IIRD: Fast Trainable Multidimensional IIR Filter Layers in TensorFlow

IIRD provides trainable 1D, 2D, and 3D IIR filter layers for TensorFlow/Keras.

Copyright 2025 Kishore Kumar Tarafdar. Licensed under the Apache License, Version 2.0. See LICENSE.

Capabilities

• IIR1D, IIR2D, and IIR3D Keras layers.
• Trainable feedforward and feedback coefficients.
• Batched multichannel TensorFlow tensors.
• Frequency-response plotting utility.

Minimal Example

import tensorflow as tf
from IIRD import IIR1D, IIR2D, IIR3D

x1 = tf.random.normal([1, 64, 1])
y1 = IIR1D(Delays=2, filters=4)(x1)

x2 = tf.random.normal([1, 32, 32, 1])
y2 = IIR2D(Delays=2, filters=4)(x2)

x3 = tf.random.normal([1, 16, 16, 16, 1])
y3 = IIR3D(Delays=1, filters=2)(x3)

print(y1.shape, y2.shape, y3.shape)

Module Commands

python -m IIRD.IIR1D
python -m IIRD.IIR2D
python -m IIRD.IIR3D

Citation

This software is released for broad research, educational, and engineering use. If this package helps your work, please cite the following paper:

@misc{tarafdar2026interpretablefrugallearningsystems,
      title={Interpretable and Frugal Learning Systems Employing Multiresolution Pyramids and Volterra Kernels},
      author={Kishore Kumar Tarafdar},
      year={2026},
      eprint={2606.15011},
      archivePrefix={arXiv},
      primaryClass={eess.SP},
      url={https://arxiv.org/abs/2606.15011},
}