IIR coefficient greater than 1

Peng Yu

Prodigy 35 points

Hello,

I have tried to designed a second order IIR filter using MATLAB.

The coefficients are 0.9427 -1.8856 0.9428 1 -1.8848 0.8863(sos format). 2 coefficients are greater than 1.

I know that in dsplib the function iircas5 uses Q.15. So what should I do to implement the filter?

Does anyone know how to fix this? Thanks!

over 14 years ago

0 Hyun Kim over 14 years ago

TI__Genius 14965 points

hi,

All coefficient scaled down by 2 and later multiply by 2 after filter applied.

Regards,

Hyun

0 Jeff Hayes over 14 years ago

TI__Expert 6850 points

A similar discussion on IIR filters and coefficient generation is located in the following thread. Perhaps this would help, as well:

http://e2e.ti.com/support/dsp/tms320c6000_high_performance_dsps/f/115/p/68558/253954.aspx?PageIndex=1

0 Peng Yu over 14 years ago in reply to Hyun Kim

Prodigy 35 points

In dsplib, function iircas5 only has 5 adjustable coefficients: b0, b1, b2, a1, a2. I can only scale down these five, but the coefficient a0 is fixed to 1. So the output must be wrong.

0 Anton Proskurnin over 14 years ago in reply to Peng Yu

Prodigy 40 points

I have the same issue. For a proper result the coefficients must be greatern then [-1,1]. So are there any additional library functions which are able to handle for example a Q3.12 data format.

0 Glenn Kubota over 8 years ago in reply to Anton Proskurnin

Prodigy 70 points

I also have the same problem with the C55xx DSPLIB. I can scale the coefficients so that N1 is less than 1, but if D1 is greater than 1, I don't know what to do. In other TI products, it looks like the N1 and D1 coefficients are divided by 2, but that doesn't seem to have been done in the DSPLIB code.

Do I need to modify the DSPLIB code, or is there some way of manipulating the biquad coefficients to make D1 less than 1 without changing the response?

0 Lalindra Jayatilleke over 8 years ago in reply to Glenn Kubota

TI__Mastermind 30365 points

Glenn,
We are looking into your questions and will get back to you.

Lali

0 Lalindra Jayatilleke over 8 years ago in reply to Glenn Kubota

TI__Mastermind 30365 points

Glenn,

Might not need to modify DSPLIB code. You could try to scale the coefficients before presenting them to the DSPLIB and compensate for the introduced gain or loss. If you scale down D by 2 the input signal should be scaled down by the same amount. Similarly if you scale down N by 2 the output signal should be scaled up by the same amount.
Keep in mind that the consequence of this method is the loss of precision and more cycles spent in the scaling of the input/output.

The other way to do it is to modify DSPLIB code to use an arbitrary Q format for filter coefficients. Hope this helps.

Lali

0 Glenn Kubota over 8 years ago in reply to Lalindra Jayatilleke

Prodigy 70 points

Lali,

I can scale the N0, N1, and N2 coefficients if one of them is greater than 1. However, I cannot do the same thing for D1 and D2, because all the coefficients are normalized so that D0 is 1. The iir32 and the iircas5 functions take only five coefficients (N2, N1, N0, D2, D1). If the D2 and D1 coefficients are scaled, the response would be changed. I suppose the DSPLIB code could be modified to take all six coefficients rather than assuming that D0 is equal to 1.

Glenn

0 Lalindra Jayatilleke over 8 years ago in reply to Glenn Kubota

TI__Mastermind 30365 points

Glenn,

I see. Yes, the DSPLIB code would need to be modified since 6 coefficients aren't supported.

Lali

Processors

Processors forum

IIR coefficient greater than 1