Whilst working on some other experimentation, I noticed that the auto-generated FIR filter co-efficients for the bandpass filters seem to introduce a lot of attenuation, and are probably not 'optimal' for the required operation. That is no real surprise, as they are generated by a quick dynamic piece of code, and calculating optimal FIR coefficients is compute intensive.
The FIR filters happen early in the chain, which means later stages are probably not getting as 'full' a bitwidth signal as they should. This probably affects their performance.
I have some experimental code already that tries to measure the gain of the coefficients at their 'centre peak' intended frequency. Let's see if that improves anything.
(If you want to look at this etc., probably best ask me first so we don't duplicate)
Whilst working on some other experimentation, I noticed that the auto-generated FIR filter co-efficients for the bandpass filters seem to introduce a lot of attenuation, and are probably not 'optimal' for the required operation. That is no real surprise, as they are generated by a quick dynamic piece of code, and calculating optimal FIR coefficients is compute intensive.
The FIR filters happen early in the chain, which means later stages are probably not getting as 'full' a bitwidth signal as they should. This probably affects their performance.
I have some experimental code already that tries to measure the gain of the coefficients at their 'centre peak' intended frequency. Let's see if that improves anything.
(If you want to look at this etc., probably best ask me first so we don't duplicate)