In this paper, we provide a software-defined radio implementation for our transmission scheme proposed in [1], [2] to secure the wireless transmission on a wiretap channel with a passive eavesdropper. We show that a message modulated with our BICM-IPM scheme can be correctly decoded at the legitimate receiver. However, the decoding of this message at the eavesdropper results in a near random bit stream.

Low Power Wide Area Networks (LPWANs) enable cost-effective, low power consumption, and long-range IoT applications but often operate under low Signal-to-Noise Ratio (SNR) conditions. Conventional deep learning models for signal recognition struggle to generalize in such dynamic environments. We propose a lightweight framework combining ensemble learning, a Mixture of Experts (MoE) with uncertainty-aware soft gating, and meta-adaptation using the Almost No Inner Loop (ANIL) method. Expert models trained at distinct lowSNR levels share frozen base weights, while only the classifier head is adapted during few-shot learning for rapid specialization. The uncertainty-aware gating mechanism produces sparse, temperature-controlled expert weights, enhancing robustness and decision reliability. Experiments demonstrate improved accuracy and generalization across both seen and unseen SNR levels, highlighting the framework's effectiveness for real-world wireless signal technology recognition in spectrum monitoring systems.

In this paper, we propose a noisy and dynamic-index partitioned modulation (IPM) to secure a quadrature amplitude modulation (QAM) constellation transmission over a non-degraded wiretap channel. Unlike approaches that rely on multi-antenna systems or tight amplitude-phase tracking, IPM partitions QAM into multiple disjoint subsets that are separately indexed by a dynamic key, known at the transmitter (Alice) and the legitimate receiver (Bob), but not at the eavesdropper (Eve). The proposed IPM maps the information bits into multiple sequences, each one lying in a different partition of the QAM constellation space. This mapping is performed through a cross-bit labeling that we define to increase the confusion at the eavesdropper, while minimizing the bit error rate (BER) at the legitimate receiver. As the eavesdropper is not aware of the dynamic index pointing to the different partitions, the IPM creates larger Voronoï detection zones around Bob’s symbol compared to Eve. To induce further confusion at Eve, we inject random uniform noise into IPM symbols. The noise varies in a domain that is fully included in Bob’s larger detection zone, but it exceeds Eve’s detection zone. The performances of the noisy IPM scheme in terms of error rate and secrecy rate are analytically and numerically evaluated. Our results show that the IPM scheme creates on the Eavesdropper’s link an error floor, independently of its signal-to-noise ratio (SNR). However, the IPM scheme preserves the legitimate link, for which the BER decreases as the SNR increases. Indeed, the secrecy rate remains positive for all SNR values and achieves an asymptotical constant plateau value.