Skip to content

Numerical instability ("randomness") in FFT in sigproc #468

@brettviren

Description

@brettviren

In some unrealted development, Claude happened to identify what may be some "randomness" in sigproc due to FFT instability.

One source arises in an ill-advised double configure which I hope to not have to keep while the other seems more inherent.

When we get some free cycles, or if this manifests in a real world program, hopefully people find this issue and dig deeper.

Here's the message from Claude:

  • ±2 ADC in ≲0.02% of samples: systematic from double-configure() on AnodePlane/WireSchemaFile in shared mode. These components store
    floating-point geometry; re-parsing produces sub-LSB differences that propagate through OmnibusSigProc's Wiener filter.
  • ±1 channel: one channel sits right at OmnibusSigProc's sparse=true threshold; tiny differences from the above tip it in or out. Not a physics
    difference

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions