Wireless Systems Laboratory

Deep learning regression network for Joint DOD/DOA Estimation (Single‑Target) in MIMO radar systems

A lightweight deep learning‑based regression network is proposed, which directly maps MIMO virtual array snapshots to direction-of-departure (DOD) and direction-of-arrival (DOA) without covariance estimation, spectrum generation or peak search. On simulated datasets, the model achieves improved RMSE compared to conventional 2D-MUSIC, especially at low SNR. The model is realised with ~O(H(H+MN)L) complexity, enabling near real‑time inference on edge hardware.

RMSE: ~1 °

SNR: 0–10 dB

Params: 0.62 M

Deep Learning‑Based OFDM Demodulator for Under‑sea RF Communiation

A modular deep learning-based OFDM demodulator is designed to learn the characteristics of undersea RF channel, and perform a robust demodulation of the received OFDM symbol, replacing conventional channel estimation and equalization. In an undersea RF channel considering direct and lateral waves along the seabed (Rician fading, K≈6 to 10 dB, 1 MHz operating frequency, distance 3~7 m), the proposed method marked improved BER over conventional equalization across 0–15 dB SNR.

BER@10dB: ↓ 35% vs. ZF

FLOPs/sym: ≈ 3.1×10⁶

Params: 1.1 M

Read Article 1 Read Article 2

Welcome

Deep learning regression network for Joint DOD/DOA Estimation (Single‑Target) in MIMO radar systems

Deep Learning‑Based OFDM Demodulator for Under‑sea RF Communiation