I feel for the folks who started this thread -- they didn't realize it would be taken to a different level.
Will bring it down to something more prosaic -- if you take the basic Nat Semi (now TI) circuit and follow the imperatives in the data sheet with respect to signal routing you'll have a pretty good sounding device. If you take Bob's circuit with or without the compensation it goes to an entirely better level of performance.
For the record, it's a lot easier to measure the output impedance of an amplifier (vs frequency) and phase angle to measure stability -- no loop breaking necessary.
Re small signal analysis on a power device () who's model is it? was it extensively proven with lots of measured data?
As a simple test, couldn't one very slowly lower the closed loop gain to some point where it just starts to break into oscillation, maybe after the input was perturbed ( DC offset step )? would the result of one test case prove anything regards model?
As a simple test, couldn't one very slowly lower the closed loop gain to some point where it just starts to break into oscillation, maybe after the input was perturbed ( DC offset step )? would the result of one test case prove anything regards model?
using the TINA model from TI it appears -- TINA seems to be a very nice system. A full featured version is available.
I built my own model for the LM3886 and as it compared well to the one TI published, I feel more comfortable with the one the manufacturer puts out.
My bad... 😱 🙂
True. If you have the amp rigged in an inverting configuration, you can also measure the error voltage (voltage at the inverting pin of the op-amp) as you sweep the frequency. The magnitude is the inverse of the loop gain (so multiply the numbers in dB by -1 to get the loop gain). As I recall, the phase measured is the PM.
~Tom
Last edited:
The model I used is the one published by TI. See the LM3886 product page.
The models published by the manufacturer are not perfect, but they come pretty darn close to reality. See my comments about model validity in Post #66. They obviously can't include the effects of board parasitics, supply impedance, etc. as those vary widely from application to application. You as the designer of the end application can model your board parasitics and include them in the simulation to get even closer to reality.
Lowering the closed loop gain until the point where the amp oscillates would give you the gain margin. You could make an educated guess about the phase margin from measurement of the overshoot on the step response. The challenge is to raise the closed loop gain without causing the poles and zeros in the feedback network to shift in frequency.
~Tom
Could be due to a number of reasons.
here's some:
Too many components.
Too easy for the operator to mess it up.
Too difficult to find the advice in a JLH project.
Not easily provable to 100% of the listeners, that squaring the squarewave is equal to good sound reproduction.
Not many builders accept that squarewave testing is applicable to audio amplifiers.
Not many builders are aware that overshoot can give rise to audible differences in sound reproduced.
here's some:
Too many components.
Too easy for the operator to mess it up.
Too difficult to find the advice in a JLH project.
Not easily provable to 100% of the listeners, that squaring the squarewave is equal to good sound reproduction.
Not many builders accept that squarewave testing is applicable to audio amplifiers.
Not many builders are aware that overshoot can give rise to audible differences in sound reproduced.
I am really disappointed!!! I thought we're going to see if filtering a signal using the feedback loop was a bad thing to do. Then I thought we're all going to find out if we can do without the optional components. I didn't even get to see if the chip oscillates or not. I know when Jimi is oscillating. Now Daniel wants all amps to have dials. I can see a global chain of AMP SERVICE STATIONS, where you bring in your amps to give them their annual tune-up.
Please ignore any of the above that didn't make you laugh.
All I wanted to ask, does the TI model of the LM3886 have pin 7 missing?
otherwise great thread.
Please ignore any of the above that didn't make you laugh.
All I wanted to ask, does the TI model of the LM3886 have pin 7 missing?
otherwise great thread.
Yes. Pin 7 (GND) is not included in the model. As you can see from attached schematic (from page 7 of the data sheet), the GND pin is only used by the mute circuit. I am guessing that the mute functionality in the model just sets the output voltage to mid rail, so if you aim to model the circuit performance in the mute condition with GND set to some other voltage than mid rail, you won't get accurate results. The omission of the GND pin has no impact on the model performance when the amplifier is active (i.e. NOT muted).
~Tom
Attachments
Thanks tomchr,
I was doing some sims, trying to work out if pin 7 should be connected to AGND or PGND and to where pin 8 should be connected. I used a different model.
I was doing some sims, trying to work out if pin 7 should be connected to AGND or PGND and to where pin 8 should be connected. I used a different model.
Pin 7 (GND) can connect to AGND or PGND. Personally, I would connect it to PGND to keep AGND clean. The quality of the "ground" at the GND pin only affects circuit performance in mute. I'm not so concerned about the sound quality when the amplifier is muted... 😛
How to treat Pin 8 (MUTE) is described pretty well in the data sheet. The current out of the MUTE pin needs to exceed 500 uA for the chip to get out of the mute state. As you can see in the schematic, there are three diode drops (Vbe and two diodes) and a 1 kOhm resistor in series with the MUTE pin. So you can calculate the mute resistor like this:
Rmute < (|Vee|-3*Vd)/Imute - 1000 = (|Vee|-2.4)/0.0005 - 1000.
Vd is the diode drop (I use 0.8 V as a worst case number) and |Vee| is the absolute value of the negative supply voltage.
So if you're running on a +/-28 V supply, |Vee| = |-28| = 28. In that case, Rmute should not exceed (28-2.4)/0.0005 - 1000 = 50200 ohm. I would use 47 kOhm in that case.
~Tom
How to treat Pin 8 (MUTE) is described pretty well in the data sheet. The current out of the MUTE pin needs to exceed 500 uA for the chip to get out of the mute state. As you can see in the schematic, there are three diode drops (Vbe and two diodes) and a 1 kOhm resistor in series with the MUTE pin. So you can calculate the mute resistor like this:
Rmute < (|Vee|-3*Vd)/Imute - 1000 = (|Vee|-2.4)/0.0005 - 1000.
Vd is the diode drop (I use 0.8 V as a worst case number) and |Vee| is the absolute value of the negative supply voltage.
So if you're running on a +/-28 V supply, |Vee| = |-28| = 28. In that case, Rmute should not exceed (28-2.4)/0.0005 - 1000 = 50200 ohm. I would use 47 kOhm in that case.
~Tom
I came to the same conclusion about pin 7. I was not expecting to see the audio signal at pin 7.
As for Rmut, I think you need to know the rail voltage at full power before calculating Rmut. In some of the sims I was able to lose more than 1/3 of the rail voltage and get the amp to mute. A diode in series with Rmut seemed to help some.
As for Rmut, I think you need to know the rail voltage at full power before calculating Rmut. In some of the sims I was able to lose more than 1/3 of the rail voltage and get the amp to mute. A diode in series with Rmut seemed to help some.
No. Pin 7 is a ground connection used for switching the bias between the input stage (active state) and the error amplifier used for the mute function (mute state) in the LM3886. It needs to be tied to ground. See the circuit schematic for the LM3886 (image attached to Post #89).
~Tom
~Tom
I connect pin 7 to power supply GND using a separate wire (P2P) I think I will have to rethink this. I connect Rmute to power supply negative. This is also probably wrong.
I think its a choice between power supply symmetry and good grounding practice. I want both.
I think its a choice between power supply symmetry and good grounding practice. I want both.
Last edited:
It sounds like you're doing it right. You don't need to run pin 7 (GND) through a separate trace, but there is no harm in doing so. Rmute needs to go to a negative voltage. The negative supply rail is the most obvious choice.
~Tom
Hi mjf,
Thanks for trying that, I guess another of my long-time assumptions falls off the shelf. Oh well, you're never too old to learn new stuff right? The reason I believed that allowing the input to roll off at a lower frequency than the feedback network was an improper design technique was from an experience I had years (decades 🙁 ) ago. I had an amp that had weak bass due to a high pass Fc around 30Hz as I recall, and I wanted to improve that. So I replaced the input coupling cap with one that was a lot larger and "better quality." The first time I tried the amp it made some horrifying noises and blew fuses. So I checked everything over and found no shorts or other issues. So decided to put the original caps back in and carefully try it again. It worked just fine. That got me curious as to why? I took another look at the amps' circuit and discovered that the AC coupling on the feedback network also had a high-ish Fc. So I reckoned that since most solid state amps need to operate at some minimal gain to be stable, that having the input and feedback Fc's arranged bass-ackward was making the amp operate at a lower gain than the compensation would allow at lower frequencies, and since then it's been one of my automatic rules of thumb. So I guess, live and learn, right? Thanks again.
Mike
tube amps with output transformer, and input/interstage coupling C can have a low frequency gain intercept that could give stability problems depending on feedback details - the low frequency oscillation was called motorboating
the 3886 amplifies down to DC with negligible phase shift much below 100 Hz - a single feedback DC blocking C with time constant < 20 Hz won't give any problems
few op amp circuits (a category that includes the 3886) have the low frequency gain intercept issues - the gain doesn't drop to below the feedback network defined gain - so there isn't a low frequency loop gain intercept
the 3886 amplifies down to DC with negligible phase shift much below 100 Hz - a single feedback DC blocking C with time constant < 20 Hz won't give any problems
few op amp circuits (a category that includes the 3886) have the low frequency gain intercept issues - the gain doesn't drop to below the feedback network defined gain - so there isn't a low frequency loop gain intercept
Last edited:
h1/h2
That's good stuff!
Let's look at an error: The datasheet shows 1k with 22u, so that more can go into the amplifier than can come out. This error, too much like constipation, causes foldback, a warm but muddy 2nd order harmonic distortion bass effect. A little bit is useful and possibly vital, but a lot of that is extremely disappointing. If one were to hook up an eq/baxandall, the bass knob is utterly defeated with a big warm muddy booming blur instead of the expected bass. That "Boom-Only" bass sounds cheap, like entry level retail.
Let's look at the opposite error: Surely, clean is a good first step, but there is the possibility of TWO practical problems with a technically near-perfect RC value. Firstly, we'd have to raise the resistor value so that a battleship size nfb-shunt cap doesn't store enough charge to knock out the inputs. Here are 4 technically near-perfect combinations for clean bass:
3k3+220u
2k7+270u
2k2+330u
1k5+470u
Okay, so the first problem of cap far too big, was easily solved by changing the resistor value (so as to require a smaller cap). The second problem with clean bass is "Thud Only" non-musical bass. It means you did a perfect job, but alas not perfectly practical.
Sometimes, there is not enough bass distortion for expected rendering. There are many reasons to add bass distortion, but I think that the right reason is so the amplifier is able to play both thuds and booms rather than constrained to a monotony of only thuds or only booms. So, to corrupt the technical perfect values only far enough to avoid monotony, we decrease the cap size by 1/3rd (or more).
Like this,
For example, 2k2+220u--This is not wide open enough to cause thud-only rendering, yet not constipated enough to cause boom-only rendering, but rather an even and suitable blend of both is allowed to pass through the filter.
SO, 2k2+220u can be used for the nfb-shunt-RC and 56k can be used for the feedback resistor. These slightly imperfect values are far cleaner and more practical than anything in the datasheet. Option: If you want an even warmer sound, you could swap the 220u for a 150u or 100u and/or increase the size of the +input cap. Just don't allow the warming effect to reach/congest pitches as high as the voice band.
Other thoughts: May all of your effects be small and transparent.
That's good stuff!
Let's look at an error: The datasheet shows 1k with 22u, so that more can go into the amplifier than can come out. This error, too much like constipation, causes foldback, a warm but muddy 2nd order harmonic distortion bass effect. A little bit is useful and possibly vital, but a lot of that is extremely disappointing. If one were to hook up an eq/baxandall, the bass knob is utterly defeated with a big warm muddy booming blur instead of the expected bass. That "Boom-Only" bass sounds cheap, like entry level retail.
Let's look at the opposite error: Surely, clean is a good first step, but there is the possibility of TWO practical problems with a technically near-perfect RC value. Firstly, we'd have to raise the resistor value so that a battleship size nfb-shunt cap doesn't store enough charge to knock out the inputs. Here are 4 technically near-perfect combinations for clean bass:
3k3+220u
2k7+270u
2k2+330u
1k5+470u
Okay, so the first problem of cap far too big, was easily solved by changing the resistor value (so as to require a smaller cap). The second problem with clean bass is "Thud Only" non-musical bass. It means you did a perfect job, but alas not perfectly practical.
Sometimes, there is not enough bass distortion for expected rendering. There are many reasons to add bass distortion, but I think that the right reason is so the amplifier is able to play both thuds and booms rather than constrained to a monotony of only thuds or only booms. So, to corrupt the technical perfect values only far enough to avoid monotony, we decrease the cap size by 1/3rd (or more).
Like this,
For example, 2k2+220u--This is not wide open enough to cause thud-only rendering, yet not constipated enough to cause boom-only rendering, but rather an even and suitable blend of both is allowed to pass through the filter.
SO, 2k2+220u can be used for the nfb-shunt-RC and 56k can be used for the feedback resistor. These slightly imperfect values are far cleaner and more practical than anything in the datasheet. Option: If you want an even warmer sound, you could swap the 220u for a 150u or 100u and/or increase the size of the +input cap. Just don't allow the warming effect to reach/congest pitches as high as the voice band.
Other thoughts: May all of your effects be small and transparent.
- Status
- Not open for further replies.