CCS/TMS320C5535: 32 Bit Filters

Hi there,

We've been working with these filters and the bit quantity issue is not an issue anymore. Right now, we're trying to get some filters working but, at the moment, we still haven't figured it out. Our problem now is with coefficient scaling or sample scaling. We have the following code:

/******************** IIR FILTER **********************************************/
 
    for (i = 0; i < NX; i++) {
        xaux[i] = (x50hz[i] / (1024 ));
    }
 
    errorcode = fltoq15(xaux, xaux1, NX);
 
    for (j = 0; j < NX; j++) {
        haux[j] = (h5[j] / 2);
    }
 
    errorcode2 = fltoq15(haux, haux1, 5);
    q15tofl(haux1, haux2, 5);
 
 
    // clear
    for (j = 0; j < NX; j++)
        r[j] = 0;                     // clear output buffer (optional)
 
    for (i = 0; i < (2 * NBIQ) + 1; i++)
        dbuffer[i] = 0;  // clear delay buffer (a must)
 
    // compute
      success = iircas5(xaux1, haux1, r, dp, NBIQ, NX);

Both xaux1 and haux1 are values between +1 and -1, which shouldn't raise any problems, but the filter is overflowing. Our input is a 50hz sine wave with amplitude between 1024 and -1024 and the first for scales it down to 1/-1. The coefficients were calculated using matlab and are values between -2 and 2, which are a problem. In this implementation we decided to just divide them by 2, being fully aware that's not the right way to do it. Still, the iircas5 overflows.

The iircas5 routine only stops overflowing when we change the first for to:

 for (i = 0; i < NX; i++) {
        xaux[i] = (x50hz[i] / (1024*128 ));
 }

So, my question is: how can we solve this ? Since the coefficients and data are both in the +1/-1 range, shouldn't that avoid overflowing? If not, how can we scale the coefficients in order to stop overflowing the filter?

Thanks