package com.test.test2;

public class FFT {
    public static final int FFT_N_LOG = 10; // FFT_N_LOG <= 13
    public static final int FFT_N = 1 << FFT_N_LOG;
    private static final float MINY = (float) ((FFT_N << 2) * Math.sqrt(2)); // (*)
    private final float[] real, imag, sintable, costable;
    private final int[] bitReverse;

    public FFT() {
        real = new float[FFT_N];
        imag = new float[FFT_N];
        sintable = new float[FFT_N >> 1];
        costable = new float[FFT_N >> 1];
        bitReverse = new int[FFT_N];

        int i, j, k, reve;
        for (i = 0; i < FFT_N; i++) {
            k = i;
            for (j = 0, reve = 0; j != FFT_N_LOG; j++) {
                reve <<= 1;
                reve |= (k & 1);
                k >>>= 1;
            }
            bitReverse[i] = reve;
        }

        double theta, dt = 2 * 3.14159265358979323846 / FFT_N;
        for (i = 0; i < (FFT_N >> 1); i++) {
            theta = i * dt;
            costable[i] = (float) Math.cos(theta);
            sintable[i] = (float) Math.sin(theta);
        }
    }

    /**
     * 用於頻譜顯示的快速傅立葉變換
     *
     * @param realIO
     *            輸入FFT_N個實數，也用它暫存fft後的FFT_N/2個輸出值(複數模的平方)。
     */
    public void calculate(float[] realIO) {
        int i, j, k, ir, exchanges = 1, idx = FFT_N_LOG - 1;
        float cosv, sinv, tmpr, tmpi;
        for (i = 0; i != FFT_N; i++) {
            real[i] = realIO[bitReverse[i]];
            imag[i] = 0;
        }

        for (i = FFT_N_LOG; i != 0; i--) {
            for (j = 0; j != exchanges; j++) {
                cosv = costable[j << idx];
                sinv = sintable[j << idx];
                for (k = j; k < FFT_N; k += exchanges << 1) {
                    ir = k + exchanges;
                    tmpr = cosv * real[ir] - sinv * imag[ir];
                    tmpi = cosv * imag[ir] + sinv * real[ir];
                    real[ir] = real[k] - tmpr;
                    imag[ir] = imag[k] - tmpi;
                    real[k] += tmpr;
                    imag[k] += tmpi;
                }
            }
            exchanges <<= 1;
            idx--;
        }

        j = FFT_N >> 1;
        /*
         * 輸出模的平方(的FFT_N倍):
         * for(i = 1; i <= j; i++)
         * realIO[i-1] = real[i] * real[i] + imag[i] * imag[i];
         *
         * 如果FFT只用於頻譜顯示,可以"淘汰"幅值較小的而減少浮點乘法運算. MINY的值
         * 和Spectrum.Y0,Spectrum.logY0對應.
         */
        sinv = MINY;
        cosv = -MINY;
        for (i = j; i != 0; i--) {
            tmpr = real[i];
            tmpi = imag[i];
            if (tmpr > cosv && tmpr < sinv && tmpi > cosv && tmpi < sinv)
                realIO[i - 1] = 0;
            else
                realIO[i - 1] = tmpr * tmpr + tmpi * tmpi;
        }
    }

    public static void main(String[] args) {
        FFT fft2 = new FFT();
        float[] realIo = { 1, 2 };
        fft2.calculate(realIo);
    }
}