Hooray! Everyone is getting together and learning. This is what makes the internet such an amazing resource, and almost makes up for all the bad stuff it can generate.
Tom, I hope that among all your other projects you can build a MyRef and let us know results of your measurements (I agree that they're important) as well as listening impressions (which are also important).
Peace,
Tom E
Tom, I hope that among all your other projects you can build a MyRef and let us know results of your measurements (I agree that they're important) as well as listening impressions (which are also important).
Peace,
Tom E
I've committed to Fremen that if he sends me boards, I'll have the boards (or at least a board) assembled by the end of summer (so early September). He requested that I followed the build instructions. I will do that. It is a completely reasonable request. No worries there.
I'm curious as well. Ahhh..... Data. Useful for making design decisions.
Cheers.
Tom
I'm not aware of any replacement for the LM4780. I do have ten LM4780 in a tube from Newark if you need one (PM/email me if interested), but no more boards.
When I designed the Parallel-86, I did try two LM3886es in parallel as that's much more convenient mechanically than the LM4780. As the LM4780 is basically two LM3886es in one package, I figured the two circuits should be similar. They were not. The THD with two LM3886es was considerably higher than with the two channels of the LM4780 in parallel. I suspect two LM3886es have higher mismatch between their offset voltages. I don't know if the two channels of the LM4780 are tighter matched or if the fact that they're closer to the same temperature helps out. Whatever the root cause, I decided to move on and just use the LM4780 as it provided the best performance and tightest PCB layout.
Now that the LM4780 is being obsoleted by TI, the story changes. I'll still need to find a solution to get the two channels to play nicely. I have some ideas, but we'll see. That's a back burner item right now.
Tom
I'm looking at various medium power, low noise, low distortion amplifiers for my next project. My speakers have fairly high impedance and sensitivity (8-16 ohm, 96-99 dB). I currently run them in a hybrid configuration with an active crossover to the basses and a passive between tweeter and midrange, but I want to go fully active and get rid of the passive crossovers. The basses will be driven by Hypex NCore NC400/SMPS600, but that seems overkill power-wise for the mids and tweeters.
I have a "lifetime supply" of LM4780's in a drawer, and I might end up printing a small run of PCB's for my own use. For space reasons, I also want to run the new amplifiers off SMPS's. Given the moderately high efficiency of the speakers, I want to run the amplifiers at 17-20 dB voltage gain (preferrably without stability issues!)
I find the composite amplifier approach very interesting, and I'm trying to understand the design tradeoffs between the differential input, single ended Modulus-86 and the fully differential, bridge mode TP Sympatico. I understand the current limitations of a bridged amplifier vs a paralleled one, but also the benefit of drawing on both sides of a power supply and the built-in reduction of even order distortion products. I have seen the extremely impressive measurements of the Modulus-86 and its late parallel sibling, but I have not been able to find similar measurements of the Sympatico. My first guess would be that these should be similar within the voltage and current envelope where I will be operating.
Is it possible to say anything in general about the pros and cons of these two approaches if peak current delivery is not really an issue, but THD+N and gain structure is important?
I have a "lifetime supply" of LM4780's in a drawer, and I might end up printing a small run of PCB's for my own use. For space reasons, I also want to run the new amplifiers off SMPS's. Given the moderately high efficiency of the speakers, I want to run the amplifiers at 17-20 dB voltage gain (preferrably without stability issues!)
I find the composite amplifier approach very interesting, and I'm trying to understand the design tradeoffs between the differential input, single ended Modulus-86 and the fully differential, bridge mode TP Sympatico. I understand the current limitations of a bridged amplifier vs a paralleled one, but also the benefit of drawing on both sides of a power supply and the built-in reduction of even order distortion products. I have seen the extremely impressive measurements of the Modulus-86 and its late parallel sibling, but I have not been able to find similar measurements of the Sympatico. My first guess would be that these should be similar within the voltage and current envelope where I will be operating.
Is it possible to say anything in general about the pros and cons of these two approaches if peak current delivery is not really an issue, but THD+N and gain structure is important?
Last edited:
Being brutally honest, if you cannot work out the trade-offs for yourself, a home designed composite is not for you. They are challenging to get stable and well behaved.
For that sensitivity and and active do you need more than 40W? Realistically is 40W is not enough, then the next step is 400W to get any meaningful increase in volume as 10dB is 'roughtly' twice as loud.
For that sensitivity and and active do you need more than 40W? Realistically is 40W is not enough, then the next step is 400W to get any meaningful increase in volume as 10dB is 'roughtly' twice as loud.
Appreciate the direct feedback.
Being an EECS PhD from MIT (but in a different subject than audio circuits) I do have some confidence in my ability to learn things I may not already know. This seems like a reasonable challenge and a good excuse to run out and buy some test equipment.
I actually estimate that I need approx 3 W for the mids and 1 W for the tweeters, peak voltage perhaps 10 V (mids), peak current into 16 ohms - not much.
But I thought it would be interesting to hear Tom's view on this, beyond the current limitation he already commented on.
Being an EECS PhD from MIT (but in a different subject than audio circuits) I do have some confidence in my ability to learn things I may not already know. This seems like a reasonable challenge and a good excuse to run out and buy some test equipment.
I actually estimate that I need approx 3 W for the mids and 1 W for the tweeters, peak voltage perhaps 10 V (mids), peak current into 16 ohms - not much.
But I thought it would be interesting to hear Tom's view on this, beyond the current limitation he already commented on.
Last edited:
Ah fairy nuff. Sorry, you sometimes have to make a snap judgement on how to pitch an answer. This does result in foot in mouth .You should get within 10dB of the mod-86 performance within a few months. You read Walt Jung's early papers on composite and it seems easy. Talking to Tom and Twest820 I realised I had no chance so just bought the boards. You should have more luck.
.You should get within 10dB of the mod-86 performance within a few months. You read Walt Jung's early papers on composite and it seems easy. Talking to Tom and Twest820 I realised I had no chance so just bought the boards. You should have more luck.
The references listed at the bottom of my Modulus-86 page should get you started. You may also want to brush up on stability analysis and frequency compensation as those subjects will come in handy. In addition to regular small-signal stability, you need to look at stability as the output swings near the supply rails. As the driver transistor (VAS) in the LM3886 approach saturation, the gain plummets, which causes trouble in the overall loop response.
I suggest that anyone who is considering bridging two LM3886es or the two halves of an LM4780 to read the LM3886/LM4780 data sheet and do the math on the output current. The LM3886 is guaranteed to be able to deliver 7 A. Put two of them in series (bridge) and you still only get 7 A. This means a bridged amp will not be able to deliver more power than a single LM3886, which defeats the purpose of bridging.
Put the two halves of an LM4780 in parallel and you'll get 14 A of output current. That's more fun. My Parallel-86 amp is able to drive a 2 Ω load at 225 W with vanishingly low THD. Unfortunately I can no longer deliver boards for that project as I can't get the LM4780s anymore. The LME49710NA used in that project has been discontinued by TI as well. Grr...
I'd be willing to make an R1.1 of the Parallel-86 where the DIP-8 footprint for the LME49710 is switched to a TO-99 (the LME49710HA is still in production) and get another board order going if I can get eight boards pre-ordered. PM me if interested. You'd be on your own for the LM4780, though.
Putting two LM3886es in parallel is not trivial if you want good performance. You need to minimize or match their input offsets to ensure that they don't set up a current from one amp to the other. That's easy to do with a DC servo, but that also sets up a zero at DC, which breaks the composite loop, so you need to find a different solution. It's a solvable problem. I solved it and I'm sure you can too.
Best of luck.
Tom
I suggest that anyone who is considering bridging two LM3886es or the two halves of an LM4780 to read the LM3886/LM4780 data sheet and do the math on the output current. The LM3886 is guaranteed to be able to deliver 7 A. Put two of them in series (bridge) and you still only get 7 A. This means a bridged amp will not be able to deliver more power than a single LM3886, which defeats the purpose of bridging.
Put the two halves of an LM4780 in parallel and you'll get 14 A of output current. That's more fun. My Parallel-86 amp is able to drive a 2 Ω load at 225 W with vanishingly low THD. Unfortunately I can no longer deliver boards for that project as I can't get the LM4780s anymore. The LME49710NA used in that project has been discontinued by TI as well. Grr...
I'd be willing to make an R1.1 of the Parallel-86 where the DIP-8 footprint for the LME49710 is switched to a TO-99 (the LME49710HA is still in production) and get another board order going if I can get eight boards pre-ordered. PM me if interested. You'd be on your own for the LM4780, though.
Putting two LM3886es in parallel is not trivial if you want good performance. You need to minimize or match their input offsets to ensure that they don't set up a current from one amp to the other. That's easy to do with a DC servo, but that also sets up a zero at DC, which breaks the composite loop, so you need to find a different solution. It's a solvable problem. I solved it and I'm sure you can too.
Best of luck.
Tom
Thanks!
Given 16 ohm, 96 dB+ speakers, current is not really a constraint, but noise at idle and undistorted voltage swing will be more important. My mids are PHL 1130 and 3440. If I force 1 A of current through them, I think the ceiling will be coming down. Also, I'll hit the +/- 25 V rail voltages long before 7 A into 16 ohms. The tweeters are Beyma TPL150's. "Only" 5 ohm impedance, but 99 dB sensitivity and an impedance plot almost as flat as a 5 ohm resistor. More than 7 A into them is not going to happen, either. The basses are 2x 8 ohm Beyma 12P80Nd in a cardioid configuration, BTW, but they are driven by a pair of Hypex NC400's since the power and current needs are quite different in the bass. I'm only looking at the low mid and up here.
The noise floor from the signal processor/crossover (DEQX HDP4) is high enough that I'll pay close attention to the gain structure in the chain, so low voltage gain in the amps is also important. And I understand that low gain and high load inductance could lead to stability issues, and that Zobel networks on the outputs are a pretty good idea.
Reading the datasheet for the LM4780 and the AN-1192 application note, it seems that the bridged solution starts out with slightly lower noise and distortion than the parallel solution. I was wondering if that might carry over to a "first watt" advantage in the composite circuit. Also, using the OCM reference of an OPA1632 or THS4131 to handle DC offset seems like an elegant solution to me. The part I do not like about the Sympatico boards is that they have on-board power supplies. I'd prefer to use a separate SMPS, for instance the Connex Electronics SMPS300RE, which also gives a regulated +/- 15 V driver supply. That would allow an extremely small amplifier board.
I'll give it some more thought. If I give up, there will be an order for a few Modulus-86's coming.
Given 16 ohm, 96 dB+ speakers, current is not really a constraint, but noise at idle and undistorted voltage swing will be more important. My mids are PHL 1130 and 3440. If I force 1 A of current through them, I think the ceiling will be coming down. Also, I'll hit the +/- 25 V rail voltages long before 7 A into 16 ohms. The tweeters are Beyma TPL150's. "Only" 5 ohm impedance, but 99 dB sensitivity and an impedance plot almost as flat as a 5 ohm resistor. More than 7 A into them is not going to happen, either. The basses are 2x 8 ohm Beyma 12P80Nd in a cardioid configuration, BTW, but they are driven by a pair of Hypex NC400's since the power and current needs are quite different in the bass. I'm only looking at the low mid and up here.
The noise floor from the signal processor/crossover (DEQX HDP4) is high enough that I'll pay close attention to the gain structure in the chain, so low voltage gain in the amps is also important. And I understand that low gain and high load inductance could lead to stability issues, and that Zobel networks on the outputs are a pretty good idea.
Reading the datasheet for the LM4780 and the AN-1192 application note, it seems that the bridged solution starts out with slightly lower noise and distortion than the parallel solution. I was wondering if that might carry over to a "first watt" advantage in the composite circuit. Also, using the OCM reference of an OPA1632 or THS4131 to handle DC offset seems like an elegant solution to me. The part I do not like about the Sympatico boards is that they have on-board power supplies. I'd prefer to use a separate SMPS, for instance the Connex Electronics SMPS300RE, which also gives a regulated +/- 15 V driver supply. That would allow an extremely small amplifier board.
I'll give it some more thought. If I give up, there will be an order for a few Modulus-86's coming.
Last edited:
The Zobel networks (both R+C and L||R) are mandatory for stability with an even mildly capacitive load.
Interesting observation. In a composite amp, the noise and distortion will be determined by the op-amp up front, though. This means the Sympatico is dominated by the noise and THD of the OPA1632 just as my Modulus-86 is dominated by the noise and THD of the LME49710. The PCB layout plays a big role in the THD and hum performance as well.
Not likely. My guess is that the primary driver in the Sympatico design was the "balanced" circuit. I put balanced in quotes as the power section is actually not balanced. It is two single ended circuits running in opposite phase. That's as close to balanced as you can get with available parts. I'm not dissing it. I'm just pointing out that it isn't as balanced as many would like to believe.
It would be interesting to see measurements of it.
I've tested the Modulus-86 with the SMPS300RE. It performs as well on the SMPS as it does on a linear supply or a well-regulated lab supply (HP 6643A).
Taking the supply off the main amp board makes it much easier to avoid hum from the charging pulses going into the supply caps. You can mitigate that some with good layout, but my preference is still to have the supply separate both for flexibility and to avoid hum.
I'll give it some more thought. If I give up, there will be an order for a few Modulus-86's coming.
Works for me
You know where to find me. With your high-efficiency speakers, there's really no need for the high output power of a parallel amp. I'd optimize for noise and THD in your design.Tom
Thanks again. I was thinking that with high impedance, high inductance speaker drivers (many windings on the voice coil) and short cable runs, the Thiele network (L||R) might not be necessary, but perhaps it is. If so, I'd probably place it off the amp board to save on space there.
To me, there is something attractive about the symmetric nature of the Sympatico, and the bridge mode favors voltage swing over current delivery, which is fine for my application. But I have noted your comment that "the layout is the circuit" for these things, and would really like to see measurements of that one as well.
My guess is that the differential design has potential to be at least as good as the Mod-86. There is lower THD+N in both the bridged LM3886/LM4780 vs a single or paralleled chip and in the OPA1632 vs the LM49710, albeit at very low levels. The voltage noise specification is 1,3 vs 2,7 nV Hz^(-1/2) and the minimum THD+N 0,000022 % vs 0,00003 % at 3-4 V and 1 kHz into 2 kOhm load, but the 20 kHz THD+N is about the same for both at 0,00009 % and 0,00011 %. Using a LM4562 as a differential input buffer instead of the THAT1200 could also drop the distortion by a few points (0,00003 vs 0,0005 % data sheet headline specs). Probably not audible differences, but still. The circuit can be made simpler and more compact too, given that the OCM feature in the OPA1632 seems to eliminate the need for an elaborate DC servo. A 1" x 2" 100 W amplifier board - how cool is that?
Still, the likelihood that a design is able to reach its ultimate performance potential without extensive build-measure-build-cycles seems pretty low. The few subjective listening comparisons that have been described on this forum for the Sympatico vs Mod-86 seem to indicate this as well. There should really not be any audible difference between them, but the Mod-86 is described as more "detailed" and the Sympatico as "warmer", which could indicate significant differences in distortion levels.
But it might be a lot cheaper for me just to buy some amplifiers than to acquire an Audio Precision rig and equipment for PCB production.
To me, there is something attractive about the symmetric nature of the Sympatico, and the bridge mode favors voltage swing over current delivery, which is fine for my application. But I have noted your comment that "the layout is the circuit" for these things, and would really like to see measurements of that one as well.
My guess is that the differential design has potential to be at least as good as the Mod-86. There is lower THD+N in both the bridged LM3886/LM4780 vs a single or paralleled chip and in the OPA1632 vs the LM49710, albeit at very low levels. The voltage noise specification is 1,3 vs 2,7 nV Hz^(-1/2) and the minimum THD+N 0,000022 % vs 0,00003 % at 3-4 V and 1 kHz into 2 kOhm load, but the 20 kHz THD+N is about the same for both at 0,00009 % and 0,00011 %. Using a LM4562 as a differential input buffer instead of the THAT1200 could also drop the distortion by a few points (0,00003 vs 0,0005 % data sheet headline specs). Probably not audible differences, but still. The circuit can be made simpler and more compact too, given that the OCM feature in the OPA1632 seems to eliminate the need for an elaborate DC servo. A 1" x 2" 100 W amplifier board - how cool is that?
Still, the likelihood that a design is able to reach its ultimate performance potential without extensive build-measure-build-cycles seems pretty low. The few subjective listening comparisons that have been described on this forum for the Sympatico vs Mod-86 seem to indicate this as well. There should really not be any audible difference between them, but the Mod-86 is described as more "detailed" and the Sympatico as "warmer", which could indicate significant differences in distortion levels.
But it might be a lot cheaper for me just to buy some amplifiers than to acquire an Audio Precision rig and equipment for PCB production.
Last edited:
With a short wire run and a voice coil driver you're probably OK without the Thiele network. With longer wire runs or electrostatic speakers you definitely need the Thiele network. I've always found that winding an inductor is preferable to replacing an expensive speaker, so that's what I recommend.
You should in theory get some harmonic cancellation from the symmetric circuit. That would lower THD. You also get higher output current as each amp half "sees" half the load impedance. Higher output current --> higher THD. Which effect is strongest? Does it matter in a composite? Those are both good and relevant questions that could be answered with measurements.
I agree. It has the potential - at least for 8 Ω*loads, assuming ±28 V supply rails. With a 4 Ω*load, each half of the LM4780 "sees" a 2 Ω*load. To stay within the 7 A output current limit, you need to lower the output swing to 14 V peak, so you're looking at ±15-16 V supply rails. In my view, that's why bridging LM3886es or LM4780 halves is unattractive.
The OPA1632 has lower voltage noise and lower current noise than the LME49710, so if the opamp is the dominant noise source (it isn't in the MOD86), you will see an improvement.
The OPA1611 has lower noise as well. While the OPA1611 is unity-gain stable, it has a really nasty phase wobble around 1 MHz where its internal compensation kicks in. That may disqualify it as a candidate for a composite amp based on the LM3886. Indeed I have gotten some rather interesting behaviour in the MOD86 when I tried the OPA1611. Fun begins when the output voltage approaches clipping and the gain of the LM3886 drops. That seems to slide the crossover frequency down near than phase wobble and suddenly you have a power oscillator.
The "Composite Game" is a game of loop gain. You want as much loop gain at 20 kHz as you can get without running into stability problems. The loop gain at 20 kHz determines the THD at 20 kHz (assuming you're using a modern opamp front-end).
The THD+N near DC is determined by the noise floor or the THD of the op-amp, whichever is higher.
In the MOD86, the THAT1200 brings the THD+N at frequencies below a few kHz up above the noise floor of the APx525, even if just barely. That's due to the noise of the THAT1200, which is actually a pretty quiet chip.
More importantly, it drops the noise, thus improving the THD+N. The raw THD (no +N) of the THAT1200 is 0.0000xx % as you can see from my THAT Receiver.
Usually "warmer" implies some amount of second order distortion. It could also be that any residual hum from the on-board rectifier and caps in the Sympatico sets up some 60 Hz IMD products that some perceive as warmth. Measurements would reveal this in about ten seconds on the AP...
If precision is what you're after, the MOD86 is second to none. I can, of course, easily be accused of bias so the usual disclaimers apply. That said, it's the most precise amp I've ever built. That was my design goal, so I'm happy.
But it might be a lot cheaper for me just to buy some amplifiers than to acquire an Audio Precision rig and equipment for PCB production.
A full-fledged stereo MOD86 amp can be built for about $400, including chassis, connectors, and everything. That's 30 dB less than the cost of an APx525.
Tom
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.