80dB now but still needs improvement at 1KHz

If testing with sine waves that all fit with an integral number of cycles per window, the FFT noise floor will be determined by the precision of the arithmetic used.

But there is often a splice error at that join because the fit into the window is approximate, which is a good reason to use a window function. The default choice in the radar world is Taylor.

Joe Gwinn

I wouldn't trust LT Spice for simulating a low-distortion oscillator.

Run a sim and zoom up on the top of some sine wave. It looks all segmented and ratty.

I was just simulating something and had a KHz range loop oscillation that refused to be compensated. Turns out I had a BI behavioral current source driving an inductor, and that oscillated all by itself. A G current source does the same thing.

At really low distortion, the opamp models can getcha too.

Oh yeah. I don't use spice simulations unless I'm designing a circuit, which is rare these days.

Mostly, I use Wolfram Mathematica, often with Bessel window functions (for phase linearity across the passband).

Joe Gwinn

But if Raspberry Pi does achieve world domination, Wolfram will be forced to charge money for Mathematica there. So do it very slowly -- maybe they won't notice in time.

Joe

EDN article, if it's any good I don't know: