On Jun 27, 9:39=A0am, tonyjian...@[EMAIL PROTECTED]
wrote:
> I implemented the matlab simulation model according to hardware
> architecture.The impulse response is ok when the impulse length is
> short.But when I increase the length of impulse stimulus, I found
> increasing spurs. I don't know whether this is caused by the
> instability of the filter or some other reasons. Attached is the
> matlab code,and at the end of code I added the matlab internal
> function mfilt.cicdecim for comparison.
>
> %Filter Stimulus
> %data_in =3Dsin((1:2^15)*pi/2^10);
> data_in =3D[1 zeros(1,4*6000-1)];
>
> %Filter Parameters
> R=3D 4;%decimation factor
> M=3D1;%differential delay
> N=3D5;%cascaded numbers of section
>
> %Integrator Section Registers Initialization
> intg_reg1(1) =3D data_in(1);
> intg_reg1_d(1) =3D0;
> intg_reg2(1) =3D 0;
> intg_reg2_d(1) =3D0;
> intg_reg3(1) =3D 0;
> intg_reg3_d(1) =3D0;
> intg_reg4(1) =3D 0;
> intg_reg4_d(1) =3D0;
> intg_reg5(1) =3D 0;
> intg_reg5_d(1) =3D0;
>
> for i=3D2:length(data_in)
>
> =A0 =A0 intg_reg1_d(i) =3D intg_reg1(i-1);
> =A0 =A0 intg_reg1(i) =3D intg_reg1_d(i) + data_in(i);
>
> =A0 =A0 intg_reg2_d(i) =3D intg_reg2(i-1);
> =A0 =A0 intg_reg2(i) =3D intg_reg2_d(i) + intg_reg1_d(i);
>
> =A0 =A0 intg_reg3_d(i) =3D intg_reg3(i-1);
> =A0 =A0 intg_reg3(i) =3D intg_reg3_d(i) + intg_reg2_d(i);
>
> =A0 =A0 intg_reg4_d(i) =3D intg_reg4(i-1);
> =A0 =A0 intg_reg4(i) =3D intg_reg4_d(i) + intg_reg3_d(i);
>
> =A0 =A0 intg_reg5_d(i) =3D intg_reg5(i-1);
> =A0 =A0 intg_reg5(i) =3D intg_reg5_d(i) + intg_reg4_d(i);
>
> end
>
> % Decimation by R
>
> deci_data=3Dintg_reg5_d(R:R:end);
>
> %Comb Section Registers Initialization
> diff_reg1(1) =3D deci_data(1);
> diff_reg1_d(1) =3D 0;
> diff_reg2(1) =3D 0;
> diff_reg2_d(1) =3D0;
> diff_reg3(1) =3D 0;
> diff_reg3_d(1) =3D 0;
> diff_reg4(1) =3D 0;
> diff_reg4_d(1) =3D 0;
> diff_reg5(1) =3D 0;
> diff_reg5_d(1) =3D 0;
> diff_reg6(1) =3D 0;
>
> for ii=3D2:length(deci_data)
>
> =A0 =A0 diff_reg1(ii) =3D deci_data(ii);
> =A0 =A0 diff_reg1_d(ii) =3D diff_reg1(ii-1);
>
> =A0 =A0 diff_reg2(ii) =3D diff_reg1(ii)-diff_reg1_d(ii);
> =A0 =A0 diff_reg2_d(ii) =3D diff_reg2(ii-1);
>
> =A0 =A0 diff_reg3(ii) =3D diff_reg2(ii)-diff_reg2_d(ii);
> =A0 =A0 diff_reg3_d(ii) =3D diff_reg3(ii-1);
>
> =A0 =A0 diff_reg4(ii) =3D diff_reg3(ii)-diff_reg3_d(ii);
> =A0 =A0 diff_reg4_d(ii) =3D diff_reg4(ii-1);
>
> =A0 =A0 diff_reg5(ii) =3D diff_reg4(ii)-diff_reg4_d(ii);
> =A0 =A0 diff_reg5_d(ii) =3D diff_reg5(ii-1);
>
> =A0 =A0 diff_reg6(ii) =3D diff_reg5(ii)-diff_reg5_d(ii);
> end
> data_out =3D diff_reg6;
>
> data_in =3D[1 zeros(1,4*10000-1)];
>
> hm=3Dmfilt.cicdecim(4,1,5);
> cic_ir=3Ddouble(filter(hm,data_in));
I got it! In fact double type in matlab is represented with
54(including 1 sign bit) ,so the so-call floating point simulation is
fixed-point theoritically. Thus overflow is inevitable with the
increase of integration length."Wrap-around" is needed according to
Hogenauer's paper in order to avoid overflow error.
|
