Efficient Fixed-Point Implementation of Mitaka

What it is

A fixed-point C implementation of Mitaka signing (Zhang, Kirchner, Lin, Yu; ePrint 2026/2521, a preprint). Mitaka is a Falcon (FN-DSA) variant with a simpler, parallelizable signer. The paper says MitakaZ, the variant that avoids floating point, had not previously been implemented in practice. This is the first such implementation.

How it works

• MitakaZ uses a wide integral Gram root decomposition, which costs more Gaussian samples. The authors replace it with a fixed-point square Cholesky decomposition, which they say substantially reduces the cost of integer Gaussian sampling.
• They prove bounds on intermediate values and representation errors, without empirical assumptions about magnitudes.

Measured results

The page gives these as average signing costs. It states no comparison against Falcon or other implementations.

Caveats

• The distributional analysis models each integer Gaussian call as exact at its represented inputs. Security of the output distribution therefore depends on the underlying integer Gaussian sampler being correct and, as far as the abstract says, is not proven for the sampler's own implementation.
• It is a preprint with no stated peer review. The page mentions no code release, constant-time analysis, or side-channel evaluation, so none should be assumed.
• Only signing is covered. Key generation and verification costs are not reported.
• Mitaka is not the standardized FN-DSA signer. Treat this as evidence that floating-point-free lattice signing is feasible on Cortex-M4-class parts, not as a drop-in FN-DSA replacement.