In my (old) book, balanced connections are useful when long lengths of cabling (meters and meters...) are required, for example in a recording studio or pa set ups... Though it has recently been converted by objectivist reviewers into a fashionable trick to improve sinad (basically by cancelling H2 products which do no harm at all...) and impress picky users of compact desktop set ups with low technical culture, and convince them that unbalanced connections suck...😊
Last edited:
You perfectly indicate the limitations of measurements: "known metrics". My statement: telling you can describe the sound of a system based on a few measurements creates perception bias by those who are not understanding that measurements are only particular viewpoint.
The 20Hz-20kHz times
When I was about 18 years old, as a true technical oriented person was convinced that a 20Hz-20kHz flat range with a 96b dynamic range was sufficient knowledge to describe the sound of a device sufficiently. When my first CD player (Akai CD-D1) broke a couple of times (laser tracking errors), I exchanged it for a Sharp CD player. It sounded different, less engaging, but I was convinced I was fooling myself. 20Hz-20kHz, right? Also the Sharp was not very stable, and after a few repairs I got a Sony CDP-<something>, also sounding different (better than the Sharp). But you know, 20Hz-20kHz, right? OK, there was oversampling and pass-bands and phase distortion, and you need to take care how to connect your ground planes between analog and digital, but this was all understood and engineered in. All the above players were oversampling 4x, so I must have been fooling myself, right? Because my believe in measurements was stronger than what I thought I heard.
Après 20-20kHz
After I started my PhD (in electronic science, I'm a data-driven rational guy), I went into an audio shop to look for new speakers. There was a demo with some "high-end" stuff, with a separate DA convertor. "What a nonsense, if your CD player is designed properly etc.:". The sound was nice and good, but there was some problem, so the DA convertor was detached, and the CD player was directly connected. The difference was stunning, and when investigating in the background, the reported measurements all looked the same. My curiosity started. I wanted to explore and design my own DA convertor.
Over the years. the world of EMC, PCB design and impedance matching, non-ideal components and jitter opened up. All of those things mattered, despite that for all of those the 20Hz-20kHz measurements were the same. I was blinded by a religious belief that a specific set of measurements would characterise the sound of a device. Don't get me wrong, measurements are a crucial and important tool to understand a device on certain aspects, and know what you are doing, but they are not defining.
Jitter
One step further: despite all the pre-cautions to prevent jitter at the DA convertor IC, we could still hear (yes, hear) differences when different type of transports were connected. Measuring jitter at different places in the design (spectrum and time domain analysers) did not give much info. So the measurement guru will tell you that you are affected by the perception bias ghost, not? Well, if you cannot measure an absolute phenomenon, you can also measure differential signals, so as there was a PLL in the design to stabilise a clock from an SPDIF signal, we have a phase comparator. So we could measure (and listen) to the difference between a recovered SPDIF clock signal and the average constructed stable clock to do the conversion. When music was started, you could hear the music in the differential signal, indicating there was music correlated anomalies in the clock. The loop was slow (by intention), and with some transport we slowly heard the music disappear, and only noise remained. With other transports the music stayed audible in the clock difference signal. Whatever it tells about the quality, transports did have an effect on the clock signal. With this method, you could even measure different spectra with pink noise recorded on different brands of CD-R, played with the same CD transport. Repeatable, and consistent...
Moral of the story
Explaining the taste of a pizza by its salt-sour-sweetness balance and a spectrography to guess about its ingredients is probably not the best way to tell how it tastes. Yes you can measure everything, but you need to know which measurement to do to explain a phenomenon, not the other way around. This is how science works. If you suspect your ears give unreliable results for some test, do double blind tests (which also does not work everywhere). If two devices measure equal with a certain set of measurements, they don't necessarily sound equal. That's reversing the implication arrow, and is a scientific flawed reasoning.
The listening test
The test was done at my house, with my Ayre QB-9 Twenty DA convertor. Fedde was interested to compare, as another friend had less good experience with the perceived sound from another Topping DAC. The differences were not small, and tested with the same track with obvious characteristics. E.g. a track with small subtle metallic chime bells in the background, that with the Ayre sound metallic (like bells...) and resonating, flanging and decaying (like a bell), and was hanging in a holographic space in the middle and behind the speakers. The Topping sounds muffled, shut-in, voices are not pinpoint-able, the bass is wooly, and there is a lack of separation because the texture is not explicit, and the 3D image was much flatter. We swapped to some different filters in the Topping, but that was not bringing the devices closer together.
From experience with other equipment (e.g. Sennheiser HD 650 or 660S, with and without Harman Curve applied, and on a variety of headphone amplifiers, or with a Grimm LS1 and many other loudspeaker systems), I know those chime bells should sound metallic (in case you want to know, Jeanette Lindstrom, Walk, track 2: This Time).
The Topping is a nice product for its price, but if you want to get an ultimate view into your music collection, I wasn't too impressed with what I've heard.
Btw, the concept of transparence in the mouth of objectivist measurements only believers is a tremendously misleading one..
But i also understand It eases online sales a lot...
But i also understand It eases online sales a lot...
Hey everyone I hooked up my Russet potato and the bells were so bell like I new I had reached Nirvana. This was in comparison to my previous reference red-skinned potato with mayo which was in comparison grainy, darker. Back in the day I used to buy first one potato, then two potato, then three, and the store owner told me that they were all excellent spuds, but I know better now.
See the problem here?
See the problem here?
Combining USB potatoes running in adaptative mode was way more fun, costed less, and did not sound so bad... 🤣
Yes I do see a problem. Your only contribution so far is to try to make your point by open and unproven assertions, try to make topics sound ridiculous by taking things fully out of scope, and a content-based reply is fully missing. I will definitely not make that my problem.
Last edited:
In the meantime, I also did a comparison with the Okto research dac8 PRO. Also here, I am afraid the DM7 was falling behind. The Okto DAC sounds more natural and live, while still being detailed (though it costs roughly 1k extra, so it should sound better as well). So I decided the DM7 is not for me and I sent it back. Pity. Not sure what is actually wrong inside the DM7. There must be some low level degradation somewhere in DM7 that does not show up in the ASR measurement set.
Thats unlikely. Given few can even hear distortion under -60dB, a DAC with -120dB HD3 and IMD, cant be expected to have any perceptible distortion signature. As noted above, current measurements do not predict timbral characteristics of well engineered and exemplary measuring DACs.
Only his ears can know the truth, of course with some help by his eyes and with a dose of bias.
My DM7 should arrive early next week.
My DM7 should arrive early next week.
Maybe you meant "ASR measurements" instead of "current measurements". There are many other measurements than what ASR is using.
But more importantly even if this "timbral characteristics" is not seen in measurements there is no way to tell whether or not Okto is better in this respect than DM7. It may as well be that Okto has low level "timbral characteristics" degradation somewhere but some find it more pleasing.
The only way @TNT your debunking could actually happen is if there is some parameter or characteristic that could be measured, and that then could be correlated in double blind testing with "good" and "bad" examples. So, that would be again measurements.
TBH I have always wondered why there is not more time domain characterization of DAC waveforms under a variety of conditions, e.g. waveform type, transient conditions, etc. instead of frequency domain measurements that do not capture this info...
TBH I have always wondered why there is not more time domain characterization of DAC waveforms under a variety of conditions, e.g. waveform type, transient conditions, etc. instead of frequency domain measurements that do not capture this info...
Seems like there are two main problems with that: (1) We need instrumentation that is more SOA the the DUT, or we at least need some way to differentiate artifacts of the two systems so we know what to attribute to the DUT, and (2) we need data analysis results in a form that makes sense to humans.
Regarding (2), say, for one example, assuming we have instrumentation more accurate than the DUT, we can digitize transient input and output data for the DUT. We could then run an FFT on the data, we could view the data in the time domain, we could look at modulation envelopes using Hilbert Transforms, etc. There are various ways we could try to analyze the data.
However, let's say for example we look at the FFT of a transient event. There will be a spray of frequencies, but what do they mean? How to interpret if the result is good or bad as compared to some other DUT? Its hard to make sense of what we see in a way we can easily understand.
OTOH, a steady-state HD FFT is trivial for a human to interpret. If there are harmonically related spurs that aren't supposed to be there, that's obviously HD.
In the case of time-domain analysis, we are looking for tiny aberrations in our acquired data to make sense of. If you have even looked at magnified time-domain sample points in a DAW, how can we make sense out of a few samples that aren't quite the precise amplitude they should be? Compared to another DUT, how do figure out which is better in a way that makes sense to us and that correlates well with blind listening tests?
Last edited:
Honestly if you just need to see the time domain performance a DSO should do the trick. No?
For example:
http://www.pavouk.org/hw/dacshapes/en_index.html
Steady state FFT into frequency domain cannot capture any information about short time events, e.g. time resolution is very poor when frequency resolution is high.
For example:
http://www.pavouk.org/hw/dacshapes/en_index.html
Steady state FFT into frequency domain cannot capture any information about short time events, e.g. time resolution is very poor when frequency resolution is high.
Last edited:
A DSO is typically good to about 8-bits. Do you find the sound of 8-bit dacs to be audibly transparent? Why should you expect an 8-bit scope to resolve enough low level details to correlate well with blind listening tests of 24-bit dacs? Doesn't make sense that it should, does it?
Regarding FFTs, if phase information is not discarded, then no information is lost. An inverse FFT can restore all the time-domain information. Therefore, its really a question of whether or not you know how to make sense of the data in a particular representation.
BTW, this issue of making sense, of being data being comprehensible and understandable, isn't only a problem in audio, its a much more general problem spanning all the way from how business decisions are made to how likely it is that an AI will ever be able to design a even smarter AI. In its general form, it a big problem.
Regarding FFTs, if phase information is not discarded, then no information is lost. An inverse FFT can restore all the time-domain information. Therefore, its really a question of whether or not you know how to make sense of the data in a particular representation.
BTW, this issue of making sense, of being data being comprehensible and understandable, isn't only a problem in audio, its a much more general problem spanning all the way from how business decisions are made to how likely it is that an AI will ever be able to design a even smarter AI. In its general form, it a big problem.
Last edited:
Some things you could see with 4-bits, but so what? That isn't a general solution to the measurement of audible differences in 24-bit dacs.
To put it another way, with an analog scope you can see down to around 1% crossover distortion. Is that good enough to evaluate a Benchmark AHB2 amplifier?
To put it another way, with an analog scope you can see down to around 1% crossover distortion. Is that good enough to evaluate a Benchmark AHB2 amplifier?
Well the signal I showed has almost zero distortion and a smooth frequency response. Unless you look at the signal in the time domain you would never have a clue that this sort of waveform is appearing. So, yes, I think a scope or other time domain signal capture device is needed, and signals from DUTs need to undergo some series of trials with the results inspected in the time domain.
When there is extremely low distortion and smooth, extended frequency response via the current methods, perhaps it is time to look elsewhere for clues?
When there is extremely low distortion and smooth, extended frequency response via the current methods, perhaps it is time to look elsewhere for clues?
- Status
- Not open for further replies.