Ever since I came across Douglas Self's quirky and playful NE5532 parallel op amp amplifier (also see this thread and this one), I've felt the itch to build such a massively parallel ‘OpAmplifier’. However, in the emitter follower output stage of the amplifier instead of the NE5532 I'd like use the slightly better performing LME49720 op amp. I have a few questions that hopefully some people on this forum can help me figure out. I’ll also use this thread to post progress updates.
An overview of the schematic so far has been attached to this post as a PDF:
View attachment 20171020 Schematic.pdf
The DAC I’m using has a balanced analog output, so the amp will be outfitted with a balanced input along these lines:
From left to right, there’s some EMC RF filtering action, followed by non-polarized DC blocking capacitors and an instrumentation amplifier. The DAC that I’m using has an output level of 2Vrms. So, after the instrumentation amplifier, which is configured here with a gain of 2, the ground-referenced signal going into the output stage is 8Vrms max. With my speakers that’s good for about 102dB SPL. As I tend to listen at lower levels, that should be plenty, with room to spare.
I’ll mostly be using surface mount parts; that should also help keep the PCB size within reasonable bounds. I’ve still got a tube with 10 or so LME49990 op amps lying around, but single op amps such as the OPA1611, OPA827 or OPA1641 should also do just fine for U1, U2 and U3. The op amp rails for U1, U2 and U3 are outfitted with a low-pass filter (100Ω thin-film resistor, 22µF polymer tantalum capacitor and a 1µF X7R multilayer ceramic capacitor).
After the balanced input there's an option to hook up a volume control with a unity gain buffer directly behind it (depending on the volume control, R141 might not need to be populated):
I’ve got a relay-based stepped attenuator that could be used here at a later stage. J2 is a 10 position 2.54mm IDC male socket, so that the attenuator board can be connected using a flat ribbon cable. This would let me adjust the system volume outside of the digital domain—somehow I prefer the feel of a solid metal dial over the volume control in Windows.
Finally, we get to the meat of this amplifier in the form of 40 LME49720 dual op amps grouped into 4 arrays:
Each array combines 20 op amp channels followed by 1Ω isolation and current sharing resistors. As Self explains, grouping the op amps like this should make it easier to track down defective ones. 8Vrms into my 6Ω speakers requires 1.3A. I’ve added a few more op amps than strictly necessary so as not to stress them too much. Here, each op amp channel would have to supply 17mA, which is about two-thirds of the typical output current specified in the datasheet.
Each array has its own filtered rails:
Directly next to each op amp is a small 1µF decoupling capacitor.
For stability, at the amplifier output there’s the usual Zobel network and output inductor:
For the rails I’ll probably use a regulated power supply (±17V, each rail approx. 3A).
As you can see, I’ve left out quite a few elements from the Self amp like the built-in options for bridging and paralleling of amplifier boards (not needed here), the unbalanced input (using balanced input from DAC instead), and the DC fault protection mechanism (might still include one, but it doesn’t feel like a big risk here; I will include a MCU that controls a pair output relays to prevent turn-on and turn-off thumps and the like; it would also operate the relay-based volume control mentioned earlier).
Now, for some of the questions:
An overview of the schematic so far has been attached to this post as a PDF:
View attachment 20171020 Schematic.pdf
The DAC I’m using has a balanced analog output, so the amp will be outfitted with a balanced input along these lines:

From left to right, there’s some EMC RF filtering action, followed by non-polarized DC blocking capacitors and an instrumentation amplifier. The DAC that I’m using has an output level of 2Vrms. So, after the instrumentation amplifier, which is configured here with a gain of 2, the ground-referenced signal going into the output stage is 8Vrms max. With my speakers that’s good for about 102dB SPL. As I tend to listen at lower levels, that should be plenty, with room to spare.
I’ll mostly be using surface mount parts; that should also help keep the PCB size within reasonable bounds. I’ve still got a tube with 10 or so LME49990 op amps lying around, but single op amps such as the OPA1611, OPA827 or OPA1641 should also do just fine for U1, U2 and U3. The op amp rails for U1, U2 and U3 are outfitted with a low-pass filter (100Ω thin-film resistor, 22µF polymer tantalum capacitor and a 1µF X7R multilayer ceramic capacitor).
After the balanced input there's an option to hook up a volume control with a unity gain buffer directly behind it (depending on the volume control, R141 might not need to be populated):

I’ve got a relay-based stepped attenuator that could be used here at a later stage. J2 is a 10 position 2.54mm IDC male socket, so that the attenuator board can be connected using a flat ribbon cable. This would let me adjust the system volume outside of the digital domain—somehow I prefer the feel of a solid metal dial over the volume control in Windows.
Finally, we get to the meat of this amplifier in the form of 40 LME49720 dual op amps grouped into 4 arrays:

Each array combines 20 op amp channels followed by 1Ω isolation and current sharing resistors. As Self explains, grouping the op amps like this should make it easier to track down defective ones. 8Vrms into my 6Ω speakers requires 1.3A. I’ve added a few more op amps than strictly necessary so as not to stress them too much. Here, each op amp channel would have to supply 17mA, which is about two-thirds of the typical output current specified in the datasheet.
Each array has its own filtered rails:

Directly next to each op amp is a small 1µF decoupling capacitor.
For stability, at the amplifier output there’s the usual Zobel network and output inductor:

For the rails I’ll probably use a regulated power supply (±17V, each rail approx. 3A).
As you can see, I’ve left out quite a few elements from the Self amp like the built-in options for bridging and paralleling of amplifier boards (not needed here), the unbalanced input (using balanced input from DAC instead), and the DC fault protection mechanism (might still include one, but it doesn’t feel like a big risk here; I will include a MCU that controls a pair output relays to prevent turn-on and turn-off thumps and the like; it would also operate the relay-based volume control mentioned earlier).
Now, for some of the questions:
- For sine-wave testing at 8Vrms into 6Ω the amp would probably need some active or passive cooling added. Accounting for the voltage drop in the RCRC filter I reckon that the amp would need to dissipate somewhere in the region of 15-20W per channel. I wouldn’t trust the bare PCB with the op amps on it to be able to shed more than 4W or 5W. But what about when playing music? What would be a realistic estimate when playing, for example, electronic music or jazz at max volume?
- Should I connect the signal ground (XLR pin 1) at the speaker output (or perhaps somewhere else)? Or should I not link it to the amplifier ground at all, and only connect signal ground to the chassis? I’ve found seemingly opposite advice on this from people that all seem like they know what they’re talking about.
- Where should I ideally connect the grounded leg of R13 (near U3)? Would the connection to ground between R7 and R8 be a good location?
- Would it be worthwhile to nest the output stage in a feedback loop? I noticed that Self did not do this in his design.
Last edited: