I've recently bought Sennheiser HD600 and I also bought an Asus Xonar U7 USB sound card to have a high quality portable source for it. I also own a Sound Blaster Omni Surround 5.1 USB sound card which I use at work (the PCs have terrible on-board audio).
I was curious to compare the two sound cards - how well would they drive the HD600.
So I've bought an Y jack splitter, downloaded RMAA and started testing (using my MSI GE60 laptop running off battery, WiFi disabled). The first "problem" I encountered was that the only usable line-in was the SB Omni's one. The laptop and the U7 have only a microphone input and I was not going to risk it and connect an amplified headphone output to a 5V phantom-powered mic input (i.e. two voltage sources against each other). Amplifiers should have an output capacitor in series which should cut off the DC phatnom voltage but I'm not an audio specialist and I don't know if it's safe.
So the tests I ran were:
1) Xonar U7 headphone out - splitter - Omni's line-in and HD600 in parallel
2) Omni headphone out - splitter - Omni's line-in and HD600 in parallel
3) Laptop's Realtek headphone out - splitter - Omni's line-in and HD600 in parallel (this was unplanned, did it just for fun)
All cards were set to 48kHz/24 bit in windows playback device options (and recording for Omni's line-in), all effects disabled.
Setup 1) (http://www.hydrogenaud.io/forums/index.php?act=attach&type=post&id=8104)
Setup 2) (http://www.hydrogenaud.io/forums/index.php?act=attach&type=post&id=8105)
(How on earth do I show pictures uploaded to HA directly? It tells me "Sorry, dynamic pages in the [IMG] tags are not allowed")
I didn't take a picture of the Realtek test - I just plugged the jack into the laptop headphone output.
The results are quite surprising:
http://www.hydrogenaud.io/forums/index.php?showtopic=107821 (http://www.hydrogenaud.io/forums/index.php?showtopic=107821)
Screenshot (http://www.hydrogenaud.io/forums/index.php?act=attach&type=post&id=8106)
I did not perfectly level-match the 3 cards. However, I set both U7 and Omni very close to each other. Xonar U7 was set to highest of the 3 amp levels and windows volume at 74%, Omni - having a more powerful amp - was set to 56% and Realtek was set to 90% (a bit louder than the other two, maybe giving it a bit of advantage on the SNR test).
Quite unexpectedly, Xonar U7 was only marginally better than the Realtek. Omni ended up a class or two above both.
Normally, people RMAA loopback of a single card. In the above scenario, the only difference was the source (output). The input was always the same with the same settings (I just switched the black cable to a different output and selected that output in RMAA). The only explanation I can think of (besides Xonar U7 being a very mediocre product) is that there was some grounding issue between the two cards (I tried swapping the devices between various USB ports on my laptop which had no effect on the results). Or is there some other reason why analogue loopback on the same card should have such huge advantage? I mean if Omni could feed it with such a good SNR why couldn't U7?
I did not test any lower-impedance headphone because the output level I needed for line-in would have fried it (Omni's line in was set to 100% volume). Even the HD600 was painfully loud at that volume (the RMAA test is not very pleasant to listen to, anyway )
Out of curiosity, I also tested my portable players (see the uploads thread). They both seem to beat the U7 except for the L/R crosstalk where they suck. I had to do it at 44/16 because the players produced atrocious resampling artifacts for 48kHz input.
Note that I do not claim to be able to hear any of the differences I measured.
I will try to figure out how to test with low-impedance headphones (without having to blast them with ridiculous voltage). I expect all the measured devices to perform significantly worse with e.g. 16 ohm headphones.
Set the clip to 48k output if you want to test 48k files.
I wouldn't be surprised if the xonar is mediocre. A lot of those USB sound cards are pretty bad.
You shouldn't have a problem driving most headphones at line level voltage or maybe a little low if you want to be safe.
The Xonar U7 headphone amp is mediocre at best, or at least it seems that people on Headfi think so
IIRC it has a pretty high output impedance and can't power hungry headphones, though it should have pretty good specs as a DAC, unfortunately there's few measurements done on that card, only a few RMAA in PC gaming websites and a test done by a guy who uses the card for SDR, though those are for the ADC
Well, that sucks. I mean U7 is twice as expensive compared to the Clip+ and considerably more expensive than the SB Omni, yet they fail to equip it with a low noise quality headphone amp circuit?
Taken from their U7 marketing material http://www.asus.com/Sound_Cards/Xonar_U7/ (http://www.asus.com/Sound_Cards/Xonar_U7/)
Integrated headphone amplifier pumps up every sound detail
Xonar teams integrate a headphone amplifier on Xonar U7 to boost immersive headphone output.
The above sure does not seem to imply that U7 is no better than an on-board Realtek when it comes to driving headphones... Deceitful bastards.
You really shouldn't buy an amp or DAC without actually verifying that the specs are good.
The Xonar U7 headphone amp is mediocre at best, or at least it seems that people on Headfi think so
IIRC it has a pretty high output impedance and can't power hungry headphones, though it should have pretty good specs as a DAC, unfortunately there's few measurements done on that card, only a few RMAA in PC gaming websites and a test done by a guy who uses the card for SDR, though those are for the ADC
This one paints a pretty good picture:
Xonar U7 Review (http://www.reviewstudio.net/191-asus-xonar-u7-review-surround-portability/features-amp-performance)
I did my best before buying it. I have no idea how the reviews measured the sound card to achieve those numbers (I can only guess it was line-out to line-in). Also the last two charts in the review Arnold linked look quite suspicious, especially the crosstalk. Why would the line-out have a worse cross-talk than the headphone-out? It does not make any sense to me.
I did my best before buying it. I have no idea how the reviews measured the sound card to achieve those numbers (I can only guess it was line-out to line-in).
I believe that the tests are exactly based on on line-in to line-out. They are indications of the performance of the card in that mode of operation, but they would require additional tests with simulated or real headphone loads to be relevant to that mode of operation.
Also the last two charts in the review Arnold linked look quite suspicious, especially the crosstalk. Why would the line-out have a worse cross-talk than the headphone-out? It does not make any sense to me.
Most crosstalk is due to signal leakage between the channels, which is often greatest when higher impedances and higher frequencies are involved. Thus one need not be concerned when the testing of a higher impedance output shows e more high frequency cross-bleed than that of a lower impedance output.
I don't think that many people have actually done listening test to determine requirements for controlling cross-bleed or crosstalk in order to avoid undesirable alterations to the sonic scene. Its been a long time since I did any such thing, but my recollections is that someplace between 20 and 40 dB separation, audible alterations of the sonic scene vanish. IOW 60 or 80 dB separation are just two different numeric permutations of sonically perfect performance.
The tests in that link are unloaded tests and so not very useful unless one is also planning on buying another amp. Of course if you were going to do that I'd suggest that you buy an amp with a quality DAC built in ; )
The tests in that link are unloaded tests and so not very useful unless one is also planning on buying another amp. Of course if you were going to do that I'd suggest that you buy an amp with a quality DAC built in ; )
I see the possibility of using a good inexpensive headphone amp such as the Topping NX1 to address this audio interfaces weak performance with low impedance headphones.
The tests in that link are unloaded tests and so not very useful unless one is also planning on buying another amp. Of course if you were going to do that I'd suggest that you buy an amp with a quality DAC built in ; )
I see the possibility of using a good inexpensive headphone amp such as the Topping NX1 to address this audio interfaces weak performance with low impedance headphones.
That would absolutely work, but for what they charge I'd expect a quality amp built in. Fiio will sell you that for the same price for instance. Not like these things are expensive or difficult.
The point was having only a single piece of hardware. Maybe I'm a stupid or something but I did not really expect a soundcard with a built-in headphone amp to be so much different from a headphone amp with a built-in DAC.
At least not at that price point. Or does common sense no longer apply today?
I have no idea how the reviews measured the sound card to achieve those numbers
- you need the highest quality ADC that you can get your hands on;
- when trying to measure the SNR of a device, you need to do it unloaded, and set its volume to the maximum (and the ADC needs to be able to record it without clipping);
- when doing the measurements with headphones attached, you need to set the volume to a "loud" level that won't make you insta-deaf and that won't significantly raise distortion figures. The measured SNR will be lower, but you already measured it with the unloaded test.
I see the possibility of using a good inexpensive headphone amp such as the Topping NX1 to address this audio interfaces weak performance with low impedance headphones.
That would absolutely work, but for what they charge I'd expect a quality amp built in. Fiio will sell you that for the same price for instance. Not like these things are expensive or difficult.
I haven't seen any evidence so far that the headphone amplifier in the Xonar U7 is bad.
I did not really expect a soundcard with a built-in headphone amp to be so much different from a headphone amp with a built-in DAC.
They are the same thing.
I haven't seen any evidence so far that the headphone amplifier in the Xonar U7 is bad.
I'm under the impression that it provides a source impedance that may be too high for really low-impedance headphones. The number 10 ohms seems to be rattling around in my head.
The online spec sheet says: Headphone output : 1.3 Vrms (3.677 Vp-p) which is a little low for some really high impedance headphones resting on ears who like it loud.
I was apparently interested in a loaded headphone-out SNR figure since this was what I was going to use. I didn't need a high quality ADC, just ADC with figures better than what I was measuring. The U7's headphone output was "so bad" that SB Omni's line-in was good enough.
Every device I have used was able to drive HD600 at a volume higher than what I needed.
Just look at my measurements, there you have evidence that for HD600, this sound card is no better than the laptop's integrated Realtek.
I was apparently interested in a loaded headphone-out SNR figure since this was what I was going to use. I didn't need a high quality ADC, just ADC with figures better than what I was measuring. The U7's headphone output was "so bad" that SB Omni's line-in was good enough.
Every device I have used was able to drive HD600 at a volume higher than what I needed.
Just look at my measurements, there you have evidence that for HD600, this sound card is no better than the laptop's integrated Realtek.
The presence or absence of a headphone load is unlikely to materially affect SNR.
The presence of a load might affect THD or more importantly FR.
The SNR in rmaa usually does depend on load because it plays a tone and measures the noise floor. Thd and other effects end up factored into the noise.
Although looking at this test:
http://rmaa.dfkt.tk/Comparisons/Comparison...600%20Loads.htm (http://rmaa.dfkt.tk/Comparisons/Comparison%20-%20Cowon%20D2%20-%2016%2C%2080%2C%20300%2C%20600%20Loads.htm)
The entire noise floor seems to rise with increasing load. Maybe more power supply noise is coupled in at higher currents?
The SNR in rmaa usually does depend on load because it plays a tone and measures the noise floor. Thd and other effects end up factored into the noise.
Although looking at this test:
http://rmaa.dfkt.tk/Comparisons/Comparison...600%20Loads.htm (http://rmaa.dfkt.tk/Comparisons/Comparison%20-%20Cowon%20D2%20-%2016%2C%2080%2C%20300%2C%20600%20Loads.htm)
The entire noise floor seems to rise with increasing load. Maybe more power supply noise is coupled in at higher currents?
Power supply noise could have something to do with the increased noise. Noise from the power supply sometimes tends to increase when the amplifier is pretty heavily loaded.
What was the power source - batteries, a wall wart, or other?
Cowon D2 is an old mp3 player, so its battery powered.
Cowon D2 is an old mp3 player, so its battery powered.
Well, that may explain why the increase in noise seems to have taken the form of random noise. If it was powered by the power line the usual situation is that the increased noise while driving low impedances has strong components related to either the power line or the operational frequency of any switchmode power supply. When you put a heavy drain on batteries you increase chemical activity which may create more random noise.
Cowon D2 is an old mp3 player, so its battery powered.
A little research seems to show that it was available with a line-operated power supply:
http://anythingbutipod.com/2007/04/cowon-iaudio-d2-review/ (http://anythingbutipod.com/2007/04/cowon-iaudio-d2-review/)
I intentionally ran the test on battery power. Maybe there is another source of power noise in the laptop (it's a powerful gaming laptop). Anyway, components in the less expensive SB Omni had absolutely no problems with noise so why couldn't U7...
Setups 1) and 3) show copious amounts of what looks to be ground loop noise spoiling the results. Not sure why the Omni is better here - maybe someone bothered to implement an (electronically) balanced input? USB port in a different location vs. U7? Some custom cabling to break the loop should fix that if need be.
IMD on the Realtek also indicates that it could use a driver update (should be 2011, early 2012 vintage now). Should be similar to U7 then, except still higher in output impedance / FR deviation.
I see the possibility of using a good inexpensive headphone amp such as the Topping NX1 to address this audio interfaces weak performance with low impedance headphones.
That would absolutely work, but for what they charge I'd expect a quality amp built in. Fiio will sell you that for the same price for instance. Not like these things are expensive or difficult.
I haven't seen any evidence so far that the headphone amplifier in the Xonar U7 is bad.
I measured a source impedance of 17 ohms on mine, and that makes it iffy for low impedance 'phones.
PS:
Setups 1) and 3) show copious amounts of what looks to be ground loop noise spoiling the results. Not sure why the Omni is better here - maybe someone bothered to implement an (electronically) balanced input?
I missed the obvious, namely that the Omni is the one recording. It thus has the smallest ground loop of them all by far in loopback, and most importantly said ground loop does not extend over USB power ground into the notebook and back! I would advise retesting the others while using their own inputs. Easier than making a ground-loop-breaking cable or even a balanced input circuit, anyway.
Using one of these combo mic/line inputs is not ideal (lowish input impedance) but tends to yield quite decent results IME. You just need to disable preamplification (it probably defaults to something like +20 dB), then the input ought to accept 1-1.4 Vrms - or maybe things can be switched altogether (jack retasking). Mic bias voltage tends to be ~3 V and applied through resistors of 1-2.2 kOhms, so it's not particularly dangerous.
Once that is sorted, I'd be interested to know how well the Omni does with 44.1 in
playback.
Is there any reason why you need 44.1kHz? It does 48kHz very well so why not just set that in the Windows 7/8 playback device options and forget about it? Foobar2000 has a quality resampler if you are concerned with the windows mixer's SRC quality.
I did a test of U7's own loopback and it changed everything. Thanks for the suggestion, stephan_g. I have uploaded the results in the Uploads thread. The result is excellent.
Is there any way how I can mark the thread as SOLVED? I can't edit my initial post.
That's much better. You can clearly see that the opamp used to drive the output seems to break a bit of a sweat - the dominant odd-order distortion is telling (though -90 dB is nothing much to worry about yet, especially since levels appear to be near maximum), as well as the increase towards higher frequencies in the swept sine test. You would see it even better at 96 kHz - distortion actually does not drop above 6 kHz and change, it's just that the dominant 3rd harmonic is being attenuated by the anti-alias filter then.
Apparently the Xonar U7 uses a TI / NatSemi LME49726MA opamp - a type with substantial output current that is suitable for lowish operating voltages like +5 V. At just 0.7 mA per amplifier typical (1.4 mA for a stereo pair), substantial amounts of crossover distortion are not at all surprising. For illustration, the classic NJM4556A (for higher voltages) consumes 4 mA per side - even if only half of that is output stage quiescent current, it's still about 4 times as much as I'd expect for the NatSemi part, possibly even more. The NatSemi part can't make up for it in GBW either, as it actually is slightly slower.
If we are seeing dominant 3rd into an easy 300 ohm load, I would expect results to be less than pretty with insensitive 32-50 ohm cans. It might have been a better idea to put up with the notoriously diva-like AD8397 or somesuch instead.
I would assume that the Omni fares better in this respect, at least if it uses the same MAX97220 found in other modern-day Creative cards. My Audigy FX shows a slight increase in swept IMD towards 10 kHz when driving ~1 Vrms into a HD580 at the front output in 24/96, but otherwise seems hardly bothered. Things still are under -80 dB at this point. With some extrapolation, the U7 would seem to barely hit -60 dB.
About the 44.1kHz playback... The Omni does not even have any 44.1 options in the Windows Playback or Recording device configuration. There are only 4 options - 16/24 bit 48/96kHz. I guess it is not possible to test native 44.1kHz without Windows resampling it to 48 or 96 kHz (since the device does not have that capability).
I updated the uploads thread with some low-impedance load measurements. It seems the amplifier in Omni is somewhat better (less affected by the load) than in U7.
Unfortunately, both devices seem to further boost (+1dB) the already overblown base of Sennheiser HD518. The L/R separation suffers from low impedance equally on both devices.
That's much better. You can clearly see that the opamp used to drive the output seems to break a bit of a sweat - the dominant odd-order distortion is telling (though -90 dB is nothing much to worry about yet, especially since levels appear to be near maximum), as well as the increase towards higher frequencies in the swept sine test. You would see it even better at 96 kHz - distortion actually does not drop above 6 kHz and change, it's just that the dominant 3rd harmonic is being attenuated by the anti-alias filter then.
Apparently the Xonar U7 uses a TI / NatSemi LME49726MA opamp - a type with substantial output current that is suitable for lowish operating voltages like +5 V. At just 0.7 mA per amplifier typical (1.4 mA for a stereo pair), substantial amounts of crossover distortion are not at all surprising. For illustration, the classic NJM4556A (for higher voltages) consumes 4 mA per side - even if only half of that is output stage quiescent current, it's still about 4 times as much as I'd expect for the NatSemi part, possibly even more. The NatSemi part can't make up for it in GBW either, as it actually is slightly slower.
If we are seeing dominant 3rd into an easy 300 ohm load, I would expect results to be less than pretty with insensitive 32-50 ohm cans. It might have been a better idea to put up with the notoriously diva-like AD8397 or somesuch instead.
I would assume that the Omni fares better in this respect, at least if it uses the same MAX97220 found in other modern-day Creative cards. My Audigy FX shows a slight increase in swept IMD towards 10 kHz when driving ~1 Vrms into a HD580 at the front output in 24/96, but otherwise seems hardly bothered. Things still are under -80 dB at this point. With some extrapolation, the U7 would seem to barely hit -60 dB.
What do you see that you consider to be reliable evidence of crossover distortion?
A considerable increase in HD towards higher frequencies when driving anything resembling a load is very suspect, for one thing. If you extrapolate towards 10 kHz, it's far from where I'd expect in line out (quasi-unloaded) operation in a competent soundcard.
The very low quiescent current draw makes things even more suspect. Besides, -90 dBr @ 1 kHz is a far cry from what the LME49726 datasheet claims for a buffer driving 3.5 Vrms into 10 kOhms (0.00008%, i.e. <-120 dBr).
Thanks for asking though (even if the fullquote would not have been necessary) - otherwise I wouldn't have realized that the dominant odd-order distortion profile is likely to indicate a push-pull type opamp circuit, most likely push-pull AB (given the shoestring current budget). Unfortunately I'm a bit out of the loop when it comes to literature showing such circuitry in CMOS (you don't get to within 4 mV of the rails with BJT output). Some bipolar implementations mentioned in the classic Tietze-Schenk include OP467, LT1352 and LM7171; Groner (http://www.sg-acoustics.ch/analogue_audio/ic_opamps/index.html) also mentions LT1354-LT1363; his OP467 measurement results show the dominant odd-order distortion very clearly, especially in the output linearity test with 600 ohm loading.
Qualitatively similar results are also seen in the Clip+ (http://nwavguy.blogspot.de/2011/02/sansa-clip-measured.html), another MOS IC affair. You can set output stage bias levels for its headphone driver (3 amps, 1 being a virtual ground driver) in software, IIRC the standard value is 0.5 mA per amp. Upping this to a nominal 0.75 mA yields a very modest but measurable reduction of high-frequency harmonics when driving loads (BTDT). I don't think they're selling the headphone driver section separately somewhere.
A considerable increase in HD towards higher frequencies when driving anything resembling a load is very suspect, for one thing. If you extrapolate towards 10 kHz, it's far from where I'd expect in line out (quasi-unloaded) operation in a competent soundcard.
The very low quiescent current draw makes things even more suspect. Besides, -90 dBr @ 1 kHz is a far cry from what the LME49726 datasheet claims for a buffer driving 3.5 Vrms into 10 kOhms (0.00008%, i.e. <-120 dBr).
Thanks for asking though (even if the fullquote would not have been necessary) - otherwise I wouldn't have realized that the dominant odd-order distortion profile is likely to indicate a push-pull type opamp circuit, most likely push-pull AB (given the shoestring current budget). Unfortunately I'm a bit out of the loop when it comes to literature showing such circuitry in CMOS (you don't get to within 4 mV of the rails with BJT output). Some bipolar implementations mentioned in the classic Tietze-Schenk include OP467, LT1352 and LM7171; Groner (http://www.sg-acoustics.ch/analogue_audio/ic_opamps/index.html) also mentions LT1354-LT1363; his OP467 measurement results show the dominant odd-order distortion very clearly, especially in the output linearity test with 600 ohm loading.
I would never expect the analog output of a battery-powered IC-based headphone amp to be anything but push-pull. If for no other reason than power efficiency. But p-p also supressed even order distoriton.
I always remember that when we look at measurements of these devices we can't separate the converter performance from the analog buffer performance without opening the box, and perhaps not even then.
The 100% evidence of crossover distortion is a discontinuity in the devices ouptut voltage near Output Voltage= Zero. This is usually verified by playing a small signal and enlarging it so that such a thing could be seen on a Voltage versus Time plot. Sometimes partially notching out the test signal can improve its visibility, if it exists.
Qualitatively similar results are also seen in the Clip+ (http://nwavguy.blogspot.de/2011/02/sansa-clip-measured.html), another MOS IC affair. You can set output stage bias levels for its headphone driver (3 amps, 1 being a virtual ground driver) in software, IIRC the standard value is 0.5 mA per amp. Upping this to a nominal 0.75 mA yields a very modest but measurable reduction of high-frequency harmonics when driving loads (BTDT). I don't think they're selling the headphone driver section separately somewhere.
I can't see where the author of that report called out the Clip for having crossover distortion. I have a Clip and a couple of Fuzes and the necessary testing hardware and software, and if it weren't that my day job is making heavy demands right now, I'd do some specific tests. Whatever the cause of the distortion is, it is still like 70 dB down at worst, and lots of luck detecting that in a proper listening test! That all said I listen to my Fuze via a NX-1 buffer for reasons of dynamic range and relatively insensitive IEMs, so I don't have a dog in this fight. ;-)
Mostly CDs picked up at Estate Sales. Rip to Lame 256K because MP3 works just about everywhere and 256K is plenty good for these 568 year old ears.
Cross over distortion would show up in those DAC linearity tests right? He doesn't show the first harmonic in that test, but presumably if there was a huge peak there, he would have commented on it.
Cross over distortion would show up in those DAC linearity tests right? He doesn't show the first harmonic in that test, but presumably if there was a huge peak there, he would have commented on it.
Crossover distortion as well as other low level nonlinearities show up in the Dynamic Range test which is based on a test tone that is 60 dB below FS. Delta Sigma DACs are the rule and they are inherently linear. In practice they are not perfectly linear but their nonlinearity comes primarily from their analog side.
When someone shows a THD test with all (both even and odd) harmonics comparably represented but uniformly decreasing with order, the wave to think about is a Sawtooth wave being superimposed on the signal. If you notch out the test signal, that's the basis of what is left over.
People with CEP or Audition can generate these at will by generating a Triangle/Sawtooth wave with a Flavor of 0% or 100%. The FFT analysis will show every harmonic with decreasing amplitudes, reminiscent of the odd harmonics in a Square Wave.
I would never expect the analog output of a battery-powered IC-based headphone amp to be anything but push-pull. If for no other reason than power efficiency. But p-p also supressed even order distoriton.
Especially if it extends right up to the input stage, which is a point that possibly I should have been stressing a bit more. I would not be surprised to see a fully Class AB ("power on demand") circuit like what I mentioned.
Cross over distortion would show up in those DAC linearity tests right? He doesn't show the first harmonic in that test, but presumably if there was a huge peak there, he would have commented on it.
We're talking analog here, not DAC nonlinearity. Some of the latter is there in the Clip+, but the measured deviation (-90 dBFS tone at -91.6dBFS) computes to 0.05% of distortion re: -40 dBFS. So even in the quiet parts of a fairly dynamic classical recording it would still be about 60 dB down - noise level is expected to be higher than that in many cases. IOW, nothing I'd lose much sleep over, particularly since volume control is analog and follows after the DAC, easing up on DAC requirements by orders of magnitude (which saves
heaps of power).
Contrary to popular belief, analog crossover distortion in properly-designed AB amplifiers actually decreases when reducing amplitude. At some point it tends to reach sort of a plateau, and eventually proceeds to drop further as Class A operation is reached. Actually even the "plateau" is a numerical thing, as you'll see the dropoff towards higher-order harmonics varying quite considerably - a hearing-weighted measure of harmonic distortion would be anything but flat.
If you look at the Clip+ 400 mV into 16 ohm measurements, you can see that the decay in higher-order harmonics is very slow. Halve the amplitude, and you would see them decay much more quickly, until at some point the distortion profile becomes benign.
Now if you've got a real
discontinuity / dead zone there (Class B or C, depending on whose definition you follow), then things do in fact tend to get nasty. This tends to result in amplifiers whose distortion profile at full output isn't so bad, but once the output amplitude approaches the order of dead zone width, distortion becomes pretty awful (preceding circuitry desperately jumps between the ends of the dead zone in an attempt to maintain linear output, taxing its slew rate). Hence why some people emphasize "first watt" performance so much, and for good reason - 1 W tends to be good for >80 dB SPL after all.
Several years ago I simulated a headphone amplifier circuit using a 5532-ish opamp plus underbiased Class B Darlington booster within the feedback loop (~1.8 V bias, so maybe 0.4-0.6 V missing), and obtained:
(10 kHz THD, 300 ohm load - worse at lower impedance)
6 Vrms: 0.15%
0.6 Vrms: 0.7%
0.06 Vrms: 0.6%
0.006 Vrms: 0.02%
Yikes. (And from what I remember, mid-level distortion spectra weren't exactly looking pretty either.) The real deal seemed to measure better than that, but I still didn't like its sound much when driving 100 ohm cans.
Similar bias-less circuitry has also been looked at in practice in a small speaker amp (https://web.archive.org/web/20131116080840/http://ludens.cl/Electron/audioamps/AudioAmps.html). Unsurprisingly, opamp slew rate becomes a major factor. (Yes, the TL071 should theoretically be slewing faster than the NE5534, but its high internal output impedance means that it's losing a lot of voltage internally, especially around the deadband region where buffer transistor beta is low. This type is notorious for its poor output driving abilities, worse than the '741's, while a '5534 is pretty rugged in that regard.)