FM multipath filter coefficient optimization
Closed this issue · 4 comments
Observations from @rsuzuki0 of the 72-stage experiment results in this video file:
- The reference position (where initial coefficient = 1 + 0j) should be moved to a later position so that more previous (early time) samples (reflected waves) and less later samples (early-reflected waves) should be affected to the calculated output. Currently, the length ratio is 1:1. A new ratio of 3:1 or even 4:1 should be considered.
- Phase of the previous (early time) samples looks much more stable (converged) than the later samples.
New algorithm: c9d36ed
A result video: https://youtu.be/J-6Q_1_bPEM
The 3:1 ratio (or simply reducing the number of stages to 36 for the later samples) works well.
While the 3:1 ratio is somewhat reasonable, ignoring late-coming (later) signals might be risky in some cases when the major component of the signal comes after the late-coming signals, especially in the case of the cable and direct waves interfere with each other at the same frequency bandwidth.
We'll keep the heuristics as is for the time being, but extending the late-coming (later) stages might be considered in the future.
Current -E100 is feasible even for Raspberry Pi 4. I'll keep the design as it is now unless any show-stopper shows up. Issue closed.