In essence this just means that op-amp macro models are not suited for this type of analysis. It does not mean discrete version works better in reality.
I would agree with that. This range (appx 200Hz - 2kHz)is where the fundamentals and formant harmonics of instruments reside. I was going to post a big philosophical veer on measurements vs. what we hear, but I lack time and energy. I need to rebuild my PC with i5-10600K and new cooler and then overclock it.
In short, at the current state of affairs I view the objective measurements of signal fidelity impairments and the perception of audible fidelity impairments as MOMA, or "Minimally Overlapping MAgisteria", following the NOMA concept:
"Non-overlapping magisteria (NOMA) is the view, advocated by palaeontologist Stephen Jay Gould, that science and religion each represent different areas of inquiry, fact vs. values, so there is a difference between the "nets" over which they have "a legitimate magisterium, or domain of teaching authority", and the two domains do not overlap."
So in this case minimal overlap suggests that there is a certain degree to which current limited objective tests (moreover even more limited by only testing a limited set of available tests most of which derive from marketing vehicles) can describe the audible side, but for large parts of the whole the two exist in separate - non-related magisteria or domains.
Why the difference is sound quality between multibit and Delta Sigma?
I would point straight at the operating principles and at the limitations of linear PCM.
DS essentially represents the waveform as an average of many pulses and ultrasonic noise. It is in effect a noisy line with no discrete steps or discontinuity.
It has, implemented correctly and end to end (or with sufficient interpolation from PCM) a behaviour more akin to analogue tape (which ultimately is of course digital as we have distinct magnetic dipoles and we flip more or fewer of them to express the waveform.
I personally like to call the DS system "non-deterministic" as at any given discrete point in time we cannot predict the actual output state (before lowpass) from the analogue input, but using a large enough time window we can average to something that is still not deterministically tied to signal state but progressively approaches it, the larger the time window.
Another way is to say that a DS system is always wrong, but on average it approaches "right". It is also notable that this kind of model of operation is better the less actual signal is present, large scale signals pose challenges.
By comparison an ideal PCM system is absolutely deterministic, any input value will be reliably and deterministically mapped into a numerical space and it absolutely predictable (to better than 1LSB) and the output from the system in turn is again deterministic.
Our biggest issue is that we are mainly deal with 16/44.1 source material. This was acceptable when the source being digitised was an analogue 15/30ips master tape. Once we progress into direct to digital recordings and editing, as Decca found, at least 48/18 was needed to get past audible fidelity impairments. I would posit that had CD been made either 66.15k/18Bit or 72k/18Bit LPCM we would have no debate.
Another issue is that LPCM is LINEAR (that is every step is identical is size) and hearing is logarithmic. Thus PCM has the most resolution at high levels and moderate frequencies.
LogPCM might have been a way to get more resolution out of 16 Physical bit's, but while today a LogDAC (and ADC) is trivial, it was not in the late 70's.
So, where does that leave us?
PCM and DS sound different and debating why in a meaningful way means we first must dramatically increase the overlap in our "MOMA's".
Thor
Last edited:
Quite right, that was also my conclusion in #3009 & #3013, especially the Opa2210 model behaved far worse as the real one.
That’s why I swiched to a discrete version.
But also take into account that not everybody likes the sound of opamps, so there are always discrete alternatives.
That’s just another way of viewing at it.
Hans
P.s. also keep in mind that the whole exercise was an attempt to replicate the odd spectral lines that were caused by modulation.
When this had been possible, it would have been easier to test alternatives.
But despite all efforts, it did not bring anything,
but is was fun trying.
Last edited:
I got the same feeling, but it does not comport with what I hear on electrostatic speakers, along with the rest of my system here. Horn speakers and NGF amps might give a very different impression. Part of it is surely human preference though.
EDIT: Forgot to mention, my dual SE Andrea dac is different from the one listened to at the session described by @Joseph K
The one here uses two SE dac boards and dual power supplies to get stereo Vref. Also, found a trick adjusting damping of the BCK lines to the dac boards. In addition, I am using SOA clocks, not Accusililcon at the moment.
Last edited:
As You might noted, I am talking about a specific, different DSD dac model (without Andreas clocks, different topology, no FiFo, etc..) Did you have the opportunity to listen to that specific design? Which is a very promising, liked what I heard, just not the same sound profile which is warmer, more impactful & vivid in both the R2R and DSD 'classic' versions.
Large size electrostats are throughout most of their operating range chaotic mode resonators, you cannot reliably predict the exact position of a specific point on the diaphragm reliably with the input signal.
Together with many other challenges they are neither low coloration or low distortion (objective Magisteria) despite being hawked as the ultimate in transparency.
FWIW, here I listen to horns, in the main system Technics "Linear Phase" E-100's:
The Amplifier is "super high GNFB" non-switching Class AB (read Class AA) solid state amplifier with build in TDA1541 DAC fitted with a dual PLL for clock recovery and a short memory buffer (at the factory) driven from a USB-2-SPDIF converter from an micro PC (Windows etc.).
No room treatment, mostly bare room, modern minimalist design, well stuffed sofa. Speakers are toes in at 45 degrees.
Thor
Attachments
I did try the USB board and Accusilicon clocks with the stereo Vref SE dacs. That was the impetus for designing a clock board.
Of course, we all make choices. I am not unfamiliar with large Electrostats.
Back in the 90's a friend had multiple home made systems. One of the biggest achievements in his designs was to gain relatively good control over the "chaotic resonator" nature of large ES Panels. It was never commercialised.
The System also featured transformer direct drive of the panels and a rather different approach to Biasing etc.
In this system, on listening, there was a general preference for multi-bit DAC's over Delta Sigma among a wide range of listeners and a TDA1541 + Tube Out in this system beat out the resident Ultra-Analogue DAC Module based HDCD DAC (also a system with a dual loop PLL).
And I would take similar descriptions of the sound as you gave.
As the system was hosted in a large extension that was used as rehearsal space for early/classical music (replete with a baby grand piano, organ and several spinet's, we often had occasion to do recordings in this space followed by direct replay.
Multibit ADC and DAC was judged more realistic than involving any DS. There was a vividness that DS lacked. Note, in these days recordings were 48/24 PCM with ADC/DAC being a variety, including AKM/Cirrus.
As said, this now decades ago, my friend long has passed out of this plane of eXistenZ and the system was broken up.
Thor
It shouldn't be, when the ADC and all requantizers are properly dithered.
https://en.wikipedia.org/wiki/NICAM
ESL 63's have a fairly flat response and a low distortion at most frequencies.
I don't know if it is still the case, but the main issue with (DIY as well as commercial) electrostatic loudspeakers used to be that many were designed by people who didn't know or understand Walker's equation. They would try to drive electrostatic loudspeakers with a frequency-independent voltage, leading to a +20 dB/decade far-field response. They would then try to correct for that by not damping the fundamental resonance and by using much too large radiating surfaces for the high frequencies, leading to cancellation effects between the signals coming from different parts of the diaphragm.
As ESL panels go, recent vintage Sounds Labs are probably about as good as it gets. There is sort of a lay person's white paper at their website. Don't laugh too much though, its not written for engineers.
https://www.soundlabspeakers.com/wp-content/uploads/2019/07/The-Complete-White-Paper.pdf
File photo:
https://www.soundlabspeakers.com/wp-content/uploads/2019/07/The-Complete-White-Paper.pdf
File photo:
Almost unmatched by any other speaker as long as you don’t need a booming bass reproduction.
The only problem is that cats don’t like them.
Hans
My present cat did like them, but used them for climbing rather than listening. It would have been no problem if he had been a bit more careful when jumping off.
My previous cats never felt the urge to climb on them. One of them urinated against an ESL 63, but that happened only once in over 12 years.
My previous cats never felt the urge to climb on them. One of them urinated against an ESL 63, but that happened only once in over 12 years.
Clearly shown with my exercise was that opamp models in LTspice are not suited for this type of mixed analysis.
However in general discrete designs mostly are processed quite correct in LTSpice.
The fact that in real life the OPA2110 and the OPA1632 showed modulation products with low level signals proves that they are susceptable to HF modulation.
The discrete 797 like version in LTSpice with double clock speed did not even show the least modulation products with a whopping 60dB higher spectral content around fs/2, and processed the signal faithfully.
IMO this could well be a sign that a discrete version is much better armed against HF modulation than an opamp but of course it’s no prove.
Hans
This is with 1/3rd octave smoothing.
Low distortion - I found that the distortion has many spikes (even ESL-63 which were quite common in London).
Yes, this applies not only to DIY Electrostats.
This problem of partial resonances patterns from a large undamped diaphragm and the problematic horizontal directivity was what my friends designs addressed.
The downside was that below somewhere between 80...200Hz it was necessary to hand over to dynamic woofers. For the largest system there was a 4-Way setup with a crossover at ~120Hz and 480Hz. Below 120Hz a "M" shaped dipole with dual 18"Pro Audio drivers and the final extreme low bass handled by 18" drivers venting into the basement of the house taking over at around 30Hz IIRC (yes, this system could reproduce a 32' organ pipe and even a 64'one).
I doubt a purely ESL system with a single Panel would be able to compete.
Even then I preferred horns at my owm place. How to say, I respected this system and appreciated what it could do, but it was more of a cerebral than visceral experience of music.
Thor
With a classic Multibit ADC or DAC there is no dither and no requantitisation.
If you used Pacific Microsonic Model 1/2 at 88.2/96k or 176.5/192k you got in effect a direct recording via a 20 Bit Multibit ADC and playback via 20 Bit Multibit DAC.
The ADC and DAC modules used were direct Voltage In/Out.
The units are still legendary:
https://reverb.com/item/14637550-pacific-microsonics-model-two-hdcd-processor
In a TRUE PCM system there is no dither and no requantisation.
Nicam used transcoding from 14 Bit 32k LinPCM into 10 Bit LogPCM to fit a digital audio signal into the analogue Colour TV Signal. The DAC used however was not a LogDAC and the source signal was 14Bit/32KHz linear PCM.
So NICAM is a CODEC, not a recording system. NICAM actually survived and is now in effect the main high quality Bluetooth Codec - aptX.
Thor
Yu
In the attachment the left ESL63 recorded in my living room without any correction.
One curve is taken on axis from 84cm distance, the other curve at my listening position.
I have the same recording from the right speaker, practically identical.
Hans
What a ridiculous and unrepresentative frequency response with -30dB at 20Khz.
In the attachment the left ESL63 recorded in my living room without any correction.
One curve is taken on axis from 84cm distance, the other curve at my listening position.
I have the same recording from the right speaker, practically identical.
Hans
Attachments
The pic is an example from Sound Lab's website. Don't know what the amps are. My particular amps at the moment are a pair of Marantz MA9S2. They handle the loads quite well.
BTW, the Sound Lab speakers here are good down to around 28-30Hz. Then a pair of REL subs extends down to 14Hz.