zoukankan      html  css  js  c++  java
  • FFT的物理意义

    FFT是离散傅立叶变换的快速算法,可以将一个信号变换
    到频域。有些信号在时域上是很难看出什么特征的,但是如
    果变换到频域之后,就很容易看出特征了。这就是很多信号
    分析采用FFT变换的原因。另外,FFT可以将一个信号的频谱
    提取出来,这在频谱分析方面也是经常用的。

        虽然很多人都知道FFT是什么,可以用来做什么,怎么去
    做,但是却不知道FFT之后的结果是什意思、如何决定要使用
    多少点来做FFT。

        现在圈圈就根据实际经验来说说FFT结果的具体物理意义。
    一个模拟信号,经过ADC采样之后,就变成了数字信号。采样
    定理告诉我们,采样频率要大于信号频率的两倍,这些我就
    不在此罗嗦了。

        采样得到的数字信号,就可以做FFT变换了。N个采样点,
    经过FFT之后,就可以得到N个点的FFT结果。为了方便进行FFT
    运算,通常N取2的整数次方。

        假设采样频率为Fs,信号频率F,采样点数为N。那么FFT
    之后结果就是一个为N点的复数。每一个点就对应着一个频率
    点。这个点的模值,就是该频率值下的幅度特性。具体跟原始
    信号的幅度有什么关系呢?假设原始信号的峰值为A,那么FFT
    的结果的每个点(除了第一个点直流分量之外)的模值就是A
    的N/2倍。而第一个点就是直流分量,它的模值就是直流分量
    的N倍。而每个点的相位呢,就是在该频率下的信号的相位。
    第一个点表示直流分量(即0Hz),而最后一个点N的再下一个
    点(实际上这个点是不存在的,这里是假设的第N+1个点,也
    可以看做是将第一个点分做两半分,另一半移到最后)则表示
    采样频率Fs,这中间被N-1个点平均分成N等份,每个点的频率
    依次增加。例如某点n所表示的频率为:Fn=(n-1)*Fs/N。
    由上面的公式可以看出,Fn所能分辨到频率为为Fs/N,如果
    采样频率Fs为1024Hz,采样点数为1024点,则可以分辨到1Hz。
    1024Hz的采样率采样1024点,刚好是1秒,也就是说,采样1秒
    时间的信号并做FFT,则结果可以分析到1Hz,如果采样2秒时
    间的信号并做FFT,则结果可以分析到0.5Hz。如果要提高频率
    分辨力,则必须增加采样点数,也即采样时间。频率分辨率和
    采样时间是倒数关系。
      假设FFT之后某点n用复数a+bi表示,那么这个复数的模就是
    An=根号a*a+b*b,相位就是Pn=atan2(b,a)。根据以上的结果,
    就可以计算出n点(n≠1,且n<=N/2)对应的信号的表达式为:
    An/(N/2)*cos(2*pi*Fn*t+Pn),即2*An/N*cos(2*pi*Fn*t+Pn)。
    对于n=1点的信号,是直流分量,幅度即为A1/N。
        由于FFT结果的对称性,通常我们只使用前半部分的结果,
    即小于采样频率一半的结果。

        好了,说了半天,看着公式也晕,下面圈圈以一个实际的
    信号来做说明。

        假设我们有一个信号,它含有2V的直流分量,频率为50Hz、
    相位为-30度、幅度为3V的交流信号,以及一个频率为75Hz、
    相位为90度、幅度为1.5V的交流信号。用数学表达式就是如下:

    S=2+3*cos(2*pi*50*t-pi*30/180)+1.5*cos(2*pi*75*t+pi*90/180)

        式中cos参数为弧度,所以-30度和90度要分别换算成弧度。
    我们以256Hz的采样率对这个信号进行采样,总共采样256点。
    按照我们上面的分析,Fn=(n-1)*Fs/N,我们可以知道,每两个
    点之间的间距就是1Hz,第n个点的频率就是n-1。我们的信号
    有3个频率:0Hz、50Hz、75Hz,应该分别在第1个点、第51个点、
    第76个点上出现峰值,其它各点应该接近0。实际情况如何呢?
    我们来看看FFT的结果的模值如图所示。

                          图1 FFT结果
        从图中我们可以看到,在第1点、第51点、和第76点附近有
    比较大的值。我们分别将这三个点附近的数据拿上来细看:
    1点: 512+0i
    2点: -2.6195E-14 - 1.4162E-13i 
    3点: -2.8586E-14 - 1.1898E-13i

    50点:-6.2076E-13 - 2.1713E-12i
    51点:332.55 - 192i
    52点:-1.6707E-12 - 1.5241E-12i

    75点:-2.2199E-13 -1.0076E-12i
    76点:3.4315E-12 + 192i
    77点:-3.0263E-14 +7.5609E-13i
       
        很明显,1点、51点、76点的值都比较大,它附近的点值
    都很小,可以认为是0,即在那些频率点上的信号幅度为0。
    接着,我们来计算各点的幅度值。分别计算这三个点的模值,
    结果如下:
    1点: 512
    51点:384
    76点:192
        按照公式,可以计算出直流分量为:512/N=512/256=2;
    50Hz信号的幅度为:384/(N/2)=384/(256/2)=3;75Hz信号的
    幅度为192/(N/2)=192/(256/2)=1.5。可见,从频谱分析出来
    的幅度是正确的。
        然后再来计算相位信息。直流信号没有相位可言,不用管
    它。先计算50Hz信号的相位,atan2(-192, 332.55)=-0.5236,
    结果是弧度,换算为角度就是180*(-0.5236)/pi=-30.0001。再
    计算75Hz信号的相位,atan2(192, 3.4315E-12)=1.5708弧度,
    换算成角度就是180*1.5708/pi=90.0002。可见,相位也是对的。
    根据FFT结果以及上面的分析计算,我们就可以写出信号的表达
    式了,它就是我们开始提供的信号。

        总结:假设采样频率为Fs,采样点数为N,做FFT之后,某
    一点n(n从1开始)表示的频率为:Fn=(n-1)*Fs/N;该点的模值
    除以N/2就是对应该频率下的信号的幅度(对于直流信号是除以
    N);该点的相位即是对应该频率下的信号的相位。相位的计算
    可用函数atan2(b,a)计算。atan2(b,a)是求坐标为(a,b)点的角
    度值,范围从-pi到pi。要精确到xHz,则需要采样长度为1/x秒
    的信号,并做FFT。要提高频率分辨率,就需要增加采样点数,
    这在一些实际的应用中是不现实的,需要在较短的时间内完成
    分析。解决这个问题的方法有频率细分法,比较简单的方法是
    采样比较短时间的信号,然后在后面补充一定数量的0,使其长度
    达到需要的点数,再做FFT,这在一定程度上能够提高频率分辨力。
    具体的频率细分法可参考相关文献。

    [附录:本测试数据使用的matlab程序]
    close all; %先关闭所有图片
    Adc=2;  %直流分量幅度
    A1=3;   %频率F1信号的幅度
    A2=1.5; %频率F2信号的幅度
    F1=50;  %信号1频率(Hz)
    F2=75;  %信号2频率(Hz)
    Fs=256; %采样频率(Hz)
    P1=-30; %信号1相位(度)
    P2=90;  %信号相位(度)
    N=256;  %采样点数
    t=[0:1/Fs:N/Fs]; %采样时刻

    %信号
    S=Adc+A1*cos(2*pi*F1*t+pi*P1/180)+A2*cos(2*pi*F2*t+pi*P2/180);
    %显示原始信号
    plot(S);
    title('原始信号');

    figure;
    Y = fft(S,N); %做FFT变换
    Ayy = (abs(Y)); %取模
    plot(Ayy(1:N)); %显示原始的FFT模值结果
    title('FFT 模值');

    figure;
    Ayy=Ayy/(N/2);   %换算成实际的幅度
    Ayy(1)=Ayy(1)/2;
    F=([1:N]-1)*Fs/N; %换算成实际的频率值
    plot(F(1:N/2),Ayy(1:N/2));   %显示换算后的FFT模值结果
    title('幅度-频率曲线图');

    figure;
    Pyy=[1:N/2];
    for i=1:N/2
     Pyy(i)=phase(Y(i)); %计算相位
     Pyy(i)=Pyy(i)*180/pi; %换算为角度
    end;
    plot(F(1:N/2),Pyy(1:N/2));   %显示相位图
    title('相位-频率曲线图');

    #include <math.h>
    #include "fftfunc.h"


    const int16 stc_fftrealcoef[FFT_XBNUMS][FFT_N] = {
    {0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0},
    {-14703,565,1108,1609,2048,2408,2676,2841,2896,2841,2676,2408,2048,1609,1108,565,0,-565,-1108,-1609,-2048,-2408,-2676,-2841,-2896,-2841,-2676,-2408,-2048,-1609,-1108,-565,14703},
    {-7103,1365,2239,2772,2882,2554,1837,841,-284,-1365,-2239,-2772,-2882,-2554,-1837,-841,284,1365,2239,2772,2882,2554,1837,841,-284,-1365,-2239,-2772,-2882,-2554,-1837,-841,7387},
    {-4400,2048,2841,2676,1609,0,-1609,-2676,-2841,-2048,-565,1108,2408,2896,2408,1108,-565,-2048,-2841,-2676,-1609,0,1609,2676,2841,2048,565,-1108,-2408,-2896,-2408,-1108,4965},
    {-2925,2554,2772,1365,-841,-2554,-2772,-1365,841,2554,2772,1365,-841,-2554,-2772,-1365,841,2554,2772,1365,-841,-2554,-2772,-1365,841,2554,2772,1365,-841,-2554,-2772,-1365,3766},
    {-1949,2841,2048,-565,-2676,-2408,0,2408,2676,565,-2048,-2841,-1108,1609,2896,1609,-1108,-2841,-2048,565,2676,2408,0,-2408,-2676,-565,2048,2841,1108,-1609,-2896,-1609,3057},
    {-1229,2882,841,-2239,-2554,284,2772,1837,-1365,-2882,-841,2239,2554,-284,-2772,-1837,1365,2882,841,-2239,-2554,284,2772,1837,-1365,-2882,-841,2239,2554,-284,-2772,-1837,2594},
    {-663,2676,-565,-2896,-565,2676,1609,-2048,-2408,1108,2841,0,-2841,-1108,2408,2048,-1609,-2676,565,2896,565,-2676,-1609,2048,2408,-1108,-2841,0,2841,1108,-2408,-2048,2272},
    {-201,2239,-1837,-2239,1837,2239,-1837,-2239,1837,2239,-1837,-2239,1837,2239,-1837,-2239,1837,2239,-1837,-2239,1837,2239,-1837,-2239,1837,2239,-1837,-2239,1837,2239,-1837,-2239,2038},
    {184,1609,-2676,-565,2896,-565,-2676,1609,2048,-2408,-1108,2841,0,-2841,1108,2408,-2048,-1609,2676,565,-2896,565,2676,-1609,-2048,2408,1108,-2841,0,2841,-1108,-2408,1864},
    {506,841,-2882,1365,1837,-2772,284,2554,-2239,-841,2882,-1365,-1837,2772,-284,-2554,2239,841,-2882,1365,1837,-2772,284,2554,-2239,-841,2882,-1365,-1837,2772,-284,-2554,1733},
    {774,0,-2408,2676,-565,-2048,2841,-1108,-1609,2896,-1609,-1108,2841,-2048,-565,2676,-2408,0,2408,-2676,565,2048,-2841,1108,1609,-2896,1609,1108,-2841,2048,565,-2676,1634},
    {994,-841,-1365,2772,-2554,841,1365,-2772,2554,-841,-1365,2772,-2554,841,1365,-2772,2554,-841,-1365,2772,-2554,841,1365,-2772,2554,-841,-1365,2772,-2554,841,1365,-2772,1560},
    {1170,-1609,0,1609,-2676,2841,-2048,565,1108,-2408,2896,-2408,1108,565,-2048,2841,-2676,1609,0,-1609,2676,-2841,2048,-565,-1108,2408,-2896,2408,-1108,-565,2048,-2841,1506},
    {1302,-2239,1365,-284,-841,1837,-2554,2882,-2772,2239,-1365,284,841,-1837,2554,-2882,2772,-2239,1365,-284,-841,1837,-2554,2882,-2772,2239,-1365,284,841,-1837,2554,-2882,1469},
    {1393,-2676,2408,-2048,1609,-1108,565,0,-565,1108,-1609,2048,-2408,2676,-2841,2896,-2841,2676,-2408,2048,-1609,1108,-565,0,565,-1108,1609,-2048,2408,-2676,2841,-2896,1448},
    };

    const int16 stc_fftvircoef[FFT_XBNUMS][FFT_N] = {
    {0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0},
    {1448,2841,2676,2408,2048,1609,1108,565,0,-565,-1108,-1609,-2048,-2408,-2676,-2841,-2896,-2841,-2676,-2408,-2048,-1609,-1108,-565,0,565,1108,1609,2048,2408,2676,2841,1448},
    {2155,2554,1837,841,-284,-1365,-2239,-2772,-2882,-2554,-1837,-841,284,1365,2239,2772,2882,2554,1837,841,-284,-1365,-2239,-2772,-2882,-2554,-1837,-841,284,1365,2239,2772,728},
    {2352,2048,565,-1108,-2408,-2896,-2408,-1108,565,2048,2841,2676,1609,0,-1609,-2676,-2841,-2048,-565,1108,2408,2896,2408,1108,-565,-2048,-2841,-2676,-1609,0,1609,2676,489},
    {2401,1365,-841,-2554,-2772,-1365,841,2554,2772,1365,-841,-2554,-2772,-1365,841,2554,2772,1365,-841,-2554,-2772,-1365,841,2554,2772,1365,-841,-2554,-2772,-1365,841,2554,371},
    {2375,565,-2048,-2841,-1108,1609,2896,1609,-1108,-2841,-2048,565,2676,2408,0,-2408,-2676,-565,2048,2841,1108,-1609,-2896,-1609,1108,2841,2048,-565,-2676,-2408,0,2408,301},
    {2299,-284,-2772,-1837,1365,2882,841,-2239,-2554,284,2772,1837,-1365,-2882,-841,2239,2554,-284,-2772,-1837,1365,2882,841,-2239,-2554,284,2772,1837,-1365,-2882,-841,2239,255},
    {2184,-1108,-2841,0,2841,1108,-2408,-2048,1609,2676,-565,-2896,-565,2676,1609,-2048,-2408,1108,2841,0,-2841,-1108,2408,2048,-1609,-2676,565,2896,565,-2676,-1609,2048,224},
    {2038,-1837,-2239,1837,2239,-1837,-2239,1837,2239,-1837,-2239,1837,2239,-1837,-2239,1837,2239,-1837,-2239,1837,2239,-1837,-2239,1837,2239,-1837,-2239,1837,2239,-1837,-2239,1837,201},
    {1864,-2408,-1108,2841,0,-2841,1108,2408,-2048,-1609,2676,565,-2896,565,2676,-1609,-2048,2408,1108,-2841,0,2841,-1108,-2408,2048,1609,-2676,-565,2896,-565,-2676,1609,184},
    {1667,-2772,284,2554,-2239,-841,2882,-1365,-1837,2772,-284,-2554,2239,841,-2882,1365,1837,-2772,284,2554,-2239,-841,2882,-1365,-1837,2772,-284,-2554,2239,841,-2882,1365,171},
    {1448,-2896,1609,1108,-2841,2048,565,-2676,2408,0,-2408,2676,-565,-2048,2841,-1108,-1609,2896,-1609,-1108,2841,-2048,-565,2676,-2408,0,2408,-2676,565,2048,-2841,1108,161},
    {1212,-2772,2554,-841,-1365,2772,-2554,841,1365,-2772,2554,-841,-1365,2772,-2554,841,1365,-2772,2554,-841,-1365,2772,-2554,841,1365,-2772,2554,-841,-1365,2772,-2554,841,154},
    {960,-2408,2896,-2408,1108,565,-2048,2841,-2676,1609,0,-1609,2676,-2841,2048,-565,-1108,2408,-2896,2408,-1108,-565,2048,-2841,2676,-1609,0,1609,-2676,2841,-2048,565,148},
    {696,-1837,2554,-2882,2772,-2239,1365,-284,-841,1837,-2554,2882,-2772,2239,-1365,284,841,-1837,2554,-2882,2772,-2239,1365,-284,-841,1837,-2554,2882,-2772,2239,-1365,284,145},
    {422,-1108,1609,-2048,2408,-2676,2841,-2896,2841,-2676,2408,-2048,1609,-1108,565,0,-565,1108,-1609,2048,-2408,2676,-2841,2896,-2841,2676,-2408,2048,-1609,1108,-565,0,143},
    };

    /////////////////////////////////////////////////////////////////
    //////////////////////////FFT////////////////////////////////////
    /////////////////////////////////////////////////////////////////
    void CFFTCalc::FFT_Calc(int16 xbno,int16 *pu,int16 *pustart,int16 *puend)
    {
    if(xbno<0 || xbno>=FFT_XBNUMS)
    {
    m_realvalue = 0;
    m_virvalue = 0;
    return;
    }

    if(xbno == 1 && pustart == NULL && puend == NULL)
    {
    FFT_CalcJB(pu);
    return;
    }

    long long ta=0,tb=0;
    const int16 *curpu = pu;
    const int16 *currealcoef = stc_fftrealcoef[xbno];
    const int16 *curvircoef = stc_fftvircoef[xbno];
    for(int16 i=0;i<FFT_N;i++)
    {
    ta += ((int32)(*curpu) * (int32)(*currealcoef)) ;
    tb += ((int32)(*curpu) * (int32)(*curvircoef)) ;
    currealcoef++;
    curvircoef++;
    curpu++;
    if(puend != NULL)
    {
    if(curpu>puend)
    {
    curpu = pustart;
    }
    }
    }
    m_realvalue = ta>>16;
    m_virvalue = tb>>16;
    }

    void CFFTCalc::FFT_CalcJB(int16 *pu)
    {
    long long ta=0,tb=0;

    ta += -14703*((int32)pu[0]-pu[32]);
    ta += 565*((int32)pu[1]+pu[15]-pu[17]-pu[31]);
    ta += 1108*((int32)pu[2]+pu[14]-pu[18]-pu[30]);
    ta += 1609*((int32)pu[3]+pu[13]-pu[19]-pu[29]);
    ta += 2048*((int32)pu[4]+pu[12]-pu[20]-pu[28]);
    ta += 2408*((int32)pu[5]+pu[11]-pu[21]-pu[27]);
    ta += 2676*((int32)pu[6]+pu[10]-pu[22]-pu[26]);
    ta += 2841*((int32)pu[7]+pu[9]-pu[23]-pu[25]);
    ta += 2896*((int32)pu[8]-pu[24]);

    tb += 1448*((int32)pu[0]+pu[32]);
    tb += 2841*((int32)pu[1]-pu[15]-pu[17]+pu[31]);
    tb += 2676*((int32)pu[2]-pu[14]-pu[18]+pu[30]);
    tb += 2408*((int32)pu[3]-pu[13]-pu[19]+pu[29]);
    tb += 2048*((int32)pu[4]-pu[12]-pu[20]+pu[28]);
    tb += 1609*((int32)pu[5]-pu[11]-pu[21]+pu[27]);
    tb += 1108*((int32)pu[6]-pu[10]-pu[22]+pu[26]);
    tb += 565*((int32)pu[7]-pu[9]-pu[23]+pu[25]);
    tb += -2896*((int32)pu[16]);

    m_realvalue = ta>>16;
    m_virvalue = tb>>16;
    }


    int16 CFFTCalc::FFT_CalcAngle()
    {
    iq15 TempValue = 1;//IQ15atan2(IQ15(m_realvalue),IQ15(m_virvalue));
    TempValue = TempValue * IQ15(57.29578);
    return IQ15FLOAT_ROUND_TO_INT(TempValue);
    }

    int16 CFFTCalc::FFT_CalcMagValue()
    {
    return IQint(IQmag(IQ(m_realvalue),IQ(m_virvalue)));
    }

  • 相关阅读:
    染色法判定二分图
    Kruskal算法求最小生成树
    Prim算法求最小生成树
    Floyd算法求多源最短路
    spfa判断负环
    java 线程的使用
    java IO基础
    数据库 EXISTS与NOT EXISTS
    数据库 何为相关查询和不相关查询?
    数据库的基础知识
  • 原文地址:https://www.cnblogs.com/darren-715/p/3759036.html
Copyright © 2011-2022 走看看