LTspice and simulating tube circuits?

[markusA]

I've recently ran into something very odd.
The plots doesn't seem to match the fft?
Look at this and tell me what you see?

The left simulation appears to have a ragged response, really ugly.
The fft shows much better than the sim on the right with the smooth nice looking curve?
What am I missing here?

Ok, I'll give a hint...
The left sim uses a smps @ 23kHz while the right one uses idealized voltage sources.
Wich would sound better? Things like this makes my head spin.

[Robert McLean]

What you are missing is the sharp spike at 20KHz.
Forget about the fft stuff 60 or 80 db down, you wont see that in the top curves.
But it is that 20KHz spike that is giving you the wavy ripples in the left hand curve.

[markusA]

I guess I can treat you to another fft as well, this one represents the previous sim on the right but with a simmed smps @ 23kHz. (The smps model is ran with the tube circuit)

Would the superimposed 23kHz ripple be audiable in any way?
I don't have the right terms for it, harmonics or modulations of sort lower in frequency but generated by the switching frequency?

If the 23kHz ripple is harmless the left sim is one magnitude better than the right one, if it's harmful the right one (with a really good psu) might be a better solution?

(The plot in this post shows the worst of both cases, the right sim with a 120mVpp 23kHz ripple on the B+.)

[sawreyrw]

How do you expect anyone to give you useful feedback when we have no idea what you are doing?

[markusA]

I'm asking on how to interpret graphs since the circuits themselves are protocircuits nothing more.
If I can't read the output the sims will mean very little.

What am I doing...
I'm comparing a couple of different topologies for DC-coupled tube buffers with smps.
I have a long term project building a digital preamp with dsp and tube output stages.

Since it'll be 8 channels and fully balanced, the difference beween using one triode or 4 per channel will be substantial, both the workload and the financial part.
I guess I'm trying to figure out if the cheap and simple solution is good enough?

Is -60dB good enough? (I have a feeling I should be aiming for -80dB if I want silence)
Will the HF ripple be audiable in any way?
Lots of questions but I figured the plots were a good place to start.

[TheGimp]

It might be helpful if you posted the circuits.

[markusA]

It'll be messy but I'll try.
Please be gentle, I'm just starting to learn this stuff.
I'll try to get them posted later tonight.

[jcx]

the 23KHz could cause problems several ways, most "audible" will involve intermodulation distortion prodcuts which can be below the 23 KHz

try a 20 KHz sine with the circuit showing 23 KHz ripple - you will most likely see a 3 KHz IMD difference product, simialrly with 10 KHz

[Distorque]

Try just doing an FFT on a sine wave. You'll be surprised by the inaccuracy of the FFT. I can easily get LTSpice to show -60dB harmonics with a pure sine wave. With a few tweaks, you can get that substantially lower, but I still wouldn't trust harmonics below -60dB or so.

Each of these changes improves the accuracy of the FFT:
Increase the simulation time.
Reduce the maximum timestep (1u is a good starting value.)
On the FFT menu, increase the "Number of data point samples in time" by 2 or 3 clicks.

[jcx]

I find the fft highly accurate with good sim, fft settings - a sine source itself should have better than -150 dB floor

add the .option plotwinsize=0 directive

always use an integer number of periods of the stimulus, easy if you sim to a integer ms length, say 10 mS with integer KHz input

Blackman window works well but requires >4x periods to completely resolve frequencies

sometimes settling time is required so you may need to take the fft over a specified time, or change the start saving data time in the .tran to allow for settling

max time step should be smaller than any gain device Ft, feedback loop gain intercept, and for high fft resolution give >10k points per period
there is no need to exactly calculate the max step since the sim will take variable steps as required by accuracy setting anyway - the point used for the fft are interpolated to the exact sampling time - the order of the interpolation filter is what you're setting with the fft Binomial Smoothing "points" setting