Posted on January 24, 2014 at 16:20I don't know how much special effort it takes to bottom post responses so they are invisible, but PLEASE STOP IT. Use Ctrl-A clear out the post.
To your filter issue, build some unit-tests with data patterns you understand, and whose results you have a handle on.
This will allow you, and others, to evaluate the code without worrying about the sampling, and what's getting sampled.
Time those operations, and validate your solution.
Optimize your code, right now it looks pretty woeful and slow.
Get constant computation outside of the implementation, ideally in a fashion the compiler can optimize further.
Don't use DOUBLES in code you expect to be fast. If you must use floating point at all here, use FLOATS as this is the only type supported by the FPU, and execute rapidly.
Ideally don't use doubles/floats in interrupt code, it can be done, but if you have to push FPU context onto the stack it's still overly expensive for most tasks.
Do 32-bit math, 16-bit math may look more efficient, but it's almost always going to be slower on a part designed to do 32-bit math natively.
