Q&A with Charles Hansen of Ayre Acoustics

Charles Hansen of Ayre with their QB-9 asynchronous USB DAC, Stereophile's 2009 Product of the Year

Ayre Acoustics makes a complete line of electronics including amplifiers, preamplifiers, integrated amplifiers, disc players, DACs, ADCs and more. Our focus today will be on their digital side and specifically look at issues that affect file-based playback. The idea for this Q&A came about from a series of email exchanges I had with Charlie Hansen, Ayre's Founder and Designer, related to my recent post which contained a list of NOS DACs. You'll see that we touch on this topic and it was my hope that we could come to a better understanding of some of the underlying issues involved in the D/A process in general which might lead to a better understanding of why a list of NOS DACs is about as useful a grouping as four-legged animals when looking for the ideal DAC or pet. I'd like to thank Charlie Hansen for his time and very informative and detailed answers.

Could you give us some history of your involvement in digital component design?
Ayre was founded in 1993. I had a non-compete agreement with Avalon to not make loudspeakers, so we started with electronics. Our first product was a power amp, for no particular reason. We followed that up with a preamp for obvious reasons. We wanted to do a digital product, but the rumors of a new format (eventually named DVD) were flying around, so we decided to wait until that was format available.

"We have always been pushing the envelope, and doing something completely new with every product we build."

When the DVD spec was announced in December of 1996, they had included 96/24 at the last minute, probably at the behest of Pioneer who were huge supporters of high-resolution audio. When I found this out I immediately let Mike Hobson of Classic Records, David Chesky of Chesky Records, Kevin Halverson of Muse Electronics, and Jeff Kalt of Resolution Audio know about it. I knew that we needed a consortium of both hardware and software providers to make it successful. Theta Digital and Mark Levinson Audio (actually their Proceed division) worked on it independently. A year later at the 1998 CES, we all had prototype players and about a half a dozen discs. It was quite the talk of the show that year.

We have always been pushing the envelope, and doing something completely new with every product we build. A list of our digital accomplishments includes:

  • World’s first audiophile grade DVD player to provide support for 96/24 audio discs.
  • World’s first disc player to provide user-selectable digital filter responses, including “slow roll-off” algorithm with improved transient response.
  • World’s only DVD player to provide total galvanic isolation between audio and video circuits.
  • World’s first production progressive-scan DVD player.
  • World’s first DVD player with 14-bit video DACs.
  • Consultant to Analog Devices on bringing 12- and 14-bit video DAC chips to the mass market.
  • World’s first audiophile grade CD player to use a computer ROM transport mechanism—joint project with Resolution Audio.
  • World’s first CD player to use both “upsampling” and oversampling for a data rate of 1.4112 MHz at 24 bits.
  • World’s first audio-only universal stereo player to play all available audio optical disc formats.
  • World’s first disc players to provide user-selectable “Minimum Phase” digital filter responses, including both “slow roll-off” algorithm with improved transient response and “apodizing” algorithm for removal of ringing from digital filters used to produce the disc.
  • World’s first disc players to implement 16x oversampling in a single-pass path rather than the conventional cascade of 2x filters.
  • World’s first solid-state USB D/A converter with asynchronous data transfer control for zero interface-induced jitter.
  • World’s first all-format (video and audio) disc player with asynchronous USB audio input.
  • World’s first A/D converter with all-discrete, fully-balanced, zero-feedback analog circuitry.
  • World’s first A/D converter with moving-average digital low-pass filters to achieve perfect transient response with zero overshoot, pre-ringing, or post-ringing.
What are the different elements of a DAC chip from input to output?
In the early days of digital audio, a DAC chip was just a DAC chip. You fed digital data to the input and it would output a current that was proportional to the input data.

"In general, the more bits it has, the better the performance will be."

As time went on, the main demand has been for smaller, cheaper DAC chips with lower power consumption. This is due to the iPod craze. There is still a market for high-performance audio DAC chips, but there are only a handful left. All but one (the Burr-Brown PCM1704) use some form of a delta-sigma design that typically has only one to six bits, and relies on oversampling and noise shaping to attain reasonable performance.

In general, the more bits it has, the better the performance will be. However with a ladder DAC, all of the bits beyond 18 or so are called "marketing bits" as there is no audio-grade ladder that can exceed 18 bits of resolution. For example when Burr-Brown replaced the "20-bit" PCM1702 with the "24-bit" PCM1704, not one single specification changed. The only difference was that you could feed it digital words that were 24 bits long.

A typical modern high-performance DAC chip will include:

a) A data reformatter to change the serial input data into parallel words for conversion. It will also separate out the left and right data, as all modern DAC chips have at least two channels.

b) An equalizer to correct the frequency response of discs that were made with pre-emphasis. This is a relic from the very early days of CD and it is extremely rare to find a disc like this, but this equalizer is required for proper playback of all discs.

c) A digital oversampling filter, most commonly 8x in modern chips.

d) An interpolator to bring the oversampling rate up to the rate of the modulator, which is where the 24 (or whatever) bits are truncated to the 1 to 6 bits of the actual DAC and it's digital feedback loop.

Embarrassingly enough, almost all "interpolators" simply repeat the same data over and over again. For example a common modulator will operate at 64 x Fs (Fs = the sample rate). The 8x oversampling filter gets us half-way there, so then most chips will simply repeat each of those new words 8 times in a row to get 8 x 8 = 64 times oversampling. Not very sophisticated, but remember that cost is the main driver here.

e) The output of the modulator will have between 1 and 6 bits running at (most commonly) 64 times the rate of the original input signal. This is fed to various types of ways to change this into an output voltage or current proportional to the input data. Each different way has its own advantages and disadvantages, and are often proprietary to each chip maker.

f) If it is a current-output DAC chip, you are done. The output is a current and an external current-to-voltage converter must be added by the manufacturer of the final product. 99% of the time, this is done with an op-amp, which is probably the worst way to do it. It looks great on paper, but in the real world, there is no op-amp that can keep up with the high-speed transitions generated by the DAC chip.

g) If it is a voltage-output DAC, it is simpler for the manufacturer of the final product to implement, but it also means that you have turned over one of the most critical elements that contributes to the sound of the unit to the hands of the chip manufacturer. That is not something that I would ever care to do. They don't have audiophiles designing these chips, that is for sure. Instead they will do one of two things:

1) Just put an op-amp inside the DAC chip to perform the current-to-voltage conversion. Now you are stuck with whatever op-amp they put in there, with no way to change it or tweak it.

2) Use a switched-capacitor output stage instead of a switched-current source output stage. Switched capacitors are like everything else in life -- a series of trade-offs. In my opinion most of the trade-offs are bad ones when it comes to sound. For one example, just think of how large and expensive an audio grade polystyrene or Teflon capacitor is. Then think of what the capacitor inside an integrated circuit is like, and how it is made, and how it sounds. All I can say is that if they sounded good, high-end audio manufacturers could save a lot of money and space by using those tiny little capacitors as used inside a switched-capacitor DAC chip.

Rob McCance's picture

Great QnA Charles and Mike. Interesting to hear the technical discussion.

I'm a EE and a Audiphile (with a music background) and I no longer even attempt to explain anything to Vinyl/Analog biggots who will drone on forever about things they have no clue about. I typically just nod my head now and hope that they soon pipe down so I can hear the music.

Had one the other night at a listening session try to explain to me about how digital was just ON and OFF and therefore there was no way it would ever sound as good as analog. 

Finally I said, "before you get overly excited, know that this vinyl you are listening to right now was recorded and mixed on digital gear before being squished onto that platter."

He quickly got another beer than changed the subject.

And so will I..

Currently, I'm using the Metrum Octave who (best I can tell) has tried to make a NOS DAC using extremely fast chips to tackle some of the issues. 

I'm always on the lookout for innovative designs attempting to solve the digital playback issues. Very interesting the angles Ayre are using.

slim's picture

makes we wish for a DAC that offers all the discussed options in one unit, switchable, to get an aural grip on the sonic differences implied by the choices made:

- choice of digital input format: DSD, PCM, DoP

- choice of sampling: NOS, 2-, 4-, 8-, x-fold oversampling

- choice of filter: minimum phase, linear phase, apodizing, etc.

I would not need to own such unit, it should help to narrow down the choices to what would likely suit one's preferences.

Charles Hansen's picture

We already did all of that work so that you wouldn't have to.

Regarding the input formats, there isn't any performance difference between DoP (DSD over PCM) and he traditional DSD formats. So there isn't any point in comparing these two. Our A/D converter offers the two currently existing formats, SDIF-2 and SDIF-3. The next version of the DoP format should cover A/D converters, so we will offer a software upgrade for that after that standard becomes availaable.

We listened to NOS, 4x, 8x, and 16x and 16x was the clear winner, so that's what we used. That was at the single-rate sample rate. The data is fed to the DAC chip at the same overall rate for each sample rate, so dual-rate data uses 8x oversampling, and quad-rate data uses 4x oversampling. All of these arrive at the DAC chip at the highest rate it can handle, 706.4 kHz for multiples of 44.1 kHz and 768 kHz for multiples of 48 kHz.

In all of our tests, minimum-phase filters sounded better than linear-phase filters, so we use them for all of the filter choices.

Apodizing filters require a steep rolloff (with more ringing) to filter out any pre-ringing that may be present from the recording equipment. These are incorporated into our "Measure" filters, while the "Listen" filters all use slow rolloff filters with minimal ringing. You can listen to that and ssee which you prefer, but every review I've ever read agrees with our choice of using the "Listen" filters. The only reason that switch exists is that in some countries, products that don't measure perfectly are crucified in their audio press. So we put the switch there so that the reviewers in those countries won't give our products horrible reviews if the frequency response is down -0.5 dB at 20 kHz.

Our belief is that we are not trying to suit one's preferences as much as we are trying to build the most natural and realistic equipment we can. As noted above, we do the work so that you don't have to. I think any other approach is essentially saying, "We don't really know which is best, so why don't you try it and see what you think?

slim's picture

and for having done the work so that I don't have to.

Yet, being a natural scientist way more than an audiophile, I sympathize a lot with the approach mentioned in your last statement

"We don't really know which is best, so why don't you try it and see what you think?"

I do know that "best" is relative in the audio world, and let me make clear that the product I am wishing for is a hypothetical one, sort of a "probe DAC", and I do not expect any company to ever manufacture it.

Charles Hansen's picture

There are only two ways I can think of that the average enthusiast can try various digital filters to hear what effects they have on musical playback. The first is to borrow or purchase a product with one of the recent Wolfson DAC chips. I believe that these have five different digital filters that (in some products, at least) are selectable by the user.

I am not really familiar with these DAC chips as they are voltage output devices, and as I explained in the interview, I have little to no interest in them. The critical questions would be:

a) When selecting digital filters are all other parameters held constant? If so, then a fair comparison can be made. If not, then one doesn't really know which parameter results in the change of sound quality.

b) Does the unit offer the specific comparisons in which you are interested? Perhaps you want to compare a linear-phase slow-rolloff filter against a minimum-phase slow-rolloff filter. Those two options may not be offered by the DAC chip.

When we ran our experiments, we used FPGA's to implement the digital filters. This allowed us to program virtually any type of digital filter we wanted. We would just change one variable at a time, and switch between the two filters via a toggle switch on the rear. This is the only fair way to conduct such a test.

The other way would be to purchase an older non-MP version of the Ayre C-5xe. It had two user selectable digital filters -- a sharp rolloff ("brickwall") and a slow-rolloff (for improved transient response. This would allow you to compare these two parameters. Then you can send the unit back to the factory for the MP upgrade. We had basically the same two filters, but now they were implemented as minimum-phase types (hense the C-5xeMP desgnation). This test would not quite be fair, as we also bypassed the digital filter built into the DAC chip, and the "Measure" (sharp rolloff) filter was reduced slightly in frequency to filter out any ringing from the equipment used in the recording process (ie, "apodizing" filter).

But it would give you a good idea of what the differences were. You can also search the online forums from around that time to get some responses from some of the Ayre owners who had their units upgraded to MP status, such as this one:


Good luck in your quest for better sound!

labjr's picture

Will the QB9 be upgradable to decode DSD128 ?

Charles Hansen's picture

You have to remember that the only reason that DSD even exists is that the patents for CD were expiring. This used to amount to a $1 billion per year royalty stream for Sony and Philips, and they did not want to lose this income. They needed to come up with a system that they could patent and license. It's ironic because SACD failed in the marketplace, in large part because it is not readable by computers. It is only now that Sony has released "DSD-Disc" without copy protection that will play on computers, that there is a resurgence of interest in DSD.

DSD-128 is probably what DSD should have been all along. "Normal" DSD is 64x the standard CD sampling rate. Since it has only 1 bit, they use a 7th-order noise shaper to get reasonable performance out of the system.The problem with "normal" DSD is that while the signal to noise floor in the audio band is 120 dB (roughly 20 bits of resolution), above 20 kHz the noise rises very sharply. This has two effects:

a) The practical frequency response is limited to 30 or maybe 40 kHz at the most. Above that, any audio signal is going to be swamped by the noise.

b) You can only perform one conversion to "normal" DSD without losing a significant amount of performance. There is some mathematical equation that shows that more than "x" number of conversions to and from DSD will result in more noise than signal, and "x" is something relatively small, between 4 and 8 as I recall.

So for recording in the studio, they almost are forced to use double speed DSD (DSD-128) to have the ability to convert the master recording to PCM for editing, EQ, fades, mixing, adding reverb or other effects. After that it is converted back to DSD a second time, almost invariably at the "normal" DSD rate, since there is almost no hardware that will play back DSD-128. Remember, they have to sell product to stay in business. There really isn't much use for it as a playback format currently. Some of the problems include:

a) Sending DSD disguised as PCM over the USB connection requires a 176.4 kHz connection. Going to DSD-128 would require a 352.8 kHz connection. While USB 2.0 Class 2 Audio can easily handle this, it puts more stress on the signal chain. Everything has to be completely up to snuff, including the USB cable itself.

b) Download times are doubled, as are storage capacity requirements.

c) I don't know if the DAC chips Ayre uses (Burr-Brown DSD1792A in the DX-5 and  DSD1796 in the QB-9) will accept and decode a DSD-128 signal. They were designed before this format was in use and the data sheets are very spotty in this regard. We could try it and see, but I kind of doubt that they would work.

d) There is barely any DSD-64 software available at this time. DSD-128 is even rarer. I think 2L has a demo track or two but I don't think that this is something that will ever gain widespread support.

The best use for DSD (in my opinion) is as a digital playback format for analog tapes made before the digital era. But as I noted in my interview, I think that quad-rate PCM, when done properly, gives all of the sonic advantages of DSD with none of the practical disadvantages.

So the short answer to your question is "Maybe, but most probably not."

CG's picture

Charlie has a really great point.  Much of audio realism seemingly has to do with the implicit and explicit filtering in the overall system.  How much energy is devoted to that aspect of the design process?


One wonders how Charlie's observations might apply to Pono.  What kind of digital conversion will they be using?  How about all those filters?

At one time people raved about HDCD recordings.  Were they better because of the math?  Or because of the really high quality ADC needed for the process designed by Keith Johnson for Pacific Microsonics?  Same questions here.  I hope they get it right.

Charles Hansen's picture

We don't know much about the details of PONO yet, so all we can do is speculate. As far as HDCD goes, I would  guesstimate that probably at least 80% of the sound improvement was due to Keith Johnson's superb ADC box.

There were three other features. The Low-Level Extension only kicked in at levels below -40 dBFS. This only happened rarely, during a few very quiet passages in classical music. The Peak Extend was a compansion system, but was not recommended for use with pop music becausse undecoded, it would further compress already over-compressed music to make the overall sound worse. And there were two different anti-alias filters used that automatically switched in depending on the spectral contect of the music being recorded, but there was only one playback filter for both record filters.

The Pacific Microsonics ADC's are still highly prized for their sound quallity, indicating just how advanced Keith Johnson's ADC design was.

pbarach's picture

Really interesting article that left me (a non-engineer) with a couple of questions:

1. Why does a sharp filter produce more ringing than a gentle filter?

2. What's wrong with op-amps? I see a lot of criticisms of them, but I'm not sure exactly what negative effects they are supposed to have on sound quality.

Charles Hansen's picture

These are two really great questions, Thanks for asking them.

I was having a tough time with the first one. Richard Feynman, the late Nobel Laureate in physics, was once asked by a Caltech faculty member to explain why spin one-half particles obey Fermi Dirac statistics. Rising to the challenge, he said, "I'll prepare a freshman lecture on it." But a few days later he told the faculty member, "You know, I couldn't do it. I couldn't reduce it to the freshman level. That means we really don't understand it."

(However, he also once said, "If I could explain it to the average person, I wouldn't have been worth the Nobel Prize." But since this isn't worth the Nobel Prize, I'll give it a shot.)

At the website for the University of Oregon is a page that talks about electrical resonances:


It says, " Resonance in electrical circuits is a difficult and very non-intuitive subject. The way it is typically taught is through analogies to mechanical systems which exhibit the same kind of behavior." and shows an animated GIF picture of a simple resonant mechanical system:

Physical Analogy

In this mechanical analogy to an electrical resonance, the moving block has mass (which represents inductance), the spring has compliance (the opposite of stiffness, which represents capacitance), and the friction of the block sliding on the surface (also called damping, which represents resistance).

The compliance of the spring together with the mass of the block will determine the frequency of resonance. But in this discussion we are concerned with the sharpness of the "knee" between the flat part of the frequency response and the part that rolls off. The sharpness of this knee is set by the damping or resistance. A lower resistance in the electrical circuit creates a sharper knee, which corresponds to less friction in the mechanical system. In this case, one can intuitively see that if the block is disturbed, that it will oscillate (ring) for a longer period of time than if the friction is high (high electrical resistance). I am hopeful that this will help to understand why a sharper "knee" or "corner" in the frequency response will lead to more ringing in a filter.

The way to make the slope of the electrical rolloff steeper is to add additional sections to the circuit. In the mechanical analogy, this would be like adding additional springs and blocks to the first set, all coupled together like railroad cars in a train. Again, I am hopeful that once can see that if there are multiple springs and blocks oscillating that they will oscillate for a longer period of time before the movement dies down to zero.

(If this explanation is unclear, please let me know and I will try again.)

Regarding integrated circuits (IC's) in general (of which op-amps are a sub-type), there are many advantages to them, which is why they are so widely used in modern electronics. They are small, light, and cheap compared to a circuit constructed from discrete parts (eg, transistors, resistors, and capacitors). However, very few of these advantages have anything to do with performance. When an IC is used, it is similar to baking a cake from a box of Betty Crocker mix. There is very little that one can do to optimize the values and quality of the parts inside. In contrast, a master chef making a cake from scratch can select the quality of the ingredients and adjust the proportions to attain better results than an off-the-shelf box. And just as with cake mixes, some IC's offer higher performance (taste better) than others. But just as with cakes, the highest performing circuits are typically made from discrete parts, where the designer can select the precise values and quality of the individual components.

Most audio components use a special type of IC called an "operational amplifier", or op-amp for short. As the full name indicates, these circuits were first used in the '50s for analog computers before digital computers became practical (with the advent of IC's). The most basic operations that an op-amp can perform are to add (sum) two signals or subtract (take the difference) of two signals. Almost universally it is the latter that is done in audio circuits, where the output signal of the op-amp is compared to the input signal to the op-amp.

Since no circuit is perfect, there will always be noise and distortion of the output signal compared to the original input signal. By comparing the two, an error signal is generated and the error is subtracted from the input signal in a way that tends to cancel steady-state distortion. This is called "negative feedback" and is used almost universally in all audio products, whether based on tubes, discrete transistors, or IC's. But an op-amp uses negative feedback as its fundamental mode of operation. It is not possible to use an op-amp without also using negative feedback.

From a mathematical (theoretical) perspective, this works perfectly (as long as certain rules are obeyed). But many listeners prefer the sound of circuits with little or no feedback, despite what the theory says. At this point nobody has any proof as to why this should be. One argument (and in most cases it will just lead to further arguments!) is that negative feedback can only correct an error after it has occurred. Obviously this is true, but the theorists will claim that the correction (when done properly) happens so quickly that it is impossible for the human ear/brain to hear it.

But the theorists also claim that all cables, amplifiers and CD players sound the same, too. So in the end the best thing to do is let your own ears be your guide. It is always best to listen to music you are familiar with, and in a system and listening room that you are familiar with. This is most easily done if you have a quality dealer nearby that will let you borrow a component for a short while to listen to at your home. (Please note that it is considered bad manners to borrow equipment from a dealer and then purchase it elsewhere.)

I hope this helps, and have fun discovering which components bring you the greatest musical satisfaction!