2025-05-20 7:45 AM
Hi all
I have an IIR filter design with 7 coefficients. (For feed-forward and for feed-back)
Regularly all the coeffs are between -1 to +1
So I Just multiply each by 2^15 to get it in q1.15 format.
However, now, the feedforward coeffs are very small (0.0000000001 or something like that)
and the feedback coeffs are up to 30.
How to initialize the system to get it work?
2025-05-20 1:34 PM
Is chatgpt broken today?
2025-05-21 10:37 AM - edited 2025-05-21 10:39 AM
Seriously?
I was thinking that the toxicity is only stack-overflow privilege.
Do you really want me to go through all the BS it gives me just to get the conclusion
that I need to scale the factor to get them in values greater than 1 but
it is impossible due to the fact they are TOO SMALL (as I said in the post)
Take a look here:
https://chatgpt.com/share/682e0de8-9490-800a-9a07-e690a4711dc9
2025-05-21 1:30 PM
Fixed point probably not going to cut it then, could you use the FPU and VMLA.F32, with 32-bit floats?
2025-05-21 7:34 PM
Just curious what the filter is actually being used for - maybe there are alternative approaches that can be taken...
2025-05-25 7:32 AM
That's what I doing right now. with a for loop...
2025-05-25 7:33 AM
I have to filter velocity derived by position[n] - position[n-1]
The sample freq is 50K and the cutoff should be less than 350 Hz