• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required as is 20 years of participation in forums (not all true). Come here to have fun, be ready to be teased and not take online life too seriously. We now measure and review equipment for free! Click here for details.

Opamp Rolling, Does It Work?

jackenhack

Active Member
Joined
Oct 25, 2018
Messages
173
Likes
367
Location
Stockholm, Sweden
#1
Many hobbyists like to do "Opamp Rolling", by changing the different opamps in audio gear. There's much discussion on the internet audio forums about the best opamp to use. However, many just read about a specific opamp that is the current flavour of the month and change it willy nilly. When an engineering team designs a product, they (hopefully) first do simulations, then measures and listen to the prototype and compare prices before deciding which opamps to use. There are a lot of different kinds of opamps for a specific purpose, and some of the examples of changing opamps on the forums are plain crazy.

So let's see if we can measure the difference by changing opamps by measuring instead of waffling about how much more air the sound has, or better bass response you get by using more expensive opamps.

To be able to do this, I've decided to use a well known and most importantly, well-documented headphone amplifier. I chose to use the JD Labs O2 headphone amplifier because it's a good design, but also because there is a schematic available. I also had one laying around. The face on the device adds an airiness to the sound, trust me.

o2-headphone-amplifier.jpg


The O2 headphone amplifier was designed by a guy who was known on the internet as NwAvGuy. He entered the audio arena, writing about audio design and made the O2 headphone amplifier to prove to people that you don't need a lot of fancy and expensive boutique items to create a transparent headphone amplifier. He released it as an Open Source hardware design and didn't want any money for it. JD Labs took the design and started to manufacture and sell it. Then NwAvGuy disappeared. My uneducated guess is that he worked for an audio company, and didn't want to take the risk of conflict of interest. However, that's just my guess. If you want to learn a lot about how audio equipment works, his blog is still online. By using the O2 amplifier as a test mule, we can make some decisions based on the design instead of just throwing in a fast opamp meant for instrumentation or something similarly crazy one you've just read about on an internet forum post.

The O2 amplifier has an NJM2068D as it's gain opamp. It's a good performing and cheap bipolar junction transistor based opamp. So if we are going to change it, we should stick with a bipolar opamp. MOSFET based opamps have very high input impedance, vanishingly low bias current, but has higher voltage noise at the input than a bipolar opamp. They both have their places, but that's a design decision when making the amplifier, so it's best to stick to the same opamp type if you feel the need to swap. You don't want the output offset to be all wonky, do you?

The output buffer opamps used in the O2 amp is the NJM4556A, an opamp that can deliver ±70mA but by adding two outputs together, you get twice the output power. That's why there are two dual channel opamps on the output. However, there's another benefit of paralleling the output of an opamp. Every time you double the number of outputs, you lower the noise by approximately -3 dB. Quite handy! Yes, you can add together 256 opamp outputs and try to measure gravity waves! :)


summoned-opamps.jpg


o2-amplifier-pcb.jpg


I've decided to change the gain opamp to the venerable OPA1612. It's a VERY good bipolar junction opamp with incredible specs. If changing the opamp to this one doesn't make a difference, nothing will.

The OPA1612 comes in a SOIC-8 package, and the NJM2068D is old-school DIL package, so I first used a small PCB to convert it to DIL. I also added decoupling and bulk capacitors on the PCB, to make sure that there was both noise filtering and power available for it. I did measure it, and it didn't make any difference to the measurements, but generally, it's good practice to have decoupling and bulk capacitors as close as you can to the IC. The IC socket doesn't make it easier to have capacitors close to the opamp. I wired the ground wire from the small PCB to the nearest GND I could find (the volume control.)

opa1612-adapter.jpg


First, let us do a base measurement of the amplifier as it comes from JD Labs. I used the amplifier at full volume on high gain setting. 1 Volt input.

High gain original gain original buffer amps.png


Ok, let's swap out the gain opamp to the OPA1612 and see if there's a difference.

High gain OPA1612 gain original buffer amps.png


Well, that was a disappointment! A 0.6 dB lower THD at 1kHz isn't much to write home about. However, we have to consider the stuff that comes after, in this case, the unity gain output buffers. If they are the limitations of the THD measurement, we wouldn't see any benefit of replacing the gain opamp.

The output buffers on the O2 amplifier are reasonably grunty opamps. The only opamps I had at hand that could deliver ±70 mA per channel was the OPA1622 which can provide even more. Unfortunately, the OPA1622 in a 5x5mm square package and is highly specialized. They even have a ground connection for god's sake! So I was left with a handful of opamps that I had laying around, and the ones with the highest output power were the LME49860. They have outstanding specs when it comes to THD and noise, but can only deliver ±30 mA. But as we are testing this swap without any load, it shouldn't be a problem. So here's the measurement with the LME49860 as output buffers.

High gain OPA1612 gain LME49860 buffer amps.png


Well, we got an improvement of -2dB THD. However, if you have a look at the frequency span between 0 to 1kHz, you can see what looks like the start of oscillation. Not all opamps are playing nice when they are summed at the output. You can rectify this by adding an output resistor to each output, but then you increase the output impedance, something you would like to avoid. There are 1 Ω resistors on the PCB, but that's for the original opamps. So here we see a typical example of why you shouldn't just replace opamps without testing. Another type of opamp could oscillate much worse. Yes, they can even self destruct if you're unlucky.

Edit: To summarize, yes, you can improve on a design. Sometimes they have to keep the cost down and use less capable opamps.

So what is the next step? I need to find a better replacement for the output opamps. They need to be able to handle the same amount of output current as the NJM4556 but have better specs. If you have any suggestions, I'm all ears. It could be fun to see if there are further gains (pun intended) to be had with another opamp.
Then it's time to test under actual load. Stay tuned...
 

Attachments

Last edited:
Joined
Nov 4, 2017
Messages
38
Likes
26
Location
East Coast, USA
#2
I swapped out the op-amp on my Little Dot I headphone amp and thought it sounded better. Who knows, I was probably so surprised that I didn't break it that it had to be an improvement. IIRC, I settled on some kind of adapter that transformed a "stereo" op-amp into two mono parts. It was fun and maybe cost $10.
 

KSTR

Active Member
Joined
Sep 6, 2018
Messages
106
Likes
179
Location
Berlin, Germany
#3
[...]
So what is the next step? I need to find a better replacement for the output opamps. They need to be able to handle the same amount of output current as the NJM4556 but have better specs. If you have any suggestions, I'm all ears. It could be fun to see if there are further gains (pun intended) to be had with another opamp.
Then it's time to test under actual load. Stay tuned...
[...]
OPA1688 is a known good headphone output, and the datasheet centers on HP amp as application. In the RME Adi-2 Pro they use 6 paralleled channels per side on +-15V rails, and this headphone output is known to drive even low-Z cans to absolutely insane levels. Two paralleled channels per side should suffice for all normal operating conditions.
 

jackenhack

Active Member
Joined
Oct 25, 2018
Messages
173
Likes
367
Location
Stockholm, Sweden
#4
OPA1688 is a known good headphone output, and the datasheet centers on HP amp as application. In the RME Adi-2 Pro they use 6 paralleled channels per side on +-15V rails, and this headphone output is known to drive even low-Z cans to absolutely insane levels. Two paralleled channels per side should suffice for all normal operating conditions.
Cool! I'll check it out. Could be fun to see how far you can push a design. JD Labs made the excellent Atom amplifier, so they probably did a lot of this and started from scratch.
 

JohnYang1997

Senior Member
Joined
Dec 28, 2018
Messages
401
Likes
119
#5
I did all types of measurements of opamp rolling and modification in circuit last year. I'll share some.
 

JohnYang1997

Senior Member
Joined
Dec 28, 2018
Messages
401
Likes
119
#6
most test used lg quadbeat 3 as load.
In short opa827 is the best.
Also i ended up bypassing the entire input stage directly into the output stage.
changed the volume pot to 1k ohm bourn pots.
changed the output resistor to 0.1ohm. These all lower the distortion and noise of the amp. And it outperforms jds el anp at higher frequency and higher volume.
 

Attachments

JohnYang1997

Senior Member
Joined
Dec 28, 2018
Messages
401
Likes
119
#7
opa1688 is going to be not good. It is not for output except to do like adi2pro'dac.
i will try opa2156 and opa828 in some time soon.
I also tested 1622 evm.
Btw opa1622 is not really very happy with the parallel configuration, offset can be higher than ideal and can get hot if the output resistor is low. Low output impedace of less than 0.1 or 0.05 is needed for reactive load. The load is not linear in real headphones. The amp need to control the behavior of the headphone instead of isolating from resistor. The stability is also important to. So high capacitance cable should be avoid.
 

JohnYang1997

Senior Member
Joined
Dec 28, 2018
Messages
401
Likes
119
#9
and test at 2vrms output is not realistic since you don't really listen at that volume. If you want headroom, the dynamic is limited in the digital domain meaning, 0db is the maximum output. So you can lower the output to 1mw 10mw and test it. That's much more realistic power output.
Also at low volume, noise is actually higher than distortion so lower the noise at 50% pot position is very crucial.
Sarcastically, people like listening their amp at 11-1 oclock position where the noise is maximum. It seems people likes noise. lol.
 

JohnYang1997

Senior Member
Joined
Dec 28, 2018
Messages
401
Likes
119
#10
Cool! I'll check it out. Could be fun to see how far you can push a design. JD Labs made the excellent Atom amplifier, so they probably did a lot of this and started from scratch.
atom is essentially cost down jds el amp. If you really read the distortion harmonics you will find el amp is really better than atom. the atom wins at thd+n only because the much better powersupply noise. which is not existing when i measured mine. amirm measured jds element and the amp section itself. but at 150 ohm.
 

JohnYang1997

Senior Member
Joined
Dec 28, 2018
Messages
401
Likes
119
#11
jds used lm49720 for el amp and njm2068 for atom. same configuration of composite amplifier with lme49600.
Neurochrome hp-1 is essentially the same but using opa1612 and paralleled the whole composite amplifier section. It is hard to get it stable so 1 ohm output resistors are added in each output instead of what i liked 0.1ohm or 0.33ohm.Btw the frequency compensation is used according to the designer in order to get it stable.
 

JohnYang1997

Senior Member
Joined
Dec 28, 2018
Messages
401
Likes
119
#12
one thing to notice is that nwavguy is very lucky/or blinded by measurements to use njm4556 because 1khz is the only frequency it performs well. Distortion rise drastically at lower snd higher frequencies. You don't need that much output current. If you just want an improvement over 4556, muses 8920 is a candidate 100ma to both rails making total 200ma. 4580 can be good too but i haven't tested 4580 yet.
 

jackenhack

Active Member
Joined
Oct 25, 2018
Messages
173
Likes
367
Location
Stockholm, Sweden
#13
Btw opa1622 is not really very happy with the parallel configuration, offset can be higher than ideal and can get hot if the output resistor is low.
Eh, I have a working amplifier with four summoned OPA1622 with only 1 Ω resistor at the output — ten prototypes up and running for over a year. So OPA1622 works perfectly in parallel.
 

JohnYang1997

Senior Member
Joined
Dec 28, 2018
Messages
401
Likes
119
#15
Eh, I have a working amplifier with four summoned OPA1622 with only 1 Ω resistor at the output — ten prototypes up and running for over a year. So OPA1622 works perfectly in parallel.
Did you use Rf instead of directly connect the output and inverting input? Also what about the input resistor? I was in the diyaudio forum 1622 thread their and saw people using that. I didn't validate that but imo the resistors introduce noise probably low as 47 ohm 100ohm is possible.
 

solderdude

Major Contributor
Joined
Jul 21, 2018
Messages
1,068
Likes
1,667
Location
The Neverlands
#16
Some constructive (and less constructive comments)

Are the opamps summoned or summed ? ;) (that was the less constructive one)

I've decided to change the gain opamp to the venerable OPA1612. It's a VERY good bipolar opamp with incredible specs. If changing the opamp to this one doesn't make a difference, nothing will.
Well if this opamp doesn't make the performance better most likely nothing will. But changing it to a lesser one will make a measurable difference.
Perhaps more harmonics are added and even to the amount it may even become audible.

What's the used load resistance ? I reckon it makes a difference in THD certainly when you want to explore the boudaries of the max output currents of op-amps as they usually have current limiters.

Can you show the output noise increases 3dB by removing one the 1 Ohm resistors ?

Consider the OPA551 (DIP version) on a single DIP converter and you can even loose the 1 Ohm.
It can provide 300mA peak on its own.
Not rail-to-Rail though.
 
Top Bottom