Metrum Octave Dac - What are the Chips used

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
That seemingly contradictory fact helped lead me to the hypothesis I had suggested in the final several paragraphs of post #120 here: http://www.diyaudio.com/forums/digi...ctave-dac-what-chips-used-12.html#post2968716

I suspect that much of the annoying subjective aspect of CD audio stems from the dynamic interaction of the MULTIPLE (at least two) sinc filter responses across the recording/playback chain. Remove either the ADC sinc response via apodising, or remove the DAC sinc repsonse via NOS and, viola', much more natural sounding reproduction results. I would add that while eliminating one of those two (at the least) sinc responses in the chain substantially improves the sound, it may also be that removing BOTH would improve the sound to it's ultimate point. Something which could be done using high sample rate audio.

If you chain linear phase brick-wall filters the result is the same sinc impulse response. You can't get any sharper than a brick-wall.

You can see and compare the effects of various filters by doing filtering "offline" in software on a PC. Try a blind test converting 192 or 96 KHz source to 44.1 or 48.

You could also test your hypothesis with a WM874x DAC that has the selectable filters and select the minimum phase apodizing filter response.

In my opinion, the fact that all the integrated filters are half-band and thus only 6 dB down at the Nyquist is worse than the ringing. Bruno Putzeys has also commented on the pre-echo caused by the in-band ripple.
 
That seemingly contradictory fact helped lead me to the hypothesis I had suggested in the final several paragraphs of post #120 here: http://www.diyaudio.com/forums/digi...ctave-dac-what-chips-used-12.html#post2968716

I suspect that much of the subjectively annoying aspects of CD audio stems from the dynamic interaction of the MULTIPLE (at least two) sinc filter responses across the recording/playback chain. Remove either the ADC sinc response via apodising, or remove the DAC sinc repsonse via NOS and, viola', much more natural sounding reproduction results. I would add that while eliminating one of those two (at the least) sinc responses in the chain substantially improves the sound, it may also be that removing BOTH would improve the sound to it's ultimate point. Something which could be done using high sample rate audio.


Interesting, time to take out the DSO again see if I can measure any difference.

I have a 99 track Denon Test CD from 1993 (EC 3991-2), track 59 features an impulse recorded at 0 dB.
I ripped it to HDD and looked at the waveform in an ancient version of Cool Edit. At first it looked like there's indeed a lot of ringing in the waveform, but then I realized that the red dots are the samples and the blue line is what the waveform would look like after DA-conversion. Cool Edit shows the reconstructed analogue waveform with the dots where the actual sample was. Double-clicking on a dot shows the sample value of that dot in the decimal system.

So, if you ignore the blue line and connect the dots with straight lines, there's still ringing, but the amplitude is nowhere near as high as it would have been after DA-conversion. In other words: AD-conversion does create ringing, but not as much as DA-conversion. This should show on a measurement with my DSO.

It does, look at the results. It think this might answer the question why NOS may sound better despite (some) ringing in the AD-conversion stage.

From left to right:
- Waveform displayed by Cool Edit as if it were already DA-converted the traditional way;
- Octave, 44.1 kHz, only the ringing from AD-conversion;
- DAC1, 44.1 kHz, added ringing from DA-conversion OS-style;
- PDR-555RW, 44.1 kHz, added ringing from DA-conversion OS-style;

A bit off-topic, but I found out this way that the Micromega DAC1 has phase inverted on the unbalanced analogue output. To correct, I used the digital phase inversion-fucntion on the DAC1.
 

Attachments

  • Image4.gif
    Image4.gif
    10 KB · Views: 281
  • oct_imp_44k.gif
    oct_imp_44k.gif
    4.6 KB · Views: 282
  • dac1_imp_44k.gif
    dac1_imp_44k.gif
    4.7 KB · Views: 277
  • 555_imp_44k.gif
    555_imp_44k.gif
    4.8 KB · Views: 255
Last edited:
That seemingly contradictory fact helped lead me to the hypothesis I had suggested in the final several paragraphs of post #120 here: http://www.diyaudio.com/forums/digi...ctave-dac-what-chips-used-12.html#post2968716

Thanks for drawing my attention back to your earlier post. I hadn't fully digested the import of what you were suggesting there :) Yet even though I've mulled over your ideas some more now, I still have some nagging doubts...

I suspect that much of the subjectively annoying aspects of CD audio stems from the dynamic interaction of the MULTIPLE (at least two) sinc filter responses across the recording/playback chain. Remove either the ADC sinc response via apodising, or remove the DAC sinc repsonse via NOS and, viola', much more natural sounding reproduction results.

Here's one of my concerns. You don't suggest a mechanism for how or where the imtermodulation between sinc artifacts occurs. In the case of a normal OS DAC with normal OS ADC this is presumably in the DAC chip and/or subsequent analog stage? Do you have any meat to put on the bones of this - for example could it be avoided without going to NOS?

Secondly, presumably in the case where some kind of ASRC has been used on a recording along with a normal ADC then there will already be two sinc responses embedded into the data. If your hypothesis is correct then ISTM those recordings won't sound truly NOS-like even when played back with a NOS DAC. Has anyone here found such a recording - one which nullifies the NOS sound?

I would add that while eliminating one of those two (at the least) sinc responses in the chain substantially improves the sound, it may also be that removing BOTH would improve the sound to it's ultimate point. Something which could be done using high sample rate audio.

I can think of another interesting experiment - find out if any recordings use NOS ADC. I'm also curious to discover whether a non-OS apodizing filter does even better than NOS with no filter. I might even try this last experiment myself :) When I listened to my own designed min-phase apodizing filter at 2X OS I found the sound not to be as sweet as pure NOS - I put this down to increased glitchiness from the DAC when running at 88k2 rather than 44k1.
 
Thanks for drawing my attention back to your earlier post. I hadn't fully digested the import of what you were suggesting there :) Yet even though I've mulled over your ideas some more now, I still have some nagging doubts...

Thanks, for recognizing it. :)

Here's one of my concerns. You don't suggest a mechanism for how or where the imtermodulation between sinc artifacts occurs. In the case of a normal OS DAC with normal OS ADC this is presumably in the DAC chip and/or subsequent analog stage? Do you have any meat to put on the bones of this - for example could it be avoided without going to NOS?

I agree with this statement. I have not suggested a fleshed out theoretical mechanism, and essentially admit as much in my comment #120. I wish that I had the mathematical analysis skills to perform a proper investigation in to what audibly occurs when either apodising or NOS are alternately implemented. My hypothesis about there being some kind of dynamic interaction, or, perhaps, some peculiar intermodulation between the recording and playback sinc filter responses is mostly based on logical deduction stemming from my empirical listening experiments with OS, NOS, and apodising filters.

Secondly, presumably in the case where some kind of ASRC has been used on a recording along with a normal ADC then there will already be two sinc responses embedded into the data. If your hypothesis is correct then ISTM those recordings won't sound truly NOS-like even when played back with a NOS DAC. Has anyone here found such a recording - one which nullifies the NOS sound?

Yes, I should think that the only instances where more than two independent sinc filter responses would be involved is for sample rate conversion. I'm just now thinking that, perhaps, the independent aspect of the separated ADC and DAC sinc filters may be critical here. While most chip based OS filters are comprised of a chain of multiple smaller FIR units, such filter chains are coherently designed and implemented as a unified architecture. This is only conjecture, however, on my part.

This is an intriguing question. It may just be, that recordings which have been subjected to an ADC filter + ASRC filter + DAC filter won't sound particularly natural, even upon apodising or NOS.:D

I can think of another interesting experiment - find out if any recordings use NOS ADC. I'm also curious to discover whether a non-OS apodizing filter does even better than NOS with no filter. I might even try this last experiment myself :) When I listened to my own designed min-phase apodizing filter at 2X OS I found the sound not to be as sweet as pure NOS - I put this down to increased glitchiness from the DAC when running at 88k2 rather than 44k1.

I believe that we share opinions regarding apodising vs. NOS. While I hear the same type of natural, or relaxed quality via apodising as I do via NOS, overall, apodising lacks treble energy, more so even than NOS. I suspect that you may find the same results for 44.1ksps apodising as for x2 OS apodising, depending on the stop-band frequency you are using. To effectively remove the ADC sinc filter ringing with a DAC apodising filter requires, from my observations, that FULL stop-band rejection occur by 19kHz-20kHz, not the 22.05kHz Nyquist frequency typically touted by vendors promoting their commercial apodising filters.

Having the stop-band at no higher than 20kHz means that the transition band will typically extend down to around 18kHz. This produced a much more obvious high frequency curtailment than did the -3dB@20kHz of NOS. While naturalness, clarity, and effortlessness was about equivalent between them, apodising's more obvious high roll-off tended to rob life from instrument harmonics, rendering the overall subjective presentation slightly inferior when compared to NOS.
 
Last edited:
I can think of another interesting experiment - find out if any recordings use NOS ADC.

I may be mistaken, but aren't SACDs (DSD) recorded sans any ADC anti-alias filtering, and played back sans any DAC digital reconstruction filtering? If so, it may explain the positive subjective qualities of SACD. One would think that some audiophile oriented label would have released some 96k sample rate high-res. tracks recorded without any ADC anti-alias filtering. Anti-alias filtering wouldn't seem to be necessary with a 48kHz frequency-domain signal bandwidth.

It's too bad that Sony chose such an unecessarily low quantizer resolution, given that the x64 oversampling ratio also chosen really isn't very high, even using high-order noise-shaping. But that is another can of worms.
 
Last edited:
Hi,

abraxalito said:
I can think of another interesting experiment - find out if any recordings use NOS ADC.

Many early CD's where made with Sony Processors that operated at 1 X FS sample rate. Apogee got their start in digital audio by making replacements for the elliptical LC filters for these machines. Other processors often worked the same. It too a fair bit of time for oversampling to gain wide use in the industry. Audiophile Labels where the early adopters. Sadly there is no way to tell for sure.

I may be mistaken, but aren't SACDs recorded sans any ADC anti-alias filtering or DAC digital reconstruction filtering? If so, it may explain the positive subjective qualities of SACD. It's too bad that Sony chose such an unecessarily low quantizer resolution, given that the oversampling ratio also chosen isn't that high at x64.

SACD is basically the DS Modulator output from a generic ADC originally meant for transferring analogue recordings for later release on CD. On a fundamental level the principle is flawed, both due to the DS nature and the fact that for the claimed dynamic range and bandwidth the required dither would overload the modulator.

This is generally shown by the simple observation that most (all? - but I have not measured all) SACD Players have a rising noisefloor in the audio range and by 20KHz have only comparable SNR to CD. While I am not in general a fan of adding dither, in principle CD can achieve a similar result using suitably noiseshaped dither during the mastering.

I suspect "pure" SACD primarily sound better than their PCM counterparts because editing and applying effects to DSD is even now not well and widely supported, compared to the proliferation of such tools in the "PCM Universe", where so many dynamic range and other effects are now routinely applied that to my "old skool analogue minimalist recording" ears even recordings that are by todays standards nearly untouched sound sound over-produced and overcooked.

With SACD it is actually very hard to get the same kind of sound (unless mastering and applying FX in PCM and then converting to DSD) so the sound of DSD based recordings often more closely resembles reality.

THAT SAID, really good CD-Recordings using really high grade CD-Replay, to my ears at least (and as I demonstrated in the early 2K's at a few London HiFi Shows to many others) can achieve the same or greater subjective sound quality as that offered by SACD.

Ciao T
 
Many early CD's where made with Sony Processors that operated at 1 X FS sample rate. Apogee got their start in digital audio by making replacements for the elliptical LC filters for these machines.

However in those days, dither was not really well understood amongst the broad sweep of designers (I'll pre-empt Thorsten saying 'nothing's changed there then'). I recall someone (I think it was Paul Frindle) telling me that the replacements were initially lower noise but sounded worse. Until someone woke up to the fact that the circuit noise was providing the necessary dither to linearize low-level performance. Increasing the noise from the earliest Apogee filters improved the sound. (I could have got this story a little bit garbled over the intervening time - so if somebody's got a less fogged version please chime in :)).

Also generally ADC linearity sucked then too. Crystal only came out with its fiendishly clever self-calibrating CMOS switched-capacitor ADC designs in the mid 80s I seem to recall and even then they started out at 12bits.
 
Hi,

I was thinking of getting hold of one of ADI's newest breed of ADCs based on SAR (rather than S-D) for my projects. They look really good spec-wise and have become quite affordable - http://www.analog.com/static/imported-files/data_sheets/AD7988-1_7988-5.pdf

These look nice.

Personally I'd probably want to stack 8 pcs per channel run interleaved with a suitable following summing logic based NOT on classic digital filters. This should allow effective 16 X Oversampling at 192KHz, or the addition of 8 more encodable levels, giving us 18-19 usable bits, which should suffice where 16 Bit may not.

Ciao T
 
I listened to my own designed min-phase apodizing filter at 2X OS I found the sound not to be as sweet as pure NOS - I put this down to increased glitchiness from the DAC when running at 88k2 rather than 44k1.

One final thought about the subject of apodising not sounding as sweet. Try reversing the absolute phase when listening to the apodising filter versus NOS. It may make no difference, but I regularly find that the absolute phase needs to be reversed (much to my annoyance) when switching from one digital reconstruction filter algorithm to another, else the sound be grainy or less sweet.
 
Last edited:
I can think of another interesting experiment - find out if any recordings use NOS ADC.

I found out that with Cool Edit you can edit individual samples, so I should be able to eliminate the ringing that occurred during the recording-stage. Cool indeed. Later today I will give it a go...

No doubt the Octave will show the difference, but I'm curious it if has any effect on the OS DAC. Will ringing be less? Anyone care to speculate?
 
Last edited:
SACD is basically the DS Modulator output from a generic ADC originally meant for transferring analogue recordings for later release on CD. On a fundamental level the principle is flawed, both due to the DS nature and the fact that for the claimed dynamic range and bandwidth the required dither would overload the modulator.

This is generally shown by the simple observation that most (all? - but I have not measured all) SACD Players have a rising noisefloor in the audio range and by 20KHz have only comparable SNR to CD. While I am not in general a fan of adding dither, in principle CD can achieve a similar result using suitably noiseshaped dither during the mastering.

I suspect "pure" SACD primarily sound better than their PCM counterparts because editing and applying effects to DSD is even now not well and widely supported, compared to the proliferation of such tools in the "PCM Universe", where so many dynamic range and other effects are now routinely applied that to my "old skool analogue minimalist recording" ears even recordings that are by todays standards nearly untouched sound sound over-produced and overcooked.

SACD has the fame of being airy and sweet, which i correlate to lots of HF material (and a great deal of distortion up there), plus the typical "sweetness" of S-D.
BUT often the SACD recordings are mastered much better than the equivalent CD layer -another important factor why many audiophiles says that it sounds better :)

THAT SAID, really good CD-Recordings using really high grade CD-Replay, to my ears at least (and as I demonstrated in the early 2K's at a few London HiFi Shows to many others) can achieve the same or greater subjective sound quality as that offered by SACD.

JVC "24 bit" ones for instance are amazing.
 
You are talking smak about SACD just to feel good about yourself? "Lots of HF noise"? Did you ever listen to a SACD? There is no audible noise present in audio band, the sounds come from a dark background, the noise at 20kHz is some -110dB. If you average the noise with "A" curve you get something lower than that. CD can only dream of -96dB at digital zero (not using "mute").
Sure, some SACD are not native DSD recordings and I can easily spot the ones that are PCM-based. Usuall are releases of old albums or live concerts.
This is a real-live (Sony SCD-1) SACD player noise spectrum:
Scdfig16.jpg
 
Last edited:
I found out that with Cool Edit you can edit individual samples, so I should be able to eliminate the ringing that occurred during the recording-stage. Cool indeed. Later today I will give it a go...

No doubt the Octave will show the difference, but I'm curious it if has any effect on the OS DAC. Will ringing be less? Anyone care to speculate?

Here are the measurements with the ringing removed from the impulse.

While I was editing the samples, Cool Edit already "predicted" what the impulse would look like on a traditional DAC. As you can see in the first pic, the ringing is shorter in length (time) but higher in amplitude. This, shorter but stronger ringing shows in the measurements from the PDR-555RW, but not from the DAC1. The signal from the Octave confirms that the ringing was indeed removed from the recording.

@Ken Newton, comparing these measurements with my previous impulse measurements with the ADC ringing still present (post 282), I'm beginning to think that those formed during the DA-conversion overshadow those from the AD-conversion. I'm even thinking that, with the ones from the AD-conversion still present, ringing is actually slightly less severe. Wether they interact or not is hard to tell from the measurements.

From left to right:
-Waveform in Cool Edit predicts ringing analogue domain after OS DA-conversion despite removal of ringing from AD-conversion;
- Octave, 44.1 kHz, confirms removal of ringing;
- DAC1, 44.1 kHz, only DAC-ringing;
- PDR-555RW, 44.1 kHz, only DAC-ringing.
 

Attachments

  • Image1.gif
    Image1.gif
    16.8 KB · Views: 231
  • oct_editedimp_44k..gif
    oct_editedimp_44k..gif
    4.5 KB · Views: 230
  • dac1_editedimp_44k..gif
    dac1_editedimp_44k..gif
    4.8 KB · Views: 231
  • 555_editedimp_44k..gif
    555_editedimp_44k..gif
    4.9 KB · Views: 235
Last edited:
Here are the measurements with the ringing removed from the impulse.

@Ken Newton, comparing these measurements with my previous impulse measurements with the ADC ringing still present (post 282), I'm beginning to think that those formed during the DA-conversion overshadow those from the AD-conversion. I'm even thinking that, with the ones from the AD-conversion still present, ringing is actually slightly less severe. Wether they interact or not is hard to tell from the measurements.

Thanks, for taking the time to conduct and show these graphs. I can see what you are suggesting about the ADC sinc response maybe not being nearly as important as the DAC filter. However, it appears that a proper DAC apodising filter produces a subjective sound quality which is very similar to what one hears via NOS. Keep in mind that a 'brick wall' apodising DAC filter produces the same amount of impulse repsonse ringing during playback as any other 'brick wall' filter. The only difference is where the stop-band rejection begins. This is intriuguing. It probably should be said that a filter doen't HAVE to be minimum phase in order to remove the ADC sinc filter response. It could just as well be a linear phase apodising filter.
 
Hi,

You are talking smak about SACD just to feel good about yourself?

Whoever you directed this remark at aside, it is seriously out of order. Not that it matters though, but please refer to the forum rules.

Now let's get on with facts instead of discussing rude boyz lingo.

"Lots of HF noise"? Did you ever listen to a SACD? There is no audible noise present in audio band, the sounds come from a dark background, the noise at 20kHz is some -110dB. If you average the noise with "A" curve you get something lower than that. CD can only dream of -96dB at digital zero (not using "mute").

Well, well well.

Incidentally, CD does not have to dream 'of -96dB at digital zero (not using "mute")' (BTW, hint, the correct technical term to use is digital silence). It actually easily exceeds this, if DAC's with 16 Bit equivalent performance (such as the TDA1541A and other similar ones) are used. For example the TDA1541 is rated as -110dB noise with digital silence. The noisefloor is much increased by the 16-Bit quantisation noise if for example a -90dBFS signal is applied.

But, how about we instead compare SACD and CD with some signal, shall we (after all, no-one listens to digital silence, well, no-one I know anyway)?

Here the graph from the same test set at stereophile you appropriated yours from:

Scdfig15.jpg


I do not wish to to be rude, but looking at this I though to myself: "W. T. F. is THAT!?"

The first thing we observe is that there is actually more HF-Noise (HF here used in the technical sense of "High Frequency" as compared to AF which means "Audio Frequency") than there is signal. At least I personally would call that "Lots of HF noise"...

What you should really see is the same plot at 0dBfs. A beaut, 10dB more HF-Noise than actual signal! I know a fair few Amplifiers that will not be very comfortable with such a signal. In fact almost any Solid State Class AB Amplifier is going to have a hissy fit if you feed it THAT kind of signal. My own Amp's actually would not mind, but that is another story.

Another way would be to call all this HF stuff "200% added HF distortion". Of course, added HF distortion of as much as 30% is the very crime that "Non-OS" DAC's are routinely charged with... :D

The second thing we observe is the shape of the -60dB Peak, which is smeared all over so that -100dB it already covers well over two octaves. I like to call the cause of what this this graph shows "Fuzzy Distortion" (as opposed to the distinctly separate phenomena of jitter) and it is indeed inherent to the use of a 128 Level at 44.1KHz (that is a system that without noise-shaping is equivalent to 7 Bits sampled at 44.1KHz).

To be honest, any PCM System that measured like what is shown above I would consider "defective by design". Of course, it is not "revoltingly bad PCM" but instead "Super Audio CD", so it has to be good, after all it SO SUPER compared to CD with it's 16-Bit's at 44.1KHz and anyway, what is a 9dB shortfall among friends?

Let us compare this with a "State Of The Art" result for CD standard PCM, that is state of the art in 1989, in the form of the TDA1541 equipped Philips LHH-1000 (which is incidentally one of my references):

666PHILFIG2.jpg


While this shows -90dBFs, the noisefloor of this system does not change with signal levels above 1LSB, we can see it offers around 6dB improvement over SACD at 20KHz. Moreover, we would find a much, much narrower peak at -60dB than with SACD,

This should actually not surprise anyone who understand a minimum about digital audio.

As for why do some people prefer the sound of SACD? No idea*.

Ciao T

* If one was rude and witty, one may, hypothetically speaking, be tempted to suggest that they have not been talking, what was mentioned earlier, but instead have been taking it. But that would be rude and against the rules.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.