vecfit

Pure-Rust implementation of relaxed vector fitting for rational approximation of frequency-domain data. Fits scalar, vector, and matrix-valued responses.

[dependencies]
vecfit = "0.1"

Quick start

use vecfit::{Csv, Options};

let csv = "\
freq_Hz,re_f1,im_f1
1.0,0.95,-0.31
3.0,0.55,-0.62
10.0,0.22,-0.78
30.0,0.08,-0.45
100.0,0.02,-0.15
";

let model = Csv::from_csv(csv)?.fit(Options::new().poles(3))?;

Column formats are auto-detected from headers: re_*/im_* for rectangular, |*|/ang_* for magnitude/phase.

Input formats

CSV from a file:

use vecfit::{Csv, Options};

let model = Csv::from_path("data.csv")?.fit(Options::new().poles(4))?;

Also supports TSV (from_tsv), SSV (from_ssv), and custom delimiters (from_delimited). For matrix-valued data, chain .matrix(rows, cols)? before .fit(...).

Touchstone (.s1p, .s2p, ...):

use vecfit::{Touchstone, Options};

let model = Touchstone::from_path("network.s2p")?.fit(Options::new().poles(6))?;

Supports S/Y/Z/G/H/T parameters, RI/MA/DB formats, and all standard frequency units.

Closure fitting

For analytic transfer functions, pass a closure directly:

use num_complex::Complex64;
use vecfit::{Model, Options, complex};

let s: Vec<Complex64> = (0..80)
    .map(|k| Complex64::new(0.0, 1.0 + k as f64))
    .collect();

let model = Model::fit(
    complex(&s),
    |s| 0.05 + 1.2 / (s + 3.0) + 0.4 / (s + 15.0),
    Options::new().poles(4),
)?;

Closures can return scalars, arrays, or nested arrays — the shape is inferred automatically. See the API docs for c(re, im), axis wrappers (hz, rad, real), and matrix fitting.

Axis wrappers

Examples

cargo run --example fit_scalar          # scalar closure fit
cargo run --example fit_vector          # multi-channel vector fit
cargo run --example fit_csv             # CSV parse-and-fit
cargo run --example json_roundtrip      # JSON serialize/deserialize
cargo run --example export_matrix_emt   # real-section + state-space export
cargo run --example plot_scalar_report  # plotted report (PNG + Markdown)

Scalar fit (10 poles, RMSE 2.2e-4)

Vector fit (18 poles, 3 channels, RMSE 2.3e-4)

Matrix EMT fit (10 poles, 2x2, RMSE 1.9e-4)

Author

Luke Lowery · Google Scholar

Related projects

• gspx — graph signal processing in Rust, uses vecfit for its fitting backend
• sgwt — spectral graph wavelet transform in Python (docs)

License

Licensed under either of Apache License 2.0 or MIT, at your option.