September 2013 - Ayre QA-9 ADC

t: 01727 865488 e: [email protected] w: www.symmetry-systems.co.uk

Symmetry, Suite 5, 17 Holywell Hill, St Albans, Hertfordshire AL1 1DT

The QA-9from Ayre Acoustics

Press ReleaseAugust 2013

Some of the stuff is the "standard" Ayre stuff - things that were radical when we implemented them, but as more and more people copy these ideas and our equipment becomes more common, people forget how truly radical our designs are.

Words by Charles Hansen:

We have toyed around with the idea of building some "pro" (recording) equipment for years, even though it is pretty much outside of our marketing world, and we are not sure how to make money on it. However, there were two reasons for us to do it:

a) If we build a great (say) preamplifier, then one audiophile will be able to bliss out on the music. On the other hand, if we build a great A/D converter, then literally millions of people could benefit from the improved sound. And as I wrote in the Audio Asylum article, this has a very real effect on the state of one's consciousness.

b) Several professionals own Ayre equipment, usually for their home systems. Somewhat typical but most notable is Rick Rubin. He says that Ayre makes the best sounding digital playback gear and he has been after us for years to build equally good sounding recording gear.

We have been thinking about it for quite a while and finally got around to it.

" When we started Ayre, there was only one company that had ever built zero-feedback solid state equipment (to the best of my knowledge). "

It was a small direct-mail order only company called Wingate Audio, run by Steve Wingate back in the mid-'80s (now deceased, I believe). He actually patented his designs and you can easily pull them up on the internet now, but back then it was relatively difficult (although cheap -- $2 as I recall) to get copies of patents. I looked at his schematics a couple of years ago, and they were quite good. He got a very nice review in TAS but was out of business within two years (as seems to be the fate of almost everyone that tries to sell direct).

Ayre QA-9 IntroductionEqually versatile in a recording studio or home system, the QA-9 opens up the world of high resolution digital to your analog sources.

Accepting analog inputs, the QA-9 offers both USB and AES/EBU digital outputs depending on your system's configuration. Featuring an output of up to 24 bit / 192 kHz, recording with the QA-9 captures every element you love in your analog sources. Whether you are recording your favourite vinyl, preserving master tapes, or creating files from your mixing board - capture the essence of your music with the Ayre QA-9.

UK PRICE: Standard £2,995* / Pro £3,495*Silver chassis as standard *add £180 for black chassis

TYPE: USB ADC

AVAILABLE FINISHES: Silver or black

DIMENSIONS: w 21.5cm x d 30.5cm x h 7.5cm

WEIGHT: 2.3kg

AVAILABILITY: Now

MANUFACTURER: Ayre Acoustics

WEB SITE: www.ayre.com



Page 2

Also when we started Ayre, the only consumer audio company building equipment with balanced connections was Rowland. And again since then, more and more people copy our approach until it seems relatively pedestrian.

" To my knowledge, nobody has ever built an ADC with fully discrete, fully balanced, zero feedback analog circuitry. "

In my experience, the single most important factor in getting good sound in digital audio is with the *analog* circuitry, so that was reason enough to start the project. Of course, we at Ayre are never content to rest on our laurels, so we also incorporated a slew of other innovations.

Now that you know some of what goals were part of the design of the QA-9, please allow me to describe some of the details:

1) There is a small company called Arda that does various consulting work for big contractors. They are audiophiles, and in their spare time designed and built a new A/D chip that clearly outperforms anything from the competition. Specifically the out of band noise is ridiculously low, and this gives us a lot of freedom to do "think different" (as Apple un-grammatically said a couple of decades ago.)

2) Many, many people prefer the sound of DSD to PCM. Whether it is real or imagined, it seems that the vinyl crowd

leans much more to the DSD side than does the general audiophile population. Therefore, for home transcription of vinyl, it seemed imperative to offer a DSD output option.

Currently, the DSD can only be output on BNC connectors, either in SDIF-2 or SDIF-3 format. Andreas Koch is working on version of 1.2 of DoP (DSD over PCM) for USB and we will offer a (free?) update for that when the standard is finished.

" Then DoP will be available in the lower-priced "standard" version without the word-clock outputs. "

Of course, ever since DSD was announced it has received uniformly positive comments on the sound quality, while (at least for the first few years) hiding as much of the working details as possible.

" It is too bad that Sony won the DSD vs. DVD-Audio "war", but such is life. "

One of the big advantages they had was that they hired Ed Meitner and Andreas Koch to design and build the original recording hardware. In contrast, Mike Hobson hired Kevin Halverson to make the 96/24 converters used by Classic Records for their 96/24 "Digital Audio Disc" (DAD) releases.







Page 3

Literally, these were just the stock Texas Instruments (who had just purchased Burr-Brown) eval boards! My understanding is that these were not even put into chassis, so they were completely unshielded. Who knows what kind of power supplies were used. (Eval boards typically don't come with a power supply, as it assumed that they will be used with "bench" power supplies.) It is surprising that they sounded as good as they did.

So of course it was no contest. Kevin H. is a really smart guy, but not a terribly good listener, and he didn't even try to improve the TI eval boards. Hobson was in such a rush to be "first to the market" and so the first 96/24 discs received mixed reviews at best. (Another big part of the problem is that there weren't any audiophile-grade DVD players back then!). As they say, "Haste makes waste!".

But to be honest we were basically baffled by Sony's original marketing material. We couldn't figure out how the system worked, let alone why it sounded good. But over the years, not only did more and more information leak out, but our understanding of digital audio at Ayre grew rapidly.

" In short, the reason that DSD sounds as good as it does is because there is NO filtering done on the record side. "

The playback side requires a filter (per the Scarlet Book specs), but compared to the brickwall filters used in typical PCM products, this is a *much, much* gentler affair.

" When we developed the QB-9 we spent nearly four months performing listening tests to digital filters. It was clear that both the anti-aliasing (record) and reconstruction (playback) filters had a very strong influence on the sound of digital audio (remember how the "non-oversampling" DACs were quite popular

for a few years?) and so we wanted to really understand what was going on. "

Previously we had done some hit-and-miss investigations (eg, which of the four DSD filters built into the Burr-Brown DAC chips sounded best), but we had never done a systematic investigation of what wasgoing on.

" It turns out that one of the biggest problems was the fact that recording engineers make purchasing decisions based almost purely on printed specifications. "

Almost nobody other than a few (such as Tony Faulkner) even bother to listen to the stuff they use. The other factor that influences the purchasing decisions of recording engineers is what the "stars" use. In fact pro audio is probably the only field worse than musical instruments themselves in that regard. (Jimi Hendrix played a Fender Stratocaster, so I probably should too. Jascha Heifetz played a Stradivarius, so I probably should too. Rick Rubin uses ProTools 10, so I probably should too.)

The problem is that it is trivially easy to get ridiculously good specs from digital audio equipment. Of course, there is no such thing as a free lunch, and the elephant in the room is the HORRIBLE transient response of the digital filters used by 99.9% of all pro audio equipment.

" Consumer playback equipment got a boost decades ago when Wadia introduced their various "cubic spline" filters and so forth, which are all just variations of "slow rolloff" filters that offer improved transient response at the expense of worsened frequency response. "

These in turn were popularized by Burr-Brown, which incorporated them even into their low-cost DAC chips. And of course the entire problem can be laid at the







Page 4

5

feet of Sony and Philips who chose the sampling rate of 44.1 kHz for digital audio. In their defence, they were only trying to equal the performance of cassette tape, but unfortunately it became the standard that we were all stuck with for decades.

" Of course, the relatively recent advent of using the personal computer as a way to store and play back music files has changed the game entirely. "

Now we are free to invent almost any format we want, and it doesn't take much work to make recording and playback software to make it work. Now it is a reality that we can buy, store, and play back music at quad-sample rates (or even higher) if so desired.So everything is different now. When one records at quad-sample rates (ie, 176.4 kHz or 192 kHz), everything changes. No longer do we have to worry about aliasing. There aren't many instruments with significant amounts of energy above 90 kHz to 100 kHz. Nor are there many microphones with any significant response this high. Or mic preamps. Or mixing boards. Or whatever.

But the problem is that digital audio equipment is still designed by digital audio engineers. (BAD engineers! Somebody slap their wrists!). And if they are using a 192 kHz sampling rate, they are going to make their equipment with flat frequency response to 96 kHz and then just put a brickwall filter on it.

" Fortunately, I am not a traditional digital audio engineer! I therefore have the freedom to ask "Why?", and I do that a lot! "

So for the QA-9, we decided that the goal was to make the converter operating at the quad-sample rate to perform more or less like a perfect 30 ips analog tape machine.

The frequency response is down about -3 dB at 50 kHz, but it goes down to around 1 Hz (more on that later) with no "head bumps" to worry about. There is *zero* wow and flutter, and the distortion and noise is about an order of magnitude better (ie, 20 dB) better than the best analog tape machines.







Page 5

Yes, the anti-aliasing filter (in "Listen" mode) is only down about -20 dB at the aliasing point of 96 kHz, but is that *really* a problem? Do any of your recording actually have any meaningful amount of musical energy up at 172 kHz, where there would be aliasing that would fold down into the audio band? Of course not!

So the first thing that we did was to use a completely different type of digital filter at the output of the delta sigma converter. (Every audio ADC chip made for the last 20 years has been a delta sigma type, as the competing successive approximation devices died off long ago. I have an idea about how to build a "super delta sigma" ADC chip from scratch, but that is a different story altogether.)

Instead of using the normal low-pass FIR to turn the output of the delta-sigma DAC into PCM, we use a moving average filter. This doesn't just allow for *improved* transient response, but actually *perfect* transient response. And since we are starting with a 256 Fs sample rate, we don't have the problems exhibited by "non-oversampling" DACs that also have "perfect" transient response.

" There is no pre-ringing, no post ringing -- no ringing whatsoever. "

(Again this is in the "Listen" mode at both the quad and double sampling rates. It is not possible to use this trick at the single sample rate, where instead we use a more conventional FIR low-pass filter, but of course it is a slow rolloff design to minimize ringing, and

also is a minimum phase design, so that all of the ringing occurs *after* the transient with no un-natural "pre-echo" before the transient occurs.)

So now we are starting to get somewhere -- perfect transient response, combined with zero-feedback, fully discrete, fully analog circuitry.

Another thing that we wanted to avoid that is ubiquitous with ADC chips is the *high-pass* brickwall filter that is intended to keep DC out of your system. If you look at the curves for the top-of-the-line Burr-Brown PCM 4222 A/D converter: http://www.ti.com/product/pcm4222, on page 15 of the datasheet it shows the curves for the high-pass DC filter. Figure 36 shows that it is truly a brickwall filter, going from >-140 dB to -10 dB





Page 6

in a very small fraction of an octave.

Now, it is good to keep DC out of your recordings for a variety of valid reasons, but adding a brickwall filter will create ringing no matter whether it is a low-pass or a high-pass filter. The Arda ADC chip contains its own similar DC filter, but we chose to implement ours from "scratch" in the FPGA instead. (In fact, the FPGA is so "full" of circuitry, that when you switch from PCM to DSD, the unit has to reload the entire FPGA function with new instructions!)

If you tried the conventional approach of using a FIR, you would need several bazillion taps to make a filter that low. So instead what people normally do is use an IIR, where the output signal is recirculated through the filter to get the desired response. This has the side effect of turning the filter into a minimum-phase type, which would normally be good, but in this case, one does not want to introduce phase shift in the bass range. Also, I do not like to use FIRs as the inevitable errors accumulate as the signal recirculates through the filter. The goal is that the signal is smaller than one LSB by the time that the error is equal to one LSB, but it doesn't always work as planned.

" You may remember an AES paper from ten or fifteen years ago where some speaker manufacturer tried to "prove" that low frequency phase shift cause all sorts of audible problems, but I actually think that it was just a case of a vented woofer that was ringing more than a sealed woofer. (More or less the same explanation, but one is much easier to understand!) "

At any rate we had to come up with a different technique to avoid these problem, so we send the audio data through two parallel paths. One is the undisturbed audio and the other goes through a low-pass IIR with a cutoff in the range of 0.1 Hz. This basically only allows DC through. Then we take this

signal and subtract it from the main music signal. Then any DC offset that may be present is subtracted out of the music signal without having to send the main signal through any filters.

" By doing this slowly, it avoids any clicks or pops. By using a first-order IIR highpass filter, it is (once again) transient perfect, with no ringing, pre-ringing, overshoot, or echo. Essentially what we have done is exactly emulate the way an analog servo works and implement it in the digital domain. "

The "Listen" position of the dual-sample rate filter basically emulates a 15 ips analog tape machine. Everything works the same as at quad-rate, but since it is running at half speed, the frequency response is lower. I think it is down -3 dB at around 18 kHz or 20 kHz, just as an analog 15 ips machine would be. Again it is transient-perfect and sounds great.

If you are in a situation where you need totally flat frequency response to the highest possible frequencies (eg, recording bats or dolphins), the "Measure" position allows for that. For those filters, we use the moving average to bring the sample rate down from 256 Fs to either 8 Fs or 4 Fs and then use an FIR lowpass field to do the final division by two. While these are minimum phase slow rolloff types they still will introduce a cycle or two of ringing while the "Listen" filters are absolutely perfect in regards to transient response. (No free lunch! Well actually if we boosted the sample rate to octal rate, we could probably get absolutely perfect sound, but the performance at quad rate is close enough!)

For the single rate filters, we can't do any special tricks like the moving average filter. Here we just bring the sample rate down from 256 Fs to 4 Fs with the moving average filter and then do the final division by four with the minimum-phase, slow rolloff FIR filter. (The "Listen" and "Measure" position simply use different







Page 7

coefficient sets for that final division.) Again there is a cycle or two of ringing with the "Listen" position, and many cycles of ringing with the , but at this sample rate we also have a few dB of rolloff at 20 kHz.

The lower the sample rate, the more compromises you have to make and there is just no way to get the top level of performance from the thing. Instead, you really need to record at least 2x and preferably 4 x to get the best sound. One can always down-convert at a later point in time using whatever method is preferred. As Keith Howard pointed out several years ago, the simple averaging method produces the best sound, although it doesn't measure as well.

" We also include a DSD output for the hard-core DSD guys. We thought this would be trivially easy -- until we found out that this only operates at 4x the normal 64x sample rate of normal DSD! "

Some people are advocating and using the double rate of 128x, which solves a lot of the problems of DSD. With "regular" DSD (64x), they use 7th order noise shaping and the FFT noise floor is -120 dB up to 20 kHz but then sharply rises to around -30 dB at 100 kHz.

This is why the "Scarlet Book" mandates a low-pass playback filter to limit the out-of-band playback noise.

We once had a customer with a C-5xe and he had some VAC tubed preamp. Normally a tube preamp is pretty immune to out-of-band noise, but this unit was built with (poorly designed) transformers. Stereophile had tested a similar model and it had something like a 10 or 12 dB peak at 150 kHz! The result was some high-frequency "warbling" sounds caused by the modulation with the out-of-band noise.

Anyway, going to double-rate DSD moves the noise to twice the frequency, so it doesn't start rising until 40 kHz.

Our unit can go all the way to quad-rate DSD and the noise doesn't start rising until 80 kHz. This more or less solves the problem because our analog circuitry has a fixed low-pass filter at 80 kHz for all sampling rates (with additional poles at higher frequencies).

" The problem here is that very little existing equipment can handle quad-rate DSD! "

The thing that was a pain was going from quad-rate DSD down. I thought we could just average samples or discard every other sample, but the Arda guys said that would not work well. Instead we had to build our own 7th-order single bit remodulator to bring the 6-bit 256 Fs output from the Arda modulator down to the 64 Fs 1 bit rate. Rather convoluted if you ask me, and another reason that I believe the PCM will sound better than the DSD (I haven't had a chance to do listening tests myself yet...) All of this remodulation uses the entire power of the FPGA, which is the most powerful one they make without going to a BGA package. So when you change from PCM to DSD (or vice versa), it has to"reboot" the FPGA and load a completely new memory program. That took some more doing to get that to work seamlessly...

But of course I think you will find that the best solution from either a sonic or an operational standpoint is simply to use the quad-rate PCM. There is NO out-of-band noise (well, actually -148 noise floor on the FFT), the signal is trivially easy to process with any existing software, and it sounds even better than DSD because we don't need filters or remodulators for either record nor playback. Add in the zero-feedback, fully-balanced, fully-discrete circuitry and I think it may be the best sounding ADC on the planet today.

Right now Rick Rubin is evaluating one. Next in line is Kevin Gray and Joe Harley. I will probably also send demo units to Todd Garfinkle (M-A Acoustics) and David Chesky. So hopefully we will be headed towards an era with better sound than ever -- even than the classic stereo recordings from the late '50s on RCA and Mercury. We shall see....







Page 8

" Alan Silverman of ARF! Mastering has bought one and he loves it: http://www.arfmastering.com/ "

The only place that you may run into troubles is with the word clock and DSD outputs. There is no standard for either one! We are big believers in galvanic isolation. This not only keeps RFI from one unit out of the rest of the system, but also prevents the formation of ground loops. So all of the BNC outputs are transformer coupled. (The USB output is already galvanically isolated and runs off of the +5 volts supplied from the computer.)

If the receiving circuit is properly designed (also with a transformer), there won't be any problems. But even the "best" digital gear is designed by digital engineers, and they simply aren't aware of the problems that a direct connection causes. So if you try to connect to a word clock input or some DSD inputs without success, let me know.

" We can send some adapters that include AC coupling capacitors and will allow the QA-9 to function with just about anything. If necessary, we will even get the schematics from the manufacturer and make a custom adapter. "

About AyreAyre Acoustics, Inc. has been designing and building superior quality, award winning audio and video equipment since 1993.

They are recognized as a world leader in the industry, and our full line of audio and video components sets new standards in innovation, design, and performance.

Discover Ayre and lose yourself in the moment… intimately experiencing music and cinema that is perfectly timed and naturally alive.







Page 9

FeaturesUp to 24 bit /192 kHz digital output. Variable gain circuitry for input level control. Optional word clock / DSD output board (Pro version) Separate left and right recording level meters.Zero-feedback, fully-balanced discrete circuitry. Equilock circuitry for active gain devices.

SpecificationsInput Impedance 2 MΩ balanced (1 MΩ per phase)

Analog Input Sensitivity 10.0 Vrms - at minimum gain 0.75 Vrms - at maximum gain

Analog XLR Input Polarity Pin 1 = Ground Pin 2 = Non-inverting (Positive) Pin 3 = Inverting (Negative)

Frequency Response DC - 20 kHz (44.1 kHz sample rate) DC - 22 kHz (48 kHz sample rate) DC - 40 kHz (88.2 kHz sample rate) DC - 44 kHz (96 kHz sample rate) DC - 60 kHz (176.4 kHz sample rate) DC - 66 kHz (192 kHz sample rate) DC - 20 kHz (DSD64)DC - 40 kHz (DSD128)

DC - 80 kHz (DSD256)

USB Audio Output Signal Class 2 44.1, 48, 88.2, 96, 176.4, and 192 kHz at 24 bits Class 1 44.1, 48, 88.2 and 96 kHz at 24 bits

AES/EBU Output Signal 44.1, 48, 88.2, 96, 176.4 and 192 kHz at 24 bits

Word Clock Output Signal 44.1, 48, 88.2, 96, 176.4 and 192 kHz

DSD Output Signal SDIF-2: 64x, 128x, and 256x SDIF-3: 64x and 128x

AES/EBU XLR 110Ω, transformer-isolated

Word Clock / DSD BNCs 75Ω, transformer-isolated

Power Consumption 25 watts

Dimensions 8.5"W x 12"D x 3"H (21.5cm x 30.5cm x 7.5cm)

Weight 5 pounds (2.3 kg)





Page 10

September 2013 - Ayre QA-9 ADC

Documents

ayre equipment

ayre qa

pro recording equipment

standard ayre stuff

ayre acousticsweb site

dtthe qa

recording studio

good sounding recording