iir滤波器c语言程序,请教C语言做iir滤波器问题

最近在做轴承故障检测,C语言编程,在滤波部分遇到了问题,由于本人对滤波知之甚少,想求助:

56860a9f6a056cb72852af58e4093133.png

这是测得的振动信号,在这个基础上我分别用C语言与LabVIEW对其做了滤波,都是用iir巴特沃兹滤波,频域上两者结果一样(我就不放图了),但时域差距很大。

60ee2fef080a756037b7e0741bcec5a6.png

这是LabVIEW做得滤波之后的时域图像看上去很正常,iir巴特沃兹滤波3阶

431108054b00ee6c99b8c5adfb9f925a.png

这是C语言同样采用iir巴特沃兹滤波的时域结果,非常不像。为什么会出现这样的问题。C语言是用的双线性变换

void iirbcf(short ifilt,short band,short ns,short n,double f1,double f2,double f3,double f4,double db,double b[],double a[])

{

short k;

double omega,lamda,epslon,fl,fh;

double d[5],c[5];

if((band==1)||(band==4)) fl=f1;

if((band==2)||(band==3)) fl=f2;

if(band<=3) fh=f3;

if(band==4) fh=f4;

if(ifilt<3)

{ switch (band)

{ case 1:

case 2:

{ omega=warp(f2)/warp(f1);

break;

}

case 3:

{ omega=omin(bpsub(warp(f1),fh,fl),bpsub(warp(f4),fh,fl));

break;

}

case 4:

{ omega=omin(1.0/bpsub(warp(f2),fh,fl),1.0/bpsub(warp(f3),fh,fl));}

}

lamda=pow(10.0,(db/20.0));

epslon=lamda/cosh(2*ns*cosh1(omega));

}

for(k=0;k

{ switch(ifilt)

{ case 1:

{ chebyi(2*ns,k,4,epslon,d,c);

break;

}

case 2:

{ chebyii(2*ns,k,4,omega,lamda,d,c);

break;

}

case 3:

{ bwtf(2*ns,k,4,d,c);

break;

}

}

fblt(d,c,n,band,fl,fh,&b[k*(n+1)+0],&a[k*(n+1)+0]);

}

}

double cosh1(double x)

{ double z;

z=log(x+sqrt(x*x-1.0));

return(z);

}

double warp(double f)

{ double pi,z;

pi=4.0*atan(1.0);

z=tan(pi*f);

return(z);

}

double bpsub(double om,double fh,double fl)

{ double z;

z=(om*om-warp(fh)*warp(fl))/(warp(fh)-warp(fl)*om);

return(z);

}

double omin(double om1,double om2)

{ double z,z1,z2;

z1=fabs(om1);

z2=fabs(om2);

z=(z1

return(z);

}

void bwtf(short ln,short k,short n,double d[],double c[])

{ short i;

double pi,tmp;

pi=4.0*atan(1.0);

d[0]=1.0;

c[0]=1.0;

for(i=1;i<=n;i++)

{

d[i]=0.0;

c[i]=0.0;

}

tmp=(k+1)-(ln+1.0)/2.0;

if(tmp==0.0)

{ c[1]=1.0;

}

else

{

c[1]=-2.0*cos((2*(k+1)+ln-1)*pi/(2*ln));

c[2]=1.0;

}

}

void chebyi(short ln,short k,short n,double ep,double d[],double c[])

{ short i;

double pi,gam,omega,sigma;

pi=4.0*atan(1.0);

gam=pow(((1.0+sqrt(1.0+ep*ep))/ep),1.0/ln);

sigma=0.5*(1.0/gam-gam)*sin((2*(k+1)-1)*pi/(2*ln));

omega=0.5*(1.0/gam+gam)*cos((2*(k+1)-1)*pi/(2*ln));

for (i=0;i<=n;i++)

{ d[i]=0.0;

c[i]=0.0;

}

if (((ln%2)==1)&&((k+1)==(ln+1)/2))

{ d[0]=-sigma;

c[0]=d[0];

c[1]=1.0;

}

else

{ c[0]=sigma*sigma+omega*omega;

c[1]=-2.0*sigma;

c[2]=1.0;

d[0]=c[0];

if(((ln%2)==0)&&(k==0))

d[0]=d[0]/sqrt(1.0+ep*ep);

}

}

void chebyii(short ln,short k,short n,double ws,double att,double d[],double c[])

{ short i;

double pi,gam,alpha,beta,sigma,omega,scln,scld;

pi=4.0*atan(1.0);

gam=pow((att+sqrt(att*att-1.0)),1.0/ln);

alpha=0.5*(1.0/gam-gam)*sin((2*(k+1)-1)*pi/(2*ln));

beta=0.5*(1.0/gam+gam)*cos((2*(k+1)-1)*pi/(2*ln));

sigma=ws*alpha/(alpha*alpha+beta*beta);

omega=-1.0*ws*beta/(alpha*alpha+beta*beta);

for (i=0;i<=n;i++)

{ d[i]=0.0;

c[i]=0.0;

}

if(((ln%2)==1)&&((k+1)==(ln+1)/2))

{ d[0]=-1.0*sigma;

c[0]=d[0];

c[1]=1.0;

}

else

{ scln=sigma*sigma+omega*omega;

scld=pow((ws/cos((2*(k+1)-1)*pi/(2*ln))),2);

d[0]=scln*scld;

d[2]=scln;

c[0]=d[0];

c[1]=-2.0*sigma*scld;

c[2]=scld;

}

}

void fblt(double d[],double c[],short n,short band,double fln,double fhn,double b[],double a[])

{

short i,k,m,n1,n2,ls;

double pi,w,w0,w1,w2,tmp,tmpd,tmpc,*work;

pi=4.0*atan(1.0);

w1=tan(pi*fln);

for(i=n;i>=0;i--)

{ if((c[i]!=0.0)||(d[i]!=0.0))

break;

}

m=i;

switch (band)

{

case 1:

case 2:

{ n2=m;

n1=n2+1;

if(band==2)

{ for(i=0;i<=m/2;i++)

{ tmp=d[i];

d[i]=d[m-i];

d[m-i]=tmp;

tmp=c[i];

c[i]=c[m-i];

c[m-i]=tmp;

}

}

for(i=0;i<=m;i++)

{

d[i]=d[i]/pow(w1,i);

c[i]=c[i]/pow(w1,i);

}

break;

}

case 3:

case 4:

{n2=2*m;

n1=n2+1;

work=malloc(n1*n1*sizeof(double));

w2=tan(pi*fhn);

w=w2-w1;

w0=w1*w2;

if(band==4)

{ for(i=0;i<=m/2;i++)

{ tmp=d[i];

d[i]=d[m-i];

d[m-i]=tmp;

tmp=c[i];

c[i]=c[m-i];

c[m-i]=tmp;

}

}

for(i=0;i<=n2;i++)

{ work[0*n1+i]=0.0;

work[1*n1+i]=0.0;

}

for(i=0;i<=m;i++)

{tmpd=d[i]*pow(w,(m-i));

tmpc=c[i]*pow(w,(m-i));

for(k=0;k<=i;k++)

{ ls=m+i-2*k;

tmp=combin(i,i)/(combin(k,k)*combin(i-k,i-k));

work[0*n1+ls]+=tmpd*pow(w0,k)*tmp;

work[1*n1+ls]+=tmpc*pow(w0,k)*tmp;

}

}

for(i=0;i<=n2;i++)

{d[i]=work[0*n1+i];

c[i]=work[1*n1+i];

}

free(work);

}

}

bilinear(d,c,b,a,n);

}

double combin(short i1,short i2)

{ short i;

double s;

s=1.0;

if(i2==0) return (s);

for(i=i1;i>(i1-i2);i--)

{ s*=i; }

return (s);

}

void bilinear(double d[],double c[],double b[],double a[],short n)

{short i,j,n1;

double sum,atmp,scale, *temp;

n1=n+1;

temp=malloc(n1*n1*sizeof(double));

for(j=0;j<=n;j++)

{

temp[j*n1+0]=1.0;

}

sum=1.0;

for(i=1;i<=n;i++)

{

sum=sum*(double)(n-i+1)/(double)i;

temp[0*n1+i]=sum;

}

for(i=1;i<=n;i++)

for(j=1;j<=n;j++)

{ temp[j*n1+i]=temp[(j-1)*n1+i]-temp[j*n1+i-1]-temp[(j-1)*n1+i-1];

}

for(i=n;i>=0;i--)

{ b[i]=0.0;

atmp=0.0;

for(j=0;j<=n;j++)

{ b[i]=b[i]+temp[j*n1+i]*d[j];

atmp=atmp+temp[j*n1+i]*c[j];

}

scale=atmp;

if(i!=0) a[i]=atmp;

}

for(i=0;i<=n;i++)

{ b[i]=b[i]/scale;

a[i]=a[i]/scale;

}

a[0]=1.0;

free(temp);

}

这是C语言滤波程序,我只用了里面的巴特沃兹

  • 0
    点赞
  • 0
    收藏
  • 0
    评论

“相关推荐”对你有帮助么?

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
©️2022 CSDN 皮肤主题:数字20 设计师:CSDN官方博客 返回首页
评论
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、C币套餐、付费专栏及课程。

余额充值