embedded-dsp 0.4.0

A no_std Rust digital signal processing library for microcontrollers, embedded systems, and real-time signals.
Documentation

embedded-dsp

crates.io docs.rs CI License: MIT OR Apache-2.0

A #![no_std] Rust Digital Signal Processing library designed for microcontrollers (Cortex-M, RISC-V, AVR, Xtensa), embedded systems, and real-time signal processing. Neural-net and classical ML inference live in embedded-nn.

Features

  • #![no_std] First: Pure core compatibility for bare-metal targets with zero dynamic allocation required.
  • libm, defmt, & serde Integrations: Optional formatting logs via defmt, model serialization via serde, and floating-point math routines in #![no_std] environments via libm.
  • Per-module Cargo features: Every algorithm module is optional. full (in default) enables them all; default-features = false plus the modules you want keeps firmware images small.
  • Fixed-Point & Floating-Point: CMSIS-style f32, f64, q31, q15, q7, and q63 saturating arithmetic, plus Q16.16 (fixed-point) and a 256-entry sin/cos LUT (lut).
  • 22 Core DSP Modules:
    1. Basic Math: Elementwise add, sub, mult, negate, offset, scale, shift, dot_prod, clip, bitwise operations.
    2. Complex Math: Complex vector addition, multiplication, magnitude, conjugate, dot product.
    3. Fast Math: Trigonometric sin, cos, tan, sin_cos, sqrt, vsqrt, divide, log, log10, exp, atan2.
    4. Filtering: FIR filters, Biquad IIR cascade (DF1 and transposed DF-II, f32/q15/q31), LMS / leaky LMS / NLMS (f32/q15), 1D convolution & correlation, FFT fast convolution, 1D conditional/thresholded median filters (f32, q15, q31), single-pole recursive low/high-pass filters (SinglePoleFilter / SinglePoleFilterQ15), Q15 DC blocker (DcBlockerQ15), O(1) recursive moving average (RecursiveMovingAverage<N> / RecursiveMovingAverageQ15<N>), and const-generic real-time CircularBuffer<T, N>.
    5. Filter Design: Biquad Low-Pass, High-Pass, Band-Pass, Notch, Peaking EQ, All-Pass, multi-stage Butterworth design, multi-stage Chebyshev Low-Pass/High-Pass design, continuous-to-discrete Bilinear Transform with cutoff frequency pre-warping, Windowed-Sinc FIR design (Low-pass, High-pass, Band-pass, Band-stop), and arbitrary-response FIR design via frequency sampling (fir_custom_frequency_sampling).
    6. Audio: Goertzel single-frequency detector (GoertzelDetector / GoertzelDetectorQ15), peak/RMS envelope followers (PeakEnvelopeFollower/RmsEnvelopeFollower and Q15), Mel filterbank (mel_filterbank_f32), and MFCC feature extraction (mfcc_f32).
    7. Spectral Analysis & PSD: Welch's method power spectral density estimation (averaged periodograms), single-segment periodograms in linear and dB scale.
    8. Spatial & 2D Signal Processing: 2D DCT-II / IDCT-II, 2D spatial convolution with normalization, 2D non-linear filtering (Min/Max/Median), Sobel edge detection, 2D histogram binning, MSE, and PSNR.
    9. Resampling & Multi-rate: Cascaded Integrator-Comb (CIC) Decimator & Interpolator, linear fractional resampler, spectral 2:1 sinc zero-padding interpolation.
    10. Kalman Filtering: 1D/2D helpers, const-generic linear KalmanFilter<N, M>, and trait-based Extended Kalman Filter (EkfModel), with _with_input variants for models driven by an exogenous input outside the state.
    11. Const Generics: Compile-time fixed-size FirFilter<N>, FirFilterQ15<N>, BiquadCascade<COEFFS, STATE>, BiquadCascadeQ15<COEFFS, STATE>, and Matrix<R, C, N>.
    12. Transforms: In-place Complex FFT (cfft), Real FFT (rfft, packed rfft_q15/rfft_q31 and irfft_q15/irfft_q31), Discrete Cosine Transform (dct4), Fast Walsh-Hadamard Transform (fwht_f32/fwht_i32), Fixed-Point FFT (cfft_q15/cfft_q31), Haar Transform (haar_transform_f32/haar_transform_i32), self-inverse Hartley Transform (hartley_transform_f32), and a generalized wavelet transform (wavelet_transform_f32 with the built-in Daubechies-4 filter).
    13. Matrix & Regression: Matrix addition, subtraction, multiplication, scaling, transpose, Gauss-Jordan inversion, and weighted polynomial least-squares curve fitting.
    14. Controller: PID motor controller (f32/q15/q31), Clarke and Park transforms (f32/q15).
    15. Statistics: Mean, variance, standard deviation, RMS, power, min/max, entropy, KL divergence, logsumexp.
    16. Support, PRNG & Noise: Array copy/fill, zero-allocation sorting (sort_f32), format conversions (q15 ↔ f32 ↔ q31), rounded FIR tap quantizer (fir_taps_f32_to_q15), XorShift64 PRNG, uniform and Box-Muller Gaussian noise generators.
    17. Interpolation: Linear, Bilinear, and Cubic Spline interpolation.
    18. Quaternions: Norm, normalization, quaternion product, conjugate, inverse, rotation matrix conversion.
    19. Window Functions: Hanning, Hamming, Blackman, 4-term Blackman-Harris, Bartlett, Welch, Flat-top generators (f32), plus Q15 Hanning/Hamming/Blackman/Bartlett and apply_window_q15.
    20. Distance Metrics: Euclidean, Cosine, Chebyshev, Manhattan, Minkowski, Jaccard, Hamming, Canberra, Bray-Curtis.
    21. Filter Analysis: FIR/biquad-cascade frequency response (DTFT) evaluation, magnitude/phase/dB helpers, FIR group delay, and pole-based IIR stability checks.
    22. Companding: µ-law and A-law curves (mu_law_compress_f32 / a_law_*) and ITU-T G.711 bytes (linear_to_ulaw / ulaw_to_linear, linear_to_alaw / alaw_to_linear).

Quick Start

Add embedded-dsp to your Cargo.toml:

[dependencies]
# For standard std environments (all modules)
embedded-dsp = "0.4.0"

# For bare-metal #![no_std] with libm and every algorithm module
embedded-dsp = { version = "0.4.0", default-features = false, features = ["libm", "full"] }

# For bare-metal, only the pieces you use (example: FIR/biquad + Q15 math)
embedded-dsp = { version = "0.4.0", default-features = false, features = ["libm", "filtering", "basic-math"] }

types and math (FloatMath) are always compiled. Other modules map 1:1 to Cargo features (filtering, transform, kalman, fixed-point, lut, …). Enabling kalman also pulls matrix (and thus basic-math); enabling audio or psd also pulls transform. FFT-backed helpers (fast_convolve_f32, fir_custom_frequency_sampling, spectral_interpolate_2x_f32) need transform as well.

Minimum supported Rust is 1.88 (edition 2024).

Migrating from 0.3.0

  • Naive Bayes and SVM live in embedded-nn, not this crate.
  • default-features = false no longer compiles every module. Use features = ["libm", "full"] or list modules.
  • cfft_q15 / rfft_q15 (and q31) are integer FFTs with about 1/n scale versus the f32 transforms.

Basic Example

use embedded_dsp::*;

fn main() {
    // 1. Vector Operations
    let a = [1.0f32, 2.0, 3.0, 4.0];
    let b = [10.0f32, 20.0, 30.0, 40.0];
    let mut vec_out = [0.0f32; 4];
    add_f32(&a, &b, &mut vec_out);

    // 2. Q15 Fixed-Point Saturating Addition
    let q15_a = [20000i16, 25000];
    let q15_b = [15000i16, 10000];
    let mut q15_out = [0i16; 2];
    add_q15(&q15_a, &q15_b, &mut q15_out); // Output: [32767, 32767] (clamped at i16::MAX)

    // 3. Filter Design & Biquad Execution
    let coeffs = biquad_lowpass_coeffs(1000.0, 48000.0, 0.7071);
    let mut biquad = BiquadCascade::<5, 4>::new(coeffs);
    let mut dst = [0.0f32; 4];
    biquad.process(&a, &mut dst);

    // 4. Kalman Sensor Filtering (1D helper or generic N×M)
    let mut kf = KalmanFilter1D::new(0.0, 1.0, 0.01, 0.1);
    kf.predict(0.0);
    let _filtered_reading = kf.update(10.2);

    let mut kf2 = KalmanFilter::<2, 1>::from_variances([0.0, 0.0], 1.0, 0.01, 0.1);
    kf2.predict(&[[1.0, 0.1], [0.0, 1.0]]);
    let _ = kf2.update(&[[1.0, 0.0]], &[1.0]);

    // 5. 64-Point Complex FFT
    let mut fft_data = [0.0f32; 128]; // 64 complex pairs [re, im, ...]
    cfft_f32(&mut fft_data, 64, 0, 1);
}

Running Included Examples

# Run basic usage example
cargo run --example basic_usage

# Run performance comparison benchmark (libm vs embedded-dsp)
cargo run --release --example perf_comparison

License

The contents of this repository are dual-licensed under the MIT OR Apache 2.0 License. That means you can choose either the MIT license or the Apache 2.0 license when you re-use this code. See LICENSE, LICENSE-MIT, or LICENSE-APACHE for more information on each specific license.