Ayre Acoustics QX-5 Twenty – The Digital Hub

[austinpop]

7 minutes ago, Charles Hansen said:

 

Hello AustinPop,

 

Thanks for the input. It is nice to get feedback from the customers. There is always the problem of "price, features, performance - pick any two". Ayre could offer all of the above with an increased price - then we would have to balance the number of increased sales from adding the features versus losing sales from the higher price. (We refuse to compromise performance.)

 

Hi Charles,

 

I am grateful that you took the time to dig into my feedback. Not that I didn't already hold Ayre in high regard, but your willingness to listen to feedback only adds to my regard. I appreciate your questions, and will endeavor to clarify.

 

7 minutes ago, Charles Hansen said:

- MQA support - will address this in another post.

 

I look forward to it.

 

7 minutes ago, Charles Hansen said:

- Native DSD support has no standards. It requires ASIO drivers and installation. We've not even looked into it and have no idea what the costs would be. As far as I can tell the only advantage is the ability to go past DSD-256. I fail to understand what the need for that would be. How many titles are available at DSD-256? How many titles are available at DSD-512? If you want that feature to use "upsampling" from an external program like HQ Player, all you are doing is inserting an extra digital filter in the chain ahead of the digital filter built into the DAC. If you prefer the sound of this to the Ayre digital filter, then either we need to learn how to design better sounding digital filters, or you prefer to color your system to compensate for some deficiency elsewhere. I've no idea which is the case. If you would elaborate, that would be helpful.

 

Yes, I agree with you on HQPlayer. I have already experimented and determined that for the Codex, HQPlayer does precisely 0. I always get the best results feeding the Codex with the native sample rate. This is what I would expect from a well-designed DAC.

 

Regarding DSD512, let me address it below.

 

7 minutes ago, Charles Hansen said:

- Ethernet support past 192kHz. Again, what is the need for this? The only content past 192kHz is solely from minor audiophile labels such as 2L. They offer zero music that I am interested in listening to. Is that what you are after? Or is it again some sort of external "upsampling" external digital filter? There is a very, very small case for DSD-128, as there are perhaps a few dozen titles recorded at that rate. The current processor on the ConversDigital module will not handle higher rates, so that will be at least a year or more away, as explained in a different post.

 

I may be atypical, but as a classical music lover, a significant chunk of my library is DSD64. These files started life on SACDs, and have been ripped. However, the advent of higher-resolutions is nigh. Some examples:

1. Unlike you, I love some of the 2L recordings. Have you heard Arnesen's Magnificat? Played at DXD, even on my humble Codex, it is mesmerizing. 
2. The new Mahler 3rd by Ivan Fischer on Channel Classics is my 2nd purchase at DXD resolution. It's SQ is stunning at DXD. They also give you DSD resolutions up to 256, but guess what? My Codex tops out at DSD128, due to the overhead of DoP.
3. I am buying more music at sites like Native DSD, where DSD256 titles are already cropping up.

Again - I may be atypical, but this requirement - to support DSD256 and beyond - is real for me, not just specmanship. I am not tied to native DSD, as long as higher DSD resolutions can be delivered. The issue is the space inefficiency of DoP. It would require PCM 768 to support DSD256. and PCM 1536 for DSD512.

 

7 minutes ago, Charles Hansen said:

- I2S is silly in its current implementation as used by a handful of companies with HDMI cable. The clock should be in the DAC and not in the transport.I see little reason to introduce yet another proprietary interface that would only work with some possible future unannounced Ayre transport, and Ayre does not like to do things in sub-optimal ways just because the people who started doing this didn't think it through properly.

 

- AES/EBU (and all other S/PDIF variants) cannot transmit DSD except as DoP. That is one of the reasons I helped form a group to develop DoP. I know of no S/PDIF receivers that will go any faster than 192kHz, and Ayre already uses the very highest performing one on the market, so this will request will likely never be physically possible.

 

Over on these threads:

• https://www.computeraudiophile.com/forums/topic/30376-a-novel-way-to-massively-improve-the-sq-of-computer-audio-streaming/?do=findComment&comment=721466
• https://www.computeraudiophile.com/forums/topic/31857-mutec-ref-10-masterclock/?do=findComment&comment=720312

We are finding that the sound quality of a digital audio chain can be greatly enhanced by feeding an extremely low-phase-noise clock like the Mutec Ref 10 to all components in the digital chain, including DACs. Since most DACs, other than the really high-end Esoteric and dCSes, do not accept reference clocks, people are finding that the benefits can be conferred via synchronous inputs like I2S and AES.

 

I realize this is a fraught subject, on which I suspect you have strong views, so please don't shoot the messenger! I will only say that I am curious about whether a mythical Codex, which accepted a low phase noise reference clock input and inherited the "goodness" of said clock -whether it would sound better, and by how much!

 

7 minutes ago, Charles Hansen said:

Thanks for letting us know what you are looking for. Again if you could elaborate your reasoning for wanting things (other than "specsmanship"), it would be very helpful.

 

Best,

Charles Hansen

 

I truly appreciate your attention to my feedback.

[Charles Hansen]

10 minutes ago, austinpop said:

I may be atypical, but as a classical music lover, a significant chunk of my library is DSD64. These files started life on SACDs, and have been ripped. However, the advent of higher-resolutions is nigh. Some examples:

1. Unlike you, I love some of the 2L recordings. Have you heard Arnesen's Magnificat? Played at DXD, even on my humble Codex, it is mesmerizing. 
2. The new Mahler 3rd by Ivan Fischer on Channel Classics is my 2nd purchase at DXD resolution. It's SQ is stunning at DXD. They also give you DSD resolutions up to 256, but guess what? My Codex tops out at DSD128, due to the overhead of DoP.
3. I am buying more music at sites like Native DSD, where DSD256 titles are already cropping up.

Again - I may be atypical, but this requirement - to support DSD256 and beyond - is real for me, not just specmanship. I am not tied to native DSD, as long as higher DSD resolutions can be delivered. The issue is the space inefficiency of DoP. It would require PCM 768 to support DSD256. and PCM 1536 for DSD512.

 

 

Over on these threads:

• https://www.computeraudiophile.com/forums/topic/30376-a-novel-way-to-massively-improve-the-sq-of-computer-audio-streaming/?do=findComment&comment=721466
• https://www.computeraudiophile.com/forums/topic/31857-mutec-ref-10-masterclock/?do=findComment&comment=720312

We are finding that the sound quality of a digital audio chain can be greatly enhanced by feeding an extremely low-phase-noise clock like the Mutec Ref 10 to all components in the digital chain, including DACs. Since most DACs, other than the really high-end Esoteric and dCSes, do not accept reference clocks, people are finding that the benefits can be conferred via synchronous inputs like I2S and AES.

 

I realize this is a fraught subject, on which I suspect you have strong views, so please don't shoot the messenger! I will only say that I am curious about whether a mythical Codex, which accepted a low phase noise reference clock input and inherited the "goodness" of said clock -whether it would sound better, and by how much!

 

 

I truly appreciate your attention to my feedback.

 

Hello AustinPop,

 

Point one is fair enough. The problem is that one has to pick and choose one's battles. Spending months of R&D and an unknown amount of money to implement an ASIO driver (for both Windows and Mac) simply so that a handful of files can be played has to be weighed against other customer demands. We currently have about several hundred customers who want us to build a higher performance phono stage - just as one example. At this time I'm afraid that there are several other projects ahead in line - especially as the only input that could currently even accept the higher rate would be the USB. Realistically I wouldn't look for something like that until we upgrade the Ethernet module, the USB module, or both. My sense is that it would make much more sense to tie in all of the upgrades together at once rather than adding them piecemeal. But I've not looked into what would be involved in adding an ASIO USB driver, so I can't say for sure.

 

Drivers are scary things as they require continual updating as the OS's are "upgraded". Normally the only feasible way to do it is to pay a third party (eg, Thesycon) to write and maintain them. That is expensive enough that we would need to have count-on-able additional sales of a product to make sure we didn't lose money on the deal.

 

As far as external clocks go, the internal clock in the QX-5 is equal or better than any external clock on the market - even the $20,000 ones. Adding an external clock input to the QX-5 would actually degrade its performance. As far as less expensive products, adding features like that only drives up the price for an extremely small number of users who would even want such a feature. And it would require Ayre to build an outboard clock to ensure the proper performance (most do it wrong, just as the current I2S implementations do it wrong). Building a separate external clock is something that would really be more useful in the pro market, and we currently have no distribution there. It simply makes no sense to develop, sell, promote, and support a product which might only sell a handful of units per month world-wide. Again, we have to pick and choose our battles. Even then 99% of the Codex customers would be paying for a feature they would never use.

 

The QX-5 has a patent pending asynchronous S/PDIF input that completely eliminates all jitter from that interface. In fact it is the best sounding input, simply because there is no computer in the system to introduce EMI and RFI. But of course then you lose all the convenience advantages of computer playback. But it's a great check to let you know how good your computer setup is. The closer your computer sounds to an S/PDIF input, the better your computer setup is.

 

Sorry for the delay on the MQA topic. I was just finishing my reply when something wacked out in my browser. I suspect HTML 5 is the problem. They keep adding garbage "features" that make it more complex and more unreliable....    Please be patient and I'll re-type it.

 

Best regards,

Charles Hansen

[austinpop]

1 hour ago, Charles Hansen said:

 

Hello AustinPop,

 

Point one is fair enough. The problem is that one has to pick and choose one's battles. Spending months of R&D and an unknown amount of money to implement an ASIO driver (for both Windows and Mac) simply so that a handful of files can be played has to be weighed against other customer demands. We currently have about several hundred customers who want us to build a higher performance phono stage - just as one example. At this time I'm afraid that there are several other projects ahead in line - especially as the only input that could currently even accept the higher rate would be the USB. Realistically I wouldn't look for something like that until we upgrade the Ethernet module, the USB module, or both. My sense is that it would make much more sense to tie in all of the upgrades together at once rather than adding them piecemeal. But I've not looked into what would be involved in adding an ASIO USB driver, so I can't say for sure.

 

Drivers are scary things as they require continual updating as the OS's are "upgraded". Normally the only feasible way to do it is to pay a third party (eg, Thesycon) to write and maintain them. That is expensive enough that we would need to have count-on-able additional sales of a product to make sure we didn't lose money on the deal.

 

As far as external clocks go, the internal clock in the QX-5 is equal or better than any external clock on the market - even the $20,000 ones. Adding an external clock input to the QX-5 would actually degrade its performance. As far as less expensive products, adding features like that only drives up the price for an extremely small number of users who would even want such a feature. And it would require Ayre to build an outboard clock to ensure the proper performance (most do it wrong, just as the current I2S implementations do it wrong). Building a separate external clock is something that would really be more useful in the pro market, and we currently have no distribution there. It simply makes no sense to develop, sell, promote, and support a product which might only sell a handful of units per month world-wide. Again, we have to pick and choose our battles. Even then 99% of the Codex customers would be paying for a feature they would never use.

 

The QX-5 has a patent pending asynchronous S/PDIF input that completely eliminates all jitter from that interface. In fact it is the best sounding input, simply because there is no computer in the system to introduce EMI and RFI. But of course then you lose all the convenience advantages of computer playback. But it's a great check to let you know how good your computer setup is. The closer your computer sounds to an S/PDIF input, the better your computer setup is.

 

Sorry for the delay on the MQA topic. I was just finishing my reply when something wacked out in my browser. I suspect HTML 5 is the problem. They keep adding garbage "features" that make it more complex and more unreliable....    Please be patient and I'll re-type it.

 

Best regards,

Charles Hansen

 

Thanks for the response. It wasn't the answer I wanted, but I accept I may not be your target customer.

 

It is disappointing that going from the Codex to the QX-5 (and presumably the new QX-8), there is no advancement in the max sample rate - still PCM 384 and DSD 128. Especially over Ethernet, where there is no need to worry about DoP. Hopefully you can find a less-crippled Ethernet interface than the ConversDigital module you mentioned a few weeks ago.

 

I look forward to your views on MQA. I think I know what to expect, but perhaps I'll be surprised.

[AmusedToD]

Looking forward for @austinpop to compare the SOtM clocking ability through the “trifecta” driven by the external Cybershaft master clock to the single built in superior clock as the one in the QX-5. 

 

I have read somewhere that it’s possible to order a stripped down QX-5 without some of the modules (for example without the USB module) and have a lower overall price of the unit, that should be interesting.

 

And finally, a question for @Charles Hansen, which type of clock will be used in the upcoming QX-8 and what kind of performance should we expect compared to the Morion one in the QX-5?

[austinpop]

13 hours ago, Charles Hansen said:

 

As far as external clocks go, the internal clock in the QX-5 is equal or better than any external clock on the market - even the $20,000 ones. Adding an external clock input to the QX-5 would actually degrade its performance.

 

That's great to hear! It certainly makes sense to have an outstanding internal clock, as you wouldn't have to deal with the degradation of clock cables and possible impedance mismatches.

 

However, could you clarify if the Morion clock is also used for the non audio clocks? By these I mean:

- the USB Input, typically 24 MHz

- the ARM processor

- the Ethernet interface, typically 25 MHz.

 

Are these clocks also benefitting from the low phase noise Morion? This is where we are finding the big SQ gains in the other thread.

[austinpop]

 

 

15 hours ago, Charles Hansen said:

As far as external clocks go, the internal clock in the QX-5 is equal or better than any external clock on the market - even the $20,000 ones. Adding an external clock input to the QX-5 would actually degrade its performance.

 

Hi Charles,

 

Someone on another thread pointed me to this video above. I must say, your description of how you chose the Morion crystal oscillators, and the design choices, was fascinating! And i can completely relate to the improvements you observed with the low phase noise crystals.

 

So I would love to know if you considered (and listened) to the Morion crystals for the non data clocks - i.e. USB, ethernet, and processor - for the QX-5? Because this is what we are finding so interesting - replacing/improving these "system" clocks with ultra-low phase noise clocks also yields that "aha" experience.

[Charles Hansen]

9 hours ago, AmusedToD said:

Looking forward for @austinpop to compare the SOtM clocking ability through the “trifecta” driven by the external Cybershaft master clock to the single built in superior clock as the one in the QX-5. 

 

I have read somewhere that it’s possible to order a stripped down QX-5 without some of the modules (for example without the USB module) and have a lower overall price of the unit, that should be interesting.

 

And finally, a question for @Charles Hansen, which type of clock will be used in the upcoming QX-8 and what kind of performance should we expect compared to the Morion one in the QX-5?

 

Hello Amused,

 

There is a lot of hype going around about clocks these days, and unfortunately it is all almost all just marketing hype. All of this talk about "femto clocks" and so forth is just that - talk. There are a lot of things that affect clock performance, but the question is "Which clock parameters correlate with audio quality?".

 

It turns out that a single jitter number is completely useless, and actually almost always misleading. When a "jitter spec" is given, it is the sum of the phase noise over a particular frequency range offset from the carrier (operating) frequency of the oscillator. For all telecommunications work, this range is specified as 12kHz to 20MHz offset. Please refer to:

http://community.silabs.com/t5/Timing-Knowledge-Base/The-12-kHz-to-20-kHz-Jitter-Bandwidth/ta-p/114177

for information as to how this standard came about.

 

It turns out that this number is almost meaningless for audio performance. Instead it is close-in phase noise - no more than 20kHz (the audio band) offset from the carrier. Low levels of lose-in phase noise are very difficult to achieve. Factors that affect this include the crystal, the oscillator circuit, and the power supply to the oscillator itself. For crystals there are two types used for audio. By far the most common is called an AT strip-cut. Depending on the care in manufacture, these can range from fair to excellent in performance. The oscillator circuit itself can also affect phase noise dramatically, but depending on the circuit, the effects may either be in the important close-in range or in the unimportant (for audio) large offset from the carrier. Finally there is the power supply to the oscillator circuit. Most oscillator circuits have very low power supply rejection ratio (PSRR). This means that the power supply is just as important as the oscillator circuit in achieving low levels of close-in phase noise.

 

Measuring phase noise is extremely difficult. Most crystal and oscillator manufacturers use an HP (Agilent) phase noise analyzer, but it costs $100,000 brand new. If you are lucky you can find a used one for half of that price. One problem is that there is so much variation from crystal-to-crystal that in phase noise performance. An even bigger problem is that crystals vary in phase noise performance over time. Getting a good measurement on an oscillator can literally take a week or more as (apparently) surface contaminants on the quartz surface "burn-off" (evaporate) as the oscillator is used. There are two things that help here. One is to only by from high-quality sources with outstanding quality control, and the other is to keep the oscillator running all the time. One of the big things that causes a sound improvement in digital products as they "warm up" is the fact that the oscillator is settling down and the phase noise is lowering.

 

Probably the most important thing is the power supply for the oscillator. All of the IC-based voltage regulators have quite a lot of low-frequency noise that directly translates into close-in phase noise - no matter how good the oscillator circuit and crystal used, as all oscillator circuits have poor rejection from power supply noise. This is one area where Ayre has a huge advantage. We build our own discrete regulators, not only for the audio circuitry, but also for everything that is in the clock path from the oscillator to the DAC chip itself. Our discrete designs typically have 10x lower noise at low frequencies than even the best IC-based voltage regulators.

 

When it comes to the crystal itself, there is only one more thing that will help lower the close-in phase noise. That is to use an SC-cut crystal. This can provide a 10dB improvement in close-in phase noise compared to even an excellent AT strip-cut crystal. One problem is they are much more difficult to use as they always have spurious resonances close to the main carrier. If the oscillator circuit is not designed properly, the circuit will "hop" from the desired frequency to the spurious frequency - often many times per minute (or even second), which translates into massive amounts of low-frequency phase noise. The other problem is cost. Not only does an SC-cut crystal cost over 100x as much as an AT-strip cut, but they also generally require an "oven" - a small, sealed metal box with a thermostatically controlled heater to hold the crystal at a constant temperature. That adds another 5x to the price. This is what is usually found in the $5000 (internal option) to $100,000 and even $20,000 external clocks.

 

In the QX-5 Ayre worked with a crystal company to develop an SC-cut crystal that has the same low phase noise, but does not require the oven. This (barely) allowed us to get the very best possible performance available in a real-world priced unit. The less expensive Ayre products use an AT-strip cut crystal, but from specially selected manufacturers that can control the quality of the crystals and provide consistent performance. The biggest difference however is the ultra-low noise power supply regulators that we use. This allows the less expensive Ayre products to exceed the performance of what many other manufacturers call "femto clocks" (because the jitter from 12kHz to 20MHz offset (which is meaningless for audio applications) is less than 1 picosecond and so would normally be measured in tens or hundreds of femtoseconds.

 

Don't be fooled though. A "femto second" oscillator can have far higher amounts of phase noise at low frequencies - within 1kHz or less offset from the carrier. Ayre has found that even very low offsets - down to 10Hz, 1Hz, and even 0.1Hz can have large effects on the sound quality of digital playback. The next step is delivering that low phase noise signal to the DAC chip itself, where it really counts. Ayre uses impedance matched traces and isolation stages help out here. Plus there is internal logic in the DAC chip itself that needs ultra-low noise phase chip before it reaches the actual conversion stage. It then becomes just as important that power that supplies the clock pins also have extremely low noise.

 

As you can see, there is far, far more to having a converter free from jitter problems than just paying $10 or $20 more for one of the Crystek oscillators that are advertised as being "femto clocks". Those are simply pre-canned oscillators employing AT strip-cut crystals that have a reasonably quiet power supply - for an IC-based internal power supply regulator. Ayre has found suppliers of crystals with equal or better quality but combine that with our custom ultra-low noise discrete power supplies to achieve even better performance. To achieve the ultimate requires using an extremely expensive SC-cut crystal, so that must be reserved for our more expensive products. But even there by developing a way to eliminate the need for the oven, Ayre can afford to use that very highest level of performance clock without having to spend multi-kilobucks for just a clock or a complete DAC.

 

Hope this helps,

Charles Hansen

Charles Hansen

 

 

[rickca]

25 minutes ago, Charles Hansen said:

As you can see, there is far, far more to having a converter free from jitter problems than just paying $10 or $20 more for one of the Crystek oscillators that are advertised as being "femto clocks".

Thanks, Charles.  That's an excellent post and really puts things in perspective.

[Charles Hansen]

6 hours ago, austinpop said:

 

That's great to hear! It certainly makes sense to have an outstanding internal clock, as you wouldn't have to deal with the degradation of clock cables and possible impedance mismatches.

 

However, could you clarify if the Morion clock is also used for the non audio clocks? By these I mean:

- the USB Input, typically 24 MHz

- the ARM processor

- the Ethernet interface, typically 25 MHz.

 

Are these clocks also benefitting from the low phase noise Morion? This is where we are finding the big SQ gains in the other thread.

 

Hi Austin,

 

The difficulty is that the audio clocks operate at frequencies that are totally unrelated to everything else. Just to achieve the best audio performance requires two separate clocks - one for the 44.1kHz family and another for the 48kHz family. The problem is that Ethernet clocks run at 25MHz, while USB clocks run at either 12MHz or 24MHz, again two new frequencies that are unrelated to anything else. Using super expensive SC-cut crystal oscillators for those parts of the signal receiver would provide the ultimate in performance, but practically double the cost of the unit. People are already complaining that the $9000 (US retail) price of the QX-5 is "too high" (despite the fact that it outperforms competing products at 2x and 3x its price), so we are offering versions with fewer input options.

 

Instead we use all of the tricks described above to achieve the very best performance form reasonably-price, high-quality AT strip-cut crystals to maximize the performance of the computer inputs (as noted in the above post). Even though there are another half-dozen (or more) other manufacturers using the ConversDigital Ethernet module, they don't achieve nearly the same level of performance that the QX-5 does, as we custom build the power supply and clocking circuitry ourselves and just plug in the ARM processor with its onboard firmware. Our USB and Ethernet clocks provide better performance than what many sell as "femto clocks", while our power supplies have lower noise than any other company I know of.

 

Maybe some day we will build a silly expensive product with Morion oscillators for the computer inputs, but I think there is a lot more available by focusing on other areas of noise rejection. Once of the nice things about the QX-5 is that one can very easily get a fix on how good their computer source is, simply by comparing it to a CD player or transport connected to one of the S/PDIF variant inputs - preferably the Toslink as it provides total electrical isolation. Ayre has developed a patent-pending asynchronous clocking arrangement for the S/PDIF inputs that eliminates of the jitter normally associated with that data protocol. All computers will generate far, far more EMI and RFI than a simple CD player, and that is a bigger problem than the phase noise of the oscillators on the computer inputs of the QX-5. Simply unplug all of your computer gear, both from the audio system and the AC mains (to ensure that nothing is getting into tthe AC mains wiring and see how close your rig can get to the sound of a "super" input without the added noise of a computer.

 

The closer you can get that to sound to the S/PDIF input, the better your computer rig is sounding. That will act as a "noise free" reference - although obviously you lose all of the convenience features of computer-based audio! I think it will take several more years before the industry gets a handle on all of the sources of degradation due to the EMI and RFI of computers (and their switching power supplies) and they stop being as tweaky as they currently are. Look how long it took us to get here with just digital audio - over three decades. We only have about one decade with computer audio. It will continue to improve.

 

Hope this helps,

Charles Hansen

[austinpop]

1 hour ago, Charles Hansen said:

 

Hi Austin,

 

The difficulty is that the audio clocks operate at frequencies that are totally unrelated to everything else. Just to achieve the best audio performance requires two separate clocks - one for the 44.1kHz family and another for the 48kHz family. The problem is that Ethernet clocks run at 25MHz, while USB clocks run at either 12MHz or 24MHz, again two new frequencies that are unrelated to anything else. Using super expensive SC-cut crystal oscillators for those parts of the signal receiver would provide the ultimate in performance, but practically double the cost of the unit. People are already complaining that the $9000 (US retail) price of the QX-5 is "too high" (despite the fact that it outperforms competing products at 2x and 3x its price), so we are offering versions with fewer input options.

 

Instead we use all of the tricks described above to achieve the very best performance form reasonably-price, high-quality AT strip-cut crystals to maximize the performance of the computer inputs (as noted in the above post). Even though there are another half-dozen (or more) other manufacturers using the ConversDigital Ethernet module, they don't achieve nearly the same level of performance that the QX-5 does, as we custom build the power supply and clocking circuitry ourselves and just plug in the ARM processor with its onboard firmware. Our USB and Ethernet clocks provide better performance than what many sell as "femto clocks", while our power supplies have lower noise than any other company I know of.

 

Maybe some day we will build a silly expensive product with Morion oscillators for the computer inputs, but I think there is a lot more available by focusing on other areas of noise rejection. Once of the nice things about the QX-5 is that one can very easily get a fix on how good their computer source is, simply by comparing it to a CD player or transport connected to one of the S/PDIF variant inputs - preferably the Toslink as it provides total electrical isolation. Ayre has developed a patent-pending asynchronous clocking arrangement for the S/PDIF inputs that eliminates of the jitter normally associated with that data protocol. All computers will generate far, far more EMI and RFI than a simple CD player, and that is a bigger problem than the phase noise of the oscillators on the computer inputs of the QX-5. Simply unplug all of your computer gear, both from the audio system and the AC mains (to ensure that nothing is getting into tthe AC mains wiring and see how close your rig can get to the sound of a "super" input without the added noise of a computer.

 

The closer you can get that to sound to the S/PDIF input, the better your computer rig is sounding. That will act as a "noise free" reference - although obviously you lose all of the convenience features of computer-based audio! I think it will take several more years before the industry gets a handle on all of the sources of degradation due to the EMI and RFI of computers (and their switching power supplies) and they stop being as tweaky as they currently are. Look how long it took us to get here with just digital audio - over three decades. We only have about one decade with computer audio. It will continue to improve.

 

Hope this helps,

Charles Hansen

 

Hi Charles,

 

Thanks for the very informative post. I'm really enjoying reading about your design and development insights.

 

You've inspired me to audition the QX-5 for myself. Casey (my local dealer) is graciously loaning me one to try out.

 

Right now, my chain comprises:

• a Roon server computer
• a tweaky, digital spaghetti chain, that has affectionately been dubbed a "trifecta,' or a "pasta medley."
• Codex DAC
• Cavalli headphone amp

What I will be looking at are at least 2 scenarios:

1. the QX-5 as a straight replacement for the Codex - i.e. how much of a sonic improvement is the QX-5 as a USB DAC?
2. the QX-5 as a replacement for everything - my pasta medley, my Codex, and my amp.

Wouldn't it be nice if 2. proves to be the equal or better of 1!! If that were the case, the incremental cost of the QX-5 for me would be minimal after I sold off the rest of the chain.

 

One specific question for you, Charles - would you say the QX-5 AES input is the better sounding one, compared to USB? My current chain is geared towards USB, but I could modify my trifecta to deliver AES as well.

 

I am very excited to try this, and will report on my findings when I have some.

[Charles Hansen]

1 hour ago, austinpop said:

One specific question for you, Charles - would you say the QX-5 AES input is the better sounding one, compared to USB? My current chain is geared towards USB, but I could modify my trifecta to deliver AES as well.

 

Hello Austin,

 

As far as I can tell, it is all about reducing RFI/EMI. With a CD player (low-powered microprocessor with no OS) there is little noise to begin with, and it can be very will isolated with Toslink. But then you have to mess with physical discs. At the factory we get much better SQ with USB than Ethernet, but that is because we are running a Melco server into the USB, which sounds far better than any computer I've heard. The Melco is running a stripped-down version of Linux that only has the Ethernet, hard drive, and USB modules installed. It's not like a computer running a full-blown operating system with a keyboard, display,  and 50 or 100 background processes running at the same time.

 

But our Ethernet has two servers and 20 computers, with the modem over 200' away running through a 100Mbs switch. Guys with better Ethernet setups get better sound on Ethernet than from a computer running into USB. By "better", there is one network dedicated for audio that has just a NUC running Roon core and some hard drives connected to the DAC. Then there is an optically-isolated connection to the rest of their network that runs their house, along with the broad-band modem. Remember that both the USB input and the Ethernet input need processors. The USB is an XMOS - not powerful enough to run an OS, but still a pretty current-hungry, high-speed embedded processor. We have isolators for that, but nothing is 100% effective. The Ethernet input is kind of the like the Melco - there is a small ARM processor running Linux, but only with the Ethernet and USB modules - not even a hard drive - only thumb drives (with a simpler file structure). When John Atkinson reviewed it, he got better sound from his very simple Ethernet setup than he did from a decent, but not great laptop using USB. But the AES/EBU was slightly better in his setup, as the CD transport generates less noise.

 

As long as there are computers involved in the source, there is no advantage to using the AES/EBU connection. The only reason it sounds better is simply because the source generates less RFI/EMI. Which of the other inputs sounds better depends on how low you can get the system noise. There are lots of threads here on optimizing both connections. As noted if you keep a CD player/transport handy you can use that as a reference to see how close your computer setup approaches that. To be fair, you should unplug all of your computer equipment from the mains and the stereo when listening to the CD player/transport. Then there won't be as much conducted RFI/EMI on the AC mains of your house.

 

We worked hard to make all of the inputs sound as good as possible. Both the Ethernet and the USB have galvanic isolation. But there is still more to learn, and I think it will be another 5 or 10 years before the industry can get it to a point where it is all figured out.

 

Hope that helps,

Charles Hansen

[firedog]

On 9/20/2017 at 6:14 AM, rickca said:

This is exactly why I want to understand what Berkeley did to implement the MQA renderer in their Alpha DAC Reference Series 2.  I really hope @mansr can get the code and figure it out.  Still more puzzling to me is why Berkeley even bothered doing it.  Chris has said he wishes I could talk to Berkeley about this.  Clearly he has done that and gotten a surprisingly positive message.

http://www.theabsolutesound.com/articles/the-politics-of-mqa/#comment-3533110452

At the link above and following comments you can see that Berkeley evaluated MQA and decided it was basically very close to a master 192k high res recording in quality.

[barrows]

9 hours ago, firedog said:

basically very close to a master 192k high res recording in quality.

So there's the rub.  Why use the "very close" of MQA when we can just use the actual 192k file instead?  MQA requires (in most cases) a new hardware purchase, so we can listen to lossy versions of 192k files, why would we want to promote that?

And answer this question for yourself: why do the purveyors of MQA bot allow full use of MQA via a playback software update alone?  

[AmusedToD]

20 hours ago, Charles Hansen said:

 

Hello Amused,

 

There is a lot of hype going around about clocks these days, and unfortunately it is all almost all just marketing hype. All of this talk about "femto clocks" and so forth is just that - talk. There are a lot of things that affect clock performance, but the question is "Which clock parameters correlate with audio quality?".

 

It turns out that a single jitter number is completely useless, and actually almost always misleading. When a "jitter spec" is given, it is the sum of the phase noise over a particular frequency range offset from the carrier (operating) frequency of the oscillator. For all telecommunications work, this range is specified as 12kHz to 20MHz offset. Please refer to:

http://community.silabs.com/t5/Timing-Knowledge-Base/The-12-kHz-to-20-kHz-Jitter-Bandwidth/ta-p/114177

for information as to how this standard came about.

 

It turns out that this number is almost meaningless for audio performance. Instead it is close-in phase noise - no more than 20kHz (the audio band) offset from the carrier. Low levels of lose-in phase noise are very difficult to achieve. Factors that affect this include the crystal, the oscillator circuit, and the power supply to the oscillator itself. For crystals there are two types used for audio. By far the most common is called an AT strip-cut. Depending on the care in manufacture, these can range from fair to excellent in performance. The oscillator circuit itself can also affect phase noise dramatically, but depending on the circuit, the effects may either be in the important close-in range or in the unimportant (for audio) large offset from the carrier. Finally there is the power supply to the oscillator circuit. Most oscillator circuits have very low power supply rejection ratio (PSRR). This means that the power supply is just as important as the oscillator circuit in achieving low levels of close-in phase noise.

 

Measuring phase noise is extremely difficult. Most crystal and oscillator manufacturers use an HP (Agilent) phase noise analyzer, but it costs $100,000 brand new. If you are lucky you can find a used one for half of that price. One problem is that there is so much variation from crystal-to-crystal that in phase noise performance. An even bigger problem is that crystals vary in phase noise performance over time. Getting a good measurement on an oscillator can literally take a week or more as (apparently) surface contaminants on the quartz surface "burn-off" (evaporate) as the oscillator is used. There are two things that help here. One is to only by from high-quality sources with outstanding quality control, and the other is to keep the oscillator running all the time. One of the big things that causes a sound improvement in digital products as they "warm up" is the fact that the oscillator is settling down and the phase noise is lowering.

 

Probably the most important thing is the power supply for the oscillator. All of the IC-based voltage regulators have quite a lot of low-frequency noise that directly translates into close-in phase noise - no matter how good the oscillator circuit and crystal used, as all oscillator circuits have poor rejection from power supply noise. This is one area where Ayre has a huge advantage. We build our own discrete regulators, not only for the audio circuitry, but also for everything that is in the clock path from the oscillator to the DAC chip itself. Our discrete designs typically have 10x lower noise at low frequencies than even the best IC-based voltage regulators.

 

When it comes to the crystal itself, there is only one more thing that will help lower the close-in phase noise. That is to use an SC-cut crystal. This can provide a 10dB improvement in close-in phase noise compared to even an excellent AT strip-cut crystal. One problem is they are much more difficult to use as they always have spurious resonances close to the main carrier. If the oscillator circuit is not designed properly, the circuit will "hop" from the desired frequency to the spurious frequency - often many times per minute (or even second), which translates into massive amounts of low-frequency phase noise. The other problem is cost. Not only does an SC-cut crystal cost over 100x as much as an AT-strip cut, but they also generally require an "oven" - a small, sealed metal box with a thermostatically controlled heater to hold the crystal at a constant temperature. That adds another 5x to the price. This is what is usually found in the $5000 (internal option) to $100,000 and even $20,000 external clocks.

 

In the QX-5 Ayre worked with a crystal company to develop an SC-cut crystal that has the same low phase noise, but does not require the oven. This (barely) allowed us to get the very best possible performance available in a real-world priced unit. The less expensive Ayre products use an AT-strip cut crystal, but from specially selected manufacturers that can control the quality of the crystals and provide consistent performance. The biggest difference however is the ultra-low noise power supply regulators that we use. This allows the less expensive Ayre products to exceed the performance of what many other manufacturers call "femto clocks" (because the jitter from 12kHz to 20MHz offset (which is meaningless for audio applications) is less than 1 picosecond and so would normally be measured in tens or hundreds of femtoseconds.

 

Don't be fooled though. A "femto second" oscillator can have far higher amounts of phase noise at low frequencies - within 1kHz or less offset from the carrier. Ayre has found that even very low offsets - down to 10Hz, 1Hz, and even 0.1Hz can have large effects on the sound quality of digital playback. The next step is delivering that low phase noise signal to the DAC chip itself, where it really counts. Ayre uses impedance matched traces and isolation stages help out here. Plus there is internal logic in the DAC chip itself that needs ultra-low noise phase chip before it reaches the actual conversion stage. It then becomes just as important that power that supplies the clock pins also have extremely low noise.

 

As you can see, there is far, far more to having a converter free from jitter problems than just paying $10 or $20 more for one of the Crystek oscillators that are advertised as being "femto clocks". Those are simply pre-canned oscillators employing AT strip-cut crystals that have a reasonably quiet power supply - for an IC-based internal power supply regulator. Ayre has found suppliers of crystals with equal or better quality but combine that with our custom ultra-low noise discrete power supplies to achieve even better performance. To achieve the ultimate requires using an extremely expensive SC-cut crystal, so that must be reserved for our more expensive products. But even there by developing a way to eliminate the need for the oven, Ayre can afford to use that very highest level of performance clock without having to spend multi-kilobucks for just a clock or a complete DAC.

 

Hope this helps,

Charles Hansen

Charles Hansen

 

 

 

Hi Charles,

 

Thank you for the time and effort in replying to my question in great detail, much appreciated!

 

So basically the QX-8 will have the AT-cut crystal which is not sourced from Morion, right?

 

 Is the clock quality and performance the only difference to expect between the QX-5 and QX-8, or are there other differences (in the PSU section perhaps)? Any hint regarding the price of the QX-8?

[firedog]

51 minutes ago, barrows said:

So there's the rub.  Why use the "very close" of MQA when we can just use the actual 192k file instead?  MQA requires (in most cases) a new hardware purchase, so we can listen to lossy versions of 192k files, why would we want to promote that?

And answer this question for yourself: why do the purveyors of MQA bot allow full use of MQA via a playback software update alone?  

Again, Berkeley claims that MQA is the best real world solution, as there is a muddle  of DA and AD converters with different filters used that results in time smearing in recordings. They claim that MQA sets real, consistent standards and deals with this time domain problem better than present day 24/192 does in reality, so they support it as the best alternative for high resolution reproduction in the real world.

[barrows]

35 minutes ago, firedog said:

Again, Berkeley claims that MQA is the best real world solution, as there is a muddle  of DA and AD converters with different filters used that results in time smearing in recordings. They claim that MQA sets real, consistent standards and deals with this time domain problem better than present day 24/192 does in reality, so they support it as the best alternative for high resolution reproduction in the real world.

Right, MQA sets a "standard" for lossy reproduction of real 192 source files, I do not want that, and no audiophile should.  different DACs have different digital filter approaches because those are the design beliefs of the engineers which make them.

For example, companies like Ayre and Chord spend a great deal of time getting their digital filters just right, to produce the sound quality that they want their products to have.  Instead, with MQA, one has to accept the "Bob Stuart as GOD" approach to digital filtering, instead of different companies having the choice to produce the sound they are looking for in the implementation of their filters.

Additionally, any audiophiles are using software programs like HQPlayer and Audirvana to create their own filters.  Seeing the responses of the MQA filters, i am certainly not convinced that they are the "last word" in digital filtering!  There is no "magic" in the MQA approach which makes it better than existing filter algorithms from other manufacturers, but they are the first implementation to ever have the gaul to suggest that their approach should be a new "standard", and that audiophiles should suddenly adapt this "standard" and pay a bunch more for it. 

[Charles Hansen]

5 hours ago, firedog said:

Again, Berkeley claims that MQA is the best real world solution, as there is a muddle  of DA and AD converters with different filters used that results in time smearing in recordings. They claim that MQA sets real, consistent standards and deals with this time domain problem better than present day 24/192 does in reality, so they support it as the best alternative for high resolution reproduction in the real world.

 

Hi Firedog,

 

I would be careful in conflating "Michael Ritter" with "Berkeley Audio". Ritter is not a digital audio engineer. He is a salesman and money raiser. When he talks about technical issues, he will use marketing terms that have no basis in science, engineering, or physics. A good example of this is the nonsense he posted about "conjugate filters".

 

There is a thing called "conjugate filters" and they are extremely commonplace. A perfect example is the pre-emphasis/de-emphasis filters used for all record players (RIAA) and tape players (NAB, IEC), and even a few early CDs that used the pre-emphasis/de-emphasis curve that is specified in the Redbook. In all of these cases the treble is boosted on the record side and the an exactly equal and opposite cut in treble is imposed on the playback side. That is truly a conjugate filter system.

 

In digital audio, there is an anti-aliasing filter on the record side and a reconstruction filter on the playback side. Both are low-pass filters. There is no way in the world for them to be "conjugate filters". That is just some marketing nonsense made up by a non-engineer who knows just enough to be dangerous.

 

If I am missing something, and there is some unknown phase or timing anomaly introduced by the anti-aliasing filter that can be corrected for by the reconstruction filter, I am all ears. In the meantime, the best thing I know of is to use filters that don't introduce errors. The only A/D converter I know of that did this was the Ayre QA-9 (but only at double- and quad-sample rates, where it used a special filter with zero phase errors, zero overshoot, and zero undershoot). When there are no errors, there is nothing to correct.

 

Caveat emptor,

Charles Hansen

[Charles Hansen]

6 hours ago, AmusedToD said:

So basically the QX-8 will have the AT-cut crystal which is not sourced from Morion, right?

 

 Is the clock quality and performance the only difference to expect between the QX-5 and QX-8, or are there other differences (in the PSU section perhaps)? Any hint regarding the price of the QX-8?

 

Hello Amused,

 

An SC-cut crystal is extremely difficult to manufacture. It is called a "double-rotated" cut, as a quartz crystal has a growth structure along a specific axis (the length) and then the arrangement of atoms grows at specific angles to this axis. Most crystal cuts (such as AT) simply cut at one specific angle to the quartz crystal (which you have no doubt seen before - they look like hexagonal rods). Just to purchase a bare SC-cut crystal will cost between $100 and $300. This compares to a complete AT strip-cut oscillator, which will typically cost between $0.50 and $1.50.

 

The Crystek parts that are used in many "high-performance" audio applications (and called "femto-clocks" by at least one manufacturer) are AT strip-cut and cost around $20. We have tested them and while they are better than most of the mass-market AT strip-cut, they are no better than certain specific brands that we have found to offer equally high performance at a much lower cost. Making crystals is part science and part art. There are many, many variables that can affect the performance of a crystal oscillator, and 99.9% of the the measurable factors only apply to the telecommunications market (see the above post with the linked reference to SONET) and has no bearing on how the oscillator will perform for audio applications. It is unclear what Crystek is doing to achieve their higher-than-average level of performance. It may just be that the parts include a low-noise power supply regulator inside the can, but that advantage is negated in the way that Ayre uses oscillators as our power supplies are even lower noise than any IC-based regulator. (By the way, when looking at manufacturer-supplied performance graphs of of oscillators, they will always use the best possible power supplies to achieve the best possible performance curves. Those power supplies may be special $20,000 supplies, but they don't tell you about that...)

 

Also be aware that there are OCXOs (Oven-Controlled X-tal Oscillators) that use AT strip-cut crystals, so the presence of a crystal oven is no guarantee that the unit uses an SC-cut crystal - nor how good the oscillator circuit is, nor how good the power supply circuit is. Trying to pin down the sonic performance of any product simply by running down a checklist of design features is about as useful as trying to pin it down on the basis of specifications - both of them simply don't work.

 

In the end it comes down to the manufacturer to make intelligent choices. Ayre can achieve much better performance from an AT strip-cut than others can when using SC-cut crystals by knowing which manufacturers provide consistently high quality crystals, using them in oscillator circuits that keep the close-in phase noise down, and (perhaps most importantly) using discrete power supplies that have much lower levels of low-frequency noise than any IC-based regulator available.

 

These are the kinds of things that allows Ayre to achieve higher levels of performance at any given price point than most other manufacturers. A good example is AustinPop. He has Ayre's `$2,000 DAC and wants to replace it with something that has more features than he currently has. To achieve a comparable (or better) level of sonic performance along with those extra features, he is looking at spending 3x to 4x the price.

 

As far as future products, I have learned it is best not to comment on them before they are actually shipping. Clearly there will be other differences besides just the change of the crystal used for the master audio clock, as otherwise there would be very little delta in the overall price. But just as the Pono Player was able to offer "much of the performance of the QB-9 DSD" (I believe those were John Atkinson's words in his Stereophile review of the Pono Player) for $399, I'm sure that people will be happy with the performance of the 8 series products. (We wouldn't bother to make them if we felt otherwise!)

 

Hope this helps,

Charles Hansen

[jabbr]

20 minutes ago, Charles Hansen said:

Ayre can achieve much better performance from an AT strip-cut than others can when using SC-cut crystals by knowing which manufacturers provide consistently high quality crystals, using them in oscillator circuits that keep the close-in phase noise down, and (perhaps most importantly) using discrete power supplies that have much lower levels of low-frequency noise than any IC-based regulator available.

 

This nonlinearity that links 1/f noise to 1/f phase offset noise is critical to understand.

[AmusedToD]

On 26.9.2017. at 3:07 AM, Charles Hansen said:

 

Hello Amused,

 

An SC-cut crystal is extremely difficult to manufacture. It is called a "double-rotated" cut, as a quartz crystal has a growth structure along a specific axis (the length) and then the arrangement of atoms grows at specific angles to this axis. Most crystal cuts (such as AT) simply cut at one specific angle to the quartz crystal (which you have no doubt seen before - they look like hexagonal rods). Just to purchase a bare SC-cut crystal will cost between $100 and $300. This compares to a complete AT strip-cut oscillator, which will typically cost between $0.50 and $1.50.

 

The Crystek parts that are used in many "high-performance" audio applications (and called "femto-clocks" by at least one manufacturer) are AT strip-cut and cost around $20. We have tested them and while they are better than most of the mass-market AT strip-cut, they are no better than certain specific brands that we have found to offer equally high performance at a much lower cost. Making crystals is part science and part art. There are many, many variables that can affect the performance of a crystal oscillator, and 99.9% of the the measurable factors only apply to the telecommunications market (see the above post with the linked reference to SONET) and has no bearing on how the oscillator will perform for audio applications. It is unclear what Crystek is doing to achieve their higher-than-average level of performance. It may just be that the parts include a low-noise power supply regulator inside the can, but that advantage is negated in the way that Ayre uses oscillators as our power supplies are even lower noise than any IC-based regulator. (By the way, when looking at manufacturer-supplied performance graphs of of oscillators, they will always use the best possible power supplies to achieve the best possible performance curves. Those power supplies may be special $20,000 supplies, but they don't tell you about that...)

 

Also be aware that there are OCXOs (Oven-Controlled X-tal Oscillators) that use AT strip-cut crystals, so the presence of a crystal oven is no guarantee that the unit uses an SC-cut crystal - nor how good the oscillator circuit is, nor how good the power supply circuit is. Trying to pin down the sonic performance of any product simply by running down a checklist of design features is about as useful as trying to pin it down on the basis of specifications - both of them simply don't work.

 

In the end it comes down to the manufacturer to make intelligent choices. Ayre can achieve much better performance from an AT strip-cut than others can when using SC-cut crystals by knowing which manufacturers provide consistently high quality crystals, using them in oscillator circuits that keep the close-in phase noise down, and (perhaps most importantly) using discrete power supplies that have much lower levels of low-frequency noise than any IC-based regulator available.

 

These are the kinds of things that allows Ayre to achieve higher levels of performance at any given price point than most other manufacturers. A good example is AustinPop. He has Ayre's `$2,000 DAC and wants to replace it with something that has more features than he currently has. To achieve a comparable (or better) level of sonic performance along with those extra features, he is looking at spending 3x to 4x the price.

 

As far as future products, I have learned it is best not to comment on them before they are actually shipping. Clearly there will be other differences besides just the change of the crystal used for the master audio clock, as otherwise there would be very little delta in the overall price. But just as the Pono Player was able to offer "much of the performance of the QB-9 DSD" (I believe those were John Atkinson's words in his Stereophile review of the Pono Player) for $399, I'm sure that people will be happy with the performance of the 8 series products. (We wouldn't bother to make them if we felt otherwise!)

 

Hope this helps,

Charles Hansen

 

Thank you very much, Charles. Your contribution here is priceless.

 

As @austinpop has just picked up the QX-5 for a listening test (and hopefully a direct comparison with his SOtM reclocking “trifecta”), I think this and the other big thread are about to become much more interesting now after he provides us with his inputs and comments. 

[George Hincapie]

I was just looking through the specs for the QX5. Why was a ESS DAC chosen over a R2R Ladder DAC? I appreciate part of the way a DAC chip performs in a device is down to implementation, but I have always found ESS DACs to be bright and painful. 

[Charles Hansen]

19 minutes ago, George Hincapie said:

I was just looking through the specs for the QX5. Why was a ESS DAC chosen over a R2R Ladder DAC? I appreciate part of the way a DAC chip performs in a device is down to implementation, but I have always found ESS DACs to be bright and painful. 

 

Hello George,

 

First of all nobody makes ladder DAC chips any more. You can make one from scratch, but the highest precision resistors you can buy are Vishay metal foil with a 0.01% tolerance and a 2 ppm temperature coefficient. If you work out the math, this would only achieve about 13 or 14 bits of resolution. There have been a couple of products tested in Stererophile that achieve more than that, and they achieve more like 19 or 20 bits, so I suspect they are doing something with oversampling, noise-shaping, and possibly even digital feedback - which is how delta-sigma DACs work.

 

The other disadvantage is that they cost many hundreds of dollars in parts cost. The question is where do you want to spend your money? If we wanted to make a $30,000 DAC, that might be worth exploring. The market for that is pretty small. Instead we use the ESS DAC chips as their modulator is specifically designed to mimic the action of a ladder DAC. The first version was called Hyperstream, and the current one is called Hyperstream 2.

 

We have found it to provide outstanding sound quality - when it is used properly. It is very critical of the power supplies. And most companies simply use it in the way shown in the manufacturer's app notes, with all of the "features" enabled and driving into an op-amp output stage. Almost all op-amps sound "bright" and "hyper-detailed", so that is likely a big source of what you are hearing. Ayre uses pure discrete zero-feedback circuitry, which does not sound "bright" and "hyper-detailed". In addition we turn off almost all of the "features of the ESS chips. The QX-5 uses the modulator and current sources and that'a about it.

 

Again, so many people look at just the surface, without realizing there are (just as with an onion) many, many deeper layers of thing that affect the sonic performance.  Choosing a DAC based on the DAC chip used is about as useful as choosing DAC based on the THD at 1kHz. Neither will tell you very much about the way the product actually sounds.

 

Hope this helps.

[George Hincapie]

5 minutes ago, Charles Hansen said:

 

Hello George,

 

First of all nobody makes ladder DAC chips any more. You can make one from scratch, but the highest precision resistors you can buy are Vishay metal foil with a 0.01% tolerance and a 2 ppm temperature coefficient. If you work out the math, this would only achieve about 13 or 14 bits of resolution. There have been a couple of products tested in Stererophile that achieve more than that, and they achieve more like 19 or 20 bits, so I suspect they are doing something with oversampling, noise-shaping, and possibly even digital feedback - which is how delta-sigma DACs work.

 

The other disadvantage is that they cost many hundreds of dollars in parts cost. The question is where do you want to spend your money? If we wanted to make a $30,000 DAC, that might be worth exploring. The market for that is pretty small. Instead we use the ESS DAC chips as their modulator is specifically designed to mimic the action of a ladder DAC. The first version was called Hyperstream, and the current one is called Hyperstream 2.

 

We have found it to provide outstanding sound quality - when it is used properly. It is very critical of the power supplies. And most companies simply use it in the way shown in the manufacturer's app notes, with all of the "features" enabled and driving into an op-amp output stage. Almost all op-amps sound "bright" and "hyper-detailed", so that is likely a big source of what you are hearing. Ayre uses pure discrete zero-feedback circuitry, which does not sound "bright" and "hyper-detailed". In addition we turn off almost all of the "features of the ESS chips. The QX-5 uses the modulator and current sources and that'a about it.

 

Again, so many people look at just the surface, without realizing there are (just as with an onion) many, many deeper layers of thing that affect the sonic performance.  Choosing a DAC based on the DAC chip used is about as useful as choosing DAC based on the THD at 1kHz. Neither will tell you very much about the way the product actually sounds.

 

Hope this helps.

 

Charles,

 

I really appreciate the response, thank you.

 

I think Schiit use the AD5791BRUZ in the Yggdrasil, although in fairness I am not sure if that is made any more. Other manufacturers like Metrum Acoustics make their own R2R chips, so it isn't true to say that R2R DACs aren't made any longer; the traditional ones perhaps aren't.

 

Choosing a DAC by chip is only part of the picture, but given the vast choice these days it is a good place to start.

 

I'd audition the QX5 - do you know who distributes/sells in the UK?                   

[jabbr]

4 minutes ago, George Hincapie said:

I think Schiit use the AD5791BRUZ in the Yggdrasil, although in fairness I am not sure if that is made any more.

 

As @Charles Hansen alludes to, the Analog Designs chips do not implement strict R2R, rather employing say 14 bits IIRC R2R along with aspects which resemble DEM and some forms of multibit SDM to get the lower order bits (i.e. corrections)

[opus101]

14 minutes ago, George Hincapie said:

I think Schiit use the AD5791BRUZ in the Yggdrasil, although in fairness I am not sure if that is made any more. Other manufacturers like Metrum Acoustics make their own R2R chips, so it isn't true to say that R2R DACs aren't made any longer; the traditional ones perhaps aren't.

          

 

The AD5791 is still 'recommended for new designs' on ADI's site, meaning its current production with no termination date. Schiit also use AD5547 in one of their cheaper models. You can pore over the internals because they're shown in the datasheet (page 12) and indeed this one's predominantly R2R with the top 4 bits segmented.