Symbol detection is critical for ambient backscatter communication (AmBC) and its optimal detection threshold is sensitive to the distribution of the signal from the ambient radio frequency source. Existing work mainly focuses on symbol detection under the assumption of a single ambient RF source (S-ARF). However, in the space-air-ground integrated networks, multiple ambient RF sources (M-ARFs), including satellites, unmanned aerial vehicles and terrestrial base stations, often reuse the same RF resources, which makes the distribution of M-ARFs signals differ from that of an S-ARFs signal. In this paper, we investigate symbol detection for AmBC in the presence of M-ARFs. To this end, we model the M-ARFs signal as an aggregate signal with log-normally distributed power. On this basis, we formulate a maximum likelihood (ML) detector and obtain a closed-form approximate threshold using Hermite-Gauss Quadrature, which is effective under specific parameter conditions. Furthermore, to enhance generality and analyze the impact of aggregate signal parameters, we construct an energy detector (ED) and propose a novel log-energy detector (LED), for which we derive the expressions for bit error rate (BER) and near-optimal thresholds. Simulation results show that the aggregate signal can significantly degrade the BER performance when directly using the detection threshold assuming an S-ARF. In contrast, the ML and ED based on the proposed detection framework can mitigate the performance degradation, and LED can achieve superior BER performance.

Ambient backscatter communication (AmBC) enables ultra-low-power, low-cost and massive connectivity. However, practical AmBC systems suffer from symbol timing offset (STO) due to propagation delay and backscatter receiver (BR) activation latency, while conventional correlation-based synchronization methods are inapplicable because ambient radio frequency sources are non-cooperative. Moreover, residual STO (RSTO) inevitably remains due to the finite synchronization sequence, which degrades symbol detection performance. To address these challenges, we first design a specialized synchronization sequence with alternating “0” and “1” bits at the backscatter device to induce observable sampling errors at the BR. Based on this, we propose a pilot-aided, sampling-error-aware maximum likelihood estimation (PSE-MLE) method for STO estimation and compensation, which exploits the statistical variations in the received synchronization signal. After STO compensation, the remaining RSTO is statistically modeled as a discrete bilateral Laplace distribution, with its parameter estimated via ridge regression. Leveraging this prior information, we further develop a Bayesian average energy detector (ave-ED) and derive closed-form expressions for both the detection threshold and bit error rate. Simulation and experimental results on a practical AmBC platform validate the effectiveness of the proposed methods.

The rise of deep learning enables spectrum sensing to make decisions only based on the observed data which brings about a lot of flexibility. However, the existing data-driven spectrum detection methods handle the data in Euclidean space well but ignore the laten structure connections of the signal. To characterize this information, we propose a graph construction mechanism based on the similarity measure of the random signal and further convert the received signal into graph topology. On this basis, we propose a graph neural network based detector that consists of the construction of graph topology, offline training and online detection. The proposed data-driven method does not need any priori knowledge of the observed signal. Simulation results demonstrate that whether there is noise uncertainty or not, the proposed method is robust and performs better than the existing deep learning based methods.

Automatic Modulation Recognition (AMR) is becoming increasingly crucial in the industrial internet, and it ensures more reliable communication for devices within this vast and intricate network. Although recent application of neural networks, notably Graph Convolutional Network (GCN) in the AMR domain shows promising results, it fails to account for temporal features within modulation signals, resulting in a loss of recognition precision. Given that modulation signals possess both time and space features, this paper proposes a hybrid deep learning model named LAGNet, which combines Long Short-Term Memory (LSTM) and GCN. The model uses an attention mechanism to map the LSTM outputs into a graph and then employs the GCN to extract the spatial features of the signal. Subsequently, the temporal features and the spatial features are combined together for modulation signal classification. Experimental results show that LAGNet not only outperforms several advanced models in classification accuracy across all signal-to-noise ratios, but also requires fewer learnable parameters.

Inspired by random matrix theory, a quantity of eigenvalue based cooperative spectrum sensing methods have been proposed. The results are based on the asymptotical assumptions in need of large numbers of users and samples, which result in inferior performance with a few users. In this paper, sensing methods based on maximum eigenvalue and minimum eigenvalue of LDLT decomposition are proposed respectively with a view to improve the accuracy of decision threshold by means of hard decision criterion. The corresponding expressions of false alarm probability are also derived. Finally, both theoretical analyses and simulations demonstrate that the proposed two methods perform better than the existing eigenvalue based sensing methods for accurate decision threshold.