Will adding zeroes to FIR filter coefficients to increase taps degrade performance?
Thread Summary
The user inquired about the impact of adding zeros to FIR filter coefficients to match the length of a longer filter. The final answer confirmed that this approach will not degrade filter performance but will add delay. The accompanying answer noted that if the original coefficients do not smoothly approach zero, adding zeros can introduce unnecessary sidelobes in the filter response due to windowing effects.AI Generated Content
We have two FIR Filters, running on same input data, generated for different
purposes.
1. FIR1 - N Taps (as generated in Matlab)
2. FIR2 - N+10 Taps (as generated in Matlab)
Case 1
======
Have 2 separate Input Data Arrays of length matching to the taps.
i.e.
Array 1 of length N for FIR1.
Array 2 of length N+10 for FIR2.
Feed the incoming input sample to both the arrays.
Case 2
======
Have a single Input Data Array of length N+10.
i.e.
Array of length N+10 for both FIR1 & FIR2.
Feed the incoming input sample to the array.
Add 5 zeroes on each side of the filter coefficients for FIR1 to make it length
/ taps as N+10.
i.e.
FIR1 = [0, 0, 0, 0, 0, N Original Generated Coefficients, 0, 0, 0, 0, 0]
FIR2 = [N+10 Original Generated Coefficients]
"Case 1" brings in complexity in our design and hence we were looking into
"Case 2" and have certain queries:
1. Will adding zeroes degrade the performance of the filter i.e. roll-off/
attenuation etc. ?
2. Will the group delay of the Filter(original generated with added zeroes)
differ from the Filter(original generated coefficients) ?
Kaushal
taps
fir
zeroes
delay
The approach will work without any performance degradation. It will however add
delay. There is also no harm in putting all zeroes at the trailing edge.
Kaushal
I will make one proviso on adding zeros to FIR filter coefficients
Any FIR filter has (implicitly) an infinite extension of zeros on either side of the filter coefficients actually stored in memory.
If the filter coefficients don't decrease smoothly to zero at each end of the sequence,
e.g. coeffs 1, 1, 1, 1, 1, 1, 1, 1, don't decrease smoothly to zero but coeffs 1/8, 1/2. 7/8, 1, 1, 7/8, 1/2 1/8 do decrease (more) smoothly,
then the response of the filter has unnecessary sidelobes on either side of the main filter pass band.
This is a windowing effect and I can go into a more theoretical signal processing discussion if you are interested
This question has been closed by the EZ team and is assumed answered.
Please log in to reply.