Do please go on Ken, I'm curious to learn what flaws you hear and on which particular DACs?
Here's what I posted at AudioAsylum on October 2, 2010. I've made some minor edits here to correct a few typos and other errors:
"Here are my extended observations of NOS via an AD1865 based DAC of my own design which allowed me to switch a typical half-band digital FIR filter in and out of circuit on-the-fly via a toggle switch.
What NOS gets right:
1) NOS delivers CD digital which is non-fatiguing or relaxing, indeed, much in the way that vinyl is. I can listen to NOS for many hours without tiring while standard digital usually has me feeling anxious and switching the music off before even a single CD has been fully played.
2) The soundstage is very open, separating what is often a rather congealed sounding mass of music from standard digital in to a much more natural and three-dimensional sounding presentation. There is a naturalness absent with the FIR filter in, well, except for what sounds like a tonal energy shift to the upper midrange (see further comment on this below).
3) I also found cymbals and bells to have a very natural tone and long shimmer decay. The FIR filter seemed to add what sounds like synthetic splashes of white noise to such higher register instruments, making them sound more homogeneous.
What NOS gets wrong:
1) There is the well known high-frequency roll-off of about 3dB at 20KHz due to the zeroth-order hold operation of R2R ladder DACs. I'm not sure that sigma-delta DACs have this problem due to their high inherent oversampling operation, pushing any such roll-off way up in frequency.
2) NOS seems to shift musical energy from the upper bass-lower midrange region to the upper midrange region, altering the tonality of most instruments and vocalists. This highlighting of the the upper midrange is initially pleasing by presenting more musical detail, but ultimately, becomes increasingly noticeable until it reaches distraction. This effect also seems to soften or loosen the impact of bass register instruments, almost as if they were no longer properly damped.
3a) Actually, I'm uncertain whether the following observation constitutes a flaw or a benefit. Along with the aforementioned shift of energy to the upper midrange I hear a large increase in the ambient field via NOS. While this greatly illuminates the upper midrange, and may even be what's responsible for creating the impression of there being more upper midrange energy in the first place, I'm not convinced it should be there. It's almost as if out of phase (inter-channel difference information) is being artificially added rather than being naturally revealed. Whatever it's genesis, I can tell you that it's real.
I performed a simple experiment where I wired the two stereo channels as a difference signal extractor (the old Hafler ambience extraction trick, where the speakers have their negative terminals wired together, but no longer to the amplifier). You can easily hear more out of phase inter-channel information via NOS than with the FIR filter. Whether this is a natural or a synthetic effect, I'm not certain.
3b) While the soundstage sounds deeper and more three-dimensional via NOS, it also sounds less wide. That may seem contradictory, but that is what I hear. The left to right spread of instrument placements was much wider with the FIR filter switched in, but was also much flatter in front to back depth and separation.
Conclusion:
NOS DACs can provide outstanding overall musicality when implemented along with some purposely warmly colored analog stage (a supposition on my part) to help balance this apparent shift of energy out of the lower midrange, as well as an anti-SINC equalizer to counter the zeroth-order hold based in-band treble roll-off".
Looking back now on that 2010 Asylum posting, I don't think I sufficiently noted the dynamic freedom and lack of strain on crescendos exhibited by my AD1865 based NOS/OS DAC. This is a huge subjective advantage which I should've better highlighted under the rubric of - What NOS gets right.
Last edited:
Excellent - thanks a lot! I really love such detailed descriptions of comparisons
Real science.Do you recall the part number of the filter - presuming it was an off-the-shelf chip you were using? Curious to know stuff like passband ripple (for time-smearing) stop-band rejection and OS ratio (4X. 8X?).
I'm totally with you on (1) non-fatiguing. More than lack of fatigue its also engaging, it draws me into the performance.
On 2) my initial experience was that it was weak - too forward and not enough ambience retrieval. But this was only on a particular TDA1543 - when I fixed up implementation details the sound stage depth improved considerably. My guess is that AD1865 has vastly superior low-level linearity which is what I (rightly or wrongly) correlate with soundstage depth.
As regards 3) yes again in agreement. What I suspect might be happening with cymbals in the OS case is that the half-band filter is aliasing above 20kHz. Bruno Putzeys has commented on how aliasing smears out HF transients so they appear to come from the speakers rather than from the recorded space. The solution is to abandon half-band filters and use apodizing ones.
On 1), were you subjectively aware of the loss of HF on NOS compared to OS?
2) I don't yet understand and haven't noticed anything similar myself to date.
3) I relate back to the half-band aliasing effects - smearing out of HF tends to mask the cues which allow us to listen in to the ambience of a recording space. The smearing could also have the effect of subjectively widening the soundstage, but this is a bit of a long-shot
This is a very interesting test which I might use when I'm developing a digital oversampling filter in future.
Lack of noise modulation is my working guess for this.
I see my remark about NOS and filtering caused a bit of a stir.
Rather than try to reply to every post seperately, I'm going to attempt to compress it into a single one.
SoNic_real_one writes about alias products with unfiltered DACs, but I'm not sure aliasing can actually occur during the DA-conversion stage.
AFAIK, aliasing occurs when (analogue) signals of a higher frequency (f) than half the sample frequency (fs/2) are sampled. So, during AD-conversion filtering remains a necessity. Once aliasing is introduced, it cannot be removed.
Traditionally it's thought of as necessary to use a low pass filter after DA-conversion to remove the image and the step like shape from the the reconstructed analogue signal. OS being used to negate the need for very steep acting analogue filters. But there is no "folding back" during DA-conversion.
The unorthodox thinking behind NOS DACs is that our ears have the needed low pass filtering by default. Perhaps with the exception of those that can hear above 22 kHz (but so far, I've only ever come across one person able to do that).
Perhaps generalizing about NOS despite only having heard this one for a few hours wasn't the best thing to do. But if this one is representative of what NOS can do, then I'm all for it.
I was auditioning the Octave through better and more revealing speakers than I have at home, for me there were no audible fake artifacts of any kind.
It was during certain music pieces that I thought that my CD-player actually sounded a little contrived.
I ended up buying the Octave yesterday, but the shop was out of stock and they wouldn't part with the demo. I can't wait until the new shipment arrives...
Rather than try to reply to every post seperately, I'm going to attempt to compress it into a single one.
SoNic_real_one writes about alias products with unfiltered DACs, but I'm not sure aliasing can actually occur during the DA-conversion stage.
AFAIK, aliasing occurs when (analogue) signals of a higher frequency (f) than half the sample frequency (fs/2) are sampled. So, during AD-conversion filtering remains a necessity. Once aliasing is introduced, it cannot be removed.
Traditionally it's thought of as necessary to use a low pass filter after DA-conversion to remove the image and the step like shape from the the reconstructed analogue signal. OS being used to negate the need for very steep acting analogue filters. But there is no "folding back" during DA-conversion.
The unorthodox thinking behind NOS DACs is that our ears have the needed low pass filtering by default. Perhaps with the exception of those that can hear above 22 kHz (but so far, I've only ever come across one person able to do that).
Perhaps generalizing about NOS despite only having heard this one for a few hours wasn't the best thing to do. But if this one is representative of what NOS can do, then I'm all for it.
I was auditioning the Octave through better and more revealing speakers than I have at home, for me there were no audible fake artifacts of any kind.
It was during certain music pieces that I thought that my CD-player actually sounded a little contrived.
I ended up buying the Octave yesterday, but the shop was out of stock and they wouldn't part with the demo. I can't wait until the new shipment arrives...
Last edited:
dont forget our ears are not the only thing in the signal chain after the dac and dont forget it is not only things we can hear directly/consciously that influence the experience of listening to music, particularly the ambient field. reflections, enclosures, amplifiers, drivers, some capacitors, rooms etc have these sounds bouncing around in them and can easily produce sidebands and diffractions and even oscillations in the frequency domain, as well as time and space. Some devices like wide-band amplifiers for instance, can react quite badly to having a HF product the circuit is not designed for injected into the signal.
if you enjoy it and are good with that caveat, fine thats totally your choice i'm not going to argue, but dont pretend its not an issue, its just an issue you are prepared to ignore. we all have compromises in our systems, some we are willing to accept, some we are not and these are different for everyone
if you enjoy it and are good with that caveat, fine thats totally your choice i'm not going to argue, but dont pretend its not an issue, its just an issue you are prepared to ignore. we all have compromises in our systems, some we are willing to accept, some we are not and these are different for everyone
Last edited:
You don't seem to have identified the items that are issues.
So we don't have to guess, what are you explicitly referring to in your 1st paragraph?
Its a good read - but his claim that:
DAC in these experiments is invariably ladder type
was made in 2007 and I've since done the NOS vs OS experiment with a TDA1387, most definitely not a ladder type. Bruno's not got an explanation yet for where the dynamic advantage of NOS comes from. Its a hypothesis I have that because he's got skin in the S-D game, he's going to ignore noise modulation. Or perhaps he has a proprietary fix for it?
DAC in these experiments is invariably ladder type
was made in 2007 and I've since done the NOS vs OS experiment with a TDA1387, most definitely not a ladder type. Bruno's not got an explanation yet for where the dynamic advantage of NOS comes from. Its a hypothesis I have that because he's got skin in the S-D game, he's going to ignore noise modulation. Or perhaps he has a proprietary fix for it?
You're talking about S-D, that is delta-sigma (D-S)?
Bruno is fully into pulse width (class d amps) and digital d-s; I met him some weeks ago, and in his (non-debatable) opinion NOS has serious flaws; maybe one more engineer who also needs to listen instead of lending his ears to the oscilloscope.
Time will tell.
Yes, because HiEnd clearly needs more "engineers" who don't know 1st year university math and design by "ear".
Moreover Bruno's approach clearly doesn't work because his designs sound terrible (pro audio and hiend market alike will agree) and not "analogue" enough.
Oh, be sure to visit a Shaman instead of a proper doctor when you have health issues (moderators please move this to the vendor's area if you feel I'm promoting my "trade"
).
yeah because designing Class D amps by ear would be really effective....
haha beaten to that line by TheShaman hehe
i'm curious about his views on convolver based DSP though, it does make me wonder if its just because he cant fit one effectively in the box. in my experience this type of DSP produces audio that is unlike anything ive ever heard.... at audio meets and shows included. i dont believe it can or should be used to completely replace competent speaker enclosure and room design, but it can sure help to augment those that live in the real and less than perfect world with less than infinite wallets
haha beaten to that line by TheShaman hehe
i'm curious about his views on convolver based DSP though, it does make me wonder if its just because he cant fit one effectively in the box. in my experience this type of DSP produces audio that is unlike anything ive ever heard.... at audio meets and shows included. i dont believe it can or should be used to completely replace competent speaker enclosure and room design, but it can sure help to augment those that live in the real and less than perfect world with less than infinite wallets
Last edited:
Yeah - some people put the sigma first, others like the delta to come before the sigma.
No, people like me need people like Bruno not to listen so as to create our own market niche. I celebrate his refusal to discuss NOS's benefits
No, people like me need people like Bruno not to listen so as to create our own market niche. I celebrate his refusal to discuss NOS's benefits
I know what you mean.
I just couldn't keep up with the more agile of my mates in basketball nor football, so I focused on body-building/gym-related stuff where I was better than most and I felt good about it.
qusp, I believe he's just trying to point out DSP shouldn't be expected to work miracles and you still have to design a proper loudspeaker and treat the room. They already utilize DSP in Hypex and Grimm.
I know what you mean.
I just couldn't keep up with the more agile of my mates in basketball nor football, so I focused on body-building/gym-related stuff where I was better than most and I felt good about it.
well played
I used to play power forward in representative and state level basketball till under 16, but then everyone became larger than me, soi worked on my outside jumpshot and still played good ball, i never did play starting 5 at state level under 18 though....
yeah i know that confused me initially; however i dont think they use convolvers, only digital crossover/EQ and maybe some delay, not full room correction. This way the XO can still be built into the box, while a convolver would mean they had to build the PC into the box, or shift the XO to the PC. the way i understand it the PC connection is just for uploading filter coefficients in their products.
I agree with it the way you state it though.
Last edited:
Excellent - thanks a lot! I really love such detailed descriptions of comparisons
I take offence in that afirmation. SCIENCE represent experiments that are controlled, measured and repeted by peers with the same results.
This is a bunch of blablabla, subjective oppinions, you can call it poetry if you like, but is NOT science.
Some would say that the Sigma block comes before Delta, so they call it Sigma-Delta. Some say that functionality is primarely Delta and Sigma is applied to the Delta product, so they call it Delta-Sigma.
Last edited:
It was an AD1896 ASRC chip, which contains a half-band FIR sinc filter that can be bypassed on-the-fly. The oversampling ratio was x4. Although, to be fair, this wasn't a totally controlled comparison between NOS and OS modes because the OS mode also features jitter suppression, which the NOS bypass mode does not.
I've always wondered about the notion of aliasing in the DAC reconstruction filter. By definition, a D/A converter can only reproduce baseband signals at the Nyquist rate of the conversion frequency. Aliasing artifacts can only occur symmetrically around integer multiples of the sampling frequency, all of which are above the baseband. Baseband signal aliasing would seem to only be possible at A/D conversion, and downsampling conversion. I believe that fully apodizing filters (fully in the stop-band by 19-20kHz) sound more natural for the same reason that NOS sounds more natural. They remove either the A/D or the D/A sinc filter response - see my further thoughts on this at the bottom.
While I was aware of the HF loss, the most obvious differences between OS and NOS seemed to involve midrange clarity and soundfield clarity, and the absense of listening fatigue even on music with little apparent upper register content.
3) I relate back to the half-band aliasing effects - smearing out of HF tends to mask the cues which allow us to listen in to the ambience of a recording space. The smearing could also have the effect of subjectively widening the soundstage, but this is a bit of a long-shot
My theory as to why redbook CD has traditionally dissapointed many audiophiles follows, taken from my recent posting on the subject in an AudioAsylum thread:
The technical problem I have with CD is that the intended signal (music) contains time-domain sensitive information. The 44.1ksps CD channel rate was designed with the frequency-domain requirements of music in mind. As far as I can tell, little to no consideration was given to the time-domain implications of the medium, either for recording or playback. The root of the time-domain problems are two-fold, as I see them. First, the fact that CD's channel bandwidth of 22.05kHz is so close to the recorded information bandwidth creates the requirement for very sharp anti-alias (recording) and anti-image (playback) filter responses. Which necessarily have a poor time-domain response, manifesting as the now fimiliar high-Q filter ringing. As it turns out, this severe ringing is fundamental to producing an accurate frequency-domain reconstruction of the original signal upon playback. Yes, I'm also aware that such ringing occurs at the edge of the ultrasonic range and should be pretty much inaudible, but that is only half the story, to which I'll shortly return. If the information to channel bandwidth ratio were wider, as it can be with high-res. digital, then both the anti-imaging and anti-aliasing filters could be much less sharp, with greatly reduced time-domain distortion. Mike Story of dCS has published a paper concluding that to be one of the reasons why high sample rate digital sounds superior to CD.
My own empirical experiments lead me to suspect that a second, non-obvious mechanism is also at work. I suspect that the time-domain problems extend to the dynamic inter-action, or intermodulation, of the near ultrasonic ringing responses of the multiple SINC filters utilized from recording through to playback. Resulting in artifacts within the audible range.
One of my self designed experimental DACs contains a programmable digital SINC filter. This programmable filter has enabled me to empirically evaluate oversampling, non-oversampling, and apodising digital filters. Here's what I heard. As is well known by now, non-oversampling, aka, NOS - which eliminates the playback SINC filter but does not affect the recording and mixing SINC filters - produces the natural and non-fatigueing sound so typically lacking in CD. Apodising - which retains the playback SINC filter, but removes the affect of the recording and mixing SINC filters - sounds equally natural and non-fatiguing as NOS. Oversampling - with all SINC filter responses in place - on the other hand, produces the typically fatiguing and course CD sound.
My hypothesis is that the primary source for what we have come to know as digititus, or traditional CD sound, is the dynamic intermodulation of the multiple SINC filter time-domain (ringing) responses of the standard CD recording and playback chain. Eliminating ONE OR THE OTHER of these sinc filter responses, so that only a single sinc filter reponse remains, greatly restores the natural and non-fatiguing quality otherwise absent. This hypothesis would also explain why high sample rate audio is often disappointingly not completely rid of unpleasant CD type artifacts. The use of SINC filters across a high sample rate chain could still produce time-domain filter response interaction which are audible. I will surmise that high sample rate recording-playback chains which take advantage of the extra channel bandwidth not for increased frequency-domain signal capture, but for utilizing anti-alias and anti-image filters having much less ringing in their time-domain responses, will subjectively provide the best high sample rate audio quality.
I've not yet developed an experiment to test this hypothesis, so it may prove faulty in so far as the exact distortion mechanism responsible is concerned. The empirical results, however, have been consistant and very obvious.
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.