我目前正在尝试在java中实现FFT算法,并且遇到了一些麻烦!我已经很好地测试了算法的所有其他部分,它们似乎工作正常。
我遇到的麻烦是,在基本情况下,它返回一个复数数组,在基本情况A[0]
内填充。在执行基本案例之后,执行for循环,其中y0[0]
和y1[0]
被发现为 null ,尽管将它们分配给基本案例,相当困惑于此。这显示在System.out.println
行
有谁能告诉我我的方式错误?
//This method implements the recursive FFT algorithm, it assumes the input length
//N is some power of two
private static Complex[] FFT(Complex[] A, int N) {
double real, imag;
Complex A0[] = new Complex[((int) Math.ceil(N/2))];
Complex A1[] = new Complex[((int) Math.ceil(N/2))];
Complex[] omega = new Complex[N];
Complex[] y = new Complex[N];
Complex[] y0 = new Complex[((int) Math.ceil(N/2))];
Complex[] y1 = new Complex[((int) Math.ceil(N/2))];
//base case
if (N == 1) {
return A;
}
else {
real = Math.cos((2*Math.PI)/N); if (real < 1E-10 && real > 0) real = 0;
imag = Math.sin((2*Math.PI)/N); if (imag < 1E-10 && imag > 0) imag = 0;
omega[N-1] = new Complex(real, imag);
omega[0] = new Complex(1, 0);
A0 = splitInput(A, 1);
A1 = splitInput(A, 0);
//recursive calls
y0 = FFT(A0, N/2);
y1 = FFT(A1, N/2);
for (int k = 0; k < ((N/2)-1); k++) {
System.out.print("k: " + k + ", y0: " + y0[k]); System.out.println(", y1: " + y1[k]);
y[k] = y0[k].plus(omega[k].times(y1[k]));
y[k+(N/2)] = y0[k].minus(omega[k].times(y1[k]));
omega[0] = omega[0].times(omega[N]);
}
return y;
}
}
以下是我的splitInput方法的代码
//This method takes a double array as an argument and returns every even or odd
//element according to the second int argument being 1 or 0
private static Complex[] splitInput(Complex[] input, int even) {
Complex[] newArray = new Complex[(input.length/2)];
//Return all even elements of double array, including 0
if (even == 1) {
for (int i = 0; i < (input.length/2); i++) {
newArray[i] = new Complex(input[i*2].re, 0.0);
}
return newArray;
}
//Return all odd elements of double array
else {
for (int i = 0; i < (input.length/2); i++) {
newArray[i] = new Complex (input[(i*2) + 1].re, 0.0);
}
return newArray;
}
}
编辑:我已根据您的建议更新了我的代码,仍然从行y[k] = y0[k].plus(omega[k].times(y1[k]));
获取空指针异常为y0
&amp;在基本情况之后,y1
仍然是null
:(任何进一步的想法?这是更新后的算法
//This method implements the recursive FFT algorithm, it assumes the input length
//N is some power of two
private static Complex[] FFT(Complex[] A, int N) {
double real, imag;
Complex[] omega = new Complex[N];
Complex[] y = new Complex[N];
Complex[] A0;
Complex[] A1;
Complex[] y0;
Complex[] y1;
//base case
if (N == 1) {
return A;
}
else {
real = Math.cos((2*Math.PI)/N); if (real < 1E-10 && real > 0) real = 0;
imag = Math.sin((2*Math.PI)/N); if (imag < 1E-10 && imag > 0) imag = 0;;
omega[N-1] = new Complex(real, imag);
omega[0] = new Complex(1, 0);
A0 = splitInput(A, 1);
A1 = splitInput(A, 0);
//recursive calls
y0 = FFT(A0, N/2);
y1 = FFT(A1, N/2);
for (int k = 0; k < ((N/2)-1); k++) {
y[k] = y0[k].plus(omega[k].times(y1[k]));
y[k+(N/2)] = y0[k].minus(omega[k].times(y1[k]));
omega[0] = omega[0].times(omega[N-1]);
}
return y;
}
}
答案 0 :(得分:2)
一些想法:
每当我看到像Math.ceil(N/2)
这样经常重复的事情时,我认为它有理由拥有自己的命名变量。 (我知道命名变量并不总是很容易,但我发现它对于易读性至关重要。)
Complex A0[] = new Complex[((int) Math.ceil(N/2))];
Complex A1[] = new Complex[((int) Math.ceil(N/2))];
请注意,N==1
时,计算结果为new Complex[0]
。我不确定这是做什么的,但我想我会在内存分配之前进行N == 1
基本情况检查。
Complex[] y0 = new Complex[((int) Math.ceil(N/2))];
Complex[] y1 = new Complex[((int) Math.ceil(N/2))];
/* ... */
y0 = FFT(A0, N/2);
y1 = FFT(A1, N/2);
我相信你可以跳过这些数组的new Complex[...]
分配,因为你从未真正存储任何数据。
Complex[] omega = new Complex[N];
/* ... */
omega[0] = omega[0].times(omega[N]);
我很惊讶这还没有爆炸 - omega[N]
应该引发IndexOutOfBounds
例外。
答案 1 :(得分:1)
跳出来的问题:
(int) Math.ceil(N/2)
您仍在进行int
除法,因此Math.ceil()
无效,而您的拆分数组可能不正确n
omega[0]
和omega[N-1]
,当您尝试访问NullPointerException
时,我会期望omega[1]
,这会在N >= 6
时出现。< / LI>
omega[N]
,sarnold也提到了A0
和A1
,然后为其分配splitInput