LM6171 input stage design

This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
The LM6171 has been very popular with audiophiles. For a voltage op-amp, it has an amazing slew rate of 3000 V/µs at "only" 100 MHz of bandwidth. Distortion performance in the data sheet looks nice, but is unfortunately not given for, say, 1 kOhm load and down to several Hz.

Looking at the simpified schematic, it becomes clear where the slew rate comes from. The non-inverting signal path is just like that of any current feedback op amp. It is the good-old four-transistor emitter follower diamond buffer. The output current gets mirrored to the gain and output stages. The output emitters of this input buffer are what would usually be the inverting input of a CFB op amp. The R_E is connected just as ususal. However, there is another buffer (which I bet is also diamond design) that buffers the inverting input. Hence, a current across R_E get mirrored, converted to a voltage in the gain stage and is current-buffered by the output stage.

My problem: the diamond buffer is essentially an open-loop structure that is outside the overall feedback loop. Due to symmetries, it is surprisingly linear, unless loaded by a low impedance, such as R_E (which will be below 30 Ohms in most CFB op amps). This is why CFB op amps cannot offer really superior THD performance and hence are not popularly used in audio.

I guess this should be true for the LM6171. To the outside, it looks like a VFB op amp as there are two high-impedance inputs. But both buffers will be outside the feedback loop whereas the input stage distortion inside a conventional (long tailed pair) VFB amp gets attenuated by the loop gain.

So how does the LM6171 reach its good distortion rating and acclaim by enthusiasts? Is there a cancellation of distortion because non-linearities are the same in both input buffers? Will this work for even and odd-order harmonics? Will it only work well for some external circuit configurations?

Curious to hear your comments...

CFB Op-amps

I have no experience with this particular part, but I have used CFB in lots of designs.

I just don't like the way they sound. Doesn't matter what application I picked, they all had the same unusual sonic characteristics. And THD was not the problem.


The Eagle 2 was the most convoluted design I ever perused. I knew guys with EE degrees that couldn't fix one. A house of cards made of resistors, capacitors, and transistors. It had very tight bass but was very solid state sounding. It was designed by Jon Iverson. Jon was the strangest designer I know of. He disappeared under strange circumstances years ago. They will probably find Jimmy Hoffa before they find Iverson.....

I assume you weren't refering to me, because I have fixed them.

Interesting topology, not so sure about the execution of it. The grounding was screwed up. And they knew it, but didn't seem to care. Claimed the right way, which involved making one wire longer, another shorter, and moving them both was too hard in production.

back to the original question...

Can anybody second or disperse my view that distortion in the open-loop input buffers will cause strange effects? >-100 dB THD at 10 kHz, 100 Ohms looks nice, but what about e.g. thermal distortion effects in the input buffers that one would expect at lower frequencies that are not given in the typical performance plots?

I considered this part originally for some very non-audio, peak-hold circuit. Out of curiosity, last night I plugged one dual (LM6172) into my AD1854 DAC in place of the OPA2604. I would not have expected any significant difference, given the facts that the I/V conversion is done by internal (yuck), CMOS (yuck), 5V single supply op amps inside the DAC which should render any subsequent distortion isignificant, that there is already a passive filter before the external amp and that the load is on the order of 2 k. However the sound was markedly different. I will have to do some more listening tests before I can give a description.

analog devices quad core is similar

The quad-core, current-on-demand input stage seems to be very similar. A schematic can be found on page 12 of Section 1 of their high speed design seminar, called "High speed operational amplifiers" by Walt Kester. Differences to the LM6171 design seem to be use of emitter degeneration resistors (which may simply be not shown in the LM6171 schematic), the use of a current mirror to couple the output stages of the input buffers and the way the current is conveyed to the voltage gain stage (steered current sources instead of current mirror). All in all, this looks still like a CFB op-amp in disguise. This quad core is used in models AD9631, 8036, 8047 and with the addition of some extra input buffers on the 8041.

some more listening results

There was definitely no oscillation on the LM6172, in spite of my use of a dip8 drilled-hole, gold-plated socket (gag!).
There was definitely a difference between the OPA2604 and the LM6172, in spite of the not too demanding application and the possible swamping of effects by other inferior parts in the chain.

I did the tests with Natalie Merchant's new Motherland album as well as Gillian Welch's Hell Among the Yearlings, both produced by T-Bone Burnett.

The OPA sounded more musical, especially when many things were coming together. However, I had the nagging suspicion that it was somewhat shallow. The LM was thinner, colder and a little hoarse. Still, the timing seemed more relaxed and more realistic (strange result for an op-amp, isn't it?). I changed back and forth many times, so I hope this is not something I was imagining.

I added a 100 nF SMD ceramic capacitor accros the leads of the dip8 package, in addition to the more than ample decoupling on already on the pcb. Now, it seems the nastyness of the LM is gone. I will have to do some more comparisons when my hears are less fatigued....

OPA2604 vs LM6172

Hmmmm.... The OPA2604 has jfet inputs, twice the supply current as the LM6172, and a reasonable gain bandwith product of 20MHz vs 100 Mhz for the LM6172. High gain bandwiths can be a real pain for decoupling, layout, and driving cables ( you had better have a resistor in series with the output). I have seen (and heard) the OPA2604 in several good sounding audio products, often as a DC servo. I don't know of any audio products with the LM6172.

LT1358 et al?

Anyone tried using the Linear Technology LT1358 or its kin (LT1354-LT1365) in an audio application? It has a topology that sounds similar to the LM6171; I haven't looked at the '6171 data sheet so I can't say for sure. The LT application engineer described it as a CFB amp in VFB drag, modest GBW but very good slew rates at modest quiescent currents.
Anyone tried using the Linear Technology LT1358 or its kin (LT1354-LT1365) in an audio application?
I have a Modwright-modified Sony 333ES that uses the LT1364 in its output stages. The 333ES (which retained the LM6172 in the i/v conversion stage) sounds very clean. Tonally, it sounds very close in A/B comparisons with an SDS Labs tube DAC, which of course uses no op amps at all. (This is in contrast to my 9000ES DVD player, which was modified with OPA627s throughout. It seems to have a "warmer" less open sound than the other units.)

Much is made about decoupling of power supplies but if board decoupling is sound and power supply of high order, surely it is valid to directly compare op-amps of high bandwidth ie 6171 versus 825 for sound quality. This is especially so if there is no measurable difference in distortion or noise at the output.

What might happen at chip level, I don't know; it takes a lot of care and equipment to measure.
You can't always do a drop-in substitution. Ironically for what was once touted as an "universal analog component", op amps have become quite specialized, and you have to use some care in choosing an appropriate device for the application.

For instance, a DC coupled gain stage or an integrator designed around a FET input device's picoamp bias current will not take kindly to the 1 uA bias current of an LT1364. Some devices are stable into any capacitive load and can be used as line drivers with impunity; others oscillate at the drop of a hat. Some can deal with large differential voltages, others will fry in the same circuit. Some will drive headphones directly without breaking a sweat, others will current limit and clip, still others will simply smoke when presented with a low-Z load.

And, to directly address your comment, some of the very fastest devices will oscillate with the wrong combination of decoupling caps! As much as geeks like me try to reduce it to a science, audio electronics is still an art.
decoupling, topology and AD link

The socket has adequate decoupling and the LM6172 has an extra capacitor soldered across the supply pins. The output is isolated with a 400 R resistor. The amp being used in a multiple feedback, second order low pass configuration, there is a 220 pF capacitor from output to inverting input, but this should be fine as the LM6172 is unity gain stable. I am going to solder a 20 R insulation resistor directly to the noninverting link when I evaluate some even faster VFB op amps.

Coming back to the topology question: Once the output has settled, both inputs are at the same voltage, no current flows in R_E and there should be no distortion in the input buffers. The question is how long the settling takes. CFB amps have been notorious for having long settling times to 0.01%, I haven't yet figured out why. There are many cases where a real VFB op amp of slightly slower bandwidth/slew rate than a comparable CFB will have a longer settling time to 0.1% but faster to 0.01%. Note that this feature is not advertised for the 6171.

I have found the AD link again, take a look at page 11:

I wonder what the advantages of coupling the input buffers through current mirrors and using voltage-controlled current sources in the voltage gain section are. Anybody "auditioned" a quad-core amp?

Take care,


I agree; but manufacturers claim otherwise. Also no two or even the same manufacturer produces data in a form suitable for direct comparison.

Surely the user can only base assessment of suitability on the literature provided and not be expected to measure each type of chip.
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.