CAREFREE ENJOYMENT: I’ve been listening almost exclusively to the ODAC for several weeks now and I’ve been smiling a lot—not so much at the ODAC but at the music. When you know your audio gear is genuinely transparent it opens a worry-free window into the music. And, while I suffer from expectation bias just like everyone else, I’ve run a second blind listening test and can report the O2+ODAC held its own against the $1600 Benchmark DAC1. They both are audibly transparent.
I’M NOT SELLING ANYTHING: Some have claimed because others may profit from the ODAC I’m profiting from it. But that’s not the case. My involvement was simply to help design and measure it--just like the DIY open source O2 amplifier. YoYoDyne and various other vendors are responsible manufacturing and selling the ODAC. They’re the ones taking the financial risks, so they’re also the only ones to profit. This is an entirely not-for-profit blog and will remain so. You can’t buy anything from me.
LIKE O2 LIKE ODAC: The O2 amplifier was created as a simple, low cost, minimalist design delivering 100% transparent performance into nearly any headphone. The ODAC takes the same minimalist approach to transparency. But it’s not quite as simple as it looks in the photo above. It has around 60 components and many of those were carefully chosen through a lot of detailed measurements and trial and error.
NO SNAKE OIL REQUIRED: Many audiophiles want to believe more elaborate or exotic DACs offer higher fidelity. The ODAC demonstrates you do NOT need any of these for 100% transparent performance:
- Asynchronous USB
- UAC2 (USB Audio Class 2) Support
- Asynchronous Sample Rate Conversion (ASRC),
- Minimum Phase Filtering (no pre-ringing)
- Non-oversampling NOS DAC chips
- Dual DAC chips
- Balanced Outputs
- Vacuum Tube Stages
- Elaborate and/or High Current Power Supplies
THE PROOF: I can confidently say none of the above are required for DAC audio nirvana—i.e. having your DAC disappear from the signal chain. Some of the above, like NOS designs and tubes, degrade fidelity. I’m confident because detailed measurements and blind listening tests verify the ODAC’s real world performance. I’ve even tried several different PCs and the ODAC’s performance is relatively consistent between them. So while some of the above might improve a few measurements, if they don’t improve the sound, it’s like taking 4 pills to get rid of your headache when 2 do the job nicely. Once the headache is no longer perceptible, more pain reliever doesn’t help anything. I encourage others to compare the ODAC blind against other DACs, at any price, that measure reasonably well.
ASSEMBLED BOARD: As explained in the ODA/ODAC and ODAC Update articles the ODAC is not DIY friendly. This is true for most 24 bit USB DACs. Neither the USB interface nor the DAC chips are available in small quantities. Both require signed OEM agreements. minimum purchase quantities, and the USB chip also requires custom programming. The ODAC has many extremely tiny 0603 surface mount components and uses a fine pitch 48 pin IC making it very tedious to build by hand. Given all that, the ODAC requires commercial automated assembled in fairly large batches. DIYers can add the ODAC board to the O2, build it into a stand-alone DAC, or add it to the upcoming Objective Desktop Amp (ODA).
NO S/PDIF: To hopefully avoid most of the “how come you didn’t add S/PDIF” or “when will you add S/PDIF” questions, please see the previous ODA/ODAC and ODAC Update articles. Sorry, but it’s not possible.
PARTNERS REVEALED: Because of the above requirements, the ODAC development has been a joint effort with YoYoDyne Consulting. George has supplied other open source USB DAC boards at reasonable prices and seems to be well respected on diyAudio. He’s taking considerable financial risk having a large batch of ODAC boards assembled and hence is coordinating all the manufacturing and distribution. I conducted all the measurements, refined the design, and helped optimize the PCB layout. Just to be clear, I’m not getting any money from the ODAC. Here are the ODAC resources so far:
- YoYoDyne Consulting ODAC Announcement (Canada)
- JDS Labs ODAC Announcement (USA)
- Epiphany Acoustics ODAC Announcement (UK)
- Head-N-HiFi ODAC Announcement (Switzerland)
- Audio Poutine ODAC Source (Canada Facebook Page)
AVAILABILITY: YoYoDyne is estimating boards will be available before the end of May assuming nothing goes wrong. It might take a few days longer to get boards to the UK and Europe. For other information please see the links above.
SCHEMATIC: YoYoDyne has received approval from Tenor to release the schematic once the production boards are verified and released to the various vendors above. YoYoDyne is also interested in a possible future version that’s more DIY friendly.
GUILT-FREE VOLUME ADJUSTMENT: The majority of USB DACs only support 16 bits over USB. That means when you turn down the volume in software you’re getting less than 16 bits of resolution. At background music levels you might only be listening to 11 or 12 bit audio. But the ODAC has a 24 bit USB interface and enough dynamic range to allow guilt-free use of software volume controls.
OTHER DETAILS: There’s a lot more in the previous ODAC articles, but in summary, the ODAC is designed to fit inside the standard O2 enclosure in place of the batteries. A few internal wires need to be soldered to connect the ODAC’s output to the O2’s input jack. Other line level sources can still be used with the O2. It will also fit inside the upcoming ODA. And it can be used standalone with either the on board 3.5mm output jack or panel mounted RCA output jacks. The USB connector is a standard USB-Mini-B as used on the FiiO products, the Sansa Clip, cameras, etc.
FOUR REVISIONS: The ODAC has been through four lengthy revisions—two of the earlier boards are shown to the right. Despite the fact we started with essentially the reference design from the datasheets, the devil was in the details. The first version played music and sounded OK. Many companies and DIYers that “design by ear” would have stopped there. But that first version didn’t come close to delivering what the DAC chip is capable of. Each revision cycle took at least several weeks, cost hundreds of dollars, and involved countless hours of work. But, in the end, it resulted in much better performance compared to where we started..
REAL WORLD PERFORMANCE: Most DACs and PC sound interfaces priced under $200 fail to come close to the published specs for the chips they use. The FiiO E10’s Wolfson DAC chip is rated at 117 dB of dynamic range but the E10 only delivers a modest 98 dB. The power supply, PCB layout, grounding scheme, I2S waveform fidelity, clock quality, and more, often degrade the performance to well below the manufacturer’s carefully optimized reference design used for the datasheet specs. This is especially true for USB sourced signals and DACs running from USB power. But the ODAC, despite being USB powered, managed to come very close to the “Holy Grail” datasheet specs. See the Tech Section for more.
THE REAL NUMBERS: Here are the real numbers (versus the April Fool’s Day numbers I published two weeks ago) and they all meet the requirements for audible transparency. The letter following many of the results is the same letter grade (A-F) I’ve used in previous reviews with A being the best, and F being a “Fail”:
|Measurement||ODAC||FiiO E10||DAC1 Pre|
|Freq. Response 10 hz – 19 Khz 24/44||+/- 0.1 dB A||+/- 0.1 dB A||+/- 0.1 dB|
|THD+N 100 hz 0 dBFS||0.0029% A||0.005% A||0.0009%|
|THD+N 20 hz –1 dBFS||0.003% A||0.004% A||0.0009%|
|THD+N 10 Khz –1 dBFS||0.003% A||0.004% A||0.0007%|
|IMD CCIF 19/20 Khz –3 dBFS||0.0011% A||0.013% B||0.0005%|
|IMD SMPTE –1 dBFS||0.0004% A||0.004% A||0.0004%|
|Noise A-Weighted dBu 24/44||--102.8 dBu A||-98.3 dBu C||-105.4 dBu|
|Dynamic Range –60 dBFS A-Wtd||--111.1 dBr A||-97.6 dBr C||-110.9 dBr|
|Linearity Error -90 dBFS 24/44||0.0 dB A||0.0 dB A||0.2 dB|
|Crosstalk 0 dBFS Line Out 100K||--93.5 dB A||N/A||-106 dB|
|USB Jitter 11025 hz J-test 24/44||Excellent||Very Good||Excellent|
|Maximum Output Line Out 100K||2.0 Vrms||1.65 Vrms||2.5 Vrms (1)|
NOTE 1: Maximum output of DAC1 is configured with internal jumpers
BOTTOM LINE: The ODAC has been released to production and will hopefully be available by the end of May. I’ll soon be publishing more detailed measurements, results of listening tests, etc. I’m confident the ODAC is audibly transparent. And, especially when it’s installed inside the O2 or future ODA, it offers a level of performance that’s difficult to find without spending substantially more. It also offers detailed measurements and blind listening tests to back up its performance which is something very few other DACs offer at any price. And, paraphrasing from credit card commercials, worry free enjoyment of music can be priceless.
SPECIFICATIONS: Here are some ODAC basics:
- Audio Formats: 16/44, 16/48, 16/96, 24/44, 24/48, 24/96
- Interface: USB Audio Class 1
- Native Driver OS Support: Windows XP & Later, OS X x86, Linux
- Operating Systems Tested: XP, Vista, Win7 (32 & 64), OS X Snow Leopard, Ubuntu 9.1 32 bit
- Line Output: Approximately 2 Vrms into 5K ohms or higher
- Dynamic Range: > 110 dB A-Weighted
- Distortion: < 0.005%
- Dimensions: 49 x 58mm (see drawing to right)
BLIND EVIDENCE: So far I’ve run two relatively informal blind tests with the ODAC. The latest one used special software on the PC to play the same track on both my Benchmark DAC1 Pre and simultaneously on the ODAC plugged into the same PC (both connected via USB and running at 24/44). The ODAC was connected to an O2 headphone amp, and a switchbox allowed the headphones to be rapidly switched between the DAC1 and the O2+ODAC. The two sources were carefully level matched (using their respective volume controls) using a test signal and wideband DMM. I tried both my Sennheiser HD650 and Denon AH-D2000 headphones with a variety of well recorded favorite tracks. One other listener and I could not reliably tell which was playing.
FUTURE BLIND TESTS: I’ll hopefully be running a more comprehensive and rigorous blind test in the future. But, ultimately, it’s best to have listeners who expect to hear a difference, and someone else who understands the technical issues to supervise the test, perform level matching, sync the sources, etc. It would be ideal, for example, to have Mike and/or Lieven from Headfonia be a listener and George from YoYoDyne oversee the test. But sometimes geography and other factors restrict what’s realistic. If anyone is willing to help coordinate such a blind test, please contact me privately with the link in the right hand column.
USB INTERFACE: There are only a few USB interface chips capable of 24 bit operation. The best option we could find is the Tenor TE7022. It’s used in the Violectric USB 24/96 and several other commercial 24 bit USB DACs. Notably, it does not require proprietary drivers to work with any popular operating system including Windows because, like the Benchmark DAC1, it’s a USB Audio Class 1 interface. It also has respectably low jitter. The XMOS solution requires an expensive license and proprietary windows drivers and offers no audible benefit. The TAS1020B is being discontinued, requires extensive firmware, and also offers no audible benefit. A custom microcontroller creates even more hurdles with no audible benefit.
THE DAC CHIP: As I’ve explained elsewhere you can get transparent performance from most of the better DAC chips on the market from a half dozen chip vendors. What’s most important is choosing one that’s best suited to the particular application. In this case, that means running from a single 5 volt USB power supply, having a buffered voltage output (to avoid needing a single-supply op amp), and operating properly without using a microprocessor. Just those three requirements narrow the choices considerably. We chose the ESS Sabre ES9023 which is used in a lot of commercial designs (it’s an improved version of the popular ES9022). Unlike TI, Burr Brown, Analog Devices, etc, ESS specializes in audio chips and they did a nice job with the ES9023’s feature set and specs.
DIRTY LIES: Many popular “boutique” DACs (especially those being sold on eBay) mislead their fans by quoting only the chip specs for their entire DAC. Basically that amounts to cheating and lying. It further implies either the company is incapable of making the proper measurements or the real measurements were bad enough they didn’t want to share them. The implication is a DACs performance is solely determined by the chip used. But the opposite is usually more true. The implementation matters far more than the chip. the FiiO E10 is just one of many examples. The HA-Info I’ll soon be comparing to the ODAC is another.
CHIP ENVY: Any serious audio engineer designing DACs and making proper measurements knows the implementation matters more than the chip used. The first version of the ODAC, which closely followed the ESS datasheet, only managed about 98 dB of dynamic range. The distortion was also much higher than listed on the datasheet and the jitter was somewhat disappointing. Unless you exactly duplicate the chip manufacturer’s reference design, right down to the PC board layout (which I’ve yet to see any manufacturer do), you really don’t know what you’ll get. To measure all the important parameters you need more than just RMAA. You need a real audio analyzer with performance substantially better than the DAC being measured.
DESIGN BY EAR: Detailed and credible published measurements are missing with most “boutique” DACs being sold including those from Schiit Audio, Audio-GD, AMB, Twisted Pear, Burson, and NuForce. Where’s the credible evidence they got it right? A lot of these companies try to claim specs don’t matter, and they instead design by ear, but that method is seriously flawed (see: What We Hear). It would be like designing a car engine without a dynamometer and having no idea how much horsepower and torque it produced, how fuel efficient it was, etc. Given all the proven problems with sighted listening, and how our ears and brains work, those who claim to design by ear are very likely getting it wrong. Put another way, they’re often designing products with far lower fidelity than they’re otherwise capable of.
LESSONS LEARNED: If this project has taught me anything, it’s that getting much better than 16 bit (96 dB) performance can be challenging. The first version of the ODAC, despite following the reference design, only had about 98 dB DNR. That’s about the same as the FiiO E10. The photo to the right shows a few dozen assorted surface mount parts that were laboriously swapped out one at a time and measurements repeated dozens of times using the dScope. Some improvements were far from intuitive. Audiophile preferred polyphenylene capacitors performed worse than less expensive types. Additional filtering on the digital power supply dramatically increased jitter. Chasing down the last few dB of dynamic range the chip is capable of proved to be especially challenging. When it was said and done, the DNR went from 98 dB to over 111 dB. That’s a huge difference and something the design-by-ear crowd would have never achieved.
THE AUDIOPHILE WAY: All too many small or “boutique” audiophile manufactures and DIYers seem to just slap trendy chips on a board, listen to their creation expecting it to sound good (so that’s what they hear), and call it good. Many don’t even follow the reference design. Instead they include a bunch of “audiophile upgrades” expecting better performance—and they hear what they expect to hear even when it’s not true. But the ODAC demonstrated those upgrades often make things worse. So instead of getting even 98 dB DNR like the first ODAC revision, those following audiophile myths and designing-by-ear probably would have ended up with something even worse. Unless you’re making the right measurements, you really have no idea what you’re getting.
THE POWER SUPPLY: For reasons explained in the earlier ODAC articles, the ODAC is USB powered. This allows it to work standalone, as an internal add-on to the O2, and in the upcoming ODA. There are many obvious advantages to USB power but it often degrades performance due to noise. To get around this, the ODAC uses split digital and analog power supplies each with their own filtering and regulator. The analog supply has additional filtering and the critical reference voltages, and negative supply for the DAC chip, are further optimized. I literally tested more than 100 variations of components, including different brands of capacitors, to get the most out of the ES9023. This level of refinement would be impossible without a serious audio analyzer.
PUMP YOU UP: The ESS chip has the huge advantage of a built-in low noise charge pump. It generates its own regulated negative power supply allowing a Redbook standard 2 Vrms output from a single 5 volt USB power supply and a direct coupled output. This is an important distinction compared to a lot of USB powered DACs. Without the charge pump, or some other negative power supply, USB DACs can’t produce the Redbook standard 2 Vrms which reduces their effective dynamic range, lowers their ENOB, and creates level matching problems. It also requires an undesirable output coupling capacitor and usually results in loud transients on power up. The AMB Gamma, and most USB powered DACs I’ve tested, don’t meet the Redbook standard. The ODAC does. And it produces only a soft click on power up.
ADAPTIVE USB INTERFACE & LOCAL CLOCK: I’ve talked about this before, but just to be clear, the ODAC is NOT clocked by the USB port. So the quality of the audio clock, and any resulting jitter, is largely independent of the PC’s USB timing. It has its own low phase noise 12 Mhz crystal controlled oscillator that’s used to generate the MCLK and SCLK audio clocks.
A NOTE FOR 24/88 FANS: Some have asked about 24/88 high resolution audio support (popular for SACD rips). While the ODAC doesn’t support 24/88, it does support the audibly identical 24/44. It’s trivial to re-sample 24/88 audio to 24/44 with no artifacts as it’s a simple divide-by-two operation (and one the operating system will perform for you automatically). I know many audiophiles probably think they’re losing something, but nobody has proven they are. Meyer & Moran demonstrated in a very in-depth study that even 16/44 audio sounded identical to SACD. Another good read is 24/192 Music Downloads. And if you refuse to believe all that, try resampling some 24/88 audio to 24/44 and compare them yourself with Foobar and the ABX add-on. It’s been done at HydrogenAudio and elsewhere always with the same result: Unless you mess up the resampling somehow, or change the levels, you can’t tell them apart.
TRANSPARENCY GUIDELINES: The What We Hear article offers information and references outlining guidelines as to what’s required for a piece of audio gear to genuinely disappear from the signal path and not alter the sound in any audible way. Here are what I believe to be relatively conservative criteria for audible transparency and the ODAC passes all of them:
- Frequency Response 20hz – 19 Khz within +/- 0.1 dB (Most DACs, due to the Nyquist limit of 22 Khz, start to roll off past 19 Khz when operating at 44 Khz sampling rate—the ODAC is down about 0.4 dB at 20 Khz). The widely accepted, but less conservative standard is +/- 0.5 dB (1 dB total variation) from 20 hz to 20 Khz.
- All Harmonic, IMD, Alias, Modulation, & Crosstalk Components Below –90 dBFS and total sum below –80 dBFS (0.01%)
- All Noise Components below –110 dB and total sum below –100 dBFS
- All Jitter Components below –110 dB and total sum below -100 dBFS
GREEN GUIDE LINES: A few months ago I introduced green guide lines on several of my measurement graphs to help show the worst case ideal performance. Some of these are slightly more lenient than the above criteria or take into account more detailed thresholds (i.e. that power line hum can be slightly higher in level than midrange noise). For now I’m keeping the green guide lines consistent with earlier reviews. But please note the ODAC meets even the tougher criteria above.
PARTIAL MEASUREMENTS: I’ve made LOTS of ODAC measurements including some things I’ve never measured before—such as true latency. For this article I’ve only shown some of the more common measurements. In a future article I’ll cover additional measurements, 16 bit operation, 24/96, as well as several comparisons to the DAC1, FiiO E10, and an HA-Info eBay headphone DAC with a well respected DAC chip. So, in the interest of getting this article done sooner rather than later, and keeping it to a manageable size, only a sampling of 24/44 measurements are shown below.
DYNAMIC RANGE: A DAC’s noise floor impacts Dynamic Range (DNR), audible noise, THD+N, and can even exceed jitter-induced distortion. If you have to pick a single number to evaluate real world DAC performance –60 dBFS dynamic range (DNR) is one of the most revealing. The guys in the white lab coats have determined DNR greater than 100 dB results in transparency under realistic conditions. And, if you want to adjust the volume in software, it’s best to have at least 110 dB DNR to keep the noise floor inaudible even if the downstream gain is left cranked way up. Anything beyond 110 dB is past the point of diminishing returns—it looks nice on paper but doesn’t help the sound quality. The ODAC is a very substantial 14 dB better than the FiiO E10. Here’s both channels of the ODAC referenced to the 2.03 Vrms at 0 dBFS. Note the channels are very symmetrical indicating a careful PCB layout:
ODAC VS ESS: As explained earlier, the DNR quoted on datasheets is often something of a Holy Grail. The chip specs are typically from a very high quality AES/EBU or I2S laboratory quality signal (as output by high-end audio analyzers like a Prism dScope or Audio Precision). And they’re typically running from expensive ultra low noise bench power supplies costing thousands of dollars. It’s safe to assume the datasheet numbers were not made with USB data while running on USB power. ESS rates the ES9023 DNR at –112 dB A Weighted. The ODAC delivers –111.1 dB A-Weighted under the same conditions. In other words, even using USB data and power, the ODAC comes within a fraction of a dB of achieving the datasheet spec! I’m fairly proud of this aspect of the ODAC. It wasn’t easy.
CCIF IMD: This 19+20 Khz twin tone is a difficult test for many DACs running at 44 Khz. Old style (NOS) non-oversampling DACs especially struggle due to aliasing problems. In addition the output buffer (or I-V stage) in many DACs contributes high frequency distortion because the RC filter can be a challenging reactive load at these frequencies. If you look back through my reviews, you’ll find lots of products struggle on this test. Even the E10 turned in a marginal result. The ODAC, however, due to careful optimization of the output filter, and the superior digital filtering of the ESS DAC, does very well here with everything in the audio band well below 100 dB (both channels shown). Note also the 19 and 20 Khz tones are visibly equal in level which is not the case for many DACs:
SMPTE IMD: This twin tone test is more revealing of low frequency problems including power supply interaction. Again, the ODAC does very well with everything well below 100 dB (both channels shown):
100hz THD+N @ 0 dBFS: This test checks for clipping of the DAC at 0 dBFS and also shows the maximum output and channel balance error. You can see the ODAC produces 2.03 Vrms which is within an in significant 0.03 volts of the Redbook standard for digital audio. And even at 0 dBFS, the distortion is still 3 times less than what’s required for 100% transparency. The channels are perfectly balanced to within 0.001 dB. This spectrum is shown all the way out to 96 Khz and you can see the ESS DAC is well behaved even above 20 Khz with all noise still below about –110 dBFS. This is excellent performance:
THD+N VS FREQUENCY: Here’s the distortion performance at –1 dBFS from 20 hz to 20 Khz into a more challenging 10K load with a measurement bandwidth of 22 Khz. At 1 Khz the distortion is only 0.0027% and it remains around 0.003% over most of the audio band with only a slight rise up to 0.0048% at 9 Khz before the harmonics fall above the audible range. This is excellent performance and both channels are very closely matched (yellow vs blue):
NEW JITTER FINDINGS: I did quite a bit more research on jitter during the ODAC’s development. I’m also using a new dScope method that shows the same spectrum as before but now the symmetrical jitter components are marked (with a white “X”) and summed to obtain a total numerical value (previously the dScope was just showing the total residual noise floor). Having the single number (-103.3 dB below) made it easier to optimize the ODAC for the lowest jitter. The objective evidence conservatively indicates if you keep all related components below -110 dB, and the total below -100 dB, the jitter will be entirely inaudible. Jitter creates dissonant distortion products in the audible band. It’s reasonable to assume if the audible effects of jitter are kept at or below the inaudible noise floor, they too will be inaudible. So the same levels of –100 dB and –110 dB that apply to noise also apply to jitter contributions. This is also consistent with various professional reviewers and their anecdotal opinions on jitter performance as well as my blind testing against the Benchmark DAC1 which has even lower jitter.
ODAC JITTER: The ODAC passes the conservative criteria with several dB to spare even on the worst-case J-Test signal. And the spread at the base of the signal (very low frequency jitter) is extremely minimal being entirely below –130 dB. It’s also worth noting the jitter here looks subjectively worse because the noise floor is much lower than most of my jitter measurements which are done with a 16 bit test signal. A 16 bit noise floors masks most of the “spikes” seen below. The ODAC also has negligible inter-channel phase error and essentially perfect pitch accuracy:
CONCLUSION: Hopefully the above provides some good evidence the ODAC delivers transparent performance. The next ODAC article will compare the 16 and 24 bit performance and I’ll be comparing it to several other DACs. I’ll be publishing many more measurements such as modulation noise, channel separation, square wave/impulse response, latency, frequency response, absolute noise, and more.