Files
rdsp-experiments/src/fft/rader2.rs

163 lines
4.2 KiB
Rust

// Implementation of raders's fft for prime sized ffts
use std::f32::consts::PI;
use crate::{
complex::Complex32,
fft::{
create_fft, is_prime, DFTAlgorithm, FFTDirection
},
};
pub struct Rader2FFT {
permutations: Box<[usize]>,
convolution_operand: Box<[Complex32]>,
convolution_fft_input: Box<[Complex32]>,
convolution_ifft: Box<dyn DFTAlgorithm>,
convolution_fft: Box<dyn DFTAlgorithm>,
output: Box<[Complex32]>,
sub_size: usize,
size: usize,
}
impl DFTAlgorithm for Rader2FFT {
fn create(size: usize, direction: FFTDirection) -> Self
where
Self: Sized,
{
assert!(is_prime(size));
// Primitive root and its powers
let g = compute_prime_primitive_root(size);
let permutations: Box<[usize]> = (0..(size - 1)).map(|i| exp_mod(g, i + 1, size)).collect();
let sub_size = next_pow2(2 * size - 3);
// Compute fourrier transform of twiddle factors
let twiddle_factors = (0..sub_size)
.map(|i| {
Complex32::cexp(
-2. * PI * direction.sign() * (permutations[i % (size - 1)] as f32)
/ (size as f32),
)
})
.collect::<Vec<Complex32>>();
let mut convolution_fft = create_fft(sub_size, FFTDirection::Forward);
convolution_fft.execute(&twiddle_factors);
Rader2FFT {
permutations,
convolution_operand: convolution_fft.get_output().iter().copied().collect(),
convolution_fft,
convolution_ifft: create_fft(sub_size, FFTDirection::Inverse),
convolution_fft_input: vec![Complex32::zero(); sub_size].into(),
output: vec![Complex32::zero(); size].into(),
size,
sub_size,
}
}
fn execute(&mut self, input: &[Complex32]) {
// Compute fft of input signal
self.convolution_fft_input[0] = input[self.permutations[self.size - 2]];
for i in 0..(self.sub_size - self.size + 1) {
self.convolution_fft_input[i + 1] = Complex32::zero();
}
for i in 1..(self.size - 1) {
// reverse sequence
let k = self.permutations[self.size - 1 - i - 1];
self.convolution_fft_input[i + self.sub_size - self.size + 1] = input[k];
}
self.convolution_fft.execute(&self.convolution_fft_input);
// Compute convolution by multiplying in freq domain
for i in 0..self.sub_size {
// Using output as staging buffer
self.convolution_fft_input[i] =
self.convolution_fft.get_output()[i] * self.convolution_operand[i];
}
self.convolution_ifft.execute(&self.convolution_fft_input);
self.output[0] = Complex32::zero();
for x in input {
self.output[0] = self.output[0] + *x;
}
for i in 0..(self.size - 1) {
// Actually compute the output
let k = self.permutations[i];
self.output[k] =
(self.convolution_ifft.get_output()[i] / (self.sub_size) as f32) + input[0];
}
}
fn get_output(&self) -> &[Complex32] {
&self.output
}
}
pub fn compute_prime_primitive_root(n: usize) -> usize {
assert!(is_prime(n));
let phi = n - 1; // Euler's totient for n prime
// Test all candidates
for i in 1..(n + 1) {
// Find multiplicative order of i
let mut val = i;
let mut order = 1;
for _ in 0..n {
if val == 1 {
break;
}
val = (val * i) % n;
order += 1;
}
if order == phi {
return i;
}
}
unreachable!("Prime must have primitive root");
}
pub fn exp_mod(mut n: usize, mut exp: usize, m: usize) -> usize {
if m == 1 {
return 0;
}
n %= m;
let mut r = 1;
while exp > 0 {
if exp % 2 == 1 {
r = (r * n) % m;
}
n = (n * n) % m;
exp >>= 1;
}
r
}
pub fn next_pow2(mut n: usize) -> usize {
if n.count_ones() == 1 {
n
} else {
let mut p = 0;
while n > 0 {
n >>= 1;
p += 1;
}
1 << p
}
}