Musings on amp design... a thread split

LTSpice has some internal tweaks that allow for faster convergence and processing. They were essential for modelling switching supplies (of course). Microcap can be quite slow. These ultra low THD numbers relate to the accuracy and shortcuts in the models. Its all a fun numbers exercize but has very little relationship to the reality in the circuit board.
I find the simulations great for checking biases operating points frequency response etc. However not useful for optimizing distortion below .1%. You can find the seriously nonliner places in the circuit but the models are just not complete enough for PPM level analysis. That may be why you can get such absurdly great numbers.
 
It is like you start metal work with primitive tools. You use them to make better tools, then you built a simple machine, with the machine you fabricate better tools to again built a even better machine. Finally you end up with tolerances of 1/000 mm.

It takes 10 years to sharpen a sword.
Chinese saying.
 
Models indeed are (sometimes gross) simplifications of behavior. It's quite obvious to some that when the numbers are far better than the specs from producer(s), the outcome is likely to be too optimistic. One cautionary measure is to test the model itself in a sim, and compare with producer(s) spec. Sometimes the difference is quite hilarious.
 
How can I get such fft graph in harmonic distortion analysis ?????
39AA9F4D-12CB-42C9-BA5B-F3A1C2490A91.jpeg
 
Still possible to do normal work on the circuit in the second running program.
No idea what "normal work" means: for me it's RF (high dynamic range - low noise) antenna amps, feedback PLL (threshold far below that of standard RX), and KISS type audio amps with "below 60 dB" low order distortion. All together slow down the system to the extent that even browsing becomes rather slow.
 
There is nothing wrong with simulations imho - as long as the underlying models represent reality accordingly. So I do share the opinion of 1audio. But with THD numbers in the region of 200dB you are far below realistic modelling - and it is impossible to prove that these numbers may become realistic at some point. Also the conclusion, that improving simulated THD from maybe -180 dB to -250dB improves your amp is nothing more than an assumption not backed up by evidence. Without a possibility of verification this whole debate is just a theoretical kind of sport without practical relevance.
 
Yep, and the reason is to keep THDN at 100 mW "low enough". This because the opamp used is VERY noisy but one of the few, allowing this topology:
View attachment 1145426
For no other reason then be able to compare LTSpice to Microsim, I simulated your circuit diagram, without knowing what Bias and what Mosfets you used.
The Mosfets I used needed a 4Volt Vgs for 250mA Bias, so I limited Vout to slightly under +/-20Volt peak to prevent clipping.
Plot that came out showed IMD somewhere at -120dB, which seems quite realistic for an opamp having a THD of of -104dB at +/-15V@1Khz.
So my impression is still that Microsim is very optimistic with distortion figures.
When you think the Mosfets that you used improve IMD, please give me de Spice models plus the bias current that you used.

Hans
 

Attachments

  • OPA552.jpg
    OPA552.jpg
    236.5 KB · Views: 116
For no other reason then be able to compare LTSpice to Microsim, I simulated your circuit diagram, without knowing what Bias and what Mosfets you used.
The Mosfets I used needed a 4Volt Vgs for 250mA Bias, so I limited Vout to slightly under +/-20Volt peak to prevent clipping.
Plot that came out showed IMD somewhere at -120dB, which seems quite realistic for an opamp having a THD of of -104dB at +/-15V@1Khz.
So my impression is still that Microsim is very optimistic with distortion figures.
When you think the Mosfets that you used improve IMD, please give me de Spice models plus the bias current that you used.

Hans
That was more interesting than I expected: I don't use those mosfets in such application but instead, use the ECW20 series.
1677012717904.png
1677012806107.png