Fluora's Miscellaneous Digital Signal Toolbox

Fluora - 2019 onwards - CC0

---

What's this?

I've made (and continue to keep making) a bunch of little Rust programs for debugging DSP projects and simulating potential techniques. They all have a few things in common:

• They're really bad and have no real error handling.
• They operate exclusively on 32-bit big-endian floating-point samples, and don't expect to find headers.
• They do about one thing each, usually by reading a file and writing to another, but sometimes by writing back to the same file.
• They're very inefficient code, and only intended to be run on small files.
• They have very little documentation and I'm really sorry.
• None of them depend on anything. There are no crates here.

What should I use these for?

You shouldn't use it at all. If you want to anyway, I can't stop you, but you should probably find a saner DSP toolbox if you want to actually get work done.

How do I compile them all at once?

You can run the build.sh script I included, which will try to compile everything in ./src/ that ends in ".rs", and put all the resulting binaries in ./target/.

Why Rust?

I like Rust.

What do they do?

Here's the fast version, covering everything included so far:

• addmix adds two signals, producing a combined output.
• mulmix multiplies two signals (like an RF mixer or ring modulator), generating harmonics at the sum and difference of frequencies.
• bomp is a plotter that generates simple images of signals that you can look at.
• termplot is another, worse plotter, which does something similar using rows of block characters in the terminal.
• correlate takes the correlation of two signals (multiply-sum for all possible time offsets).
• correlate_cyclic does the same, but makes the signals periodic.
• fsinechirp generates chirps (rising or falling tones) with linear change in frequency.
• tsinechirp generates chirps with linear change in period.
• squarechirp generates chirps using square waves.
• normalize multiplies signals by a constant value such that the maximum value becomes 1.
• rectify takes the absolute value of a signal, flipping negative values into positive values.
• peakdetect zeroes out all samples that are not local maxima.
• splitbits extracts the sign from samples, producing an output whose values are only 1 or -1.
• upsample increases the sample rate of a signal by an integer factor with linear interpolation.
• gfilter applies a gaussian low-pass filter.
• sfilter applies a sinc low-pass filter.
• swindow applies a sinc low-pass and high-pass filter, forming a rectangular bandpass filter.
• worsen adds a variable amount of random noise to a signal.
• skew simulates mismatches in sampling frequency and phase.
• sft is a naĩve, slow (O(n²)) implementation of a discrete Fourier transform. It's like an FFT, but instead of fast, it's slow.
• waterfall generates 2D frequency/time spectrum images with repeated Fourier transforms.
• zadoffchu generates Zadoff-Chu sequences, which are low-autocorrelation functions used in telecom.
• unitvex generates a sequence of complex unit vectors, using input data to set their angles.
• vecsum takes two signals and combines them with the Pythagorean theorem, e.g. to produce a power spectrum from a complex DFT.
• dcst simultaneously takes the discrete cosine transform (DCT) and discrete sine transform (DST) of a signal.
• pad adds arbitrary amounts of silence before and after a signal.
• integrate takes the integral of a signal.
• differentiate takes the derivative of a signal.
• orthosine generates orthogonal waveforms on an interval, i.e. OFDM carriers.
• echospace simulates reflections in a 3D volume.