Friday, 5 February 2016

MEASUREMENTS: Apple iPad Air 2 - Audio Output Quality

Although for the most part, I am a Windows and Android guy, I do have a little collection of Apple devices here at home. Despite taking a look at the MacBook's CoreAudio upsampling, measured the iPhone 6 when it came out, and explored various MacBook laptops over the years, I realized that I had missed out on one of the most important "line" of products from Apple - the iPad.

Last year, I needed to run an app for work that was only available on the iPad so I got one of these Air 2's. As a reminder, the iPad Air 2 is currently (early 2016) one of the latest versions of the iPad line (there are of course the parallel Mini and new Pro models). It was announced back in October 2014 and is considered the "sixth generation" iPad tablet. The model I have here is the 64GB gold-color version, Wi-Fi only, with newest iOS 9.2.1 installed. The screen is gorgeous with the "Retina" resolution of 2048x1536, nice and bright with great contrast. CPU is the Apple A8X (1.5GHz tri-core), 2GB DDR3 RAM internally. Physically, nice and thin with the standard 9.7" screen - great for surfing and media content consumption.

When it comes to the sound, we know that at least from the iPhone 6 onward, the devices do have the ability to play up to 24/48 from the internal DAC. (Of course, we can pass high resolution data including DSD through the USB digital interface with Onkyo HF Player.) I wasn't sure though whether these newer iPad's had this capability or if 96kHz was possible. Remember that other manufacturers, namely Samsung devices are already capable of 24/192 (as measured with the Galaxy Note 5 phone).

Let us then put the iPad Air 2 headphone output through some objective measurements and see what comes out!

Remember... To test the potential for high-resolution output from the device, we cannot rely on iTunes. Like I did with the iPhones before, all playback will be with the latest Onkyo HF Player 2.2.0 ($10 to activate the HD Playback Pack). No EQ or other DSP processing turned on of course.

Part 1: Digital Oscilloscope, Impulse Response, Digital Filter Composite

With a 1kHz signal and 20-ohm load, output impedance was measured/calculated at 2.2-ohms - very reasonable. Should be fine for the majority of high-quality headphones. Very low impedance headphones especially the little IEMs might see frequency variance and poor electrical damping though.

Here's the usual 1kHz 0dBFS square wave as measured with the digital oscilloscope:
We're looking at peak voltage of 1.40V or Vrms 1.0V. As you can see, it's quite clean, stable (less rippling in the peaks than the iPhone 4 or 6) with excellent channel balance.
How about the impulse response?
Minimum phase as we had seen already with other Apple "mobile" devices at least dating back to the iPhone 4. It looks like Apple likes to use these minimal phase settings for their equipment. If you compare this to the iPhones 4 and 6, the post-ringing is less prolonged suggesting that it's not as sharp of a filter as those others. Of interest I think is that Apple seems to prefer the use of linear phase filters with the MacBook laptops.

Over the last 6 months, I've been using the "Digital Filter Composite" (based on the Reis test) to have a look at how the digital filtering is being implemented with a 44kHz signal using a wideband FFT:

Not ideal behaviour, but not bad. You can compare it to something like the Samsung Note 5 and PonoPlayer. As expected, it suppresses aliasing much better than the PonoPlayer. The aliasing images are not as well suppressed as the Note 5 however, but the Apple does not "overload" as much with a complex noise signal with peaks at 0dBFS - the noise floor gets a little uneven but does approximate the noise floor. With the 19 & 20kHz signals, the loudest intermodulation distortion peak is around -70dB below primary signals; should not be an issue (I discounted the slightly higher peak around 70kHz which is unlikely to be reproduced by any transducer). Over the last while, the best digital filter overlay I've seen comes from the LH Labs Geek Out V2 which looks essentially ideal with no evidence of intersample overload in the interpolator.

Part 2: RightMark Tests

As usual, I'm using the E-MU 0404USB as previously documented to maintain consistency.

Summary chart with a few other mobiles and portable DACs.
Good results for the iPad Air 2. No surprise of course since these days any decent DAC / analogue output would have no problems handling a 16-bit signal at 44kHz. As we have seen previously, the AudioQuest Dragonfly 1.2 tends to be the worst of the DAC's I've measured in terms of objective distortion results - sure it still sounds decent, but technically surpassed by about everything else I've looked at.

The other device that stands out is the PonoPlayer. Again, as I discussed previously, this has to do with the weak filtering Ayre seems to favour resulting in increased aliasing distortion (but looks nice with the impulse response plot).

A few other graphs for completeness:
Frequency Response

Noise Floor

IMD+N sweep
We know the iPhone 6 is capable of 24-bit quality and 48kHz sample rate already. This is the case with the iPad Air 2 also:

In fact, the Air 2 and iPhone 6 look to be very close and I would not be surprised if they use the same or closely related DAC internally (I believe it's a Cirrus Logic part). The dynamic range is getting close to 18-bit resolution.

Frequency Response

Noise Floor - notice that the Squeezebox Touch is a little noisier than the others in the low frequencies and has a 60Hz hum; it's the only device not running on batteries here of course :-).

IMD+N Sweep - despite the 60Hz hum in the noise floor, the SB Touch has very low distortion on testing.

Finally, let's have a look at the 96kHz samplerate:

Frequency Response - iPad Air 2 & iPhone 6 does not surpass 20kHz.

Alas, the iPad Air 2 is just like the iPhone - no support for higher sampling rate than 48kHz (remember, I'm using the Onkyo HF Player which supports higher sample rates for USB DACs). By the way, in the numeric summary chart above, notice the objective improvement with a high quality desktop DAC like the TEAC UD-501 USB (results even better if using balanced cables).

Other graphs:
Noise Floor

IMD+N Sweep


Part 3: Jitter

No issues these days with the Dunn J-Test for mobile device playback.

Part 4: Subjective Listening

First, the built-in tiny stereo speakers located in the lower edge (below the home button) do "work"; they are there for convenience. As expected, weak bass and distorts slightly at 100% volume but for what it is, it's fine. Enough said.

As usual, I listened to a few "classics" like some Beatles, Tracy Chapman, Muddy Waters, AC/DC, Alice Cooper through some headphones as ALACs over iTunes - switching between Sennheiser HD800 and AKG Q701's. What can I say, it sounds tight and accurate. Enough power to adequately drive the Q701's for everyday use at a polite volume level only.

More aggressive music like The Prodigy's "Smack My Bitch Up" (off Fat of the Land 1997) was rendered well with plenty of detail, excellent bass definition and good dynamics (at least for music of this variety). The layering of synthetic sonics was well defined without sounding muddy. Slight (likely intentional) imperfections in the looping of samples easily audible and (again likely intentionally) jarring.

I could say the same thing as well for Tyler Bates' "To Victory" off the 300 Soundtrack, in my experience another track that easily muddies up with inferior DACs and headphones/speakers. Done right, it should sound scary :-O.

With the passing of Glenn Frey a few weeks back, I decided to have a listen to one of his solo albums I hadn't visited in awhile - Soul Searchin' (original 1988 release). It's mostly light pop fare from the late 80's before the loudness wars; plenty of that 80's sax sound in a number of tracks, a bit of a horn section here and there, and the steel drum tinged "Can't Put Out This Fire". For me it's an example of the subjective soundscape of life growing up. It sounded great on the iPad as expected and brought back great memories of those times - the magic of music and the beauty of what it is capable of doing... Having said this, the iPad was clearly capable of demonstrating the limitations of a recording from that era - some of the "synthetic" sounding drums and studio tonality leaning more on the "cool" side was evident.

Part 5: Conclusions

I think the big question for me is "Why won't Apple allow 96kHz sample rate?" among other questions for Apple including "It's 2016, where's my FLAC support?", or "Why do I need to use iTunes?", but I digress. Considering that essentially every decent DAC these days can do 96kHz, I'm a bit perplexed as to why Apple won't allow this to happen on their portable devices. There's no need to go crazy of course with big numbers (for example, I'm not a fan of 192kHz since I don't think it's necessary), but 96kHz I think would be desirable and I wouldn't be surprised if all that needs to happen is just a firmware/iOS upgrade (anyone actually know if the Cirrus Logic 338S1213 DAC chip is capable of >48kHz!?).

Be that as it may, in considering my previous post about MQA, I think  these Apple products are a good example of why streaming "hi-res audio" as lossless 24/48 makes sense. Not only would 24/48 be compatible universally without going through any proprietary decoding process, but it would also reach the hundreds of millions of users consuming audio under the Apple ecosystem (remember, they have been selling from 30-70 million iPhones alone per quarter worldwide at least since 2013)! For now, until Apple formally does something about the samplerate, 24/48 is as "high res" as it gets coming out of those analogue outputs.

Bottom line - the iPad Air 2's analogue output measures quite well and sounds excellent. The low noise floor implies >17.5-bits of resolution. The measured 2.2-ohm output impedance is fine for the majority of efficient headphones. It looks like Apple has been improving the output impedance of the iPad over the years. Ken Rockwell did similar tests with the iPad (1st gen) and got 5-ohms, while iPad 2 was 4.4-ohms. Playback samplerate is still limited to 48kHz without resampling.


Okay, to end off...

As much as some of my posts/comments may have ruffled a few feathers or as much as I have on occasion participated in "heated" arguments over audio hardware, I believe debates usually over measurements and other "more objective" topics are fair and can hopefully lead to practical insights and understanding. However, one's preferences in music, experiences with it, and emotional reactions have always been matters not really worthy of much debate for me except among friends where I can contextualize opinions.

This is of course because music itself is almost totally subjective (maybe there are some works of art either "objectively" so good to be universally lauded, or so bad as to be universally without merit). I personally would be reticent about making gross generalizations when expressing opinions in purely subjective matters.

That is not to say there is no value in artistic criticism, just that it's important to never lose sight of what is pure subjectivity and hence opinion versus that with more secure objective foundations. For me, when something is highly subjective, it's important to consider the appropriateness of expressing opinions in present company, at appropriate times, and in a reasonable venue. Considering the recent death of Glenn Frey, one of the founding members of the Eagles, I found this post by Robert Baird on the Stereophile website in terribly bad taste. Sure, he makes some good points here and there. Nothing wrong with exercising free speech, but is this really the time to be so opinionated? Let us at least demonstrate some charity of spirit; rather than drawing attention to oneself. I bet for many readers, it's just an example of "software" audio snobbery (as if high priced luxury hardware audio snobbery isn't bad enough).

BTW, although I'm not a massive fan of The Eagles, I have all their albums on digital and many of the LP's including a first pressing of Hotel California. I still think that the DTS multichannel Hell Freezes Over is one of the best live "surround" recordings I have heard - in fact that was one of my first DVD's purchased when the format came out in the late 90's. And growing up in the 70's, I still enjoy the music even after the years of radio play... Heck I even went to see the "boys" live in 2010 when they came by Vancouver on their Long Road Out Of Eden tour (a blast!). RIP Mr. Frey.

Enjoy the music everyone... Yes, even pop music - absolutely!


  1. Hi Archimago,
    You have neasured quite a few portable devices over the years, but I never saw any mesurement ir comment regarding bluetooth, both aptx and standars sbc.
    Any chance of this happening in the near future?

    1. Right. The reason is that we're unfortunately looking at lossy codecs when streaming with Bluetooth and as a result I haven't been as interested since I see them as incapable of high-fidelity. Some of the more complex test signals will not even work because the lossy codec will just filter them out.

      Having said this, I will tell you that next week I'll post on the Google Chromecast Audio since that device is capable of good fidelity :-).

    2. Archimago, you will be surprised to find how well the lossy codecs score with RMAA. Simply encode original wav file to lossy compression and then decode it back to wav. Once done perform the analysis with RMAA. This will tell you how much encoding/decoding in digital domain affects the overall performance.

    3. Yeah. Lossy codecs can certainly be looked at with something like RIAA. However, the problem is that synthetic test signals like these are usually not too difficult to encode simply because usually they're only simple test tones. Unlike real music where there's noise, all kinds of harmonics, etc. which needs to be maintained in order for it to sound good. Part of assessing the quality of a lossy codec has to do with understanding the quality of the underlying psychoacoustic model which I don't have a way to quantify. (I suppose using the digital null test would be a way.)

      Another problem is that the best test would be one where I can play a decoded digital output through a standard DAC. I see for example the Yamaha YBA-11 can be had for around $50-60 these days and has aptX. Maybe something like that could be worth testing for fidelity.

      In any case, I will look around and see if I can find an aptX device around here to try.

    4. I stream from HTC One M7 to Arcam miniBlink connected to Nubert nuPro A-200 active speakers, and with flac files I can not hear a difference compared to laptop to nuPro via usb with foobar asio playback.
      I would very much like to read about your experience with aptx in near future - after all, maybe my hearing is not perfectly accurate (I am 46).

  2. Measuring "good" lossy codecs

    Just as a side note, that it is not so "easy" to measure and evaluate the quality of a good lossy codec. I was involved for some years in the Fraunhofer working group and can report, that the behavior of the codec does change over level, over inter-channel phase relation and over time.

    A good start is to use special codec measurement files, for example the Audio Precision codec file, that is a file, that contains 21 frequencies, logarithmic equally space from 20 Hz to 20 kHz but with intentionally missing the 5 frequencies between 1k2 and 4k8 (where the ear is most sensitive).

    Measure the behavior at 0 dBFS, at - 20 dBFS, (at - 40 dBFS) and at - 60 dBFS. This with left and right In-Phase (the Mono = Mid Channel) and then with left to right Invert-Phase (only one channel has to be inverted (to each other)) to measure the Side Channel behavior (also at 3 or 4 different levels).

    And in the next step would be to measure some impuls behavior. Planing to use RMAA is not working here. Good luck.


    1. Thanks Juergen. That's pretty well what I expected. Not easy!

      I guess maybe a digital difference test might help a little bit to explore the amount of difference there is between the original and the lossy encoded version... But this might not correspond with the actual perceived quality nor allow a clear correlation between different CODECs. (Might be interesting nonetheless!)

      Don't know if I've ever heard aptX over the years but from what I see, aptX should be quite good... It does allow a 350kbps stream which is decent but I don't think it uses psychoacoustic modeling so presumably the decoding requires less processing power. I see John Atkinson made some measurements using an Arcam here:

    2. Archimago

      A digital difference test with lossy Codecs does tell you, what you are missing in the lossy file, but not, if this loss is audible or not, because on the difference file, the psychoacoustic hiding effect does not work the same way, as if you hear the "full" lossy file. I hope it is clear what I mean. But the difference test is good to train the ear, what defects you could expect.

      But judging the measurements in the FFT graph of the Codec signal measurements is not so easy and needs also a lot of listening to different codecs and you have to look always to the FFT graphs, with psychoacoustic "eyes". For example "v" shaped noise between gaps do sound "better" than rapidly digital black between gaps.

      This will be clear, when you put time measurements into account. Rapidly digital black between gaps lead to audible time smear, whereas psychoacoustic "v" shaped noise between the gaps, can have a much better time behavior.

      And this all changes with Level and also (depending on the codec) with inter-channel phase relation (Mid / Side). Good luck, but be prepared, this is not an easy task.


  3. Juergen,
    When you say "digital black between the gaps", or "V shaped noise between the gaps", are you referring to what is seen on the FFT of the Audio Precision test signal you described earlier?

  4. This comment has been removed by the author.

  5. Hello, can you test this way Lumia (735,830 or other higher models?) I would like to know if it supports 24 bit audio ....

    1. Hi Honza. A lot of this depends on whether I have access to the device - remember I do this out of curiosity :-). Having said this, I do know a friend who was into the Lumia phones so I'll ask if he still has it...

    2. Thank you. I cannot with my equipment tell whether it supports 24 bit, which means that a (small) part of my FLACs which is in 24 bit has to be resampled and dithered to use on lumia ....

  6. V Shape

    Yes. On the first sight, the V-shaped noise between the gaps looks noisier (yes they are noisier, but not audible) than the "digital black" between the gaps (sure, these black gaps are not audible at all (in the frequency domain). But when you look at the time domain, then these steady state digital black do have a lot of negative artifacts in the time domain, that the "v-shaped" gaps do not suffer.

    This explanation is a bit too simplified, and is valid only as an answer to the above written question.

    As I said, to you to look at the complete picture, to be able to know, what is going on.


  7. Differences of Lossy Codecs

    You just have to compare the above mentioned Codec Measurement File, when encoded into MP3, AAC and OggVorbis, all with same data rate (or maximum data rate) and then you clearly see the differences, that then also can be heard. You see that MP3 is a bit older and that AAC was created afterwards and that OggVorbis is the lasted "trick" of those three.


  8. Hey Archimago, where is the post about "audiophilia nervosa"?


  9. I enjoy reading G your thoughts very much, looking forward to your look at the Chromecast audio

  10. If you want to do the Julian Hirsch thing, you need to, in the beginning say how you are testing and what the hardware/software you are testing with, what is IOS is, version, blah blah blah. Keep up the good work.

  11. iPad Air 2 is best stylize and very quality full smartphone i liked ipad air 2 and i used ipad air 2 it was very fast and awesome smartphone.
    Buy iPad Air 2 in Dubai