# Found "strange" behavior in IIR Implementation of VisualAudio V2.6

Question asked by wre on Dec 15, 2011
Latest reply on Dec 20, 2011 by CraigG

Hello,

since some time we do sig proc in the MPX signal range of FM Radio. This work is done with EZ Kit ASDP-BF537 together with the Audio Extender hosting

AD1938 running at 192Khz sample frequ and using an adopted VisualAudio V2.6 SW.

The problem:

My colleague has designed  an LP filter using bi linear z-transformation with a cut-off frequency of about 95KHz, which is pretty close to fs/2 = 96KHz.

The math is done with MathCad.

The filter is designed to get a constant frequency response and a constant group delay until close to the cut-off frequency and an attenuation to until

(theoretically) 0 at fs/2.

The math was checked while using the calculated filter coefficients and put it in the filter polynominal function.

The outcome is the expected frequency response.

The implementation in BF537 VA2.6 IIR instead does not show a constant frequency response but shows a huge maximum (reaching the signal clipping level) at about 48KHz?!?

We did some tests and found that with coefficients that define a filter with a cut-off frequency of about 60 Khz the measured frequency response fits with the calculated one.

Going to a cut-off frequency of 80KHz the frequency response again shows a medium maximum at about 43KHz.

The closer we come to fs/2 with the filter cut-off frequency the higher is the frequency response maximum.

I've attached 2 diagrams

IIR-TP.pdf

Shows MathCad calculated frequency response using the previously calculated coefficients.

Frequenzgang+Gruppenlaufzeit IIR-TP.pdf

Shows the measured frequncy response with the mentioned maximum

The Filter coefficients in this diagrams are just for test and do not show the above metioned 60Khz, 80 Khz, or 95Khz filters.

b0= 0.68

b1= 0.68

a1= 0.36

I now that VA is not longer supported by ADI but

I would appreciate if someone could give us an explanation, some hints, or maybe some links to sites where we could get some more info about this behavior of the VA 2.6 II implementation.

Thanks

Wilfried