Hi ctrlx,
here are the scope captures of clipping behavior. 10 kHz.
Onset of clipping, slightly asymmetric to positive rail:
And full clipping, with rail sticking and some more nastiness at the negative rail:
Any comments from your side? My assumption is that the difference between the positive and negative sides is due to non-symmetric VAS with the bootstrap. Just a guess. But maybe the nastiness at the negative rail related to compensation? Because there was this ringing with square wave test also at the negative side.
not good clipping, but some say amps should not be used to the point where clipping occurs.
i could not get good clipping with a similar topology
i could not get good clipping with a similar topology
The solution for rail sticking is a Baker clamp on the active side(s), and similar on the bootstrap side. You can split the bootstrap resistors for a minimum boost voltage, ie not 50%-50%. You want maybe 5 to10V on the ~lower/inner resistor. Or you can also diode clamp the bootstrap to the rail, but not recommended for vertical FETs that have a significant vto gate voltage. I mentioned LPT degeneration, which may help with overdrive in general, but it is mainly used to avoid slew distortion, which is an overdrive problem.
Simulation is great way to determine the value of any circuit feature. You may find that the LTP current source is of marginal value, but the CCS and/or the current mirror are important for start-up and shutdown thumps, which you may not have simulated. Another thing you should simulate is power supply ripple. The way I simulate both is by simulating the supply and the AC input from time zero.
Simulation is great way to determine the value of any circuit feature. You may find that the LTP current source is of marginal value, but the CCS and/or the current mirror are important for start-up and shutdown thumps, which you may not have simulated. Another thing you should simulate is power supply ripple. The way I simulate both is by simulating the supply and the AC input from time zero.
I was not concerned about clipping at all because of the hard current limitation of SMPS in my amp of 5.2A peak. If both channels are driven (which is usually the case) it leaves me with just 2.6A peak per channel or about 27W into 8R max in RMS terms. The probability of clipping is extremely low in normal use in stereo, the SMPS overcurrent protection would simply shut the thing down before the output approaches rails. It was a deliberate choice with 25W as desired power and constraints of the small enclosure and heatsinks.
I hope you solved the case with your project back in 2016 in one way or another. Thanks for the response.
steveu,
very much appreciated! I am just scratching the surface of all the vastness of audio, at the moment it seems uncomprehensible for me even on a DIY level given the limited time I can devote to this hobby. But any step beyond just plain building of things designed by others is a great revelation. Lots of WOW! moments every time I begin to grasp what is going on. As for the P101, I kind of consider it done for now and I am saving your advice for my next more contemporary/challenging projects. I am undecided however, there is so much. Thanks again for putting in your time.
nonills, I am having more realistic simulation results with these models posted for K135 and J50 a long time ago. They are a bit tricky to setup in MC12 (you have to make a .lib with the subckt statements, add that to NOM.lib and then use the component editor to add the models).
DC gate voltages are a bit low, but the AC behaviour is better characterized than other models.
https://www.diyaudio.com/community/threads/spice-models.2969/
http://www.spectrum-soft.com/faq/help/faq51.shtm
@Alexandre,
much appreciated. Adding new components to MC is something I have to learn. I had tried to add one PSpice bjt model that was not present in MC library and... failed. I will work on adding the K135 and J50.
much appreciated. Adding new components to MC is something I have to learn. I had tried to add one PSpice bjt model that was not present in MC library and... failed. I will work on adding the K135 and J50.
@Alexandre,
just a short update. I applied the Hitachi models from the thread you referred me to in your post above (2SK135/2SJ50 and 2SK1058/2SJ162). Short conclusion is that these models behave worse than my reality with the Exicons. Namely, the stability margins fell and amp oscillated in simulation with square wave test into complex load of 8R || 100nF. This is not to say that the models are wrong. Quite the opposite, they are probably closer to true Hitachi laterals and make much better approximations for simulations than the respective resident models in MC12. Exicons are just somewhat better in real life (not as good as their models in MC12 though) and allow for some headroom to play with LTP current and compensation which I did.
It makes perfect sense now why Rod Elliott was cautioning about stability in the build guide of P101 and what steered his choices for LTP current and Miller cap values. He must have been experimenting with the Hitachis back then, not Exicons. Exicons seem to be good drop in replacements for the Hitachis but not the other way round, at least not without addressing stability.
BTW, importing models into MC12 turned out to be easy. "Import Wizard" or "Add Part Wizard" under Component Editor simplify the whole procedure with no need to care about the intermediary steps. I understand it was quite tricky in preceding versions of MC. I have now loaded entire Bob Cordell's library in one go. My appreciation to him for such generosity.
just a short update. I applied the Hitachi models from the thread you referred me to in your post above (2SK135/2SJ50 and 2SK1058/2SJ162). Short conclusion is that these models behave worse than my reality with the Exicons. Namely, the stability margins fell and amp oscillated in simulation with square wave test into complex load of 8R || 100nF. This is not to say that the models are wrong. Quite the opposite, they are probably closer to true Hitachi laterals and make much better approximations for simulations than the respective resident models in MC12. Exicons are just somewhat better in real life (not as good as their models in MC12 though) and allow for some headroom to play with LTP current and compensation which I did.
It makes perfect sense now why Rod Elliott was cautioning about stability in the build guide of P101 and what steered his choices for LTP current and Miller cap values. He must have been experimenting with the Hitachis back then, not Exicons. Exicons seem to be good drop in replacements for the Hitachis but not the other way round, at least not without addressing stability.
BTW, importing models into MC12 turned out to be easy. "Import Wizard" or "Add Part Wizard" under Component Editor simplify the whole procedure with no need to care about the intermediary steps. I understand it was quite tricky in preceding versions of MC. I have now loaded entire Bob Cordell's library in one go. My appreciation to him for such generosity.
The larger and more robust the MOSFET chip is, the larger the capacitance and therefore the less stable the feedback in a typical amplifier. This is exaggerated by the lack of drivers because the VAS is a high impedance source, vs the low impedance output of drivers. However, that assumes that the FETS are not the dominant pole. Normally the dominant pole is the VAS.
Did you use zobel and output inductor in the sim?
About the models, that is good news, because the laterals I have are exicons. I've played with a number of schematics in the sim, some of my own, and I've gotten decent stability even with these models. Haven't built anything yet.
One more thing, I remember reading one thread here, where the poster recommended aluminum oxide insulators, the thick ones. He said it significantly improves stability because of diminished parasitic capacitance between source (tab) and heatsink.
About the models, that is good news, because the laterals I have are exicons. I've played with a number of schematics in the sim, some of my own, and I've gotten decent stability even with these models. Haven't built anything yet.
One more thing, I remember reading one thread here, where the poster recommended aluminum oxide insulators, the thick ones. He said it significantly improves stability because of diminished parasitic capacitance between source (tab) and heatsink.
Zobel yes, but not the inductor. To my mind, the circuit must behave without the output coil both in simulator and reality with typical/regular load. I view it as a kind of "insurance" for unexpected load conditions but in general the amp has to be stable without it.
Yes, now you reminded me about those aluminum oxide insulators. I do not have any experience with them. I used kapton in P101. They are known to be very thin, so I realise now there is a risk of increased capacitance.
Yes, now you reminded me about those aluminum oxide insulators. I do not have any experience with them. I used kapton in P101. They are known to be very thin, so I realise now there is a risk of increased capacitance.
Lack of drivers is one of the known criticisms of P101. My understanding is that laterals (compared to verticals) allow a compromise since their input capacitance is low enough to be driven by VAS directly for max two pairs but too high to achieve good results if more pairs are paralleled. Hence, the FETs are to blame for the dominant pole in P101 where this compromise has been used. I'm not challenging, rather thinking aloud.