iir filter implementation

Started by Kuppala Sridhar March 19, 2007

Sorry for the trouble.
But I was able to find the solution myself.

For plotting an image , we should use the plot X_Y option. its well explained in the manual.



From: Amit Kumar Verma
To: Kuppala Sridhar ; "a..."
Sent: Thursday, 26 April 2012 11:49 AM
Subject: Re: [adsp] Re: iir filter implementation

Dear friends,

Kindly let me know whether it is possible to process 2-D arrays directly ( I mean without converting the data to 1-D variables) in visual DSP directly.

Also kindly let me know how to go about plotting 2-d data in visual dsp.

eagerly waiting for your reply.



From: Kuppala Sridhar
To: a...
Sent: Saturday, 24 March 2007 2:19 AM
Subject: Re: [adsp] Re: iir filter implementation

hello Bernhard,
as per ur  suggestions,  i have made changes with respect to the sign change of the filter coefficients. But even then my frequency response is not proper.please see the code below whether any bug is there.

#define FRAMESIZE 1   /* stereo  */
#define SECTIONS 1    /* Number of second-order sections (biquads)         */
                      /* Each channel will be filtered by SECTIONS biquads */
/*********** ********* ********* ********* ********* ********* ********* ******/
.section /dm seg_dmda;

.var  inbuf[FRAMESIZE] ;  /* All of the channels in each frame are processed
                         /*   when
the last channel arrives and the         */

.var  outbuf[FRAMESIZE] ; /*   double-buffering feeds the SPORT as needed    */
                         /*  tag data to check if ever assigned a value     */

.var  delaybuf[SECTIONS* 4]; /* Holds intermediate variables (state) for the */
                            /*   IIR.  Each of the two compute units needs  */
                            /*   storage for w[x-1] and w[x-2]
each     */
                            /*   biquad section.                            */

/*********** ********* ********* ********* ********* ********* ********* ******/
.section /pm seg_pmda;

 /* Coefficients are ordered a2,a1,b2,b1 and coefficients for PEx are      */
 /*   immediately followed by those for PEy.  Repeat the sequence for each */
 /*   successive cascaded biquad
pair.                                     */

.var   coefficients[ SECTIONS* 8] {
     1.341081473676422 ,  - 0.571654398453466,
1.994354460540368 ,   0.999999999999998 ,     

/*********** ********* ********* ********* ********* ********* ********* ******/
.section /pm seg_pmco;
_initializeAlgorith mDataStructures:

        m1=1;  /* modifiers=2 for SIMD reads & writes */

    /* Not using circular addressing with any of the pointers, so */
    /*  we must ensure that it is disabled by
setting length = 0 

        l3=0;  /* i3 points to inbuf */
        l4=0;  /* i4 points to outbuf */

        /* i0 and i1 are pointer into delaybuf */
        /*    and track w[n-1] and w[n-2]      */

        l0=4;  /* w[n-1], w[n-2] for PEx, PEy */

        l8=0;  /* i8 points to the coefficient buffer */

        /* zero the delay line */


        lcntr=r0, do
clear until
lce;     /* for each section, do:       */
            dm(i0,m2)                  /*   clear w[n-1] and w[n-2]  */
clear:      dm(i0,m2)


_initializeAlgorith mDataStructures. end:
    bit set mode1 CBUFEN;//           // Enable SIMD mode
    b0 = delaybuf;
    b1 = b0;
    b3 = inbuf;
    b4 = outbuf;
    b9 = coefficients;
    r0 = SECTIONS;

f8=dm(i3,m1) ;                               // read inbuf
    r12=r12 xor r12,    f2=dm(i0,m1) ,  f4=pm(i8,m8) ;
       lcntr=r0, do quads until lce;
            f12f4,         f8f12,       f3=dm(i0,m1) ,     f4=pm(i8,m8) ;
            f12f4,         f8f12,      dm(i1,m1)     f4=pm(i8,m8) ;
            f12f4,         f8f12,       f2=dm(i0,m1) ,    
f4=pm(i8,m8) ;
quads:      f12f4,         f8f12,       dm(i1,m1)     f4=pm(i8,m8) ;
    rts (db);
    bit clr mode1 CBUFEN ;//                  // disable SIMD mode

_Cascaded_IIR_ Filter_SIMD. end:

Holzmayer.Bernhard@ Foerstergroup. de wrote:
hai mike,
>> thanks for u
r reply.
>> i have also implemented with single biquad but this is not responding the way the frquency response in the matlab was.
>Hello Kuppala,
>here are some additional hints, which might be helpful or not:
>1) be aware that Matlab specifies the coefficient differently from how it is usually done in books or code libraries. Usually they differ in sign.
>Check the algorithm thoroughly.
>If you select coefficients with easy numbers like 0.0 0.5 or 1.0 where appropriate, you can calculate the biquad manually and check if you get the same results.
>2) Depending on the cut-off frequency and the sampling frequency, you might run into wrongly calculated coefficients (precision issue).
>For one biquad and a low-pass filter stage, keeping the relation
>Fc/Fs > 0.001 should be safe.
>Check the coefficients. Be cautious, if the values are very close to 0.0 or 1.0 or -1.0
>3) Check what
your compiler does. If
it calculates the biquad based on 16 bit wide values, the algorithm will fail or at least provide a distorted result. Even 32bit wide values might be unsufficient, depending on type and algorithm.
>Keep in mind, that IIR filters do infinite recursions on the calculated values. This means, that errors might increase at every cycle.
>If you can watch the algorithm in a debugger, look for sign changes (values around zero) and maybe saturation or rollover effects at the border of the value range (>=1.0, <=-1,0 ).
>Good luck.

Here’s a new way to find what you're looking for - Yahoo! Answers