The FFT
Đã giới thiệu| English | Ý nghĩa |
|---|---|
| FFT | Biến đổi Fourier nhanh (FFT) |
| butterfly | butterfly |
| twiddle factor | twiddle factor |
Split, twiddle, recurse
- The FFT factors an $N$-point DFT into two $N/2$-point DFTs plus $N$ combining multiplies — and recurses. Each butterfly is one multiply-and-add with a twiddle factor $W_N = e^{-j2\pi/N}$; $\log_2 N$ stages of $N/2$ butterflies.

At $N = 1024$: direct $N^2 = 1\,048\,576$ operations, FFT $N\log_2 N = 10\,240$ — 102× fewer ✓ (N = 4096: 341×).
{exercise:uni-ee1203-count}
Counting complex operations, the N = 1024 FFT beats the direct DFT by —
Độ khó 1/5 · 100 xu + phần thưởng nhỏ khi hoàn thành lần đầu tiên hôm nay
N² = 1 048 576 against N·log₂N = 1024×10 = 10 240: a factor of 102.4, and the gap widens with N.
The analyser in real time
- 1 Hz bins at $f_s = 1$ kHz: $N = 1024$ gives $0.977$ Hz ✓. At 10 ns per operation the transform takes about 100 µs — real time against a 1.024 s record. Bin 100 is $100 \times 0.977 = 97.66$ Hz: the bin-to-hertz map is $k f_s/N$.
{exercise:uni-ee1203-bin} {exercise:uni-ee1203-stages}
fs = 1 kHz and N = 1024: bin 100 sits at —
Độ khó 1/5 · 100 xu + phần thưởng nhỏ khi hoàn thành lần đầu tiên hôm nay
f = k·fs/N = 100×1000/1024 = 97.66 Hz — reading bin numbers as hertz is the classic lab error.
A radix-2 decimation-in-time FFT of N = 1024 has —
Độ khó 1/5 · 100 xu + phần thưởng nhỏ khi hoàn thành lần đầu tiên hôm nay
log₂1024 = 10 stages, each with N/2 = 512 butterflies; inputs arrive in bit-reversed order.