A high-performance, in-place FFT library written in Odin, optimized for SDR (Software Defined Radio) applications and real-time signal processing.

• In-place radix-2 DIT FFT - Memory efficient Cooley-Tukey decimation-in-time algorithm
• Optimized performance - Pre-computed per-stage twiddle factors and unrolled early stages
• SDR-friendly utilities - Built-in support for CU8 IQ data conversion and spectrum analysis
• Windowing support - Integrated Hann window with efficient application
• Spectrum analysis tools - Magnitude, power (dBFS), peak detection, and FFT shift functions
• Zero dependencies - Uses only Odin's core math library

Add to your project:

git submodule add  https://github.com/grpllyer/offt.git vendor/ffto

import fft "vendor/ffto"

// Create an FFT plan for 1024 samples at 2.4 MHz sample rate
plan := fft.fft_make_plan(1024, 2_400_000.0)
defer fft.fft_destroy_plan(&plan)

// Prepare your data (example with SDR IQ bytes)
iq_bytes := []u8{...} // Your CU8 IQ data
samples := fft.cu8_interleaved_to_cf32(iq_bytes)

// Apply windowing to reduce spectral leakage
fft.fft_apply_hann(plan.window, samples)

// Compute FFT in-place
fft.fft_inplace(&plan, samples)

// Analyze the spectrum
power_db := fft.fft_power_dbfs(samples)
peak_bin, peak_freq := fft.fft_peak_bin(power_db, plan.sample_rate)

// Center the spectrum for display (-Fs/2 to +Fs/2)
fft.fft_shift_inplace(samples)

Creates an FFT plan for size n (must be power of 2). Pre-computes twiddle factors and Hann window.

Cleans up allocated memory for the FFT plan.

Performs in-place FFT using optimized radix-2 algorithm with unrolled early stages.

Applies Hann window to reduce spectral leakage. Use before FFT computation.

Computes magnitude spectrum |X[k]| from frequency domain data.

Converts to power spectrum in dBFS (decibels relative to full scale).

Finds the bin index and frequency of the peak in the positive frequency range [0..N/2).

Shifts zero-frequency component to center for spectrum display (-Fs/2 to +Fs/2).

Converts interleaved CU8 IQ samples (common SDR format) to Complex32 in range [-1,1).

Maps FFT bin index to frequency in Hz after FFT shift.

The library includes several optimizations for real-time performance:

1. Per-stage twiddle pre-computation - Eliminates trigonometric calculations during FFT
2. Early stage unrolling - Stages 0 and 1 are unrolled for better performance
3. Inlined complex arithmetic - Reduces function call overhead in inner loops
4. Optional bounds checking disable - Use #no_bounds_check for maximum speed

Current implementation uses iterative in-place radix-2 Cooley-Tukey DIT. The code includes extensive comments about potential future optimizations:

• Split-radix FFT for ~20% fewer operations
• Mixed radix (radix-4/8) for better vectorization
• SIMD optimizations for modern CPUs
• Real-input FFT specialization
• Cache-friendly memory access patterns

• FFT Plan: O(N) for twiddle factors + O(N) for window = ~2N complex numbers
• Computation: In-place, no additional allocation during FFT
• Utilities: Output arrays allocated as needed (caller responsibility to free)

• Odin compiler
• Power-of-2 FFT sizes only
• Input data as []Complex32 or convertible formats

// Analyze a 1 kHz sine wave
plan := fft.fft_make_plan(512, 48000.0)
defer fft.fft_destroy_plan(&plan)

// Generate test signal (1 kHz sine wave)
samples := make([]fft.Complex32, 512)
for i in 0..<512 {
    t := f32(i) / 48000.0
    samples[i] = fft.Complex32{math.sin(2*math.PI*1000*t), 0}
}

fft.fft_apply_hann(plan.window, samples)
fft.fft_inplace(&plan, samples)
power := fft.fft_power_dbfs(samples)
peak_bin, peak_freq := fft.fft_peak_bin(power, plan.sample_rate)

// Process RTL-SDR data for waterfall display
plan := fft.fft_make_plan(2048, 2_400_000.0)
defer fft.fft_destroy_plan(&plan)

// Convert raw SDR samples
samples := fft.cu8_interleaved_to_cf32(sdr_iq_data)
fft.fft_apply_hann(plan.window, samples)
fft.fft_inplace(&plan, samples)
fft.fft_shift_inplace(samples) // Center for display

// samples now contains centered spectrum ready for waterfall

[Specify your license here]

[Add contribution guidelines if desired]