subtle benefit is fully isolating the digital and analog grounds - there's no galvanic connection between primary and secondary, which really simplifies board layout.
As for the audibility of different digital filters ahead of the Philips or Burr-Brown ladder converters, I dunno. I've heard the no-longer-in-production (thanks, Microsoft) Pacific Microsonics filters, but in a solid-state DAC with solid-state I/V converters, so that scrambles the comparison. Never heard the PMD-100 or PMD-200 in a DAC with the topology similar to the one I have now.
I've tempted one of my audio-friends to build a tube-based DAC based on some of these thoughts, so I'm curious how that turns out. For one thing, the Philips or Burr-Brown converters can be run in NOS mode - but at a very high speed - and the computer can do the upsampling with modern algorithms.
Last edited:
Hi Lynn and a big welcome! In a few posts, you have already provide many a great insights and I thank you for that. In brief, wrt to tube amps, what you say is so true. Most commercial amps don't really cut the mustard, even the ridiculously priced ones (some get close though) and almost all manufacturers grossly overstate the amp's driving capability. That, I think is why most tubes (esp. SET and DHT) amps get a bad name. I run a 40watts triode pp on a 95db speakers and with certain music, there's still some compression. SETs or DHTs are simple amps but to do it right takes a bit of care and money. However, with the way the audio industry works, it'll either be too expensive, too big/heavy or not enough quality parts to do an economically viable run. So most of the amps that I've heard, the good ones are usually DIY or from very boutique or custom manufacturers.
I also appreciate Mike L's refreshing and honest take on the 45's that he heard. The best tube amps will definitely gives you something into music that no SS will ever hope to achieve. I heard my 1st SET (a very good Japanese one) and got hooked ever since.
Cheers.
I also appreciate Mike L's refreshing and honest take on the 45's that he heard. The best tube amps will definitely gives you something into music that no SS will ever hope to achieve. I heard my 1st SET (a very good Japanese one) and got hooked ever since.
Cheers.
Similar threads
