gr-adapt

gr-adapt is a GNU Radio out-of-tree module that provides various adaptive filter blocks.

Overview

Adaptive Filters

The following standard filters are provided:

• Least mean squares (LMS)
• Normalized least mean squares (NLMS)
• Recursive least squares (RLS)
• Recursive least squares employing QR decomposition (QRD-RLS and IQRD-RLS)

The LMS and NLMS filters take an optional third input, which, if provided, will be used to update the filter taps. This allows the LMS and NLMS filters to be used in the filtered-x configuration (i.e. FxLMS and FxNLMS). Furthermore, adaptive filters are provided for estimating and tracking a channel under carrier and sampling frequency offsets as well as nonlinear distortions:

• Frequency offsets-compensated least mean squares (FO-LMS)[^1]
• Spline-interpolated and frequency offsets-compensated least mean squares (S-FO-LMS)[^2]

[^1]: K. Pärlin, T. Riihonen, V. L. Nir, and M. Adrat, “Estimating and Tracking Wireless Channels under Carrier and Sampling Frequency Offsets,” IEEE Transactions on Signal Processing, vol. 71, pp. 1053–1066, Mar. 2023.

[^2]: K. Pärlin, M. Turunen, A. Byman, T. Meriläinen, V. L. Nir, M. Adrat, and T. Riihonen “High-Power Cooperative Jamming with Nonlinear Known-Interference Cancellation,” IEEE Transactions on Aerospace and Electronic Systems, 2025.

Tests

There are several tests regarding performance and convergence of the provided filters in the python directory. The README in there gives an overview of those test results.

Examples

In the examples directory, examples of using adaptive filter blocks are given for the following use cases:

• Adaptive line enhancer
• Self-interference cancellation
• System identification
• Time delay estimation
• Channel and frequency offsets estimation
• Channel, nonlinear distortions, and frequency offsets estimation

Installation

Dependencies

• GNU Radio with the following components:
  • gnuradio-runtime
  • gnuradio-filter
• VOLK
• CMake 3.8
• pybind11
• Armadillo (optional*)

* RLS filters based on QR decomposition currently require Armadillo to work. If gr-adapt is built without Armadillo, those filters do nothing.

Building from source

sh $ git clone git@github.com:karel/gr-adapt.git $ cd gr-adapt $ mkdir build $ cd build $ cmake .. $ make -j$(nproc) $ sudo make install

Installing from AUR

A package description is provided in AUR for Arch Linux users.

Troubleshooting

If GRC complains that it can't find some modules like

python Traceback (most recent call last): File "/home/user/git/gr-adapt/examples/top_block.py", line 30, in <module> import adapt ImportError: No module named adapt

then most likely you used a different CMAKE_INSTALL_PREFIX for the module than for GNU Radio. Uninstall the module, then run CMake in the build directory again with the correct CMAKE_INSTALL_PREFIX and then install the module.