An open-source tool for the design of optimized multiplierless FPGA implementations of LTI filters. This tool:
- is based on a hardware implementation cost model developed by Martin Kumm in [Kumm, 2016];
- generalizes the filter design & implementation problem into an Integer Linear Programming one;
- provides fast solutions (benchmarking is in progress);
- supports features such as flexible gain and sparse implementations
An open-source end-to-end code generator for digital filters.
This tool combines numerous features, such as
- support of classic structures, e.g. Direct Forms, State-Space, Lattice Wave Digital Filters
- support of Matlab Simulink linear data-flow graphs
- reliable Fixed-Point Algorithms that guarantee that no overflow ever occurs
- rigorous a posteriori verification of implemented filters against frequency specifications
- software implementation in C
- binder with the FloPoCo tool for the generation of FPGA architectures that are computing just right
.. and many others!
Attention: the project is under re-organization. Please, contact me before use.
A C library for the reliable evaluation of the Worst-Case Peak Gain measure of a LTI filter (most commonly known as the l_1 norm of filter’s impulse response). Given the coefficients of a filter (in the form of the transfer function or as the state-space matrices) and an absolute error bound, the algorithm evaluates the Worst-Case Peak Gain such that the error bound is guaranteed to be respected.
git clone https://scm.gforge.inria.fr/anonscm/git/metalibm/wcpg.git
A C library for the reliable determination of the Fixed-Point Formats for a LTI filter. Computed formats are need to never yield an overflow. The library can also return a bound on the error of the computed output. The output of a filter implemented with the computed formats and a faithful rounding of the intermediate computations is guaranteed to respect this bound.
git clone https://scm.gforge.inria.fr/anonscm/git/fxpf/fxpf.git