Interview with Chord Electronics' Rob Watts - Part 3: Why Chord DACs Sound Better (and MQA's Issue)

Подписаться 39 тыс.

Просмотров 8 тыс.

50% 1

Congratulations to Marcel for winning the Airist X R2R DAC giveaway! Join up here for a chance to win future giveaways: / passionforsound
This is Part 3 of a lengthy interview with Chord Electronics' DAC designer, Rob Watts, a legend in the audio industry and a leader in DAC design theory. I recommend watching Parts 1 & 2 first.
In this episode, Rob gets into details about filtering approaches in DACs including his WTA filters, regular interpolation filters, and MQA. He also shares some interesting knowledge about how we perceive sound and why a DAC's filters are so crucial to the quality of the sound reproduction. It's a pretty techy episode so I chime in regularly to attempt to make sense of it all.
PASSION FOR SOUND GEAR LIST
Desktop Sources
Playback software: Roon roonlabs.com/
Primary DAC: Chord Qutest amzn.to/303FAL7
Secondary DAC for active speakers: Topping D50s massdrop.7eer.net/AVbG1 or amzn.to/2T2GExr upgraded with Muses02 op-amp amzn.to/2FtPmNj
USB cables: AudioQuest Diamond amzn.to/35AX4Qf and AudioQuest Coffee amzn.to/3059qiA
Active Speakers
Bang & Olufsen BeoLab 3 Speakers
Amps
Massdrop x THX AAA 789 Amplifier massdrop.7eer.net/J0xWa
Bottlehead Mainline DIY kit tube amplifier bottlehead.com/product/mainli...
Headphones
Open: Meze Audio Empyrean (may or may not be available at this link) amzn.to/2tDs19h
Open: Focal Clear amzn.to/37Ihgky
Closed: AudioQuest NightOwl Carbon (these may have been discontinued)
Closed: Meze Audio 99 Neo amzn.to/2FxOHKQ
Portable
Phone: Samsung Galaxy S10+ amzn.to/37MzTUA
DAC for phone: Audioquest Dragonfly Cobalt amzn.to/36AzpAB
Digital audio player: Pioneer XDP-300R amzn.to/2FwGnL5
Earphones
Noble Kaiser 10 custom IEM (discontinued) - universal version available via Drop massdrop.7eer.net/vkaWy
Campfire Audio Andromeda (may or may not be available at this link) amzn.to/39WFxpd
FitEar ToGo! 334
SIMGOT EN700 PRO amzn.to/304xO3R

Наука

Опубликовано:

11 июл 2024

Ссылка:

Скачать:

Готовим ссылку...

Добавить в:

Мой плейлист

Посмотреть позже

Комментарии : 30

@allee5480 2 года назад

Great interview series. I finally understood a little bit of what Rob Watts is doing with millions of "taps". I have an answer to your question about how it can be that all these engineers disagree on filters and DAC design. The answer is simple: in a perfect world, your DAC would take in the full piece of music, say all three minutes of a song, and for each sample create an *analogue* sinc (sin(x)/x) waveforms for the whole duration of the song. The DAQ would then add up these 3 minutes x 60 seconds x 44,100 samples per second = 8 million separate waveforms, each scaled to the height of its sample and offset for when its sample happened in time, for every point in time and have that be the analogue output of the DAC, perhaps after applying a low pass filter to keep the output to only signals less than 22.05 kHz. This would be doing exactly mathematically what Claude Shannon said was necessary to reproduce *exactly* the original bandwidth limited analogue signal that was sampled at 44.1 kHz. This is computationally very expensive, and has the downside that the song cannot start playing until the full song has been loaded into the DAC and processed because the sinc function is "heavy tailed" - has a lot of amplitude far from the central peak. Plus no piece of electronics produces perfect sinc analogue waveforms as output. All DAC designers make a compromise, doing less than the ideal DAC in producing perfectly scaled, time-off set, and summed sinc analogue wave forms. Their choice of the approximation to this ideal is where they engineers disagree. Watts has gone for adding FPGA preprocessing that takes about a second of samples (44,000 for red book CD input) around the point in time being considered, does the work of adding up digitally those sinc waveforms for about 20 interpolation points between each 44 kHz source sample (hence the 768kHz output of the Hugo M Scaler). Other engineers believe other approaches are a better approximation to the ideal. The truth is in the measurement tests and the listening tests which may find issues that instruments cannot yet measure, but are there.

@PassionforSound 2 года назад

Thanks for the detailed information. It would be so cool to see someone make that ideal DAC - perhaps one with two DAC stages so that the second track is being converted while the first one plays

@birdlandbill7867 4 года назад

This is a wonderful interview series and put forward with broader perspective and very nice "civilian" commentary. One of the most dispassionate presentations I've listened to. Congratulations and thank you for asking questions many of us have and for keeping us honest about other solutions to DAC design! Really enjoyed how you stood your ground re: R2R DACs. Love hearing discussions of DAC design at this level, particularly from the remarkably generous Mr. Watts. Truly understanding what he is saying at the maths (as opposed to EE material) level is an undertaking beyond most of us, but he provides a deeper gestalt here than I've ever seen before. Got me looking at noise shaping in greater detail. In terms of listening pleasure, I've benefited greatly with Chord's Hugo M-Scaler and Chord Dave. Started with Chord Mojo, which was such a revelation and which I still greatly enjoy for travel and more casual listening, and thought it worth while to move up the chain. Haven't regretted at all and trialed dCS, Berkeley and others and kept coming back to Chord.

@PassionforSound 4 года назад

Thanks so much, Bill! Rob is a lovely guy to speak to and as passionate about musical enjoyment as he is about the engineering behind the products. I have to admit that I was ever so slightly uncomfortable at times with his statements about others' approaches being wrong (e.g. filtering approaches), but he always has a compelling and insightful reason behind his views so I can't help feel that maybe he's just a genius in this space and others maybe are really missing some of the finer details. So glad you enjoyed the interview. Thanks for watching!

@birdlandbill7867 4 года назад

@@PassionforSound Could tell that there was some discomfort and think your interpretation of where he is coming from is spot on. Our inability to correlate unmeasurable changes with clear perception differences shows that there's more to this than software and hardware. There is wetware (brain signal processing/responses) to consider as well. There is little consensus understanding on any of these. Small wonder that we obtain sound quality improvements that outstrip our understanding/maths. You and Rob did a wonderful job of expressing this.

@tugbars4690 2 года назад

I can't thank enough for these videos. I learned a lot of things from these videos.

@PassionforSound 2 года назад

So glad you found them helpful! I've got others from Schiit, ZMF and Bottlehead with more planned in the future too

@Ray-dl5mp Год назад

The funniest part of the video is the big disclaimer right before Rob says “And the reason they’ve got it completely wrong…”. I just love how you edited that disclaimer in there perfectly to take the edge off of what Rob was saying. Funny in a good way.

@PassionforSound Год назад

Yeah, I don't tend to go for the definitive statements. Rob is incredibly knowledgeable and may well be right, but there might also be others doing it right in different ways 🙂

@chrisdeister7745 3 года назад

Amazing. Thanks for this. I really appreciate the technical discussion.

@PassionforSound 3 года назад

So glad you liked it!

@alfredallram1815 4 года назад

Very helpful and finally an understandable explanation of the use of the sinc function (a unique approach in digital audio still, strangely enough given that the principles of the Shannon sampling process have been published in 1948). Thank you "Passion for Sound" and Rob Watts!

@PassionforSound 4 года назад

You're welcome, Alfred. It seems so strange to me that others are eschewing this approach of it is indeed so effective. I still need to explore this area further...

@petekane7869 2 года назад

Great interview - your beard keeps going AWOL 😁

@PassionforSound 2 года назад

Haha. Yeah, it comes and goes a bit sometimes. Mostly it's just consistent stubble

@bunozen1 4 года назад

Loved the interview. I love the tt2, it’s already an incredible amp/dac, I don’t feel I really need more. What would the M Scaler actually do? Just more resolution? Not sure, but the tt2 perhaps is enough already. Thx. Great job man.

@PassionforSound 4 года назад

Thanks Jay!! As I understand it, the M-Scaler will add some additional resolution - specifically in the transients. It should add more texture and precision, but I don't expect it would change the general nature of the sound.

@bunozen1 4 года назад

Passion for Sound Damn, more texture? Now I am in trouble, I love that. I will wait a bit to get it I think. Thanks....

@PassionforSound 4 года назад

Based on my discussions with Rob and reviews I've read, it will be subtle, but noticeable I expect.

@GodfreyMann 7 месяцев назад

Can someone pls explain what a ‘transient’ is? Rob starts talking about how important they are here: 1:57, but not once does he explain what they are. Where on this graph is the transient?

@PassionforSound 7 месяцев назад

Transients are the leading edges of notes or the very first onset of a sound. For example the instant that a note on a piano is struck. Hopefully that makes sense and sorry if it wasn't clarified anywhere. Also, there is no transient shown on the graph - that's just representing the samples in an audio file. There will be transients all through every recording, occurring constantly from different instruments so DAC's ability to put these all together with excellent timing accuracy is the key.

@GodfreyMann 7 месяцев назад

@@PassionforSound that really helps, thx. But doesn't it clash with what Rob says about not adding more data here: 21:31? Let me explain: 44.1kHz allows for 22.7uS timing resolution between each sample, but the brain can resolve better than 4uS (250kHz) which is a resolution of 5-6x better than CD hence why high-res music should sound better... ...because Rob's implying that the transient (leading edge of a note) could lie anywhere between two sample points and CD is basically missing 5-6 sample points to keep up with the 4uS figure for our brain’s time domain resolution, but a digital hi-res file sampled at 250kHz would match it far better (let’s ignore the QuTest’s filters that seem to show the brain’s resolution is even finer than 4uS). If I understand correctly, Rob’s saying that a good DAC should be able to pinpoint the position closest to one of the 5-6 missing sample points when it’s interpolating between two samples. He says it's "preserving the information state of the original recording" but that state is lost...it's not in the data. So his algorithms are recreating those lost signals, which by definition isn't that adding data? (It's also the actual OED definition of interpolation) I understand Rob's argument that what's added is only what was contained in the original performance, but that's semantics...it's still an addition. Think of it this way: in years to come better algorithms and hardware may allow more accurate reconstruction of the interpolated data, which means his interpolation today ISN'T exactly what was performed (albeit it's very close) so it's a delta that was never in the recording (i.e. it’s definitely an addition).

@PassionforSound 6 месяцев назад

This is starting to get beyond my specific level of understanding, but I believe what he is saying is that the sample rate (e.g. 44.1kHz) is sufficient to convey all of the required frequency information to the DAC, but is somewhat vague on the timing of the signal (when thinking about the 250kHz timing accuracy required). It's not so much that the data isn't there and is being added so much that it's not as accurate as it could be and requires additional calculations to improve the timing resolution of the information. Perhaps another way to think of it is a bit like compressed files in a computer. All of the data is contained in a Zip file, but the computer has to work to extract the details of that file and make it usable.

@rossorosso 4 года назад

can you review the Koss 95X electrostatic system from Massdrop please? Thanks

@PassionforSound 4 года назад

Hi Rosso, I owned the 95Xs very briefly and was glad to be rid of them. The sound wasn't special, the build quality is horrible and there was a fain squeal in the left channel that I couldn't consistently remove. The sound of them was thin - well articulated, but not special enough to offset the lack of depth and body in the sound. Even on lighter music light classical / chamber they were uninspiring, particularly with the left channel issues. I'd say to do yourself a favour and look elsewhere.

@azzinny 3 года назад

9:46 Doesn't the Fourier Transform of the lower signal contain above 20kHz? Continuous 20kHz sinusoidal signal appears as a very sharp peak at 20Khz in the spectrum analyzer. 20kHz single cycle sinusoidal tone burst appears as a broad mountain on the spectrum analyzer, meaning that it has frequency components above 20kHz.

@PassionforSound 3 года назад

Sorry, that's beyond my knowledge and would be a question for someone like Rob directly

@allee5480 2 года назад

Yes, strictly speaking, a sharp turn on of any sinusoidal audio signal - the example could be, e.g., a 60 Hz signal, it doesn't need to be at 20 kHz), contains a lot of energy in frequencies well above 20 kHz. Rob Watts alludes to this in later comments about his 16 bit accurate interpolated outputs with 4 million taps when he mentions "assuming the bandwidth limiting on its own makes no difference to sound quality" at 21:40. A cymbal crash has transient frequencies that are far above 20 kHz which are necessary to represent the sharp turn on of the sound with the strike of a drum stick. We cannot hear these frequencies, but we can notice the start of the transient noise to an accuracy of something like 5 microseconds in the time domain, which is equivalent to 200 kHz in the frequency domain.

@azzinny 3 года назад

The important transients are believed to be all within 20kHz, right?

@PassionforSound 3 года назад

Absolutely. The transients themselves are in the audible frequency ranges. The key is the speed at which they occur and the extreme timing accuracy required by brain to perceive them as "true-to-life"