I would say that the effect is fr related. As mentioned earlier it shows its best when it starts to affect hearing range. When tuning loudspeakers, changing highs level by as little as 0.5db has very audible effect.
I would say that the effect is fr related. As mentioned earlier it shows its best when it starts to affect hearing range. When tuning loudspeakers, changing highs level by as little as 0.5db has very audible effect.
I am a bona fide professional loudspeaker designer, and changing the level anywhere by 0.5dB is indeed very audible - as it will affect a much broader FR spectrum than what we are discussing here. This is more like cutting a tiny cut off the corner of a picture and finding that the picture has been greatly affected and not quite able to figure out why, after all, it was just a few millimeters that was cut off.
Nobody is saying that we can't hear FR changes, please, we are not so ludicrous to suggest that. What we are saying that here we may have to look beyond that and it may only be a secondary effect that indicates a 'marker' since this is consistent with different delta-sigma DACs behaving so similarly.
Another important data - and that it has only been observed withd delta-sigma DACs and not Ladder type DACs.
So, again I iterate, this is not the usual FR thing. Otherwise non-delta-sigma DACs would show the same imagined (so it is being implied) outcome.
Hands up those who are just been plain skeptic and not actually gone futher than being an armchair critic...?
Hmm... I imagine I can see a fews hands going up...?
Cheers, Joe
Hmmm... this is not the first time I have not been taken seriously - there was that time about Buffers... oh well..
Another important data - and that it has only been observed withd delta-sigma DACs and not Ladder type DACs.
So, again I iterate, this is not the usual FR thing. Otherwise non-delta-sigma DACs would show the same imagined (so it is being implied) outcome.
Not under the OOB noise filtering hypthesis, no. The OOB spectrum of multibit DACs is very different from that coming from S-D DACs. It would be a serious blow to the OOB filtering hypothesis if the perceived difference was still there. This datapoint has to be explained by the 'its doing something to the DAC' theory though. What could it be doing to an S-D DAC that it doesn't do to a multibit one?
Looks to me as though the 'its all imagined' contributors have departed from the thread Joe - keep taking pops at them and they might return... 😛
To move away from arm waving arguments to some real understanding I think the following need proper testing. First since the I/v converter is intimately entangled in the circuit presented there are two possible interactions at that point. First it could be internal slew limiting in the opamp. Second it could be a perception issue from the HF rolloff. If neither of these then further research is warranted.
Proper testing of the internal slew limit possibility is pretty straightforward. It would show in a difference in HD or IM distortion. The second would require a subjective test, which is very hard work. I learned long ago that I can fool myself easily. I have since seen other professionals get trapped by believing what they want too much. This would need a controlled blind test to separate the effect described from an equivalent low pass filter further down the chain. In my experience small response changes manifest themselves as significant changes in things like space, presence, detail etc. That is why it can be really hard to test.
However, look at the phase impact of the filter. That could be significant when added to all the other low pass filters at 20 KHz in a digital chain.
I have long advocated for low pass filters on the DAC prior to any external opamp. I don't have access to any DACs with I/v stages (or ladder DACs) so I really can't do much actual testing. If some option presents itself I'll certainly try some options.
Proper testing of the internal slew limit possibility is pretty straightforward. It would show in a difference in HD or IM distortion. The second would require a subjective test, which is very hard work. I learned long ago that I can fool myself easily. I have since seen other professionals get trapped by believing what they want too much. This would need a controlled blind test to separate the effect described from an equivalent low pass filter further down the chain. In my experience small response changes manifest themselves as significant changes in things like space, presence, detail etc. That is why it can be really hard to test.
However, look at the phase impact of the filter. That could be significant when added to all the other low pass filters at 20 KHz in a digital chain.
I have long advocated for low pass filters on the DAC prior to any external opamp. I don't have access to any DACs with I/v stages (or ladder DACs) so I really can't do much actual testing. If some option presents itself I'll certainly try some options.
This datapoint has to be explained by the 'its doing something to the DAC' theory though. What could it be doing to an S-D DAC that it doesn't do to a multibit one?
Personally I see following things.
1. That such a filter (please understand I come from a loudspeaker design background) cannot ring. Have you seen the 2nd order filters on the output of Class D power amps - they are easy to make them peaky and high Q - which is absent in a 1st order filter where the termination is resistive.
2. The benefit I am hearing is very much like optimising damping in speaker design - it is so uncanny. I know that character change so well - or is that just a coincidence?
3. The filter's slope being very gradual and starting to 'bite' at a much lower frequency than high order filters, that being 1.3dB down at 20KHz is not the benefit, but a 'marker' and not in it self a plus, but is a penalty that needs to be paid and weighed against something else that is a positive? One thing (audible reduction in treble) versus... whatever is happening? That the -1.3dB is nothing more than a compromise that most of us can live with?
4. Is the fact that it is also consistent whether the DAC is voltage or current out? As long as it is a D-S DAC. Is that a clue as well?
5. Is the cap simply improving the integration of pulses into analogue, slices of voltage (voltage DAC) or current slices (current DAC)? Is the cap actually still furnishing additional need to reconstruct the audio/analog?
This may well be the most intriguing of all.
How would you describe the output on a delta-sigma DAC?
Is this it:

Is the Blue what comes out of the delta-sigma DAC and the reconstruction in Red is required in the post-DAC analogue processing/filtering?
If you agree with that - then I think this one is in the most likely tray.
-----
Final thoughts for now:
Also, I discount to a certain extent (but not totally) the internal slew rate on the basis that it is too consistent with DACs from different manufacturer. Why is -1.3dB well suited to DACs we have tried; Burr-Brown 'current' DACs, Cirrus Logic DACs ('voltage') and of course the Sabre DAC.
Re organised blind testes. I would welcome it. If anybody (not me, as I would be considered to close to it) was to arrange a blind test on this - I think it would be worthwhile. Really, it is not a subtle change in frequency response that will be heard, but something that even one totally non-audiophile heard in seconds and she didn't even know it was a blind test.
I am open to any suggestion - but that question above re what comes out of delta-sigma DACs are just zeros and ones - is that even analog? Or have I got it wrong? I can take it if I am, if the correction is made in good spirit and faith.
Looks to me as though the 'its all imagined' contributors have departed from the thread Joe - keep taking pops at them and they might return... 😛
Me "taking pops" at them - and here I thought they were "taking pops" at me? 🙂
But I do understand what you mean - and yes, hopefully the heat has died down - but insulting me, a somewhat veteran in the service of audio and I am probably old enough to be their Father - yep, perhaps even their Grandfather?
To illustrate - I know exact where I was when it was announced on the radio that Kennedy had died at Parkland Hospital.
Cheers, Joe
Last edited:
5. Is the cap simply improving the integration of pulses into analogue, slices of voltage (voltage DAC) or current slices (current DAC)? Is the cap actually still furnishing additional need to reconstruct the audio/analog?
That somehow reminds me of a couple of my comments very early in this thread:
It really sounds like it is just acting as a first-order low-pass filter that removes HF noise - but wouldn't the DAC already have a low-pass stage?
But shouldn't that already be done by a reconstruction low pass filter in the DAC?
So yes, I think some of us suggested that alternative rather early on. 🙂
It is ones and zeros (or rather 0 and full voltage/current) in a PDM pulse train - until integrated into a contiguous analog waveform by a low pass filter (capacitor).I am open to any suggestion - but that question above re what comes out of delta-sigma DACs are just zeros and ones - is that even analog?
Rather unlikely. 🙂But I do understand what you mean - and yes, hopefully the heat has died down - but insulting me, a somewhat veteran in the service of audio and I am probably old enough to be their Father - yep, perhaps even their Grandfather?
Being a northern european, JFK was never a big/memorable thing to me - unlike the space flights and ultimately Apollo 11.To illustrate - I know exact where I was when it was announced on the radio that Kennedy had died at Parkland Hospital.
I still think that it would be interesting to know who has attempted listening for the described effect, and whether they do or don't believe they hear it. If you choose to respond, please specify the DAC chip, I/V circuit input resistance, and the value of the filter cap. used in your listening test, thanks.
I would personally like to read some negative impressions from somebody who really experimented with this...
Good post for thoughts Joe...........Post #327..........
Own experience for optimising dampning on speaker drivers is very audioable, therefor i understand point 2 very well relative to listening.
Guess us diy-level people don't have equipment to measure such things, probably need very pro equipment, but of course we have our ears.
I would prefer that the example circuits shown in the device manufactures product/application notes was the optimal for the device to reproduce natural/optimal.
Little sidenote, played with simulation for voltage circuit and if going to often used differential OPamp just after the cap without buffering first, the cap makes a form of amplitude resonance up higher.
Last edited:
Guess us diy-level people don't have equipment to measure such things, probably need very pro equipment
Not really. These days cheap USB-based analog front-end devices combined with fast PC processing gives us very affordable and capable measurement systems, doing everything that we used to need very expensive scopes and spectrum analyzers for.
So yes, I think some of us suggested that alternative rather early on. 🙂
It is ones and zeros (or rather 0 and full voltage/current) in a PDM pulse train - until integrated into a contiguous analog waveform by a low pass filter (capacitor).
Rather unlikely. 🙂
Yet, this is where I do think it is likely - but you are right about it being mentioned earlier, and also dismissed somewhat.
I like the fact you call it PDM, which I kinda prefer over PWM. If this is a demodulator/integrator issue, then D for density is better for me pictorily, the way I view it in my head. Nut maybe that is just me?
Being a northern european, JFK was never a big/memorable thing to me - unlike the space flights and ultimately Apollo 11.
I am Northern European as well. It was an evening we first heard that Kennedy had been taken to hospital and about an hour and a half later they announced he was dead. I was in my bedroom shared by my brother and the radio was on in the adjoining lounge room. We lived in Mørkhøj, a suburb of Copenhagen.
Cheers, Joe
Not really. These days cheap USB-based analog front-end devices combined with fast PC processing gives us very affordable and capable measurement systems...
A lot less expensive than it used to be... and that is great. But the operator is important too.
How much does Audio Precision cost these days. My own Cliofw should not be several grand plus - but many top loudspeaker designers use it as a standard, and there is a kind of snobbery there as well. Some companies must have Audio Precision to be taken seriously - so they won't come down in price?
Cheers, Joe
I very much doubt that APs will come down in price. Second hand ones will always be more affordable though, particularly the out-dated models. The problem with the older ones is the PC dependency - the interface card issue.
In future with ADCs getting better, it'll be possible to design something much cheaper and more flexible than AP in the digital domain. That'll be a paradigm shift. I'm not sure ADC performance is quite there yet, but its hard to keep abreast of developments.
In future with ADCs getting better, it'll be possible to design something much cheaper and more flexible than AP in the digital domain. That'll be a paradigm shift. I'm not sure ADC performance is quite there yet, but its hard to keep abreast of developments.
.. but of course we have our ears.
Yes, we do. And for all the pitfalls of human hearing, and those of us who have been around in developing things and recording studios, know how easily you can be fooled, even by yourself. But that also gives you experience that helps you to deal with it honestly.
But to suggest you are fooled every time is like throwing the baby out with the bath water. We do hear things - and ultimately, when late at night and you want to play a bit of music after a long day, when you are relaxed (and maybe even by yourself) and not doing any critical listening and zero pressure to perform, that is a test that cannot be replicated by a controlled test.
I note that the doyen of controlled listening tests for decades was Martin Colloms, who did heaps of them, way back into the 70's. These days he has changed his attitude towards those tests markedly. They don't replicate the way we listen naturally - when you are walking through an environment and taking in natural sounds, you brain is not in an analytical mode. There is no stress, only your brain is processing what it hears without additional stress loads or doing a task while listening.
What has been found is that controlled listening tests produce a load on the brain (not the ear but on the ear-brain, as you cannot separate them) that is not natural. It actually lessens your ability to discern what you are hearing. It impairs judgment. You are multi-tasking and that is affecting your ability to judge.
So we should not overly elevate controlled listening tests, neither should we totally abandon them, but surely we must also take into account that it has limitations, as much as off-the-cough listening has both good and bad points.
Remember what Heisenberg said - by merely testing something, we are affecting the outcome of the measurement, hence there will always be a some level of uncertainty. Hence we have the Heisenberg Uncertainty Principle.
This totally, using a bit of lateral thinking, IMO, has proved true to auditory perception.
Little sidenote, played with simulation for voltage circuit and if going to often used differential OPamp just after the cap without buffering first, the cap makes a form of amplitude resonance up higher.
For sure - and if you use the ubiquitous 5532 opamps - it has been estimated that there has been made more 5532 opamps than the number of people who have ever lived - and they are in landfills all around us. A sort of plague upon humankind? 🙄
Cheers, Joe
Not under the OOB noise filtering hypthesis, no. The OOB spectrum of multibit DACs is very different from that coming from S-D DACs....
Yet, this is what my gut instinct tells me is at the heart of the issue.
But more than just gut...
May I hearken back to 2001, when Allen Wright and yours truly, would take the output of the VC24 Digital Filter chip (before, and hence bypassing, the DAC) and put that through a bandpass filter - filter the average DC and normalise it (a coupling cap) and then a low-pass filter followed by a UGS Buffer that gave us 0.6VRMS output.
Now, if I have the facts right, and Guido Tent may recall this, Allen (as I was told) was informed by Guido that the DSD signal coming out, discretely into Left and Right, was to be considered as a 'digital' signal that we could not brought outside the box - or else Philip's lawyers would be banging on the door (Guido was working for Philips when SACD/DSD was being developed).
That DSD signal, considered to be a digital signal by Philips (and even CDs were upsampled to DSD in Sony in the previous generation CD only players), could be reconstituted into analog by a simple passive filter. This I believe is called 'demodulation'?
My question: There really is not that much difference between between the DSD/PDM output of the VC24 chip and the PDM output of delta-sigma DACs?
Am I right? Are they not so similar as to be the same?
Sorry, but the reason I point this out, is because in my mind I remember Philips view of the DSD/PDM output as still being a digital signal.
Is the post-DAC filter being used as a demodulator and only after that function can we really say that we have an analog signal?
I am not a digital engineer, so I am quite aware of my limitations here, but sometimes an outsider sees what the insider does not see?
It would be a serious blow to the OOB filtering hypothesis if the perceived difference was still there.
That is the big question !!!
But what is an 'hypothesis' if it cannot be questioned? Surely that is what makes it a hypothesis.
I think we are dealing with OOB effects. And I think what we may be hearing is an audio signal that has not been totally extracted as an audio signal yet and the horrible and almost inconceivable thought that we need a filter that is an incursion on the sacred 'flat to 20KHz' rule - and that is why it has been overlooked?
Please note, I am just trying to ask thoughtful questions.
Niels Bohr: "Every sentence I utter must be understood not as an affirmation, but as a question."
Cheers, Joe
Last edited:
My question: There really is not that much difference between between the DSD/PDM output of the VC24 chip and the PDM output of delta-sigma DACs?
Yes and no (he says, hedging his bets). Its not digital coming out of today's chips. Single bit has problems - Philips abandoned it in their chips after the TDA1547. The next generation used multibit - but fewer bits than 16. Hence TDA1305 which has what I'd call a 'low bit' modulator. They built a 4 or 5 bit DAC with their 'continuous calibration' architecture, still fed it from a noise shaper. Thus its no longer 'digital' or PDM, its noise shaped oversampled PCM converted to analog. Other manufacturers followed suit, minus the CC.
If they were the same then it would negate Philips' reasons for designing the TDA1305.Am I right? Are they not so similar as to be the same?
That's the biggest difference - with a multilevel modulator (as found in all of today's DAC chips) it no longer is digital.Sorry, but the reason I point this out, is because in my mind I remember Philips view of the DSD/PDM output as still being a digital signal.
Sure it should be challenged, tested. But it can't be just thrown out without replacing it with another one which fits better. The only way a hypothesis gets junked is by replacement or by experimental falsification i.e. observation.But what is an 'hypothesis' if it cannot be questioned? Surely that is what makes it a hypothesis.
Your 'there's an effect on the DAC' idea isn't a testable hypothesis so far. Its in need of a proposed mechanism by which an experiment can be designed to test it.
I'd ask Niels - what question then is contained in your above words which look to me to be an affirmation about your own statements?Niels Bohr: "Every sentence I utter must be understood not as an affirmation, but as a question."
Last edited:
...My question: There really is not that much difference between between the DSD/PDM output of the VC24 chip and the PDM output of delta-sigma DACs?
Am I right?...
Joe,
A 1-bit encoding/decoding system, as is SACD/DSD, only requires the representation of two states. Call those two states high and low, or 1 and 0, or positive going and negative going, or rising and falling, or whatever, but only two states need be captured, recorded, and later reproduced. Even a simple logic gate or two-state driver could form the D/A conversion circuit for a 1-bit encoding/decoding system. Often, such simple 1-bit D/A converters are implemented within a custom ASIC, or an FPGA device, where many such simple two-state D/A circuits can be inexpensively implemented as a massively parallel array.
Having only two states available to accurately represent some source with inherently greater resolution, such as CD, will produce a huge analog amplitude errors. This error is called, quantization-noise. A 1-bit D/A converter can only be correct, by coincidence, on 2 amplitude levels out of approximately 65,000 possible levels on a CD (assuming identical sample rates). This sort of reminds me of that old joke about a stopped clock being correct twice a day.
With such a massive levels of amplitude error, the trick with sigma-delta, whether implemented as 1-bit (SACD/DSD), or as multi-bit (typically, 5 - 6 bits), is to somehow separate that massive amount of quantization noise from the desired baseband signal which is buried beneath. A digital processing technique called (quantization) noise-shaping performs this task by greatly oversampling the signal, and utilizing the resulting ultrasonic spectrum to relocate the undesired quantization noise. Now, all that needs done is to filter away that ultrasonic spectrum, and along with it goes the quantization noise - which, remember, is actually the signal's amplitude error which has now been separated from the desired signal via spectral relocation.
The lower the bit resolution of a given sigma-delta system, the more it depends on both high ratio oversampling, to open spectrum for the relocation of quantization noise, and on high order noise shaping to relocate the noise there. Sony employs something around 9th order noise shaping for 1-bit DSD. Sigma-delta systems based on higher resolution quantizing, such as 5-bits for Burr-Brown and 6-bits for Wolfson, require less extreme oversampling and noise shaping to deliver the same effective baseband resolution as DSD.
On the plus side for sigma-delta is that it's much easier (therefore, less costly) to make a smaller resolution quantizing unit linear than it is to make a full resolution multi-bit quantizing unit linear. In fact, a 1-bit quantizer features inherently perfect linearity, as two points (binary states, in this case) can only define a straight line, not a curved one. This advantage can seem counter-intuitive, because while the quantizer linearity with 1-bit is perfect, the signal amplitude error (the quantization noise) is also the highest.
But it can't be just thrown out without replacing it with another one which fits better. The only way a hypothesis gets junked is by replacement or by experimental falsification i.e. observation.
Isn't that a hypothesis to support a hypothesis? 😀
I agree it would be preferable - but not always.
(But don't you think we have such an observation? I think we do.)
Actually, on that point I agree with Richard Milton who said:
"It is the customary fate of one who delivers a fatal stroke to be called upon to replace the deceased theory with a better one. This is thoroughly illogical, quite unfair, and perfectly understandable." 🙂🙂🙂
I find myself at odds with Richard Milton about a number of things, but here he does make a good point. Left the chips fall where they may and don't shackle dissent/disagreement or where observation throws doubt.
Sometime 'science' can become too conservative - it wants to hang on to hard-fought 'gains' and is loathe to change, even thinking about such a change. Then we are in danger where a hypothesis can become dogma. Many like the comfort zone it affords them.
In Perth, there are two scientists that millions of sufferers can thank because they came up with an hypothesis that proved true in the end, and yet they were opposed every step of the way. Not now though, not after you get a Nobel Prize. They were told their patients could not have been cured of ulcers from their treatment, but they were being treated successfully.
What boils down to this, we do have, despite the doubters, an effect (or call it what you will) and there has to be a mechanism to explain it. I would like to see one.
I think it is a matter of one or two things, and those two things are related, reconstituting an acceptable analog signal and noise.
Unless I can be persuaded otherwise, that is my present position - one that I am quite open to shift if a better source can be supplied.
Call it my hypothesis.
Surely, if an hypothesis is inadequate to explain such an 'effect' (and let's not go backwards again on that score), then should a search for a new hypothesis not start?
In the quantum world we do not have certainty all the time. We cannot expect it always - but when nature behaves in a certain way, should we not at least search for the best explanation. Physics was never about figuring out how the world functions, but what we can say about nature. Einstein knew this, Bohr said it (actually Eintein did say it, just in a different way). We constantly are constantly looking for explanations (or what we can say) about all kinds of quirky behaviour.
For example, I often consider myself a mystery. 😀
(Please guys, don't take that one the wrong way.)
So let us not rely on an existing hypothesis just because we can't have an instant replacement (as we don't seem to have a spare one in our back pocket right now) - the road traveled is not always that easy.
Cheers, Joe
PS: Just got a phone call from a guy in Victoria, I sent him two small caps for his Oppo 105, which he got fitted by a very good tech down there. He immediately heard something and the first thing he said that came to mind was the word "clean" - I queried him and he clearly was hearing a lack of graininess and it was now sounded so much more fluid. He said he had only listened for a short time and was so surprised by what he heard, he picked up the phone straight away, to let me know that those two caps had made an instantly significant change - and he knew right away for the better - he has little audiophile language skills.
- Status
- Not open for further replies.
- Home
- Member Areas
- The Lounge
- DAC Filtering - the "Rasmussen Effect"