Would a speaker have a problem with a >100 kHz oscillation? You often read reports about the op-amp getting very hot even at low output. The energy is generated in the op-amp not the speaker. If you are like me and didn't do this at school, there are some nice explanations on YouTube, search "op amp loop gain".
In my third semester in college, we built an audio amp. For the oral presentation, we each had to present part of the project. I chose to present the stability analysis and the theory behind it. I chose the topic because everybody shied away from it but I found it pretty approachable and had a well developed intuitive understanding of it. I end up using that theory in nearly every block I design in my day job... I focused my education on analog electronics. That's Analog with a capital A...
Yes. The tweeter will have issues with >100 kHz oscillations. That's how some amplifiers have earned the badge, "tweeter eaters". You run a tweeter for a while at 30 W, 100 kHz and it's not going to be real happy about that. Neither will the neighborhood cats, but that's besides the point.
~Tom
Yes. The tweeter will have issues with >100 kHz oscillations. That's how some amplifiers have earned the badge, "tweeter eaters". You run a tweeter for a while at 30 W, 100 kHz and it's not going to be real happy about that. Neither will the neighborhood cats, but that's besides the point.
~Tom
Yes. The tweeter will have issues with >100 kHz oscillations. That's how some amplifiers have earned the badge, "tweeter eaters". You run a tweeter for a while at 30 W, 100 kHz and it's not going to be real happy about that. Neither will the neighborhood cats, but that's besides the point.
~Tom
bad discrete and older amps tended oscillate around the 100KHz regions, Monolithic devices badly implemented are going to oscillate in the lower MHz region, I think if you analyze it closer with stable V levels and such ( and who does?) with a proper output L//R filter and over temp Spike protection, it's a non problem for a higher power tweeter. But I wouldn't want to test it on my drivers.
Last edited:
I do know that once the LM3886 starts oscillating, it draws current like you wouldn't believe. I tripped the 4 A limiter on the lab supply more than once, even though full rail voltage into the 8 ohm load would only draw 3.5 A. I would hate to see what it does on a beefier supply. Especially if "someone" just "happened" to run it in a prototype setup without a mains fuse, because, after all, fuses are for wimps...
~Tom
~Tom
yeah I knew that for quite some time. In my experience more power op-amp oscillations of the catastrophic types ( current limit and smoke )are due to poor power pin decoupling than anything else. maybe you're saying the Zobel was saved by not doing its job effectively.
In my third semester in college, we built an audio amp. For the oral presentation, we each had to present part of the project. I chose to present the stability analysis and the theory behind it. I chose the topic because everybody shied away from it but I found it pretty approachable and had a well developed intuitive understanding of it. I end up using that theory in nearly every block I design in my day job... I focused my education on analog electronics. That's Analog with a capital A...
Yes. The tweeter will have issues with >100 kHz oscillations. That's how some amplifiers have earned the badge, "tweeter eaters". You run a tweeter for a while at 30 W, 100 kHz and it's not going to be real happy about that. Neither will the neighborhood cats, but that's besides the point.
~Tom
I love the way you approach complex analog systems and I'm following this thread with great interest
The zobel is parallel to the feedback loop and the Load. I think it compensates some HF phase shift of the feedback and Load. I know its not the normal explanation. Maybe the best solution is to try to accomplish a combined 0 degree phase shift over the complete bandwidth.
Check out my stability sim in the Improving the LM3886 thread. Posts #66, 67.
The inductance from Vout to the feedback tap point (Lout) sets up a complex pole pair with Cload. The Zobel aims to place a zero around the unity loop gain frequency. This buys you 90 degrees more loop phase so you don't get hit so hard by the complex pole pair. The improvement in PM is pretty significant as you can see in the posts mentioned above.
The resistors in the feedback network should not cause any difficulty for the frequencies we care about. Even a leaded metal film resistor will remain resistive well into the 10s if not 100s of MHz. My measurement was limited to 40 MHz as the HP 4194A I have access to doesn't go any higher. I've seen leaded parts used in radios operating in the 2 m band (145 MHz). That was common place until SMD took over. Now, parasitic inductance in the feedback network is another thing. I think it mostly matters from Vout to IN(-). You can run the sim to confirm.
~Tom
Last edited:
I measured the inductance of a piece of wire as part of a lab exercise in college. I concluded that wire has a self-inductance of about 1 nH/mm. I think the lab results said 0.7 nH/mm, but 1 nH/mm is an easy-to-remember rule of thumb. That's what we use at work as well. It's usually pretty accurate.
I use the same rule of thumb for PCB traces. You could probably get a better estimate by using one of the micro-strip calculators out there, but as 1 nH/mm is likely to be on the conservative side anyway, I usually just stick with that.
~Tom
I use the same rule of thumb for PCB traces. You could probably get a better estimate by using one of the micro-strip calculators out there, but as 1 nH/mm is likely to be on the conservative side anyway, I usually just stick with that.
~Tom
Another LM3886 schematic with data
I have some data I can add to the discussion. I built the circuit based on my past experiences with the chip and used the grounding scheme of "IN+FB Long GND" from post #30.
All data was recorded with the amplifier connected to an 8ohm dummy load and at 10V peak.
The amp is connected to a +/- 30V regulated supply.
Pic 1 - basic circuit
Pic 2 - Bode plot from 10kHz - 1MHz
Pic 3 - 50kHz Square wave response
Pic 4 - Circuit with compensation added
Pic 5 - Bode plot of updated circuit. 10kHz - 1MHz
Pic 6 - 50kHz Square wave response of updated circuit
Pic 7 - photo of the circuit.
I have some data I can add to the discussion. I built the circuit based on my past experiences with the chip and used the grounding scheme of "IN+FB Long GND" from post #30.
All data was recorded with the amplifier connected to an 8ohm dummy load and at 10V peak.
The amp is connected to a +/- 30V regulated supply.
Pic 1 - basic circuit
Pic 2 - Bode plot from 10kHz - 1MHz
Pic 3 - 50kHz Square wave response
Pic 4 - Circuit with compensation added
Pic 5 - Bode plot of updated circuit. 10kHz - 1MHz
Pic 6 - 50kHz Square wave response of updated circuit
Pic 7 - photo of the circuit.
Attachments
Oliphant, your pictures show overshoot on the negative going squarewave without your added compensation. I think if you went with the recommended zobel you would not need your added compensation. I believe Tomchrs so called instability with the "right" zobel is actually heating in the zobel resistor caused by square wave testing at high frequency and amplitude.
My blog
Consort3's Blog | Just another WordPress.com weblog
My blog
Consort3's Blog | Just another WordPress.com weblog
Instability is caused by poor phase margin or poor gain margin. This relates to the amount of open loop gain in the forward path and the gain (or in most cases, rather, attenuation) in the feedback path. Note that the inductance in the output trace and the Zobel network play in here. Lout, Cload form a 2nd order filter in the feedback loop and the Zobel network a zero.
The source of any marginal stability I saw was excessive inductance (long leads) in the Zobel network. Once that was corrected, I had no stability issues at all.
Heating of the Zobel resistor will not cause instability as long as the resistance remains within spec (i.e. as long as the resistor doesn't emit smoke).
National used 2.7R, 100nF = 589 kHz. I used 15R, 22 nF = 482 kHz. Douglas Self mentions in his Power Amps book that 100 nF, 10R = 159 kHz tends to be the sweet spot, but -- somewhat uncharacteristically does not back this up with data. From my own simulations, I confirmed that the whole point of the Zobel network is to tame the complex pole pair set up by Lout and Cload. I would need to look at it in more detail, but I would think that there is a fairly wide range of RC combinations that will work just fine here.
Cordell doesn't make a big fuzz about the Zobel either... I think he was the one pointing out that the Zobel basically works as a snubber and needs to be designed for a reasonable Q. That makes sense...
PS: I should note that I'm referring to Douglas Self, "Power Amplifiers Design Handbook", 3rd edition as that is the one I have available. It is possible that he's added more material in later editions.
Do you have the full citation (author, title, journal, year) or a link to this article? I would be very interested in reading it.
~Tom
Last edited:
In Thiele's original article[1] he designs the filter for as low a cutoff frequency as possible. His design criterion is <0.1 dB of attenuation at 15 kHz. He aims to keep the load impedance "seen" by the amp constant as function of frequency even with capacitive load.
[1]: Thiele, A.N., "Load Circuit Stabilizing Networks for Audio Amplifiers", AES 1/1976, Vol 24. No. 1, pp 20-23.
[1]: Thiele, A.N., "Load Circuit Stabilizing Networks for Audio Amplifiers", AES 1/1976, Vol 24. No. 1, pp 20-23.