ML-based performance prediction of CSRR-enhanced four-port fractal MIMO antenna for 6G communication systems

The rapid development of wireless communication systems and the growing demand for terahertz (THz) sensing have led to an increasing demand for compact antennas with high isolation, integrated with machine learning (ML), to meet the stringent requirements of the next-generation 6G networks. In this paper, a compact quad-port fractal MIMO antenna with Complementary Split-Ring Resonator (CSRR) for efficient mutual coupling suppression is proposed, along with an ML-based rapid performance prediction framework. A triangular fractal radiating structure of electrical size of 3.08λ 0 × 3.08 λ 0 on polyimide substrate along with a slotted-patch and CSRR-based isolation technique, efficiently suppresses the surface-current coupling and achieves an inter-port isolation better than −27 dB. The four port MIMO antenna operates at triple-band 6.35, 6.6, and 8.6 THz with wide bandwidths, with a maximum gain of 8.2 dB and the radiation efficiency of 89%. The MIMO performance is further validated with diversity metrics such as low Envelope Correlation Coefficient (ECC) of 0.007 and high Diversity Gain (DG) of 9.965 dB indicating good port decoupling and reliable multi-stream transmission. In order to improve the design efficiency, five supervised ML regression models were trained on CST-generated data to predict the isolation performance. The Extra Trees regressor gave the best accuracy (R2 = 95.27%, MAE = 0.2095%, RMSE = 0.5731%) compared to the other models. The results show that ML-based prediction is an effective solution to reduce the dependence on computationally expensive simulations, making the proposed design a compact, wideband, and high-performance solution for THz MIMO systems in 6G THz wireless applications.