Optimizing Custom FFT’s Speed on STM32
In this post, we’ll take the STM32 FFT algorithm from my blog “Writing Your Own FFT in Python and STM32” (link below) and focus on improving its execution speed. The goal is to minimize the number of clock cycles required to perform the FFT. To achieve this, we’ll remove unnecessary calculations, precompute the bit-reversal lookup table, enable compiler optimizations, activate the cache, and place buffers in faster memory. We’ll evaluate how each of these changes impacts performance, and compare our optimized algorithm to the CMSIS DSP library’s FFT.
Numerically Controlled Oscillator in Python and STM32
In this post, I’ll cover the basics of creating a Numerically Controlled Oscillator (NCO) from the ground up. An NCO is a digital signal generator that uses a phase accumulator and a waveform lookup table to produce precise, periodic signals—usually sine, square, or triangle waves. These waveforms are generated entirely in the digital domain, making NCOs extremely useful in systems where analog components are limited or where digital precision is required.
NCOs are widely used in a variety of applications, including digital communications, frequency synthesis, software-defined radio (SDR), and audio signal generation. Their ability to generate stable, tunable frequencies with minimal hardware makes them a valuable tool in embedded and signal processing systems.
We’ll begin by modeling the NCO in Python. This lets us explore and visualize the underlying concepts like phase resolution, tuning words, and waveform table indexing without worrying about hardware limitations. Once the behavior is well understood, we’ll transition to implementing the NCO in C on an STM32 microcontroller, where we’ll focus on efficient fixed-point arithmetic and real-time constraints.