Headphone Output - Objective Measurement Data and Audio Fidelity

[chdloc]

Introduction
I created this thread to post and discuss measurements and data related to audio fidelity of the headphone output. Note that audio fidelity does not automatically correlate with audio quality. Audio quality is a very subjective measure. Some people like their audio to be very bass heavy while others prefer accentuated mids and/or highs. Yet others (like myself) enjoy a very "neutral" sound signature. Audio fidelity -- at least the way I define it -- on the other hand can be measured: Audio data is fed to the DAC/amp and measured at the output. For perfect fidelity, the recorded output is identical to the input. Of course, this ideal cannot be achieved. Deviations from the ideal can be measured, documented, and discussed. Particularly, not all deviations from the ideal are actually audible. I predict some heated discussions on audibility in this thread going forward...


Test Equipment
The device under test (DUT) is a Pixel XL that sports a Qualcomm WCD9335 codec/amp discrete audio chip. The headphone output is split with a Y-wire. One signal path goes into the measurement rig, a Rohde & Schwarz UPV Audio Analyzer. Depending on the measurement, the other end either stays open (not connected or N/C) or connected to my Sennheiser IE800 IEMs. Those IEMs have a virtually frequency-independent impedance of 16 Ohm, a sensitivity of 125 dB/Vrms at 1 kHz, and a ruler-flat phase response. These IEMs are not only one of the best "sounding" headphones I know, but actually very easy to drive as the specs that I've listed suggest.

Test Signals
For the tests conducted so far, I have used two signals that I created with Matlab:

• stereo wave file with full-scale (0 dBFS) 1 kHz sine wave in the left channel and silence (zeros) in the right channel (16 bit, 48 kHz sampling rate)
• stereo wave file with white noise mastered to take advantage of the full dynamic range (16 bit, 48 kHz sampling rate)

These signals were played with the stock Android music player, with all audio effects disabled.

In some cases I swapped the Pixel XL with an iPhone 5s for comparison purposes. All other hardware stayed the same.

Measurement Results
This section summarizes the measurements detailed in the second post.

Output Impedance
The Pixel's output impedance was measured to be 4.8 Ohm. To compare, the output impedance of the iPhone 5s' amp is 2.0 Ohm.

Comments: The Pixel's output impedance violates the often-quoted 1/8 rule in some cases. The rule says that most amps work best with headphones that have an impedance that is at least eight times its own output impedance. According to this rule, headphones should ideally have an impedance of more than 40 Ohm. Nevertheless, the amp seems to work with my 16 Ohm IEMs without any issues (other than the level drop, see below). The reason, I think, is that my IEMs make use of a single dynamic driver rather than multiple balanced armature ones like many other IEMs. Multiple balanced armature drivers are much harder to drive and may cause minor problems (slight modification of the frequency response) with the amp in the Pixel

Frequency Response (see plots in the second post)
The frequency response is ruler flat, whose shape does not change at all once my IEMs are plugged in as a load. The gain dropped by 2.3 dB after plugging in the headphones. The reason for this drop is the interaction of the headphone's impedances with the output impedance of the amp. Higher impedance headphones will see a smaller drop in volume when plugged into the Pixel.

The channel imbalance is a negligible 0.1 dB at 1 kHz.

Comments: As discussed above, the only potential issue I can see here is that the combination of this amp with low-impedance multiple armature-based IEMs may lead to slight modifications in the overall frequency response, maybe on the order of +/- 2 dB max.

Maximum output level
When the volume on the Pixel is turned up all the way, the maximum attainable output voltage with a full-scale 1 kHz sine wave is 0.427 Vrms, without any load attached. I have found that the output volume was capped digitally to -7 dBFS in the mixer by Google. The maximum theoretically attainable output voltage of virtually any DAC/amp combo devices is 1 Vrms. Once a custom recovery has been made available for the Pixel, it should be trivial to boost its maximum output voltage to 1 Vrms.

Comments: No issues here.

Volume level steps with volume rocker
As discussed above, at max volume the Pixel delivers 0.427 Vrms, which is volume step 15.
Lowering the output volume from the maximum setting, 15, with the volume rocker results in the following steps (rounded):
Volume setting: 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
Change in gain: 0dB, -3dB, -3dB,-3dB,-3dB,-3dB,-3dB,-4dB,-4dB,-3dB,-4dB,-4dB,-4dB,-7dB,-7dB,-54dB


Noise Floor
The noise floor was measured to be 115 dB, which gives a theoretical dynamic range of (currently) 108 dB.

Comments: No issues here.


Total Harmonic Distortion + Noise (THD+N, see plots in the second post)
Measurements of THD+N were done on both iPhone 5s and Pixel at their respective maximum volume as well at a volume setting that corresponds to an output voltage of around 20 mV, with which the IE800 would produce around 90 dB SPL:

• Pixel at 20 mV: THD+N is -71.4 dB
• Pixel @ max volume: THD+N is -81.3 dB
• iPhone 5s at 20 mV: THD+N is -78.7 dB
• iPhone 5s @ max volume: THD+N is -84.1 dB

Comments: By comparing the plots in the second post, it can be seen that the Pixel has higher non-linear distortions than the iPhone 5s, especially at low output voltages. The reason for this is that Google and I believe any other Android manufacturer opts to maximize analog gain while controlling overall level with digital gain only. This is not ideal. It would be better to hit the codec with the highest digital gain possible (but no more than 0 dB), keep the analog gain low initially and control overall loudness with the analog gain only. This way the full dynamic range of the DAC can be utilized. This is the paradigm that virtually any home/car stereo systems utilizes. Based on the plots below, I'm guessing that Apple is doing it, too.


Intermodulation Distortion
Preliminary inspection of SMPTE IMD measurements suggest no problems either unloaded or loaded with my IEMs. Measurements to follow sometime next week.


Stereo Crosstalk
Stereo crosstalk very much depends on the interaction of amp, headphone, and analog circuit design of the phone. Here, at least on the surface, the Pixel does not seem to be showing strong numbers. With my headphones attached and throughout the volume range offered by the Pixel, I measured the stereo crosstalk to be 45 dB. I compared this number with the iPhone 5s. Interestingly, stereo crosstalk on the iPhone shows a strong dependence on the playback level. Again with the IEMs attached the crosstalk ranged from 43 dB at the lower volume settings up to 60 dB at the higher end. I repeated the measurements with full-size cans, the Sennheiser HD 540 (300 Ohm). With those headphones attached, the stereo crosstalk of both the iPhone 5s and the Pixel are north of 80 dB.

Comments: To put things into perspective, though, at the playback levels that in the long run do not cause permanent hearing loss (less than 90 dB SPL at the ear drum) with the IEMs, both iPhone 5s and Pixel have very similar stereo crosstalk performance, i.e. around 45 dB. Carefully controlled double-blind tests should be considered to determine at what point crosstalk actually becomes audible.

Comments on the Pixel mixer
There is one thing that strikes me as odd after having examined the Pixel mixer (/system/etc/mixer_paths.xml): The amplifier is set up as "CLS_H_LOHIFI" and not as "CLS_H_HIFI", which is the default for the WCD9335 in general and the HTC10 in particular. While I don't know what exactly the differences between those two settings are, I will play around with them as soon as a custom recovery becomes available. I'll also look into enabling hardware IIR filtering via the Qualcomm codec and update my biQuads app.

 

[chdloc]

Supporting plots done with stock mixer can be found here.

Pixel "frequency response" at 20 mV output voltage. Note that the 2.3 dB level drop to to the presence of the IE800 has been compensated for (with this output voltage my IE800 delivers around 90 dB SPL):

Pixel "frequency response" at maximum output voltage (0.427 Vrms). Note that the 2.3 dB level drop to to the presence of the IE800 has been compensated for (with this output voltage my IE800 delivers around 115 dB SPL):

Pixel THD+N at 20 mV output voltage (only the left graph is relevant):

Pixel THD+N at maximum output voltage (only the left graph is relevant):

iPhone 5s THD+N at 20 mV output voltage (only the left graph is relevant):

iPhone 5s THD+N at maximum output voltage (only the left graph is relevant):

 

[chdloc]

reserved (for future measurements based on a modified mixer)

 

[iceman4357]

Damn. Thanks for doing a deep dive on this.

 

[mariano3113]

Greatly appreciate you taking the time to do the measurements.

I'm assuming Normal Pixel would have mirrors measurements as my listening observations are in line with what you have presented.

Although on my Pixel it seems like the Single loud speaker also has the same 'Boomy' bass boost applied.

I have Campfire Jupiters and Shure SE535 with Brown Knowles dampeners and the Shures just pair better with the stock Pixel.

Looking forward to future Audio tweaks on this device.

With both headphones I have been using and brothers Shure SE215: I have noticed that the lowest volume level is still what I would consider listening level. (Where as on Note 7 sound was not audible until about 25% volume from off)

 

[bjrmd]

Thanks for providing such quality measurements and interpretation.

Since the physical structure of the plain Pixel is different, it is possible the crosstalk is not the same. Would be interesting to compare if anyone has one to test.

 

[chdloc]

I'm assuming Normal Pixel would have mirrors measurements as my listening observations are in line with what you have presented.

Since the physical structure of the plain Pixel is different, it is possible the crosstalk is not the same. Would be interesting to compare if anyone has one to test.

I can imagine that the Pixel would measure exactly the same as the Pixel XL. It would make a lot of sense for Google/HTC to use the same board layout as a cost savings measure, even though the two devices have different physical dimensions.

 

[bjrmd]

I can imagine that the Pixel would measure exactly the same as the Pixel XL. It would make a lot of sense for Google/HTC to use the same board layout as a cost savings measure, even though the two devices have different physical dimensions.

What I meant (and it was just a possibility) is illustrated here
If you look at the sony z5 regular vs compact, the crosstalk is 8 dB different. As you have mentioned, it's the grounding circuitry that determines crosstalk (not the dac) and perhaps a different physical layout makes a difference.

 

[chdloc]

What I meant (and it was just a possibility) is illustrated here
If you look at the sony z5 regular vs compact, the crosstalk is 8 dB different. As you have mentioned, it's the grounding circuitry that determines crosstalk (not the dac) and perhaps a different physical layout makes a difference.

Oh, I apologize for being so vague in my response. As reported, Google/HTC spent only 9 months in designing the Pixels. On such a short development timeframe, it would make a lot of sense to clone as much of the hardware design as possible. Provided the circuit board carrying the codec physically fits into the regular Pixel's body (a big "if" as the regular Pixel is 6.2 mm narrower), I'm willing to bet that the two phones are identical from a headphone audio perspective. Otherwise, of course, all bets are off.

 

[Reginalb124]

Introduction
Output Impedance
The Pixel's output impedance was measured to be 4.8 Ohm. To compare, the output impedance of the iPhone 5s' amp is 2.0 Ohm.

Comments: The Pixel's output impedance violates the often-quoted 1/8 rule in some cases. The rule says that most amps work best with headphones that have an impedance that is at least eight times its own output impedance. According to this rule, headphones should ideally have an impedance of more than 40 Ohm. Nevertheless, the amp seems to work with my 16 Ohm IEMs without any issues (other than the level drop, see below). The reason, I think, is that my IEMs make use of a single dynamic driver rather than multiple balanced armature ones like many other IEMs. Multiple balanced armature drivers are much harder to drive and may cause minor problems (slight modification of the frequency response) with the amp in the Pixel

Frequency Response (see plots in the second post)
The frequency response is ruler flat, whose shape does not change at all once my IEMs are plugged in as a load. The gain dropped by 2.3 dB after plugging in the headphones. The reason for this drop is the interaction of the headphone's impedances with the output impedance of the amp. Higher impedance headphones will see a smaller drop in volume when plugged into the Pixel.

Just to be clear, the output impedance is a problem. Here is the frequency response driving Unique Melody Merlins:

More than a minor problem there.

Here it is with a 32‎Ω load.

For comparison, here is a Sony ZX1 with the same two headphones used as load:

With the Merlins, the ZX1 between 30Hz-15KHz is +.5/-1.25dB with the heavy roll starting at around 7KHz, not ideal, but not much music content up there.
The Pixel XL is around +1dB/-1.5dB, but with a ton of rolling in the middle. And that's what you will see with balanced armature drivers.
Also note that the hump at around 80Hz with the PS1000 load is around double what it is on the ZX1, but the PS1000 impedance goes a little nuts right there: http://www.innerfidelity.com/images/GradoPS1000.pdf

And a comparison of the Astell&Kern AK300, Sony ZX1, and Nexus 6 with the Merlins as a load. N6 was actually the best performer of the lot!

 

[chdloc]

Just to be clear, the output impedance is a problem. Here is the frequency response driving Unique Melody Merlins:

Thanks for sharing your measurements!

For the sake of others looking at this, however, I think it is important to point out that the output impedance may be a problem and is by no means guaranteed.
Not all folks using this phone or looking to buying one use such high-demanding IEMs. Remember that my IEMs, the Sennheiser IE800, do not cause any measurable change in the overall frequency response, other than a frequency-independent level drop of 2.3 dB. I'm very satisfied how my IE800 sound with the Pixel.

As your own plots show, your IEMs cause "issues" in all configurations, even with the Astrell&Kern dedicated music player! Note that you are arguing over frequency response swing differences of less than 1 dB. Keep in mind that the amplifier-headphone system is largely linear so you can take the additional small frequency (and likely phase) swings due to the headphone-amp interaction and fold them into the already existing large swings of the headphones themselves (linear system theory).
My point is that those tiny frequency swings, +/- 1 dB, by themselves likely do not cause sufficient changes in the headphone's overall sound signature to be problematic. Unless the artificial bass boost causes the bass to become muddy.

The final word, however, will have to come after carefully conducted double-blind listening tests have been performed. I'm not going to making a lot of friends around here by saying that comparing the audio quality of two phones/devices without carefully matching the levels within a fraction of a dB and performing the tests blindly only result in subjective opinions without much merit.

 

[ChaosMinionX]

My Shure SE535's sure love this phone, sounds a bit better to me than my Nexus 6.

 

[Nitemare3219]

My SE846 with an impedance of 9 ohms sound great, although I do use the built in EQ a bit to create a v-shaped sound signature which I prefer. Maybe I'll do a comparison with my Grace m9XX DAC/amp to see if I notice any major differences with a flat EQ on both.

 

[bjrmd]

The "problem" with low impedance headphones is not necessarily the low impedance as @chdloc has indicated. It's the variability with frequency (because each driver and crossover has different impedance effects) that can mess with the frequency. The two graphs below are from a Note 4 (1.4 ohm) with a Shure se535 (variable but higher impedance) and the se846 (variable low impedance).
Here is the impedance graph of the se846-note the variability:

As can be seen, the frequency shifts are higher with the se846--but if the headphone impedance was the same throughout, it would look flat.
The se846- more exaggerated:

The se535 (much higher impedance but still variable)-not too bad:

I have measured the same curve with my Dragonfly red (.02 ohm) and it is totally flat

I would guess the exaggeration would be more with the Pixel (5 ohm). This may sound "better" to some, but not to others-as noted above.
My plan (after I get my preorder) is to remeasure the frequency response and just design some biquads to "fix it" the way I like it. But, the average user who is not adjusting this may notice a different sound signature.
It is not a deal breaker, but I was a bit disappointed in Google/HTC for not doing a better job. My Note 2 was close to 1 ohm, S6 1.4 ohm and Iphone noted above.

 

[chdloc]

The "problem" with low impedance headphones is not necessarily the low impedance as @chdloc has indicated. It's the variability with frequency (because each driver and crossover has different impedance effects) that can mess with the frequency.

It is actually a combination of a highly variable and low impedance. Higher impedance headphones with varying impedance will have less of an effect as the amplifier's damping factor -- which is a function of its output impedance -- will be able to "deal" with it.
But still, the number of users that have such high-demanding IEMs, like your SE846 or the AKG K3003, is likely very low. That's why I wanted to counter the general statement of doom and gloom expressed earlier in this thread. Having said that, you guys with demanding IEMs may want/need to resort to either utilizing correction filters, buying a different phone with a lower impedance amp, or investing in an external DAC and/or amp.

 

[kalrykh]

Thanks for the effort you put into this. I've got a couple sets of ciems from Noble Audio that I'm itching to try once my XL comes in.

 

[bjrmd]

GSMarena review up-no real surprises and basically agrees with above. Crosstalk -62 dB, not as good as HTC 10, Iphone but better than the LG V20 (special ESS dac).

All in all, with respect to audio a solid performer but certainly not spectacular.

Life is full of compromises.

 

[mariano3113]

This is from my Google Pixel (non XL)

Removed LO from LOHIFI (To my ears it seems to have reduced the 'Boomy’ bass)

About to test Elemental kernel.

Edit:
another XML edit: "RX HPH Mode" value="CLS_H_LP" compared to HTC 10 & Note 7 (Snapdragon) "RX HPH Mode" value="CLD_H_HIFI"

Before:

After:

 

[chdloc]

This is from my Google Pixel (non XL)
Removed LO from LOHIFI (To my ears it seems to have reduced the 'Boomyâ?? bass)

I've made that change, too, but I have a hard time hearing a difference (my headphones don't sound boomy to begin with).
Did you reboot after making the change? Measurements will have to wait until at least Monday.

Did you actually edit your mixer_paths.xml in /system or did you go the system-less route by binding a modified mixer in /su/etc?

BTW, biquad filtering, stereo recording, and a very first shot at dual loudspeaker playback seem to be working (the earpiece is a *lot* quieter than the main speaker, however), all done via mixer_paths.xml changes. I'm still having trouble, though, getting the biquads modification to work reliably when going the system-less route...

 

[Smallsmx3]

This is a great thread. I love getting the best out of my devices and sound has always been important to me. I'm no audiophile but I'll be keeping up with this. Thanks!