Deep Learning-Based Target Parameter Estimation in MIMO-OFDM ISAC Systems Derin Ö?grenme Tabanli MIMO-OFDM ISAC Sistemlerinde Hedef Parametre Kestirimi


Calik N., ALDIRMAZ ÇOLAK S., Goken C., Canbaz E., Ozturk C., Durak Ata L.

34th Signal Processing and Communications Applications Conference, SIU 2026, İstanbul, Türkiye, 7 - 10 Temmuz 2026, (Tam Metin Bildiri)

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Doi Numarası: 10.1109/siu71813.2026.11637044
  • Basıldığı Şehir: İstanbul
  • Basıldığı Ülke: Türkiye
  • Anahtar Kelimeler: CFAR, Deep Learning, ISAC, OFDM
  • Kocaeli Üniversitesi Adresli: Evet

Özet

In this study, the problem of estimating the target angle, range, and velocity parameters in MIMO-OFDM based Integrated Sensing and Communication (ISAC) systems is investigated. The conventional three-dimensional Cell Averaged Constant False Alarm Rate (3D CA-CFAR) algorithm, widely used in radar-based approaches, suffers from limited estimation accuracy and high computational complexity, particularly under low signal-to-noise ratio (SNR) conditions. Motivated by this limitation, a deep learning-based approach is proposed to estimate target parameters through regression by directly processing ISAC detection matrices. In this study, models based on EfficientNet-B1, EfficientViT-B1, and SE-ResNeXt-26 architectures are trained on a synthetically generated dataset covering a wide SNR range and evaluated in comparison with the conventional 3D CA-CFAR method. The results demonstrate that the transformer-based EfficientViT-B1 model achieves lower estimation error and improved robustness, especially under low SNR conditions.