[Feature Request]: Use webassembly to run ITM in the browser

[catbraincell]

Prerequisites

• I have searched existing issues to ensure this feature hasn't already been requested
• I have checked the documentation to verify this feature doesn't already exist

Problem Statement

Simulation speedup

Proposed Solution

I managed to compile NTIA/itm with emcc. Here's the repo
https://github.com/warmonkey/itm_wasm
May try to integrate

Current Alternatives

No response

Importance

Nice to have

Additional Context

No response

[mrpatrick1991]

Thanks for the feature request - I was able to get this running and the initial tests work.

The next question is if it is sufficiently performant compared to the cpp implementation.

This is going to require a major refactor of the existing site planner, but I think it will be worth it in the long run.

[loorisr]

Hello,

Someone (modimoo on reddit), made a WebGPU ITM tool: https://propagation.tools/

[peytondmurray]

Hey all, I also thought it would be an interesting side project to compile SPLAT to target WASM so that predictions run entirely on the client, so I've spent the past few weeks getting that working. It's not perfect, but it's getting close to being useful. Check it out at https://waveguide.dev, or find the source at https://github.com/peytondmurray/waveguide.dev/. Feedback welcome, I'd love to get the chance to improve!

A few interesting things:

• The SRTMGL3S dataset doesn't seem to be very complete. I pulled it from srtm.fasma.org and am hosting it on Cloudflare R2 at the moment, but I'm going to spend a bit of time implementing a toggle so that I can see what tiles are missing.
• Running SPLAT entirely in the browser doesn't always provide gigantics speedups over the current meshtastic-site-planner. Partly that's because the meshtastic-site-planner caches the SRTM tiles so that often there is no overhead in downloading, extracting, and converting from .hgt -> .sdf. Another performance bottleneck is that SPLAT is single threaded and not written to be performant from the get-go. I spent a little time chasing performance optimizations and made some progress, but the fact remains that it is a memory-hungry single-threaded tool, and it would take a bigger refactor to parallelize computations. Emscripten has pthread support, so it seems like it should be possible to make this much faster, but again it will depend on the user's hardware. I even had to do the SPLAT prediction in a web worker, otherwise the UI locks up.
• There are still a few bugs, most notably that the prediction is off from where the transmitter is centered; see the README.md about this. Not sure what I need to do differently, but it seems like it might be a problem projecting the SPLAT output.ppm onto the map.