FusionNet Based Deep Learning and Classical Detector Comparison for BER Performance in 16QAM 2×2 MIMO Systems

With the development of digital communication technology, high efficiency transmission model and effective signal detection method are necessary to develop. Combining advanced deep learning techniques, this research presents a comparative analysis of the signal detection methods under 16-quadrature amplitude modulation(QAM) in a 2×2 Multiple Input Multiple Output(MIMO) system. Five typical algorithms—Maximum Likelihood (ML), Zero Forcing (ZF), Minimum Mean Square Error (MMSE), Successive Interference Cancellation (SIC), and Semidefinite Relaxation (SDR)—were evaluated against three deep learning models: Fully Connected Network (FCN), Convolutional Neural Network (CNN), and a proposed hybrid model named FusionNet coded by Python. Simulation results show that ML achieves the best performance with a bit error rate(BER) of 0.8% at 20 dB signal-to-noise ratios(SNR), followed by MMSE (1.9%) and ZF (2.8%). Among deep learning-based models, FusionNet outperforms FCN and CNN, reaching a BER of 2.2%, compared to 4.6% for FCN and 2.8% for CNN. The FusionNet model integrates CNN's spatial feature extraction and FCN's abstract modeling capabilities using a dual-branch architecture with focal loss optimization. The results validate FusionNet's effectiveness in balancing accuracy.