OFDM Channel Estimator

Learned channel estimation for OFDM: a small neural network refines pilot-based estimates (e.g. LS) to reduce noise and interpolation error.

Motivation

In OFDM systems, channel state information is obtained from pilot subcarriers. Least-squares (LS) estimates are noisy. A lightweight NN can denoise or refine these estimates and improve BER in low-SNR regimes.

Setup

OFDM simulation: Generate OFDM symbols with pilot pattern (e.g. comb-type), pass through a multipath channel, add noise.
Pilot-based LS estimate: Classic LS on pilot positions, then interpolate (linear or spline) to get full channel.
NN refiner: MLP that takes the LS channel estimate (real/imag) and outputs a refined estimate. Implemented in run_ofdm_channel_est.py.

Usage

pip install -r requirements.txt
python run_ofdm_channel_est.py

Optional args: --n_fft, --n_pilots, --n_symbols, --snr_db, --epochs. The script runs a short simulation, trains the refiner, and saves channel_refiner.pt.

Files

run_ofdm_channel_est.py — simulation, data generation, model definition, and training.
requirements.txt — PyTorch, NumPy.

Limitations / future work

Simplified channel model (static multipath); mobility and Doppler would require a different setup.
Could extend to 2D (time–frequency) estimation with recurrent or 2D conv layers.

Author

Alireza Aminzadeh

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support