Textbooks are expensive. This one is worth every penny, though. There are also these organizations called "libraries". You can go to check out the books for free... Many libraries can request texts from college/university libraries. It's a pretty neat concept. I suggest looking into that.
For search terms, try "phase lead compensation", "phase lag compensation", "dominant pole compensation", and "loop gain reduction". Those are the four main stability tricks.
I typed "opamp compensation" into Bing and even though I mistyped 'compensation' I found this: http://www.intersil.com/content/dam/Intersil/documents/an94/an9415.pdf
I haven't read it in detail but it looks pretty useful from a first glance.
~Tom
Thanks for the search words.
I have saved the intersil pdf.
Similar to another I have read, but using a different set of words.
I have saved the intersil pdf.
Similar to another I have read, but using a different set of words.
Member
Joined 2009
Paid Member
Nice measurements.
my 2c - Providing some well balanced advice on the use of non-inverting configuration should be a sticky in this forum
The T-network does nothing for the phase/gain margins. To get higher PM at lower gain, you need phase compensation of some sort.
~Tom
~Tom
@Tom. The answer I gave does not differ from the question in any way; however, practicalities do indeed differ.
I didn't mean to say that Inverting Mode T-Net config would hit the 10x target, but rather what I meant to say is that it will get *somewhat closer* to the target gain setting.
Stability rules--arbitrary gain figure desires are not valid, except for discrete design which is conveniently not sealed shut in a tiny black box so you can feel free to make the needed (internal) adjustments, such as multi-compensation.
However, as for a chip, we can choose low (inverting) or high (non-inverting), but after that point it is most responsible to dial in whatever gain the chip happens to want to optimize it, instead of any sort of pre-determined arbitrary figure that won't be optimal.
@AndrewT: I'm sorry to report that if you really do want exactly 10x along with an output suited to high power at point blank range desktop use, then you really don't want a chip. It would be something to avoid. Yes, we can *suggest* that it use lower gain; however, if you go all the way down to 10x, the result is very likely to have an impractical tone.
Nevertheless, we certainly can try.
T-net Inverting Mode is a fair head start (it suggests low gain range), and then we can oversize the decouplers, such as 1500u Panasonic tallboy, same as audiosector or the alternative of 680u||680u per each rail in addition to ONE of Cornell Dublier Mallory SEK 4.7uF 250v set from V+ to V- at the amplifier board (the alternative disposes of less audio without disposing of less noise). That's about the best I can do to request of the LM3886 that she not sound like a female auctioneer suffering pms and recent breakup, and you can feel free to also deploy the blarebuster input mod. However, after that point, we would need to dial in the gain for practical results, not necessarily at and probably not at 10x. If this really must be done at 10x, then, in addition to the above practical steps, we must deploy either a BSC (a speaker crossover notch filter bypassed by a resistor) or Lenard Audio's ideas for Variable Current Drive, so as to dump, unload and de-power that upper midrange / lower treble peak that is a commonplace Achilles heel of solid state amplification and even worse with chips and the very worst with Overture series chips.
I still don't think that we can keep your ears intact if we do this, not even if we succeed. Indeed an inverting LM675 can meet your gain target easily and 16W is quite loud enough at point blank range of a desktop speaker. Indeed if you need more power at point blank range, then I suggest the SSA (with the optional CCS and latfets--like Nico's build).
P.S.
I suspect that AndrewT already knows everything about this topic and a great deal more. SO, just to be sporting, I suggest Edmund style compensation via capdiv. Like all other compensation attempts with a power-op-amp, this will not help you set lower gain. There's nonesuch. However, it could help it sound more attractive, and that would be practical. Indeed that might be necessary, considering 60W at point blank range with a chip. Ah, no matter how much care and no matter how much tech is put into that build, please don't really do it--I'm sure you'd prefer to avoid the whistle that never stops. When I think about this, I come up with Inverting LM675 or SSA, as somewhat safer alternatives for desktop monitor use. EDIT: Even in the case of a better amp, a good BSC is desirable to de-power the speaker at ear sensitivity peaks when used high power at point blank range. I actually don't know how to make that safe, but we could try.
I didn't mean to say that Inverting Mode T-Net config would hit the 10x target, but rather what I meant to say is that it will get *somewhat closer* to the target gain setting.
Stability rules--arbitrary gain figure desires are not valid, except for discrete design which is conveniently not sealed shut in a tiny black box so you can feel free to make the needed (internal) adjustments, such as multi-compensation.
However, as for a chip, we can choose low (inverting) or high (non-inverting), but after that point it is most responsible to dial in whatever gain the chip happens to want to optimize it, instead of any sort of pre-determined arbitrary figure that won't be optimal.
@AndrewT: I'm sorry to report that if you really do want exactly 10x along with an output suited to high power at point blank range desktop use, then you really don't want a chip. It would be something to avoid. Yes, we can *suggest* that it use lower gain; however, if you go all the way down to 10x, the result is very likely to have an impractical tone.
Nevertheless, we certainly can try.
T-net Inverting Mode is a fair head start (it suggests low gain range), and then we can oversize the decouplers, such as 1500u Panasonic tallboy, same as audiosector or the alternative of 680u||680u per each rail in addition to ONE of Cornell Dublier Mallory SEK 4.7uF 250v set from V+ to V- at the amplifier board (the alternative disposes of less audio without disposing of less noise). That's about the best I can do to request of the LM3886 that she not sound like a female auctioneer suffering pms and recent breakup, and you can feel free to also deploy the blarebuster input mod. However, after that point, we would need to dial in the gain for practical results, not necessarily at and probably not at 10x. If this really must be done at 10x, then, in addition to the above practical steps, we must deploy either a BSC (a speaker crossover notch filter bypassed by a resistor) or Lenard Audio's ideas for Variable Current Drive, so as to dump, unload and de-power that upper midrange / lower treble peak that is a commonplace Achilles heel of solid state amplification and even worse with chips and the very worst with Overture series chips.
I still don't think that we can keep your ears intact if we do this, not even if we succeed. Indeed an inverting LM675 can meet your gain target easily and 16W is quite loud enough at point blank range of a desktop speaker. Indeed if you need more power at point blank range, then I suggest the SSA (with the optional CCS and latfets--like Nico's build).
P.S.
I suspect that AndrewT already knows everything about this topic and a great deal more. SO, just to be sporting, I suggest Edmund style compensation via capdiv. Like all other compensation attempts with a power-op-amp, this will not help you set lower gain. There's nonesuch. However, it could help it sound more attractive, and that would be practical. Indeed that might be necessary, considering 60W at point blank range with a chip. Ah, no matter how much care and no matter how much tech is put into that build, please don't really do it--I'm sure you'd prefer to avoid the whistle that never stops. When I think about this, I come up with Inverting LM675 or SSA, as somewhat safer alternatives for desktop monitor use. EDIT: Even in the case of a better amp, a good BSC is desirable to de-power the speaker at ear sensitivity peaks when used high power at point blank range. I actually don't know how to make that safe, but we could try.
Last edited:
For a desktop computer speaker application, I'd probably opt for one of the buffer ICs such as the LME49600. I suspect one is probably sufficient to drive a speaker to a reasonably loud volume. Put a few in parallel if you want more oomph.
Maybe I'm misunderstanding your "T-net" thing. I'm thinking of the topology commonly used to create a high gain circuit or to lower the impedances in the feedback network to combat issues related to high op-amp input capacitance. That circuit is a T-network from the op-amp output to the op-amp input forming the feedback. It's basically like the regular resistive feedback network with a voltage divider before the feedback resistor. Is that what you're referring to?
Fig. 3.10 in this .pdf is what I'm talking about. That circuit, if used with an LM3886, does nothing to address stability issues related to gains below 10 V/V. The loop gain/phase is the same with the T-network as it would be with the regular feedback network. The only advantage of the T-network is that it reduces the impedances in the feedback network. This may help in high-gain situations where the input capacitance of the op-amp along with the high resistance feedback resistor would cause a pole in the feedback transfer function, dorking up stability.
If Andrew wants PM > 80 deg at gains below 20 V/V, he'll need to apply phase lead/lag compensation to the LM3886. No T-network will work around that fact.
~Tom
Maybe I'm misunderstanding your "T-net" thing. I'm thinking of the topology commonly used to create a high gain circuit or to lower the impedances in the feedback network to combat issues related to high op-amp input capacitance. That circuit is a T-network from the op-amp output to the op-amp input forming the feedback. It's basically like the regular resistive feedback network with a voltage divider before the feedback resistor. Is that what you're referring to?
Fig. 3.10 in this .pdf is what I'm talking about. That circuit, if used with an LM3886, does nothing to address stability issues related to gains below 10 V/V. The loop gain/phase is the same with the T-network as it would be with the regular feedback network. The only advantage of the T-network is that it reduces the impedances in the feedback network. This may help in high-gain situations where the input capacitance of the op-amp along with the high resistance feedback resistor would cause a pole in the feedback transfer function, dorking up stability.
If Andrew wants PM > 80 deg at gains below 20 V/V, he'll need to apply phase lead/lag compensation to the LM3886. No T-network will work around that fact.
~Tom
Last edited:
I just fixed my HP3577A where the AC input filter was short circuit. The filter box is replaced by a similar one. Loop gain measurement can be performed again!
I have been using a current injection probe proposed in Injection Transformers for Closed-Loop DC-DC Converter Network Analysis - Application Note - Maxim to make measurement. Any comment?
PS. Photo shows the original filter box.
Attachments
Cool. I have some ferrite cores. I'll need to play with those for a bit... I wonder if a balun would work. RF baluns are pretty wide band and not that expensive. Hmmm... Must experiment.
~Tom
~Tom
I decided to start a new thread in the Tools & Equipment forum for Loop Gain Measurement Apparati.
~Tom
~Tom
Hope this link is not out of place.
http://www.diyaudio.com/forums/chip-amps/163385-so-just-how-good-can-chip-amp-9.html#post2148761
http://www.diyaudio.com/forums/chip-amps/163385-so-just-how-good-can-chip-amp-9.html#post2148761
To me, I care more the upper bandwidth where we need to see phase response for stability, loop gain vs compensation cap value and methods, etc.
Folks,
It's worthwhile to note that the THD+N performance of the LM3886 is very dependent on the power supply. I've posted a bunch of measurements in the LM3886 P2P vs PCB thread.
~Tom
It's worthwhile to note that the THD+N performance of the LM3886 is very dependent on the power supply. I've posted a bunch of measurements in the LM3886 P2P vs PCB thread.
~Tom
Try AC-referencing everything to the neg rail instead of a center GND, sound and measurements will improve big time ;-)
A quick glance at the LM3886 DS plot for PSRR will reveal the answer. The -ve rail PSRR sucks big-time (around 50dB at 20kHz). whilst the +ve rail looks really decent (>100dB) over the audible bandwidth.
- Status
- Not open for further replies.