Has anyone managed to get a circuit model of the LM3886 to function correctly?
Using the circuit published,I find huge crossover distortion at higher frequency's - with the step response showing a horrid "Step" around the falling edge "Crossover region".
FFT measurements of LM3886 designs does show crossover components (about -110dB Ref. 1W 8R) - but the simulations show worst results - in fact looking at the internal schematic, I don't believe the transfer Gain around the crossover region is ever going to be "Smooth".
I had to slightly modify my own simulation circuit of the LM3886 - Q10 is loaded via a 5mA CCS, rather then the diode "load" indicated in the LM3886 "Equivalent Schematic"... With the Diode "load" the crossover region was even worst (And I don't believe the indicated diode connection is correct, unless some oddity of on chip device is used).
Note - I have no idea of the on chip transistor characteristics, so I used the "crummiest" discrete devices I could find for the simulation model - still I had to tweak the Cdom Cap to 18pF to match the Gain / Phase plot shown on the Data Sheet.
Does anyone have a better internal circuit? I suspect NS have deliberately not disclosed the true "circuit" in there published "Equivalent" Schematic.
I can correct OPS design to function correctly - but what's really inside the LM3886?
Using the circuit published,I find huge crossover distortion at higher frequency's - with the step response showing a horrid "Step" around the falling edge "Crossover region".
FFT measurements of LM3886 designs does show crossover components (about -110dB Ref. 1W 8R) - but the simulations show worst results - in fact looking at the internal schematic, I don't believe the transfer Gain around the crossover region is ever going to be "Smooth".
I had to slightly modify my own simulation circuit of the LM3886 - Q10 is loaded via a 5mA CCS, rather then the diode "load" indicated in the LM3886 "Equivalent Schematic"... With the Diode "load" the crossover region was even worst (And I don't believe the indicated diode connection is correct, unless some oddity of on chip device is used).
Note - I have no idea of the on chip transistor characteristics, so I used the "crummiest" discrete devices I could find for the simulation model - still I had to tweak the Cdom Cap to 18pF to match the Gain / Phase plot shown on the Data Sheet.
An externally hosted image should be here but it was not working when we last tested it.
Does anyone have a better internal circuit? I suspect NS have deliberately not disclosed the true "circuit" in there published "Equivalent" Schematic.
I can correct OPS design to function correctly - but what's really inside the LM3886?
Last edited:
I use the TI TINA spice model and a LTspiceIV model that can be downloaded at DiyAudio http://www.diyaudio.com/forums/chip-amps/98543-lm3886-macro-model.html#post3730699
Rumors,
Thanks for the reply, but the output stage design still follows the published NS circuit so will display the same Crossover irregularities...
Non Complementary OPS are notoriously non linear around the Xover region - Baxdal's (sorry for my spelling) diode goes some way to help linearising the gain, but the natively low output impedance of the upper EF stage, and the inherently high impedance of the lower half is never going allow "Complementary" performance between each phase of the OPS.
Thanks for the reply, but the output stage design still follows the published NS circuit so will display the same Crossover irregularities...
Non Complementary OPS are notoriously non linear around the Xover region - Baxdal's (sorry for my spelling) diode goes some way to help linearising the gain, but the natively low output impedance of the upper EF stage, and the inherently high impedance of the lower half is never going allow "Complementary" performance between each phase of the OPS.
The TI TINA spice is an idealised model - pretty worthless for investigating the devices behaviour during clip etc.
Bigun's "real" model again has the published OPS design (and uses idealised transistor devices) - what do you see for the falling edge step response?
Last edited:
Here is the Failing Edge step response of my earlier published LM3886 Sim circuit - showing the horrid gain step between phases of the output stage (8 Ohm load).
An externally hosted image should be here but it was not working when we last tested it.
Last edited:
I've built a 4 channel 3886 amp and think it sounds great. I noticed that your schematic on the first page shows a 20k R to Gnd off the wiper of a pot. That could be the problem right there. That R to gnd should be 10X the value of the pot, so it doesn't totally screw up the taper of the pot (unless you wanted that). That implies that he unlabelled pot would be less than 20Kohms anyway, which raises the question of what the source impedance and coupling cap is. Most sources have a coupling cap that is designed to drive anything over 10Kohms (if you're lucky), and if that source drives <10K, there can be a loss of bass, which will have psycho-acoustic effects on the rest of the frequency range. I'd change that R to gnd to 200K, and let the pot be 20K - 50K audio taper.
On the power supply bypass caps issue, I highly recommend roughly 0.1uF caps right at the 3886 chip from the power pins to the circuit ground, and in parallel with those caps, I put 100uF AE caps, but with a separate ground return to the star center gnd of the chassis. That way the charging currents of the 100uF caps won't modulate the circuit ground, but will insure a zero ohm power supply source impedance as seen by the 3886. The 0.1uF caps insure the zero ohm PS impedance as seen by the 3886 to beyond the highest frequency the chip will ever care about, making the phase margin more predictable. My 3886 amp and circuit can be viewed at my website (scroll down a bit):
Soundbar ProjectSoundbar Project
On the power supply bypass caps issue, I highly recommend roughly 0.1uF caps right at the 3886 chip from the power pins to the circuit ground, and in parallel with those caps, I put 100uF AE caps, but with a separate ground return to the star center gnd of the chassis. That way the charging currents of the 100uF caps won't modulate the circuit ground, but will insure a zero ohm power supply source impedance as seen by the 3886. The 0.1uF caps insure the zero ohm PS impedance as seen by the 3886 to beyond the highest frequency the chip will ever care about, making the phase margin more predictable. My 3886 amp and circuit can be viewed at my website (scroll down a bit):
Soundbar ProjectSoundbar Project
Last edited:
That interesting, as 76Deg is criticality damped Q=0.5 (or at least a minimal overshoot of 0.05%) - in real life an amplifiers phase margin will be deteriorated by capacitive loading and load current (especially with Bipolar OPS) so I wonder if your observations of slightly higher PM is warranted to allow for the effects of this Phase margin "Modulation" - we tend to view the transfer function under simplified operating conditions.
Last edited:
As an answer to your original question, I have worked on a reasonably optimised LM3886 design and IMO it offers an decent performance for little effort (its a simple chip based design) - but it does lack fine resolving power (detail) over a decent discrete design - so I guess you could describe it as sounding a little dark...
I believe Phase modulation as a result of the LM3886 crossover distortion could be the cause of the smeared levels of fine detail.
Thank you - however once again it still shows the same Diode loaded PNP driver to the lower output stage. TI (NS) AN-898 Application note Fig 1 shows this connection loaded via a CCS which makes more sense and improves the simulated step response WRT the Failing Edge.
On your schematic you can see the Diff pair CM and EF buffer stage switched via an NPN transistor which I presume is part of the protection / Mute circuit 🙂
Maybe they did not have a symbol for the CCD (constant current diode) at hand😀
Another detail is also interesting ... feedforward error-correction within the main differential😉
Oh - I don't see it? The second Diff Pair with 4K7 Emitter Resistors enclosed (Drawn) within the Main input pair is only enabled during Mute - Otherwise I'm missing something major? 😕
Only enclosing the Output stage within the Cdom feedback loop is a little unconventional and can lead to stability issues in practise (making it more sensitive to capacitative loads)... What have I missed?
I had another look ... you're right, it's not FFEC ... cross-coupled collectors have tricked my tired eyes😱
So let me get this straight. You're unhappy with the TINA-TI Spice model because it's "idealized" but proceed to build your own from the equivalent circuit schematic using models for various discrete devices. And you want to model distortion and clipping behavior?
Both will be dependent on device characteristics. You won't find those published by National as those are confidential company information. You can take your best guess but in the end you'll be simulating a circuit that's rather different from an LM3886, so if the performance of an LM3886 is what you're looking to model, you're really shooting yourself in the foot.
Simulators are great at giving you data with many digits for the THD. How many of those digits are valid? Not many, it turns out... Once you start pushing 0.1 % or 0.01 % THD, you really have to ensure that you're simulating the THD of the circuit and not the residual error of the circuit simulator.
Add that the circuit schematic in the data sheet is an equivalent circuit rather than the actual circuit. I guarantee you that the details of the compensation network are not shown in the equivalent circuit. I would also expect that things like saturation clamps aren't shown either.
Just saying...
~Tom
Both will be dependent on device characteristics. You won't find those published by National as those are confidential company information. You can take your best guess but in the end you'll be simulating a circuit that's rather different from an LM3886, so if the performance of an LM3886 is what you're looking to model, you're really shooting yourself in the foot.
Simulators are great at giving you data with many digits for the THD. How many of those digits are valid? Not many, it turns out... Once you start pushing 0.1 % or 0.01 % THD, you really have to ensure that you're simulating the THD of the circuit and not the residual error of the circuit simulator.
Add that the circuit schematic in the data sheet is an equivalent circuit rather than the actual circuit. I guarantee you that the details of the compensation network are not shown in the equivalent circuit. I would also expect that things like saturation clamps aren't shown either.
Just saying...
~Tom
Last edited:
Tom,
As a Chip designer myself I prefer to model "real" circuits rather then an idealised Opamp block with various parameters dropped into the model.
I'm only playing with the design trying to improve an existing product (Audiolab M-PWR) - if I was serous then I investigate the Fab process and I might even have access to models of the process characterised - but I've not bothered to make the phone calls as I'm only playing around for fun.
I prefer real models as I tend to try various external circuit tricks that a simple "OpAmp" model would offer very little information - For instance the Tina Model does not show the abrupt Falling edge "discontinuity" as the upper EF stage hands over (less then gracefully) to the lower CE stage.
Measurements on the real device at higher frequency's display this issue, so for my purposes my model is closer to the physical device then the Tina Model.
I've NEVER stated that I wanted to model THD, my THD results are based on measured hardware - I've never had much faith in simulated THD results - only using them as a "direction finder" rather then for any absolute values.
As a Chip designer myself I prefer to model "real" circuits rather then an idealised Opamp block with various parameters dropped into the model.
I'm only playing with the design trying to improve an existing product (Audiolab M-PWR) - if I was serous then I investigate the Fab process and I might even have access to models of the process characterised - but I've not bothered to make the phone calls as I'm only playing around for fun.
I prefer real models as I tend to try various external circuit tricks that a simple "OpAmp" model would offer very little information - For instance the Tina Model does not show the abrupt Falling edge "discontinuity" as the upper EF stage hands over (less then gracefully) to the lower CE stage.
Measurements on the real device at higher frequency's display this issue, so for my purposes my model is closer to the physical device then the Tina Model.
I've NEVER stated that I wanted to model THD, my THD results are based on measured hardware - I've never had much faith in simulated THD results - only using them as a "direction finder" rather then for any absolute values.
Last edited:
May I ask, what did you compare the LM3886 against? I am also looking for am amp, just unsure if LM3886 will be suitable "upgrade"
After several modding, the amp's sound character does not changed a lot.
And it does not sound as good as the LM1875 i had before.
So I sold the amp.
And it does not sound as good as the LM1875 i had before.
So I sold the amp.
what do us lesser mortals make of this?
I have some 3886, 3876, 3875, and 4780 here. Because I'm lazy i have been looking at buying some of the better known PCBs - chipamp.com audiosector.com
Although i can make the pcb, id rather avoid double sided lithography.
Am I now to think that i will be disappointed? Maybe ill cobble together a novice attempt at something using 49810 instead....
Sorry guys, but you techies really took the shine off the chipamp idea. ( And my Cambridge A1 is the nicest amp ive owned - also chipamp...)
I have some 3886, 3876, 3875, and 4780 here. Because I'm lazy i have been looking at buying some of the better known PCBs - chipamp.com audiosector.com
Although i can make the pcb, id rather avoid double sided lithography.
Am I now to think that i will be disappointed? Maybe ill cobble together a novice attempt at something using 49810 instead....
Sorry guys, but you techies really took the shine off the chipamp idea. ( And my Cambridge A1 is the nicest amp ive owned - also chipamp...)
Fair enough. And you're right. I did make some assumptions when typing up my response. It's difficult at times to gauge the intent and skill level of people here. And sometimes my own cognitive biases get in the way as well. I'm human. What more can I say? 🙂
Yeah. The clipping behavior of the LM3886 is not pretty. It also challenges the stability of the circuit. At in my non-optimized circuit with way too much parasitic inductance in the Zobel network, I saw some ringing around that discontinuity. Interestingly enough, only on the negative swing.
~Tom
- Status
- Not open for further replies.