Smartphone Audio Quality Testing
by Chris Heinonen on December 8, 2013 5:15 PM EST- Posted in
- Smartphones
- Audio
- Mobile
- Tablets
- Testing
In many of the examples you have seen so far, you notice that the Nexus 5 has a large issue with the left channel at peak volume levels. As Brian mentions in his Nexus 5 review, it is based on a similar platform to the LG G2 but it isn't identical. Because there are similarities I want to test it out and see if it has the same issue that I see on the Nexus 5.
The test that is causing the large issue on the Nexus 5 is a 1 kHz sine wave, at -0dBFS, at maximum volume. This is the loudest sound that any device will be asked to produce. If you're familiar with the trends in music mixing the past two decades you'll know that a peak of -0dBFS is not all that uncommon now. This chart at NPR shows the average and peak levels for the most popular songs over the past thirty years. Two decades ago testing for -0dBFS might not have been important but it is now. So lets look at this image from the Nexus 5 again.
Now for comparison, we will look at the LG G2.
This looks much better. However the LG G2 is still putting out 0.546528% THD+N into the left channel while only outputting 0.003338% into the right channel. So there is still some imbalance going on here. So why is the issue so much less on the G2 than on the Nexus 5?
The key to this is looking at the scale on the graphs here. While the Nexus 5 peaks are up close to 1.3-1.4V, the G2 has peaks that don't even reach 700mV. Looking at the actual numbers the G2 has a Vrms level of 475.3 mVrms while the Nexus 5 checks in at 843.6 mVrms for the left channel and 982 mVrms for the right channel. The G2 is placing far less stress on its headphone amplifier and keeping it from the output levels that cause this excessive clipping in the Nexus 5.
To look in more detail, we have THD+N Ratio charts for the stepped level sweep that we looked at earlier. First, lets look at the Nexus 5.
We see that the first three volume levels, 15-13, have THD+N distortion over 0.3% for the left ear, while they are below 0.01% for the right ear. From level 12 and below the THD+N levels are practically equal. Now to see how this data on the G2 looks.
We see the first volume step has 0.55% THD+N or so for the left ear, but the right ear is down at a similar level to level 14 on the Nexus 5. The next step drops it to 0.03% which is way, way below where it is on the Nexus 5 at that point. By step 13 they are equal.
The conclusion I pull from this is that both the G2 and the Nexus 5 have the exact same flaw right now. However, the G2 has attempted to hide it by reducing the maximum output level of their headphone amplifier. The Nexus 5 can play louder, but only with far more distortion. Given this I would expect there to be an update to the Nexus 5 at some point that lowers the maximum headphone level to something closer to the G2.
However this doesn't mean that the Nexus 5 is certainly worse to use with headphones. The top 3 settings are ones I would avoid due to the left channel issue, but I might avoid the top 1-2 settings on the G2 as well. If we consider 1% THD+N to be the maximum allowable level, that leaves 8 volume steps on the Nexus 5 that are usable. The G2 has 9 steps that are available to you, and 10 if you consider 0.03% THD+N in one ear to be OK (it probably is).
In the end, the G2 won't play as loud as the Nexus 5 will, but you don't want to play that loud anyway. It has more usable volume steps than the Nexus 5, and otherwise very similar numbers. I'll be interested to see if either of them make further changes to their maximum output levels to remove this issue.
Facebook
Twitter
RSS
188 Comments
View All Comments
DanNeely - Sunday, December 8, 2013 - link
I'm curious how well your Grado headphone's are holding up? I bought a pair of SR80's for use at work last winter; but the wire started to develop damage a month or two ago. If I move the wrong way while wearing them I can get brief bursts of static in one ear, and can mute that ear by pinching the cable just above the Y. I suspect the damage was caused by the post in the headband allowing the earcups to spin freely, combined with the unmarked cable making it hard to notice anything less than a half dozen or so revolutions of twist. I'm wondering how much of this is bad luck on my part vs poor design/manufacturing.Marovincian - Monday, December 9, 2013 - link
I had similiar concens with my Grados (actually mine are Allesandro ms1i). I sent grado an email suggesting that they put a stripe on the "Y" wires so that you could more easily straighten them out. They said that they would pass it along to their design team. Then they sent me a free T-Shirt. Classy company for sure.ManuLM - Sunday, December 8, 2013 - link
Quite a good initiative thanks, it is too hard to get these numbers nowadays.I would suggest you guys build up a database over time of phones performance (see headphoneinfo awesome job for instance).
I also suggest that you add to your test the maximum output delivered (power or voltage swing into load). This is interesting, because if a phone clips at high volumes, but its output power is 10dB above the others in average, then the normal user will simply not see the drawback (altough I admit this is initially poor job from the company in tuning the audio system).
It also helps to chase the brands which deliver lower output power, that can turn to a problem on more demanding headphones (high impedance requiring higher voltage swing). Some users will fancy some extra power on their headphone output (even if this might not be safe for their ears).
Last point, some high-end IEMs have quite low impedance, that demand fairly high current specially in the high energy low frequency, creating bass roll off. A simple frequency response check on a low impedance IEM would show this.
RandomUsername3245 - Sunday, December 8, 2013 - link
I like the idea of audio testing, but I am disappointed by the methods used in this article: why would you bother testing a device at maximum volume when you know it is clipping badly? You should reduce the volume to a setting where it does not clip and then continue the review. You can then report the maximum useable volume setting on the device.The maximum volume on an iPhone is reported to be in excess of 100 dB. Listening at this volume for even a short period (15 minutes) on a consistent basis will permanently damage your hearing. Why not test these devices at reasonable volume levels?
http://www.theatlantic.com/health/archive/2011/12/...
(hopefully not too flawed analogy follows...)
If you are comparing two overclocked computers for maximum performance, you set them to their highest stable clock rate and then benchmark. You do not set one to a clock rate that causes continual crashing, and then report that it failed several of the benchmarks. I think this is comparable to audio review for the clipping cellphones. You might argue that the device should support any user-accessible volume level, but historically it is very common for audio amplifiers to allow users to adjust the gain until the output clips. Apple is an unusual case that limits the user to only access non-clipping gain settings.
ManuLM - Sunday, December 8, 2013 - link
audio systems are tested at max performance (there are many reasons for that, including the fact that when you sell something, all usage range of the system should be good), so analogy with OC is not ideal.I agree with you though that testing at nominal volume could help, as an adder only of max volume testing
eio - Monday, December 9, 2013 - link
yes, power of drive is a good factor in a benchmark. but performances at different loads should not be compared directly.eio - Monday, December 9, 2013 - link
a ideal test may have several series of performance graphs with several steps of incrementing loads...RandomUsername3245 - Wednesday, December 11, 2013 - link
Late reply...Like I said in my previous comment, it is common for audio amplifiers to allow you to adjust the gain past where the amplifier will start to clip. You should never expect a car stereo or home theater amplifier to allow you to run at maximum gain without clipping, so why should you expect a phone's headphone amplifier to behave differently?
The proper way to run this test is to adjust the amplifier to maximum non-clipped gain and then run the test.
willis936 - Sunday, December 8, 2013 - link
The day has finally arrived? Good data with some surprising results. I think I'm mostly surprised at how well all of the devices perform. I think dynamic range is perhaps the most important test here simply because most people won't be listening at max volume on headphones and pushing the noise floor down as low as possible is important for quiet listening.Were these tests done on the AKG K701? That is well known as a difficult to drive pair of cans without an amp. If a phone can drive those loudly with good measurements then it's certainly good enough for anything I'd use it for. Testing should be done worst case and if there's time more typical cases. When using my phone as a line out I'll typically leave it 3 steps below max because I expected there to be output stage power issues (seen as dramatic clipping on LG's stuff :x) on my phone. Any lower and as you noted the static noise floor lowers the SNR.
I was a little surprised at the weak channel separation in the otherwise amazing iphone. Channel separation is already a p big issue. Even with expensive headphones it's easy to test and ballpark a crosstalk of worse than -60dB by ear just from the jack to the drivers.
I'd like to make a request for some data of testing devices (1 iphone and 1 iconic android per year?) going backwards to see a progression (or maybe lack thereof) of audio quality in smartphones over the past 4 or 5 years.
willis936 - Sunday, December 8, 2013 - link
Oh, and thanks for the excellent write up and all of your hard work! I'm looking forward to future data.