This is yet another Composite Chipamp topology, combining ideas from Walt Jung's Composte Opamp/Buffer, Walker & Albinson's current-dumping reactive bridge, and jcx's multiloop opamps posted elsewhere in this forum. It also derives some ideas from my MiniRef topology posted earlier, but drops the Howland Current Pump of the MiniRef in favour of a simple feedforward reactive bridge.
The simulated THD20 FFT plot shown is for 10W into 8R - the simulated distortion numbers are not expected to be accurate, but the notable feature is that H3 and the odd harmonics drop off rapidly, much like the MyRef and MiniRef. Distortion remains vanishingly small until it runs into clipping, at which point it gets ugly (that will need some clamps in the feedback loop to get under control).
The trade-off in practice is that a high-current air-core inductor L1 will be required here, but it completely drops the requirement for tightly-matched resistors for the Howland current pump, and saves on one low-value power resistor. Compensation is as simple as the MiniRef - a single interstage integrator, C5, which also doubles as a component of the reactive bridge consisting of (R6+R8), R5, L1 and C5.
The simulated THD20 FFT plot shown is for 10W into 8R - the simulated distortion numbers are not expected to be accurate, but the notable feature is that H3 and the odd harmonics drop off rapidly, much like the MyRef and MiniRef. Distortion remains vanishingly small until it runs into clipping, at which point it gets ugly (that will need some clamps in the feedback loop to get under control).
The trade-off in practice is that a high-current air-core inductor L1 will be required here, but it completely drops the requirement for tightly-matched resistors for the Howland current pump, and saves on one low-value power resistor. Compensation is as simple as the MiniRef - a single interstage integrator, C5, which also doubles as a component of the reactive bridge consisting of (R6+R8), R5, L1 and C5.
Attachments
Last edited:
The feed-forward current component through R5 is sizable, of the order of 0.5A or higher. The second reason is that +/- 25V rated chipamps are common, while high-voltage opamps are rare. The third reason is that dual chipamps in the same package are now relatively inexpensive and accessible - TDA7265, LM4766, LM4780, etc.
If U2 were a unity-gain buffer (a la Walt Jung), then it's possible in principle to use a high-voltage opamp for U1 and maybe resize R5 higher to keep the feed-forward current below the current limit for U1. The problem is that a lot of chipamps are unstable at unity gain.
Member
Joined 2009
Paid Member
Yup, I'd probably give it a shot with a real TDA7265 eventually - no official Spice model, but relatively inexpensive. It's a ZIP-11 package, which makes it a bit messy for a perf-board, though.
R5 goes from the output of U1 to the output node. From simulation, the current through R5 is almost exactly inverted, but of lower amplitude, than the current through L1 (most of which goes into the load RL).
I've lost you here, partly.
When I export the schematic from LTSpice into .wmf, it doesn't show junctions on wire crossings - so I use a small manual dog-leg to show wires that cross other wires without contact. If that dog-leg is absent, it denotes a junction between wires perpendicular to each other. If in doubt, please download the .asc file and browse it in LTspice.
Yes, you definitely need to present a more readable schematic if you want people to understand so they can comment. From time immemorial, connections have been indicated with a 'dot'. It used to be that the 'dogleg' was also used IN ADDITION TO THE CONNECTION DOT to make things unambiguous. These days, two lines crossing without a dot are assumed to not connect.
It's not an acceptable answer to suggest that I download and install LTspice just for the sake of reading your schematic. Again, you're the one asking folks on this forum for opinions, so make it easy on us.
It's not an acceptable answer to suggest that I download and install LTspice just for the sake of reading your schematic. Again, you're the one asking folks on this forum for opinions, so make it easy on us.
AFAIK, all versions of LTSpice and/or Wine have this bug in the .wmf image export tool. Anyway, I manually added the junctions in GIMP after exporting the image, so that should get rid of the ambiguity for now.
The simulation shown is with the LM675 model (probably a bit more realistic, with higher THD20 numbers), with ~16Vp output amplitude into 8R. The blue trace shows the THD20 FFT with the feedforward resistance R5 set to 1 Mohm (effectively open). The green trace shows the THD20 FFT with R5 set to 15 ohms (the value for balancing the bridge).
H2 is more or less the same, but H3 is lower by about 15 dB. This was similar to the earlier simulation with the LM1875 model, but it was harder to observe there since the LM1875 model was too optimistic (more unrealistic). Some R values have been tweaked to lower the DC offsets as well.
The simulation shown is with the LM675 model (probably a bit more realistic, with higher THD20 numbers), with ~16Vp output amplitude into 8R. The blue trace shows the THD20 FFT with the feedforward resistance R5 set to 1 Mohm (effectively open). The green trace shows the THD20 FFT with R5 set to 15 ohms (the value for balancing the bridge).
H2 is more or less the same, but H3 is lower by about 15 dB. This was similar to the earlier simulation with the LM1875 model, but it was harder to observe there since the LM1875 model was too optimistic (more unrealistic). Some R values have been tweaked to lower the DC offsets as well.
Attachments
Last edited:
That's jcx's idea, mainly for opamps - but of course, it will work with chipamps also. It will require 2 power resistors at the output vs. 1 inductor and medium-power resistor here. The inductor has to be linear, so it essentially has to be air-core. Maybe a Micrometals powdered-iron -2 suffix toroid will also work. Either way, an inductor is more expensive than a resistor for a given current rating, no way around it. However, in some cases an inductor is anyway required for the output series Zobel network, in which case there's no additional cost. This one also has simpler biasing.
It's a wash - prototypes have to be built and evaluated for audible sonics.
Or do a feedforward with a highly-linear DSL line driver chip and a chip amp. Basically have the DSL driver clean up the entire crossover region of the chipamp (and let the latter do the heavy lifting).
The zobel inductor is usually pretty easy to make by wrapping a few turns of magnet wire around a pencil, then again, your stated inductor isn't too egregious either.
The zobel inductor is usually pretty easy to make by wrapping a few turns of magnet wire around a pencil, then again, your stated inductor isn't too egregious either.
That is similar to what Andrew suggested in post #2, and it has some issues which I mentioned in post #3 - current drive of the opamp, lack of too many options for high-voltage opamps, and lack of unity-gain stable chipamps.
What I'm already doing with the MiniRef is a composite amp with a Howland current-pump nested inside an outer voltage-series feedback loop of a small-signal opamp, and no feed-forward. It works well, but it can't do anything about the crossover region of the chipamp, which feed-forward can.
Sounds more like it requires some honest to goodness engineering (and probably a second set of rails) to get to the crux of the problem. 
Or worth trying on a ~10 W amp first. Something like an 1875 driven at +/- 15V would allow you to drive a TPA6120 on the same rails.
Using a mediocre feedforward amp to correct another mediocre amp doesn't seem to be doing you many favors, though.
Or worth trying on a ~10 W amp first. Something like an 1875 driven at +/- 15V would allow you to drive a TPA6120 on the same rails.
Using a mediocre feedforward amp to correct another mediocre amp doesn't seem to be doing you many favors, though.
You're correct, the two chipamps don't have to be the same - I mentioned it as a convenience factor, and it's also what I simulated.
Perhaps a 10W NEC uPC1238 at +/- 15V rails can drive an LM1875, or simply 2x LM1875 (which is available as the LM1876, but it's not a commonly available part). A single STA540 quad chipamp also looks viable for a 2-channel solution.
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.