在项目中,需要画波形频谱图,因此进行查找,不是很懂相关知识,下列代码主要是针对这篇文章。
http://blog.csdn.net/xcgspring/article/details/4749075
//快速傅里叶变换 /* 入口参数: inv: =1,傅里叶变换; =-1,逆傅里叶变换 N:输入的点数,为偶数,一般为2的幂次级,2,4,8,16... k: 满足N=2^k(k>0),实质上k是N个采样数据可以分解为偶次幂和奇次幂的次数 real[]: inv=1时,存放N个采样数据的实部,inv=-1时,存放傅里叶变换的N个实部 imag[]: inv=1时,存放N个采样数据的虚部,inv=-1时,存放傅里叶变换的N个虚部 出口参数: real[]: inv=1时,返回傅里叶变换的实部,inv=-1时,返回逆傅里叶变换的实部 imag[]: inv=1时,返回傅里叶变换的虚部,inv=-1时,返回逆傅里叶变换的虚部 */ void FFT::dealFFT(double real[], double imag[], double dSp[], int N, int k, int inv) { int i, j, k1, k2, m, step, factor_step; double temp_real, temp_imag, factor_real, factor_imag; if (inv != 1 && inv != -1) return; //double *real = new double[N]; //double *imag = new double[N]; //倒序 j = 0; for (i = 0; i < N; i++) { if (j>i) { temp_real = real[j]; real[j] = real[i]; real[i] = temp_real; temp_imag = imag[j]; imag[j] = imag[i]; imag[i] = temp_imag; } m = N / 2; while (j >= m&&m != 0) { j -= m; m >>= 1; } j += m; } //蝶形运算 for (i = 0; i < k; i++) { step = 1 << (i + 1); factor_step = N >> (i + 1); //旋转因数变化速度 //初始化旋转因子 factor_real = 1.0; factor_imag = 0.0; for (j = 0; j < step / 2; j++) { for (k1 = j; k1 < N; k1 += step) { k2 = k1 + step / 2; //蝶形运算的两个输入 /* temp_real = real[k1] + real[k2] * factor_real - imag[k2] * factor_imag; temp_imag = imag[k1] + real[k2] * factor_imag + imag[k2] * factor_real; real[k2] = real[k1] - (real[k2] * factor_real - imag[k2] * factor_imag); imag[k2] = imag[k1] - (real[k2] * factor_imag + imag[k2] * factor_real); real[k1] = temp_real; imag[k1] = temp_imag;*/ //上面方法虽然直白,但效率太低,稍微改变结构如下: temp_real = real[k2] * factor_real - imag[k2] * factor_imag; temp_imag = real[k2] * factor_imag + imag[k2] * factor_real; real[k2] = real[k1] - temp_real; imag[k2] = imag[k1] - temp_imag; real[k1] = real[k1] + temp_real; imag[k1] = imag[k1] + temp_imag; } factor_real = inv*cos(-2 * PI*(j + 1)*factor_step / N); factor_imag = inv*sin(-2 * PI*(j + 1)*factor_step / N); } } if (inv == -1) { for (i = 0; i <= N - 1; i++) { real[i] = real[i] / N; imag[i] = imag[i] / N; } } for (i = 0; i<N;i++) { dSp[i] = sqrt(real[i] * real[i] + imag[i] * imag[i]); } }
一般好像需要进行下转换,即后半部分和前半部分置换,即1234变成3412.
void FFT::FFTShift(double dp[], int len) { for (int i = 0; i < len / 2; i++) { double tmp = dp[i]; dp[i] = dp[i + len / 2]; dp[i + len / 2] = tmp; } }
此时得到的应该是实部和虚部解出来的频谱图的Y轴电压值,一般频谱图Y轴为dB,因此需要进行转换
void FFT::getFFT(double dRe[], double dIm[], double dSp[], int len, int nBits, double dWorkingImpedance) { dealFFT(dRe, dIm, dSp, len, nBits, 1); FFTShift(dSp,len); //此时得到的应该是实部和虚部解出来的频谱图的Y轴电压值,还需要转换 ////dBW = 10lg(电压^2/阻抗);dBm =dBW+30,注意电压单位是V for (int i = 0; i<len; i++) { dSp[i] = 10 * log10(dSp[i] * dSp[i] / dWorkingImpedance)+30; } }
getFFT()输出之后的dp才是要的频谱图Y轴值,频谱图X轴的坐标得到通过以下方式:
//X轴精确度,采样频率/数据个数 = 步长
m_DeltaX_S = m_dataPara.nSampleFrequency / nDataNumOfPage_S;
data_SX[i / 2] = m_dataPara.nCenterFrequency + count*m_DeltaX_S - m_dataPara.nWorkingBandWidth/2;//中心频率+当前点*步长-带宽/2
在项目中,实际代码如下:
int count = 0; for (int i = 0; i < nDataNumOfPage_S * 2; i++) { if (i % 2 == 0) data_SQ[i / 2] = data_S[i] * m_DeltaY_S; else data_SI[i / 2] = data_S[i] * m_DeltaY_S; if (i % 2 == 0) { count++; data_SX[i / 2] = m_dataPara.nCenterFrequency + count*m_DeltaX_S - m_dataPara.nWorkingBandWidth/2; } } m_dataPara.nWorkingImpedance = 50; FFT fft; int nBits = log10(nDataNumOfPage_S) / log10(2);//因为参数需要是2的N次方
fft.getFFT(data_SQ, data_SI, data_SS, nDataNumOfPage_S, nBits, m_dataPara.nWorkingImpedance); LoadData_S(data_SX, data_SS, nDataNumOfPage_S);
。。。
其他参考文章:
http://blog.sina.com.cn/s/blog_65d639d50101buo1.html
http://blog.csdn.net/hippig/article/details/8778753
http://www.makaidong.com/%E5%8D%9A%E5%AE%A2%E5%9B%AD%E6%8E%92%E8%A1%8C%E6%A6%9C/20151025/365773.html