Efficient Switched-Beam Detection for Dynamic Spectrum Sharing in 6G Wireless Networks with Full Duplex Technology at the THz band

The demand for higher data rates and bandwidth-intensive applications in the digital age is driving the need for cutting-edge technologies in future 6G systems, such as terahertz (THz) and full duplex (FD) communications. These technologies promise exceptional performance enhancement when combined with a dynamic spectrum management approach based on Cognitive Radio (CR). This paper explores blind spectrum sensing (SS) methods suitable for FD-THz communications, addressing challenges posed by residual self-interference (RSI) and molecular noise, which characterizes THz communications. Both molecular re-radiation and RSI are sources of colored noise that dominate additive white Gaussian noise, complicating the operations of blind SS algorithms based on the eigenvalues of the sample covariance matrix. To address this issue, we propose a low-complexity solution using a switched beam technique, along with a whitening matrix derived from a dataset containing RSI, molecular noise, and two largest eigenvalue (LE) detectors. The LE detector is introduced as an upper bound applicable only when perfect knowledge of the variances of all the noise components is available, which is unlikely in most practical scenarios. Therefore, we propose a novel, blind, low-complexity detector based on scaled LE (SLE), which estimates the molecular noise and the AWGN variance online. The algorithm also includes a heuristic to adjust the contribution of the RSI and molecular noise components, ensuring that the detector retains approximately the constant false alarm rate (CFAR) property under typical operating conditions. Numerical simulations show that the proposed SS method is feasible and robust. Compared to the recursive least squares (RLS) adaptive filtering benchmark, it offers a more robust false alarm probability, particularly in the low-AWGN regime, with detection performance similar to or slightly lower than the RLS detector, but with significantly lower computational complexity.