Saturday, 3 August 2019

MEASUREMENTS: AudioQuest Dragonflies Reviewed! Dragonfly Cobalt, Red, and v1.2.

Audiophiles following the news cycle know by now that the new AudioQuest Dragonfly Cobalt is out. About 3 weeks back, I noticed out of the blue (no pun intended), almost all mainstream audiophile websites had a post or promotional "give away" for one of these! Clearly this made a significant splash in the collective audiophile psyche.

Over the years, I've reviewed and measured these little USB DAC devices going back to the Dragonfly v1.2 in 2014. More recently, I wrote a series on the Dragonfly Black in 2017 with a good chunk of that looking at MQA "rendering". Note that I don't have the Black v1.5 here for direct comparison and will instead refer to those older results and articles as appropriate.

Here then are the 3 Dragonflies (Dragonflys?) I have in for a listen and on the "test bench" for direct comparison. From left to right - v1.2 (released 2014), Red (2016, ~US$200), Cobalt (2019, ~US$300):


As usual, I will start with building up the objective results and then later in the post, I'll talk about subjective listening impressions and broader ideas. I know this sequence appears backwards compared to how audiophile reviews usually are structured. Most reviews typically start with background on companies, personal anecdotes, people involved in the product, rationale, etc. If one is fortunate, maybe a sidebar or graphs at the bottom of the review for objective results.

As a "more objective" audiophile, the typical sequence above is not how I would prefer to learn about a new product. There are often insights one can gain through disciplined objective evaluation one simply cannot get based on company literature or even just listening unless one were truly meticulous. Objective results apply to us all, while subjectivity is the domain of the individual. As such, technical adequacy and fidelity IMO are much more interesting and significant than a company's history, who the "guru" was behind it, or to be honest, what the reviewer "heard" or probably more often than admitted, thought they "heard"; I'd rather leave many of those items as sidebars.

Having laid out this structure, to be clear, as a "reviewer", I took turns listening to the 3 DACs with different music and headphones for about 3 days, a few hours each day, before capturing the objective results so I had some sonic impressions before seeing what the measurements looked like. Since I'm borrowing the v1.2 and Red from others, I figure that's good enough listening time to get a sense of which sounds better.

First, since it's new, let me show you the retail box and contents of the Cobalt:



As usual with AudioQuest products, we see a nice professional package. Inside, there's a carrying pouch, the DragonTail (female USB-A to male USB-C), the DAC, and an assortment of detailed instructions and both Tidal and Qobuz promotions (first 10,000 units we're told).

The DAC itself looks great:


Shiny metallic automotive-like paint job, metal body, good heft, with a nice gold-plated USB connector. The illuminated dragonfly LED changes color based on sampling rate although the color scheme varies depending on the model. The 3.5mm audio jack has a plastic rim and small amount of "wiggle". As others have complained about, there is unfortunately a little bit of a rattle when I shake it. The Red felt more sturdy in this regard but slightly longer (4-5mm), and less sleek.

I made sure the Red was updated to the latest 1.07 firmware before listening and testing. The older Dragonfly v1.2 is not firmware upgradable. The current version of AudioQuest's Device Manager doesn't recognize the Cobalt yet:


In Linux (Ubuntu), plugging the DAC into the USB port and issuing the command "sudo lsusb -v" tells me that the serial number for this Cobalt is "AQDFCB01001034**" (last 2 numbers withheld from serial number to protect the innocent).

I. Digital Oscilloscope, Digital Filter, Impulse Response

Okay, let's start shall we with the "microscopic" as I usually do with DACs. I find this generally a good way to begin looking at the analogue output and to examine for hints of anomalies.

With the digital sampling oscilloscope, let's have a peek at the 2 channels playing 1kHz sine and square waveforms at 0dBFS, 100% volume. Let's start with the Dragonfly v1.2:


Compared to the v1.2 sample from 2014 I measured, this one had excellent channel balance. Something that was present but I didn't point out in the 2014 measurements is that subtle visible distortion in the negative half-cycle at full amplitude. You can see this at the trough of the sine wave but more obvious with the square wave tracing as highlighted. At full volume, it looks like we're at the margin for clipping.

The Dragonfly v1.2 here has a measured peak voltage of 1.86Vrms (specs sheet value of 1.8V). This is higher than the Dragonfly Black 1.5 with 1.2Vrms output.

Dragonfly Red 1.0:


While I constructed the square wave to be band-limited, there is still a little bit of "aliasing" which can trigger some ringing. We see from the Red's square wave tracing that this looks like a minimum phase filter they're using - not a surprise as this is indeed what AQ did with the Dragonfly Black demonstrated a few years back.

Nice waveforms with the Red. No evidence of strain or clipping at 100% volume, 0dBFS digital signal. Excellent channel balance as well.

Here's the Dragonfly Cobalt 1.0:



Well, well, well, what do we have here?

Despite advertising it as 2.1Vrms output like the Red, it appears that the amplifier gain has been pushed a bit further when asked to play at 100% (I'm guessing it's the amplifier gain rather than at the level of the digital filter). In doing so, we're seeing some clipping of the sine wave, and this has also affected the square wave peaks and troughs. When set to the same Windows volume setting (no physical buttons on the device itself), this Cobalt is approximately 0.5-0.6dB louder than the Red. It's aiming for ~2.23Vrms at peak output but clearly not quite getting there cleanly.

This will result in distortion if you set the DAC at 100% and you're playing a lot of loud, compressed albums that hit 0dBFS. An example where this might happen is if you keep the DAC at 100% and hook it up to your sound system pre-amp expecting clean 2.1V line-level output (AQ discusses doing this in their literature).

The characteristic of the "ringing" confirms that we're also looking at a minimum phase digital filter like the Black and Red. Channel balance again is excellent.

Output impedance is important to measure especially if you're going to be using low impedance ear buds and IEMs. The technique I use is quite simple and should be an adequate estimate. Absent a low impedance load, I set the output of a 1kHz 0dBFS sine wave to 500mV, then measure the voltage drop across a 20Ω resistor, and solve for the resistance/impedance.

The Dragonfly devices have been good for low output impedance from the beginning. Stereophile's measurements of the original Dragonfly v1.0 showed a value of 0.68Ω and they listed "<1Ω" for the Red.

My results:
Dragonfly v1.2           0.49Ω
Dragonfly Black (2017)   0.61Ω
Dragonfly Red            0.53Ω
Dragonfly Cobalt         0.40
Ω
Looks great across the board with excellent sub-1Ω values. One would expect that these devices will easily accommodate any reasonable headphone.

As suggested already in the square wave tracings, AudioQuest has been playing around with the devices' digital filters over the years. Here are the impulse responses of the 3 Dragonflies, captured @ 768kHz with the RME ADI-2 Pro FS:




Notice what has happened over the "generations". We've gone from an orthodox steep linear phase filter, to a steep minimum phase filter, and now a gentler "weak" minimum phase filter. I presume this is all in the name of more "natural" sound.

All 3 have maintained "absolute polarity".

To have a better look at the filter characteristics, here are the "Digital Filter Composite" graphs (inspired by Juergen Reis' discussions awhile back) for each device at 44.1kHz:





The Dragonfly v1.2 has a steep cutoff, excellent anti-imaging suppression, but quite a bit of harmonic distortion and intermodulation products can be seen.

Notice that all 3 DACs "overload" with the 0dBFS white noise. Modest amounts of intermodulation and harmonics can be easily seen with the Red as well, but it's much less than the v1.2.

In comparison, the slow roll-off filter with the Cobalt has now allowed imaging of the 19 & 20kHz tones at 24.1 and 25.1kHz to seep through. The 19 & 20kHz signals' harmonic levels don't look bad though.

The high frequency roll-off from the Cobalt shouldn't intrude into the audible spectrum too much (certainly less than the PonoPlayer / Ayre "Listen" filter from a few years back).

II. RightMark noise and distortion results

For these measurements of noise and distortion, I used the following measurement chain:
Microsoft Surface Pro 3 with USB-powered Dragonfly DAC --> 4' shielded phono to RCA cable --> RCA-to-XLR --> RME ADI-2 Pro FS ADC --> 6' shielded USB cable --> measurement Win 10 laptop
I used the 12V Talentcell power source described last week for the RME ADC when measuring the Dragonflies. Laptops running off battery power.

Bit-perfect Windows 10 WASAPI "event" driver was used for all Dragonfly DACs and comparison devices as appropriate. Player software is Foobar. RightMark Pro 6.4.5 used here.

A. 16/44.1 "Standard" Resolution
Starting with "standard resolution" equivalent to CD, here are the summary results:


On the left we see results from the three Dragonfly devices. Then I've included the SMSL iDEA, another "USB stick" DAC based on an ESS chipset. Since these Dragonflies are advertised for smartphone use, let's also see what the old iPhone 6's headphone output can do in comparison. And finally on the far right, for reference as a flagship ESS Technology ES9038Pro DAC, we have the results from the Oppo UDP-205.

Graphs - as usual, click on image to examine the details:


Over the years, I've shown that 16/44.1 isn't much of a challenge for most DACs these days when it comes to noise level. What truly separates the great DACs from the not-so-great ones is how well the devices keep distortion levels low. The results from the v1.2 are not surprising. I had commented already back in 2014 that the results were not great. In comparison, the Red measures significantly better. It's clearly a step up from the v1.2 and I'd be satisfied with this level of fidelity along with more power to drive the headphones.

I must say that I am already disappointed in the waveform distortion shown above on the oscilloscope and now we see inexplicably high THD and IMD+N from the Cobalt! Shocking in fact that the results are merely on par with the older Dragonfly v1.2 from a few years ago. In fact, the total harmonic distortion amount was worse than the v1.2!

Results from the SMSL iDEA and Oppo UDP-205 are indicative of the low distortion levels one should come to expect from high quality ESS DACs.

As you can see, the choice of using a slow roll-off filter with the Dragonfly Cobalt has resulted in that high frequency dip, separating it from the other devices. Not a large amount, about -0.5dB at 15kHz compared to the PonoPlayer of -1.5dB.


B. 24/96 "High" Resolution
Remember that the highest resolution bit-depth/samplerate supported by the Dragonflies is 24/96. In fact, I applaud AudioQuest for their "Keep It Simple, Stupid" principle thus far by sticking with default OS drivers and maintaining only the most important PCM samplerates (44.1/48/88.2/96kHz) for this class of device. While audiophile enthusiasts might miss 192/384+kHz and DSD, the vast majority will not care about such things. Compatibility is great and I have never had any issues with simply "plugging-n-playing" these DACs on mobile devices.

Having said this, looking ahead, now that USB Audio Class 2 is ubiquitous across devices in the last couple years including Windows 10, future generations of DACs like this probably could embrace more samplerate options with less compatibility worries.

For these 24/96 results, I can't include the iPhone 6 as 96kHz isn't supported. In its place, let's put up the results from the PonoPlayer, another "audiophile favourite" portable device from the recent past.


Nice to see the excellent low noise floor of both the Red and Cobalt. Again, we see that among the Dragonflies, the Red is the clear winner for lowest distortion. The Cobalt puts up impressive noise floor and stereo crosstalk scores but this is really a minor victory in the context of the poor THD and IMD even compared to the PonoPlayer.

Notice that the Cobalt's "Frequency response (multitone)" test result looks messed up! Something about this portion of the RightMark test signal appears to be discombobulating the DAC. I wonder whether the MQA "rendering" feature might be misconstruing the data being fed. I did not have any issue with this for the Black or Red measurements. As a result, I graphed the frequency response using the "Frequency response (swept sine)" test instead:


Again, it's curious that the distortion level is so high for the Cobalt. To have a better look and for corroboration, here are the THD(+N) results using WaveSpectra running a 131k-point FFT at 48kHz. This is a standard 1kHz, -3dBFS tone at 24/48 played using the Surface Pro 3 computer with volume in Windows 10 set to 100% and at 50% (a -11dB drop from peak output for the Red and Cobalt) just in case relative distortion is a function of the output level:






As you can see, I've labeled the values of the 2nd to 4th harmonics for ease of comparison with 1kHz normalized at 0dB. With the v1.2, there is relatively strong second, third and fourth harmonics compared to the Red, hence the calculated THD is significantly higher. If you look at the Cobalt, notice that the levels are also significantly higher than the Red. In fact, the Cobalt's 2nd harmonic is even higher than the v1.2!

Another noticeable difference is the nature of the harmonic cascade. Notice that the v1.2's harmonics take the form of a predictable, relatively smooth decline with minimal higher order components. Not so with the Red or Cobalt. Higher order, odd harmonic distortions are not supposed to be good sounding, right? Well, the Cobalt and to a lesser extent Red, have quite a bit of those.

Although the SMSL iDEA is an older USB stick with its own issues (ie. sometimes it doesn't initialize well with Android and Linux, resulting in noisy sound, higher current draw), here's what the THD(+N) of SMSL's ESS DAC implementation looks like:


That is simply so much cleaner compared to any of the Dragonfly DACs! Remember, the iDEA is based off an ES9018Q2C chip with integrated headphone amp, costs <US$90, yet the distortions are so much superior to the Cobalt's brand new ES9038Q2M DAC + ES9601 headphone driver! This just seems very wrong.

One last thing before we end this section. Here's a demonstration of the clipping distortion at Windows "100%" output level with a 0dBFS 1kHz sine wave compared to "98%" showing both waveforms from the oscilloscope and resulting FFTs:

BTW, these measurements were done with Cobalt connected to Huawei MateBook X Pro rather than Microsoft Surface Pro 3 to make sure that this wasn't just an issue with the USB output from the Surface.
With this Cobalt, if I were going to plug it into a stereo system and wanted to avoid extra distortion, I'd do it with volume level at "98%" in Windows (or equivalent) with the non-clipped output peaking at ~2Vrms. If you're wondering why 98%, the Cobalt has 64 volume levels which are mapped to Windows' "100", therefore you'll often see a 2% change in Windows volume for one notch up/down in the output.

Thankfully, even at 100% volume, hitting ~0dBFS should be rare unless you're listening to lots of loud, dynamically compressed music - sadly, this is not uncommon with modern mastering techniques. In any event, clipping at maximum output IMO should not be happening from a well engineered "audiophile-grade" DAC in 2019 costing a few hundred dollars!

III. Jitter

As usual, here are the 16-bit and 24-bit Dunn J-Test graphs, a simple test of jitter:




No issue across the devices. Notice there are two pairs of sidebands with the 24-bit J-Test on the Red below -120dB and flanking the primary signal closely. Sideband levels are even lower with the Cobalt which is good.

Supposedly there's some extra "monoClock® technology" in the Red and Cobalt (and Black) to improve clock performance. Not sure about the old v1.2 which already looks great to me. Remember, jitter level hasn't been an issue with essentially every decent asynchronous USB implementation over the last 5+ years of testing. I have not seen bad jitter results from ESS DACs in general. (Plus I have doubts anyone should be worried about jitter these days anyway...)

I saw a note about this whole notion of Jitterbug technology being implemented in the device among other claims of sonic improvements by using things like iPurifiers and whatnot. Seriously folks, the noise level is actually very low already with the Red and Cobalt, and no matter what you do with these add-on USB noise "filters", it's not going to fix the distortions identified above, nor affect jitter substantially. Furthermore, isn't it cumbersome to be using these USB doohickies?

IV. Power utilization

I thought I'd throw this one out there because these are portable USB DACs often used with low-power devices like smartphones listening to music on the move. As a result, low power utilization is important.

Here's a picture of the Cobalt connected to my Microsoft Surface with the little USB Tester I have showing voltage and current while connected to my "workhorse" 1MORE Quad Driver IEMs (32Ω):


To get a sense of power utilization, here are some current draw numbers when the device is idle (nothing playing), and as a standard condition, while asking the DAC/amp to output 0.5Vrms/1kHz into a 30Ω load. I've included data from the SMSL iDEA for comparison.
                            Idle      0.5Vrms/1kHz     
Dragonfly v1.2              0.05A       0.13A
Dragonfly Red               0.04A       0.08A
Dragonfly Cobalt            0.06A       0.10A
SMSL iDEA                   0.11A       0.15A 
As you can see, given the same 0.5Vrms/1kHz output level into 30Ω, the Red and Cobalt draw less current than the v1.2 as expected.

Also, to get a sense of how much "clean" power the devices can deliver, using that same 30Ω load, I examined how high I could push the voltage before I saw visible distortion in the sine wave on an oscilloscope, then also recorded the current draw at that point. Take these just as estimates only since I'm determining this by sight rather than a measured value. I've also included a column for "max current" - the amount of current being sucked up by the device when I push the volume to 100% into the 30Ω load regardless of how distorted the sine wave became.
                 Vrms (30Ω)  before vis. distorts    Current @ 100%
Dragonfly v1.2          735mV / 0.15A                  0.22A
Dragonfly Red           900mV / 0.11A                  0.14A
Dragonfly Cobalt        950mV / 0.14A                  0.17A
SMSL iDEA               1.10V / 0.18A                  0.21A
On the whole, the Red and Cobalt are able to provide a bit more power to drive headphones without as much distortion compared to the v1.2. Remember that the Red and Cobalt share the same headphone amplifier chip although as we saw in the oscilloscope tracings above, it looks like at 100% volume, the Cobalt I have here's ES9601 is pushed a bit beyond the specified 2.1Vrms output level.

In any event, I do like that the Dragonfly Red and Cobalt have managed to keep peak current draw consistently <200mA. Neither get beyond mildly warm playing music after a few hours (in comparison the iDEA gets warmer but not hot).

V. MQA Filters

For completeness, MQA "rendering" is available in the Black/Red/Cobalt. Remember, this is not the more sophisticated "decoding" process and you'll need the playback software to "unfold" the 44.1/48kHz MQA data to 88.2/96kHz and sent to the Dragonfly. Check out the 2017 "Dragonfly Black Part 2" article if you're curious about "rendering" in general. IMO, please audiophiles, let's just retire MQA already. If you want high-resolution, get real high-resolution rather than this. IMO, MQA offers nothing "good" from the perspective of high fidelity. It's also rather clear that things like MQA-CD would deteriorate sound quality. So let's stop wasting time and money.

I did not bother capturing all the MQA filters (already published previously), but here are the impulse responses of standard Cobalt 96kHz playback, MQA 96kHz, and a motley crew of the first six MQA "filters" used when rendering as if 192kHz from the Cobalt:


As you can see, the general morphology looks about the same as the Black and Mytek Brooklyn from 2017; this time captured in higher resolution (768kHz). As far as I can tell, there are no major variations to suggest any significant "customization" for the specific DACs. From the table of impulse responses, it's also hard to see what organizing principle was involved in ordering this collection of filter "designs".

Of course, thanks to Måns Rullgård for his work and insights on MQA based on impressive reverse-engineering in order to capture these.

VI. Subjective Impressions

As I mentioned at the start, although I'm presenting my findings with objective results first, I actually made sure to listen to the DACs before putting them through the oscilloscope and ADC measurements.

I spent my time taking turns with 3 headphones - 1MORE Quad Driver (32Ω), AKG Q701 (62Ω), and Sennheiser HD800 (300Ω). While I did try quickly switching between DACs using the headphones, this did prove challenging so I ended up mostly switching DACs between the same songs to get a "feel" of the sound signature.

Let me be honest, when it comes to casually using these devices on cell phones and walking around outside with a pair of headphones on (typical scenario), I honestly would not be able to hear a difference between them. Other than blasting the volume to see which plays louder, differences are subtle and demand attentive listening in a quiet environment with good headphones. Therefore, the differences I'll be talking about now applies to listening in the quiet of my home using headphones like the AKG and Sennheiser above that I would never walk around outside with.

Punchy modern songs like Childish Gambino's This Is America sounds fine regardless of which DAC I plugged in. "Distorted by design" comes to mind as I listen to this track. Likewise, a "modern mastering" like the recent The Lion King remake soundtrack (DR8) sounds fine regardless of which device I used; I personally thought that Beyoncé's vocals were overdone on tracks like "Can You Feel The Love Tonight" - an innocent young lioness doesn't sing like that :-).

I listened to selections from Joshua Bell's Bernstein: West Side Story Suite (16/44.1 SACD rip), Keith Jarrett's J.S. Bach: The Well-Tempered Clavier Book I, Nat King Cole The Nat King Cole Story, Casey Abrams' Put A Spell On You (24/96, binaural), and Alphaville's Forever Young (2019 Super Deluxe).

The main difference between the v1.2 and either the Red or Cobalt is of course a slight difference in power output (less than the difference between the Black vs. Red/Cobalt). With the AKG Q701 for example, at 100% volume, the v1.2 sounds loud but with either the Red or Cobalt, the volume was just slightly more uncomfortable. The Red was certainly favoured compared to the v1.2 as sounding more "effortless" with the tracks I listened to and just plain "cleaner" sounding.

What is more interesting of a comparison is between the Red and Cobalt. To be honest, I certainly did not feel that the Cobalt sounded any more "natural" or "clean" even before the measurements. For example, on Casey Abrams' album playing "Georgia On My Mind", I thought the Red portrayed a better sense of space (remember, this is a binaural recording). Also, the percussive attacks/"thwak!" sounded more precise, a sense of better temporal definition. The timbre of the saxophone also sounded more "right", "clearer". I would not use the word "dull", but I just didn't think the Cobalt sounded as dynamic and "interesting" as the Red.

In the same way, during more complex passages with multiple overlaying instruments (like on Joshua Bell's "West Side Story Suite"), while the Cobalt wasn't bad, I thought the Red delineated the instruments better and the treble definition improved. Instruments like the solo violin stood out more and the harmonic complexity of the sound better realized. This difference was especially highlighted with the Sennheiser HD800 headphone.

These particular tracks stuck out more for me than the other music I listened to. I just don't think typical pop, or electronic music like the Alphaville remaster sounded much different. Both the Red and Cobalt were equivalent with deep bass reproduction. Likewise, the soothing voice of Nat King Cole sounded good on both the Red and Cobalt; but these are old recordings of limited resolution. I think this is a good reminder that the kind of music we listen to just as much as the environment and headphones used is very important in order to judge subjective sound quality.

VII. Conclusions...

Well, AudioQuest has made an interesting "family" of low-power USB DAC devices since 2012 that appear to be popular.

If we compare the generational jump from the v1.2 released in late 2013 to the Red in 2016, clearly there have been improvements in objective fidelity, higher output power, while sipping less electricity from the source device. Playback of the 0.5V, 1kHz tone into the same 30Ω load showed that the Red was able to do this with 40% less current draw from the 5V USB port compared to the v1.2. While I didn't test the Dragonfly Black this time, other than using less power, it's actually a step back in terms of output level with about the same THD and IMD distortion amounts as the v1.2 - I presume that's why they just called it "Black v1.5". I don't think there's any debate then that the jump in price from these lesser models to the Red has some merit, and that one obtained a higher fidelity product.

Interestingly, AudioQuest seems to be suggesting that there's a significant jump from the Red to the Cobalt as well. This is I think obvious when they bumped the asking price by US$100, created the new colorful name, and released various claims in the media about it being "better". I'm going to have to take the "minority report" here and absolutely disagree with all those positive reviews out there about the Cobalt - articles from Headphonia, Darko.Audio, MajorHiFi, Audio AdviceHi-Fi+, and The Audiophiliac Guttenberg with "Wowza" hype video and CNET article. (There are also the obvious advertising videos like Audio Advice and World Wide Stereo.)

As far as I can tell, this Cobalt DAC obtained from retail channels, paid for with my own money, with no evidence of damage or defect, is a step down in fidelity based on objective and subjective grounds. I cannot believe that what I'm hearing and seeing with this product is a reflection of an ESS "9038"-series component and therefore am highly suspicious that AudioQuest did a poor job with implementation.

From a hardware perspective, the Red-to-Cobalt change with upgraded DAC (ES9016K2M to ES9038Q2M) and microcontroller (PIC32MX270 to PIC32MX274) should have resulted in lower distortion readings (THD+N from the 9016K2M to 9038Q2M is supposed to be a 10dB improvement!) and lower power utilization. I saw neither. While I don't expect miracles with power utilization since it's already low, the significantly worse distortion levels are a disappointment. I have a suspicion that instead of making a slightly more compact "Dragonfly Red v1.5" where AQ incrementally made improvements to form factor and fidelity, they seemed to have decided that something "more" or "special" needed to be done to differentiate the product so as to justify the price increase. Okay, so they threw in a DragonTail, but a good quality female USB-A to male USB-C cable isn't expensive. It looks like the same headphone driver was pushed to the point of distortion in the Cobalt I have here. They fooled around with the filters (the fad these days) all in the name of sounding more "natural". Even with these changes and their anticipated effects, I'm still at a loss as to why THD has regressed back to Dragonfly v1.2/Black levels! Yes, the noise floor is very low and this might be related to the "improved power-supply filtering". That's great when nothing's playing, but what value is that when fidelity is compromised once you are actually playing something!?

Of the Dragonfly line-up, the only one I would feel somewhat okay about recommending at this point is the Dragonfly Red if you need something like this and the price is right.

I cannot recommend the v1.2/Black 1.5 models generally because sound quality really isn't better than many portable devices like the smartphones themselves these days. I am especially disappointed with the Cobalt for lack of overall value. I think it would have been better to just upgrade the Red to the smaller form factor (fix the phono jack rattling, AQ!), use the new DAC and microcontroller to lower distortion, calling it "Red v1.5" without messing around with the filter. I suppose "Red v1.5" would not sound as sexy and draw attention as much as "Cobalt" would for the advertising department, plus it'd be impossible to raise the price by $100. However, this would be honest acknowledgement of it being a simple step up assuming it actually achieved cleaner sound!

I'm curious about this 0dBFS clipping and the +0.5-0.6dB gain over the Red I'm seeing. All I can report is that I've triple checked with two 1MΩ oscilloscopes plus the RME ADC with 9kΩ input impedance. This is indeed happening on the Cobalt I have here. I asked around and an earlier serial number Cobalt did not have this issue (reported as 2.16Vrms max output). Therefore I'm not sure if this is simply a manufacturing variation due to component tolerances or an intentional change during the production. Perhaps this will be clearer when we see other test results.

I must admit that the more cynical side of me is suspicious of the slightly boosted volume level of the Cobalt compared to the Red. Everyone knows that we need to keep the volume the same when we compare equipment. Since they're both stated as delivering 2.1Vrms, one would think that if we keep the same volume setting (ie. in Windows, keeping the volume slider at 30%) would result in the same amplitude playback, right? Well, if you were to do an A/B comparison of the Red with the Cobalt tested here at the same volume, guess what, the Cobalt ends up being about +0.5dB louder and would sound "better" to many! How convenient... (Of course this is just speculation, audiophile cable companies like AudioQuest would never think about silly tricks like this. ;-)

Finally, I would be very cautious about the numerous "golden ear" subjective testimonies/reviews I have read thus far in the online media. Just look at the subjective descriptions and claims; Cobalt more "clear" than Red? Better treble extension? Sense of ease? More space between instruments? Wider soundstage? Effortless dynamics?! Who knows, maybe I have a bad copy here, or maybe they do like this kind of sound and are okay with this lower level of fidelity. Since my philosophy as an audiophile is to achieve lower distortion and ultimately better transparency, I'm not sure what many of these audiophiles are listening for in order to so quickly give a positive recommendation. Certainly, it's not impressive that none of the reviewers even hinted that maybe the Cobalt doesn't really sound as clean/undistorted as the Red!

I brought the Cobalt over to my friend who owns the Red for a quick A/B prior to completing these measurements for a second opinion with the expectation that the Cobalt "should" be better. He thought it sounded slightly "smoother" but recognized that this didn't mean that the resolution was any better. He felt there was no need to "upgrade". Yup, cannot disagree with that.

I appreciate that Headphonia actually included a disclosure that the Cobalt they tested was "sent to us free of charge in exchange for this preview article at the launch and a full review". Great to see this level of honesty which allows the reader to assess for potential biases. So how about the other reviewers?

Looking on the bright side, maybe a firmware upgrade could improve things substantially. Maybe the power output can be set a bit more conservatively and target no more than 2.1Vrms output. Maybe their digital filter coefficients needs to be tweaked for lower distortion. I was hoping that I would keep this Dragonfly Cobalt as a portable USB DAC reference given the ESS Technology latest generation "9038" heritage. Alas, I'm not willing to wait for any potential firmware tweaks given what I hear and see. I'll be returning this one back to the store. Maybe next time, AQ.

--------------------

Happy August everyone! My, how time flies.

Hope you're all enjoying the music...

46 comments:

  1. Welp, expected I guess. I was hoping from the positive reviews and impressions I'd seen online that the new DAC would be an improvement even though the amp was the same, but oh well. Sticking with my built in phone/laptop jacks + fiio a5 for portable use I guess. Sold my Red immediately upon purchasing the Geekout 1000 and hearing them side by side. Geekout is now my main desk source.

    ReplyDelete
    Replies
    1. Hi Degru,
      I must say that I was rather shocked by what I heard and saw with the Cobalt. True, I wasn't expecting miracles but you would think that objectively, there would still be some gains made over the Red.

      I really wonder how AQ designs these things. Is it purely being "tuned" by ear? If so, whose ear!?

      I look at the advertising information and see:
      "DragonFly Cobalt, our new flagship DAC, takes what music lovers around the world have come to expect from the multi-award-winning DragonFly family—naturally beautiful, seductive sound—and strips away fuzz and fog that weren't even noticeable until Cobalt removed them."

      I suppose "beauty" and "seductive" could be totally subjective. But surely this thing is not doing anything good for the "fuzz" and "fog"!

      My understanding is that there are major issues these days with those who have pre-ordered all kinds of stuff with Light Harmonic and not delivered, so I would not recommend the company. The GeekOut V2 I have and measured back in 2015 was one of the first devices to have enough overhead to not overload the 0dBFS signal. Yup, it delivered clean sound compared to the DF series.

      Delete
  2. Ooooff. Not surprising given AQ's track record. On the other hand, I've read good things about the Nextdrive Spectra X. Anyone have experience with this for a portable DAC option for Android + UAPP, or perhaps other recommendations?

    ReplyDelete
    Replies
    1. Hi Unknown,
      Don't know about the Spectra X as I have not tried. Looks like it has the ES9018Q2C and presumably will perform similar to the SMSL iDEA... Hopefully less issues with compatibility! Would be curious if anyone has tried.

      Delete
    2. Apple's usb-c is excellent for its $9. I personally own a Musiland MU2, also great!

      SMSL idea is older but as shown by archi, very good too.

      Delete
  3. "I'm going to have to take the "minority report" here and absolutely disagree with all those positive reviews out there about the Cobalt..."

    Those aren't really 'reviews'. Just a rehash of the manufacturer's spec sheet, and some subjective utterances. Pretty worthless. Thanks for your honest review, along with the time and effort you put in to measuring the products.

    ReplyDelete
    Replies
    1. Hey mp,
      Alas those look like the only "reviews" being offered online! Hard to know how to interpret who is truly independent as opposed to those offered free products or shills directly expected by the company to put in a good word.

      If I plug in "AudioQuest Dragonfly Cobalt review" in Google, we see pages upon pages of this stuff! It's really quite remarkable and not hard at all I imagine to "carpet bomb" the search engines or pay some money to direct visitors toward the advertised message.

      Delete
  4. Archimago, can I ask what ADC you using for the measurement?
    To be honest, I don't understand why people like to buy such a tiny power HPA if any 3.5mm jack in chinese Android produce a similar power and even better quality? 20-30mW, are you lidding me? ))
    PS: What about this one beast review? https://e1dashz.wixsite.com/index/9038s

    ReplyDelete
    Replies
    1. Hi. ADC used is the RME ADI-2 Pro FS.

      I guess for some Android phones there might be issues with output impedance, maybe inadequate power output.

      For me, since I'm not a big fan of being an audiophile "on the go" and don't believe I can really appreciate pristine quality walking around outside with typical headphones on, I generally would not spend hundreds on these kinds of devices. Having said this, I certainly would have appreciated having a small DAC like this as a clean audio signal generator.

      Interesting looking device in the link. Distortion results look way better than the Dragonflies!

      Delete
    2. Distortion results look way better than the Dragonflies, at the way higher power(10x times), 8x times less output Z, less noise, size and for 1/4 cost )) However, it has fully balanced 2.5mm jack output, hence, not all HP is suitable. Another drawback is idle consumption 90mA.
      Best regards, Ivan aka E1DA.

      Delete
    3. That's what I call a 'mis-design'. Most headphones are unbalanced, all balanced phones are compatible to be driven unbalanced. Most portable balanced phones out there are high sensitive IEMs that don't need more power than the phone (or Dragonfly) provides. So you designed a device that won't work for most people and is redundant for most. Yes, you are kidding us...

      Delete
    4. I'm a normal engineer, I solve the problems. When I needed tiny, precise and powerful balanced DAC for my Hifiman HE400i, I didn't found anything on the market, only overpriced junk full of marketing. I did solve my problem during 4-5 weeks. If I'll need the same sized DAC(10gram), the same power(>500mWx2ch) and same other specs(SNR -123dbA, THD .0002%@32ohm -3dbfs) but with SE 3.5mm jack output, I'll solve that task as well. I sent PCB for sample production yesterday, and point that in tweeter. Let's see how quick I'll finish this design.
      E1DA

      Delete
    5. Good luck Ivan with the design! Sounds like fun.

      Delete
    6. Hi Archimago! Let me show you a normal engineering design stile ;) #9038D (Dual rail) single-ended 3.5mm proto is working. So far idle current not so nice, and MAX power noticeable less than #9038S (Single rail, 2.5mm jack TRRS), next week it will be improved. https://www.youtube.com/watch?v=lCcQHCwjH7k
      E1DA.

      Delete
  5. FYI, this might be of interest.... Since he was troubleshooting a noise problem for me, I asked Dragonfly designer Gordon Rankin for his take on this review. He replied, "Yes I read the review... really these tests are not made with real test equipment or the knowledge of how to use them. If you look at the current measurements in the final picture he is using a $9 amazon USB tester. Does that really qualify as a true test? I don't think so!"

    ReplyDelete
    Replies
    1. s this the same Gordon Rankin of Wavelength Audio?

      LOL. Really.
      So the test is invalid because the USB cable is a generic, not an AQ model?
      Riiiight.

      This price race is a problem. If you look at, say, companies like AQ that put out headphone amp/dacs 10 times as expensive as Apple and about 1/10th as well designed.
      The relativism that rules in a market as utterly democratic as the high end allows situations like this to exist without anyone inside that market asking questions.

      AQ can put out a product that offers the newest buzzword compliant features (2018 was MQA and HD Audio) doesn't even give flat response in the audible range, let alone out to 48Khz (or real, or perceived benefit)

      For under $100 look and see what Google or Amazon can do with their Chromecast Ultra or Amazon Firestick. Audio AND video.

      All the superstar engineers left the audio scene back in the 70s/80s... now working on encoding/encryption/5G...

      Delete
    2. So would Mr. Rankin kindly show us his measurements, the ones he made when developing the Cobalt and also takes to confirm that the production examples meet the spec? It would be good to see that it is, indeed, worth 50% more than the Red and that Archimago is getting things wrong despite using an ADI-2 Pro FS. After all, the Cobalt tested here might a rogue bad one.

      As for the $9 "tester": this is just to give an approximate figure on power consumption and has no bearing on the audio performance measurements. (BTW I have one of those myself: very handy to see how much USB-powered devices use!)

      Delete
    3. //All the superstar engineers left the audio
      I agree with you, please note that we discuss a trivial device where is ZERO of engineering required. They used 2 ESS chips i.e. chipset, so need only implement datasheet's content without silly mistakes to get promised ESS distortions at the level .000*%. Such "design" is technician work but THD = .015% is 50-100 times worse, and any 3.5mm output in your home will outperform cobalt in an effortless manner. My old $300 chinese Android Nubia Z11 has .0015%(10x times better than cobalt) and SNR -118dbA at 3.5mm output and the same power at 32ohm(20mW or so).
      E1DA

      Delete
    4. Hi Unknown,
      Glad you've had discussions with Mr. Rankin. I purposely left his name out of the review / write-up because as far as I'm concerned, there's no need to bring anyone person into product performance discussions. That to me is the domain of the marketing department... It's always been about the product, not "the man".

      To read in the Audio Advice review linked above that the first "High Note" they listed was "Gordon Rankin design" is rather silly when we all know that the DAC chip itself and the headphone amp/op-amp is from ESS. The major engineering "heavy lifting" at the heart of this device should be credited to them.

      As AudioSceptic noted above, the $9.00 tester is simply for USB current draw. This did show that the Red improved on the v1.2 and Black as advertised. But did not show the Cobalt was any better. So obviously if Mr. Rankin has other data to show, I'd love to see it posted somewhere. The retail price of the ADC used for measurements is around $2000 not including the computer system or other software. Not that one needs to always correlate $$$ of course... (Which is a major problem with how the "high end" seems to "measure" sound quality.)

      The bottom line remains that this is a poor measuring DAC and in my subjective opinion is also poor sounding. I do not believe a $300 Dragonfly Cobalt DAC is showing us how good the ESS DAC/headphone driver can truly be qualitatively when implemented well. If Mr. Rankin has other metrics by which he is gauging performance by, then perhaps he'd be good enough to share this.

      Delete
  6. Hi Archimago,

    Thanh from Australia here.

    Of course after reading all your tests with 24 bit audio and 96Khz audio it reaffirmed my own experences- I don't bother with DACs and amps anymore.
    I mean sure, since the early days of digital audio recording or workstation/computer based digital audio, things have improved a lot! But the last decade of 24/32 bit audio have a approached a limit. I've owned DACs long term from innovative engineering companies like Slim Devices, Centrance, Apogee, Benchmark, and they are all converging towards what the ASIC designers at AKM or ESS can do. Even if we could get a true 24bit SNR or extended response out to 80KHz, should we be striving for it?
    (https://people.xiph.org/~xiphmont/demo/neil-young.html)

    If I want a real epiphany, I get the wax cleaned/extracted from my ears and try a new set of headphones/foam pads.

    I'll be in Vancouver/BC in the next 12 months on a working holiday- are there any Canadian/home ground equipment that you would recommend a listen into? Are there any audio clubs in the region? Looking to check out the best of what this lovely city has to offer.

    sincerely,
    Thanh

    ReplyDelete
    Replies
    1. Hey Thanh,
      Yup, can't disagree :-).

      Technology in the audio space has reached a level with many devices that for "human consumption", we can be assured that problems like noise and distortion are very much "solved" issues. And as I expressed in my "Summer Musings" post last month, at some point we all have to appreciate that there's no further need to be too impressed by even better jitter reduction, or lower noise, or even lower 0.0001 THD figures. Regardless of which "guru" said what :-).

      Sadly, no audiophile society/clubs here in Vancouver (that I'm aware of). But the city is beautiful especially on a sunny day with the water and mountains! You'll see that we have quite a few overcast, rainy days here too like in Seattle unfortunately in the autumn and winter.

      Make sure to enjoy some of the audiophile stores in town:
      Commercial Electronics, HiFi Centre, Headphone Bar, Liquid Audio, The Sound Room (recently moved to N. Vancouver) off the top of my head. Also if you like vinyl check out Red Cat, Neptoon, and Zulu Records which are the ones I'm most familiar with.

      As for "home grown" equipment, you should be able to find a selection of products from Canadian companies around the stores - Anthem, Paradigm, Classé, Bryston, PSB, Totem, Resonessence among of course the usual international brands at these stores. Remember that one of the branches of ESS Tech's R&D is in Kelowna (here in British Columbia).

      Enjoy your time here!

      Delete
    2. Oh yeah, almost forgot... You might run into some EMM Labs (Ed Meitner) products are also Canadian - based in Calgary.

      Delete
  7. wonder how the Cobalt would compare with the Apple Lightning or USB-C DAC/headphone adapter playing Red Book, blind and level-matched of course. A price ratio of 33:1!

    Going the other way, can this really be worth 1/3 the cost of an RME ADI-2 DAC, ignoring portability, of course?

    ReplyDelete
    Replies
    1. Hi Sceptic,
      As we saw with the Lightning headphone adaptor:
      http://archimago.blogspot.com/2018/01/measurements-apples-lightning-to-35mm.html

      Although the ADC used was not as good as the RME so we can't 1:1 compare with the numbers here, measurements would still show significantly lower THD and IMD nonetheless. It's not as powerful with 1Vrms output.

      Yes, a blind test with 16/44 music would be very interesting between the Cobalt and that device using a pair of sensitive headphones.

      No way the Cobalt would be worth 1/3 of the RME ADI-2 DAC for value IMO.

      Delete
    2. I wanted to mention this too. The Apple lightning dongle would be a better reference than that old trusty iPhone 6. The iPhone 6 is no longer supported with iOS updates, but the dongle is still the most convenient way to connect IEM's to any new iPhone.

      Delete
  8. Hi Archimago-san,

    DragonFly Black (2017) shares one 24.000MHz clock for USB clock and DAC clock. 48kHz clock is exact and no problem. But 44.1kHz clock is generated from clock division circuit of 1/544 (?) resulting 44.1kHz becomes 44.117kHz. because of this, sound pitch becomes slightly higher at 44.1kHz than 48kHz (it is measurable but not noticeable at least for me). Is this happens with newer Dragonflies?

    ReplyDelete
    Replies
    1. Thanks for that note Yamamoto,
      Good question, actually I would double check for you if I still have the Dragonflies with me :-). Alas the Cobalt has been returned at this point. Maybe the next time I run in to my friend with the Red, I can take it for a quick look.

      Maybe others on here can check for us...


      Delete
    2. The clocking in the Cobalt works the same way, only with a 48 MHz crystal. It's apparently a cheap crystal too. Mine is slow by about 86 ppm.

      Delete
    3. How long are your j-test recordings? Would be interesting to check the accuracy of the clock on the one you tested.

      Delete
    4. Hi Mans, alas I don't have a copy of the J-Test recording. However I do have a few seconds of the distorted 1kHz 0dBFS output @44.1kHz (recorded at 768kHz) that I'll send you over at AS.

      Have a look at that and see if there's correlation with what you've found...

      Delete
  9. Any DAC designer today would be a damn fool not to send you (or Amir at ASR) a sample to review (along with a small suitcase of cash in gratitude :-) I guess they'd rather spend that money on marketing rather than ensuring the product meets their engineering goals. As others here have stated, the industry has moved to selling the same crap in fancier packaging.

    ReplyDelete
    Replies
    1. Haha, I did send and today got the review ;) Seems my DAC has THD over 100x times lower vs cobalt )) https://www.audiosciencereview.com/forum/index.php?threads/review-and-measurements-of-e1da-9038s-bal-portable-dac-amp.8424/
      Ivan aka E1DA co.ltd.

      Delete
    2. Very nice Ivan. Very nice. Lemme know when it's out in the wild :-).

      Delete
    3. @Chron: LOL.

      I really wonder what AudioQuest was thinking releasing the Cobalt like this. Whether it's me, or Mans, or Amir, or many others who care about objective performance, surely at some point comparisons will be made... And it would not be pretty.

      One would have thought that the fact the Red was an improvement over the Black objectively, they would have at least tried to do better with the extra $100 and in their advertising appealed to both subjective and objective-leaning folks. Now we're just left with the uncomfortable question "What were they thinking?", "How in the world was this thing 'voiced'?!", and most important for them as a business "Who's supposed to be buying this?!"

      Delete
    4. IMO AudioQuest is essentially a snake oil operation which just happened on something quite good with the original Dragonfly, and then realised there was more money to be made with a genuinely better version. After that they reverted to type, making something that looks much better and costs much more, designed "by ear" to appeal to the subjectivist crowd but lacking proper engineering, which we know is not required in that market. Should we be surprised?

      Delete
    5. You're being far too generous calling it "quite good."

      Delete
    6. Point taken, but wasn't it quite a new idea, and actually worked at a basic level? Hugely overrated by the subjectivists, of course, but no worse than the beyond-belief reviews of the Cobalt by the usual suspects now.

      Delete
    7. At $100, the Black is already hugely overpriced and underperforming. The Red and Cobalt don't do any better, and in some ways they are actually worse.

      Delete
    8. "At $100, the Black is already hugely overpriced and underperforming. The Red and Cobalt don't do any better, and in some ways they are actually worse"
      Comments of this nature are neither accurate nor of any use to a prospective buyer.
      Subjective opinion is all very well but when it descends to the level of spiteful drivel, it should be ignored.

      Delete
    9. Are you Gordon? If you don't want criticism, build a better product.

      Delete
    10. Fascetious? Much? No. I am not Gordon; I am Graham Luke. I also own a Black and a Red and have thoroughly enjoyed the considerable improvement that they have both brought to my audio experience.
      I have compared them to the Chord Mojo and various ifi productsand have found they stand up remarkably well.
      Thank you for your response.

      Delete
  10. Just wanted to give a thumbs up to the Alphaville reissue! I didn't know this existed... Apple Music completely fails recommending me new albums from bands I already have in my library.
    This album was one of my first audiophile experiences, I listened to it on a noname chinese walkman with cheap on-air headphones. Loved it!!! :)

    ReplyDelete
  11. My turn: https://audiophilestyle.com/ca/reviews/objective-audioquest-dragonfly-cobalt-review-r832/

    ReplyDelete
    Replies
    1. Nice Mans :-).

      I echo the calls for you to test every piece of hardware AS reviews!

      Delete