A fully fledged lock-in amplifier running at 25 kHz sampling rate @ 12-bit on an Adafruit M4 microprocessor board with an extensive Python graphical user interface showing real-time signals, interactive filter design and real-time power spectra.
Using pyFFTW for the convolution (FIR filters) and power spectra calculations, instead of using numpy or scipy. The cpu load has dropped from 45% to 6% on my test machine.
Using custom numpy-ringbuffer, instead of collections.deque for all timeseries buffers.
Numba @njit rules.
Full OpenGL support in both Windows 10 and Ubuntu 18.04 (Linux).
And others:
Fixed CUDA support under Ubuntu.
GUI improvements: better contrast in graphs and plots, larger labels and CPU load indicator for both system and process.
Major speed improvements:
And others: