Small version of siderust-archive

[ToBinio]

Hi,

I'm playing around with a hobby project: a small desk device that can point at various celestial bodies.

I found your library and it looks extremely promising. However, I'm running into one problem: my ESP32 only has 4 MB of DROM, and just two instances of vsop::vsop87e_data already fill it completely.

Would it be possible to get a smaller version of the data with reduced precision?

Please note that astronomical positioning is completely new to me and well outside my comfort zone, so if there's something I should be doing differently, please let me know!

Thanks a lot!

[VPRamon]

Hi @ToBinio ,

Thanks for bringing this up! Your ESP32 project is a very interesting use case for Siderust, and I think it highlights an opportunity to improve support for resource-constrained devices.

One of the properties I'd particularly like to preserve is that VSOP87 coefficients are embedded directly into the compiled binary. This allows Siderust to operate completely offline, without external datasets, filesystem access, or runtime initialization, while providing fast and predictable calculations.

Rather than simply reducing the VSOP87 tables, I'd like to investigate a more general approach to compact, embedded ephemerides, balancing accuracy, binary size, and computational performance.

A few possibilities seem worth exploring:

• Truncated VSOP87: Reduce the number of coefficients while preserving the existing implementation and broad temporal coverage.
• Lightweight orbital approximations: Significantly smaller datasets, potentially suitable for applications with modest pointing accuracy requirements.
• Embedded Chebyshev ephemerides: Precompute compact polynomial coefficients from modern reference ephemerides such as JPL DE440, then embed them directly into the firmware. This could provide excellent accuracy and fast evaluation without requiring external kernels or runtime I/O.

Importantly, all three approaches could preserve offline operation and compile-time embedded data. The trade-offs would need to be validated through actual binary-size, accuracy, and performance measurements.

Before deciding which approach to pursue, I'd love to understand a few details about your project:

1. Hardware: Which ESP32 variant are you using? Is the 4 MB limitation related to the DROM region, available flash, or the application partition?
2. Targets: Which celestial bodies do you need to track? Just planets, or also the Sun, Moon, and stars?
3. Accuracy: What approximate pointing accuracy are you aiming for? Would 1 arcminute be sufficient, or could you tolerate larger errors?
4. Time coverage: Do you need broad historical/future coverage, or would a restricted interval (e.g., 2020–2050) work?
5. Memory measurements: Have you inspected the linker map or binary sections to determine how much space the VSOP87 coefficients actually occupy?

I'd also like to verify whether the linker is already removing unused coefficient tables before introducing a new representation.

My preference would be to keep the existing high-precision VSOP87 implementation intact while exploring an optional compact backend, ideally compatible with "no_std" and requiring no runtime allocations.

Once we get a clear idea we can create dedicated issues and prioritize them.

Thanks again for opening the discussion!

[ToBinio]

Thanks, answers below:

1. Hardware: the classic ESP32-WROOM-32, so 4 MB of program space
2. Targets: Sun, Moon and the planets. No stars
3. Accuracy: 1 arcminute is plenty. Even a few arcminutes would be fine.
4. Time coverage: a restricted window is fine, e.g. around 2020-2050.
5. Memory: I can't build the native ESP32 binary since it just fails because of the size limits, but in a desktop example build, only the tables that are actually used end up in the binary, so the linker does strip the rest.

Edit: I have decided to implement the position calculation myself instead of using a library, so I get a better understanding of the math / logic behind it! Still, thanks a lot for the quick answer!

[VPRamon]

Thanks again, @ToBinio!

I've opened three issues to track this work:

• [Siderust/archive#5](Generate and validate compact embedded VSOP87 planetary tables for ESP32-class targets archive#5) — Generate and validate compact VSOP87 planetary tables.
• [Siderust/archive#6](Investigate compact embedded lunar ephemeris data for 2020–2050 pointing archive#6) — Investigate compact lunar ephemerides.
• [Siderust/siderust#139](Integrate compact offline ephemeris backend for ESP32 celestial pointing #139) — Integrate a compact embedded ephemeris backend and validate it on ESP32-class hardware.

feel free to add any comment/suggestion there if you will