Enabling artificial intelligence native end-to-end systems in ultra-wideband sub-terahertz spectrum faces several challenges. The particularly complex channel variations and nonlinear behavior of analog components of the transceivers are major obstacles to the over-the-air adaptation of these systems. In this paper, we investigate an edge-based bidirectional long-shortterm memory neural network capable of predicting the channel gain variations in Non-Line-of-Sight conditions. We aim to enable end-to-end autoencoders with a predictive model for scheduling the training phase when the power is above the receiver sensitivity and there are no large fading variations. Otherwise, the training of the end-to-end system will likely fail. With only 16 BiLSTM cells our model is capable of inferring the channel gain variations with a worst-case root mean squared error lower than 0.0547 (i.e., 1.1% compared to the normalized channel gain range). Also, with lower computational complexity, our model decreased the propagation of the error compared to traditional recurrent neural networks and deep-learning-based forecasting models.

In this paper we investigate the coexistence of Cognitive Radio networks on TV White Spaces for rural and suburban connectivity. Although experimental models and laboratory measurements defined the maximum interference threshold for TV white space technologies for general use cases, our research found that in real wireless rural and suburban scenarios, severe interference to the broadcasting services might occur. This is particularly relevant when the traffic load of the telecom Base Stations (BSs) exceeds 80% of their maximum capacity. We propose a dynamic management algorithm for minimizing the interference, based on a centralized access control architecture for cognitive radio wireless networks. In an experimental emulation for assessing the impact of cognitive radio interference on the broadcasting service's QoE, our method reduced the perceived video distortion by the broadcasting users by at least 50% and 27.5% in a rural and suburban scenario respectively, while the spectrum usage is increased by just 8%.

Cell-free massive MIMO networks have recently emerged as an attractive solution capable of solving the performance degradation at the cell edge of cellular networks. For scalability reasons, usercentric clusters were recently proposed to serve users via a subset of APs. In the case of dynamic mobile scenarios, this form of network organization requires predictive algorithms for forecasting propagation parameters to maintain performance by proactively allocating new APs to a user. In this paper, we present a BiLSTM-based multivariate path loss forecasting algorithm. Thanks to the combination of dual prediction by the BiLSTM and diversity from multiple antennas, our model mitigates the error propagation typically faced by sequential neural networks for time-series forecasting. In the evaluated scenario, from 2 to 10 steps ahead, we reduce the propagation of the error by a factor of 18 compared to previous research on path loss forecasting by an LSTM time-series-based model. In contrast to parallel transformer solutions, the complexity cost of our algorithm is also significantly lower.

Most spectrum surveys conducted worldwide demonstrate that the radio-electric spectrum in use at any given location and instant of time is below 25%. Current spectrum management policies and spectrum utilization inefficiency are becoming unsustainable for the future development of radio technologies and services. In this context, dynamic spectrum access is a promising technique for improving spectrum utilization efficiency. A key scientific gap is identifying inaccurate spectrum data from hidden nodes that are not homogeneously distributed in the spatial domain and dynamically vary in time and frequency. For bridging this gap, our paper presents the research results of a spectrum feature extraction algorithm based on multi-correlation and Random Forest. Our algorithm is capable of estimating the spectrum utilization pattern in the spatial and frequency domain with a reliability up to 92% for a real heterogeneous networking scenario.