[KurtW]

I made some distortion measurements of a number of different buffers that I'd like to share. I have measured not only the THD (total harmonic distortion) but also how many dB down from the fundamental 1kHz frequency each of the first 4 harmonics are. Typically even harmonics are less bothersome than odd harmonics. But first let me state that lower distortion doesn't necessarily mean that it sounds better, it is just one of several important factors that influence the sound quality. I've made the measurements with both a 30 ohm load and a 100 ohm load with a 1.0v RMS output level. The amp used as a test vehicle was a META42 with AD8610 op amps and a regulated and filtered 24v supply, with a gain of 9. There was no attempt to optimise any of the resistor values for any particular buffer. For ease of taking the data, both channels were driven but one channel had a 30 ohm load and the other a 100 ohm load. This means that there were two different sets of buffers used, one for each load. The loads were metal-film resistors. In general I didn't see any significant relative differences between buffers when using different frequencies or other voltage levels within a reasonable range.

With a 30 ohm load:
BUFFER,% THD,2nd harmonic,3rd ,4th,5th harmonic
EL2001 x 1,0.28%,-62,-53,-78,-59
EL2001 x 2,0.21%,-66,-55,-80,-62
EL2001 x 3,0.168%,-69,-56,-83,-64
HA5002,0.129%,-81,-58,-89,-67
EL2002,0.126%,-65,-59,-83,-67
EL2002 x 2,0.067%,-70,-64,-85,-74
EL2002 x 4,0.034%,-74,-72,-80,-81
BUF634, 0.336%, -63, -51, -71, -58
BUF634 WB, 0.045%, -70, -76, -84, -76
BUF634 WB x 2, 0.037%, -74, -72, -88, -87
EL2008,0.034%,-74,-72,-87,-81

With a 100 ohm load:
BUFFER,% THD,2nd harmonic,3rd,4th,5th harmonic
EL2001 x 1,0.151%,-70,-57,-84,-65
EL2001 x 2,0.09%,-74,-61,-85,-71
EL2001 x 3,0.068%,-78,-64,-87,-75
HA5002,0.044%,-87,-69,-89,-79
EL2002,0.043%,-73,-70,-87,-81
EL2002 x 2,0.027%,-79,-80,-89,-85
EL2002 x 4,0.027%,-83,-88,-88,-87
BUF634, 0.162%, -67, -57, -78, -64
BUF634 WB, 0.031%, -78, -78, -90, -85
BUF634 WB x 2, 0.028%, -82, -90, -90, -84
EL2008,0.03%,-87,-77,-89,-84

An interesting data point, taken from the data sheets, is the static current draw (in mA) for these buffers:
EL2001 x 1, 1.3mA
EL2001 x 2,2.6mA
EL2001 x 3,3.9
HA5002,8.3
EL2002,5
EL2002 x 2,10
EL2002 x 4,20
BUF634 1.5mA
BUF634 WB (wide bandwidth mode), 15mA
BUF634 WB x 2, 30mA
EL2008,13

The most obvious observations are that the higher current buffers generally have lower distortion, and using multiple buffers lowers the distortion, especially for a lower impedance load.

Which sounds better? Well, they don't all sound the same, but which one is better is a matter of opinion and right now I'm just presenting data.

 

[kevin gilmore]

Would really like to see a buf634 under the same exact
set of conditions...

The results are very interesting. How much of this is actually
thd, and how much is thd+n?

 

[KurtW]

Perhaps I'll add the BUF634 as it is commonly used. The reason I didn't put it in initially is the ones I have are TO-220 and I didn't want take the time to do the conversion for the EL2xxx pin-out as it's not straight forward.

Kevin, the power supply I used is very clean and the amp itself is very quiet (at my measurement limit of 62uV) but the signal generator has some output in the 250-400 Hz range that adds to the THD. It measures 0.024% by itself, although that's about the limits of my meter as well. I had taken the data out to the 10th harmonic but beyond the 5th it was very close to the measurement of the generator.

 

[morsel]

Kurt, you are a god. Thanks so much for doing those tests!

Could you also give us numbers for a single EL2001 and stacked HA5002 x2 as well as the BUF634 buffers?

 

[KurtW]

Per requests, I've added a single EL2001 and several configurations of the BUF634 to the original data. The BUF634 in wide bandwidth mode is with the BW pin tied to V-. The supply current goes up by 10x and the distortion goes down by almost as much.

Can the HA5002 be stacked? I think I read somewhere that you should match the devices first. Anyway I blew up a AD8610 by forgetting to power down during one of the buffer swaps, so I'm done testing for awhile.

 

[Audio-Me]

Kurt could you please give us a sonic comparison between EL2001 and EL2002? (preferably stacked)

 

[Nisbeth]

Interesting reading.
If there are anyone other than me who find the data a bit difficult to read in the post, I've made the data into a spreadsheet and made a few simple graphs. Feel free to download and modify as needed!

/U.

EDIT: Updated spreadsheet. Corrected a few errors.

 

[morsel]

Thanks again, Kurt.

Sometime when you get your circuit working again and have the time, try stacking the HA5002 buffers 2x. I think that will work for the purposes of this test.

BTW, performing these tests at other frequencies such as 20Hz or 20kHz may give radically different results.

 

[rickcr42]

interesting . Looks like the venerable BUF634 holds up pretty well against the competition . Good thing since i use 'em a lot

 

[morsel]

Um, Rick, just because the BUF634 had favorable results for this one test does not mean it is a superior buffer. It is actually the most problematic of the tested choices.

 

[rickcr42]

I still use it extensively , of course no match for my LH0033 or 66 buffers soundwise.

BTW-you should SEE the LH0063 ! BIG ASS 8 pin TO-3 can with a custom heat sink (no way I could drill one out for 12 pins in a circular pattern) and 8 pin socket . The heat sinks are like 3X5X1.5 !
One big mutha of a buffer . Looks like a 50 wpc power amp .

The 33 is the mil spec ceramic 8 pin dip package. Not a normal size dip but much wider

Surprisingly the newer buffers spec better , the big boy only puts out 250mA current continious and 500ma peak . Considering the size you would think it output a full amp.
The 33 is 100/250.

Total internal parts

LH0033 :2 jfets,1 diode,4 transistors
LH0063 :4 jfets,3 transistors


great Grado drivers

 

[morsel]

I think one can look at the lack of output protection on the HA5002 as a blessing rather than a curse. It has lower output impedance which results in superior control over the driver and consequently better sound. A fuse, circuit breaker, or current source can be used to limit current to the buffer. I like the latter two options as they don't require replacing fuses.

 

[aos]

Kurt, great work! But, did you say that the noise on the output is less than 62uV and the amp has gain of 9? What value of the input volume pot did you use? Those values are extremely good.

 

[KurtW]

aos, I used a 10k Noble pot.

morsel, can you expand on your comment that "It (the BUF634) is actually the most problematic of the tested choices."

I checked the THD in the range of 100 to 10KHz and it was almost identical to the 1kHz reading. I personally don't have much interest in THD at 20Hz and 20KHz, as no headphone I've ever heard can reproduce 20Hz and I can't hear 20kHz, let alone the 2nd harmonic of 20KHz (I can't measure that either).

 

[morsel]

Joe, external current limiting solutions have the potential to be of superior design to intrinsic current limiting. I was just checking out Raychem PolySwitch resettable fuses today. I am not an expert in this area, but I just happen to be looking into it atm. A Burr-Brown paper I was reading talks about using back to back diodes in the feedback loop. This can result in compression. Another BB paper suggests using voltage regulators as current sources for the power rails. The regulator has to be fast or it will limit the transient response of the buffer. Sorry I can't give you any concrete advice. Maybe in a month or two I will be in more of a position to comment. Here are URLs for the papers, which were originally brought to my attention by PPL:

http://www-s.ti.com/sc/psheets/sboa042/sboa042.pdf
http://www-s.ti.com/sc/psheets/sbva011/sbva011.pdf

Kurt, Joe pointed out quite correctly that the Elantec buffers are being discontinued. I was speaking purely from an electronic perspective. The BUF634 is double the price of other buffers, has a 30-100mV DC offset, a greater dependency on stacking to sound good, and many people feel that it does not sound as good as Elantec or Intersil buffers, even when used in high bias mode. There have been many threads on this subject on both Head-Fi and Headwize. These are the most recent:

http://headwize2.powerpill.org/ubb/s...num=3&tid=2977
http://www4.head-fi.org/forums/showt...threadid=14214