New zero slowdown linear advance algorithm

[dbuezas]

Description

Linear advance reduces the max acceleration so that the la_advance_rate is kept below e_jerk. The existing implementation also suffers from some degree of quality loss due to it not accounting for the discontinuities in pressure due to both movement jerk and changes in line width (i.e the ratio of e_steps to step_count).

This WIP PR removes that restriction by handling linear advance acceleration in its own isr, and allowing it to not be perfectly in synch with movement. This should theoretically reduce quality, but in practice it doesn't seem to be a problem. In fact, since this algorithm does account for movement jerk and line width changes, the resulting quality is improved beyond the original existing algorithm, and that at higher printing speeds.

The algorithm is as follows:

• Compute the target pressure (in steps) as current_rate * k
• Compare with current pressure and accelerate/decelerate extruder to match it.
• When accelerating the extruder, avoid overshooting by ensuring that the extruder will be able to break in time to match the target with zero speed.
• When loading a new block rescale the current (residual) nozzle pressure to the new movement step rate to extrusion step rate ratio. (This handles the aforementioned discontinuities)

This is work in progress.
Optimisation so it can run in slower machines is pending (and ideas very welcome)

Requirements

Definitely a 32 bit board. For now, maybe even one with a floating point unit

Benefits

Better quality, faster prints

Configurations

#define LA_ZERO_SLOWDOWN

Related Issues

#27259

[dbuezas]

TL;DR: new algorithm is faster and has higher quality.

TOP LEFT: 300mm/s 4000mm/s2, K=0, 30 mins
TOP RIGHT: ~82mm/s K=0.15 (OLD ALGORITHM limits to ~1500mm/s2). 48 minutes
BOTTOM LEFT: 300mm/s 4000/s2, K=0.09, 30 mins. (yes k changes a lot with high speeds, non linear!)
BOTTOM RIGHT: ~82mm/s 3000mm/s2, k=0.15, 37 mins
There's a printing artefact on the top right corner of the inner hexagon on the old algorithm. It's gone with the new algorithm. I think that's the unaccounted for movement jerk and/or line width changes
Discoloration due to lower temperatures on the bottom left makes it look as if there was more blobbing than there is.

[ThomasToka]

Edit: Found it. My error. Will try it now.