The widespread adoption of wireless communication technologies has intensified spectrum demand, exacerbating spectrum scarcity despite underutilization in many frequency bands. Cognitive radio systems offer a solution by supporting dynamic spectrum access, particularly through passive spectrum monitoring for opportunistic use. Existing machine learning (ML) and deep learning (DL) based white space detectors fall short of the accuracy needed for large-scale networks. This paper presents the Autoencoder-Transformer Integrated Cognitive Radio (ATIC) framework, a lightweight model that leverages the Transformer's self-attention mechanism to effectively capture complex spectraltemporal patterns without extensive preprocessing. ATIC's robustness is validated on both field-collected 5G Non-Standalone (NSA) and synthetic datasets, achieving an accuracy of 99.77 %. Furthermore, ATIC's architecture supports parallel processing across multiple primary resource blocks (PRBs), optimizing inference time as PRB count increases and achieving a 32% reduction compared to convolutional neural network (CNN).

Interweave Cognitive radio (CR) technologies enables support of infotainment and multimedia services in vehicular ad hoc networks (VANETs), especially during accidents or traffic congestion. The dynamic nature of VANET topology and fluctuating traffic volumes makes machine learning (ML) techniques promising for spectrum sensing; however, centralized training of ML models can lead to significant drawbacks, such as high communication overhead and data privacy concerns due to the transfer of large amounts of raw data. While adaptation of federated learning (FL) can address this problem, on-device training of deep learning with a federated approach poses computational challenges for vehicles. To address these limitations, this paper proposes a novel lightweight spectrum detection framework integrating spiking neural networks (SNNs) and FL. The SNN's event-driven nature reduces computational complexity, while FL enables decentralized training across vehicles, preserving data privacy and minimizing communication overhead by transmitting model updates rather than raw data. Validation using field-collected LTE subsystem reservations of 5G Non-Standalone (NSA) confirms that the proposed approach achieves fast inference times, enhancing spectrum sensing efficiency and expanding spectral resources for vehicular communication.

In cognitive communication, detecting white space is critical for preventing interference with primary user’s transmissions. Rather than modifying the radio transceivers, several machine learning (ML)-based detection techniques have been proposed, where a model is trained offline and then employed to infer white spaces in real-time. Despite their viability for spectrum sensing, these techniques are computationally complex, especially when deep neural network (DNN) models are pursued. Moreover, training a centralized model requires transmitting massive data to the fusion center (FC), which ultimately results in congestion on the transmission channel rather than flagging opportunities for cognitive transmissions. This paper opts to fill the technical gap by proposing: (1) a federated learning model that enables effective design of spectrum monitors in a distributed manner, and (2) a spiking neural network (SNN)-based lightweight white space detector that solves the issue of computationally expensive DNN models and is suitable for resource-constrained devices. The SNN-based federated learning (SFL) model employs secondary users to train their respective SNN models using local data (spatial locality) and sends the gradient of SNN to FC. FC combines the individual SNN models and sends the aggregated model back to each edge node. Validation using live LTE data has demonstrated the effectiveness of SFL with a detection accuracy of 91.16%.

The Dedicated Short-Range Communication (DSRC) protocol has become the de facto means for supporting vehicle ad hoc networking (VANET). Yet, the current allocated spectrum for DSRC is insufficient for handling large volumes of data, particularly during accidents or high traffic congestion, where most spectral resources are dedicated to control and emergency awareness, degrading infotainment and multimedia application services. Such a challenge motivates the pursuance of cognitive radio (CR), also known as CR-VANET, where vehicles opportunistically tap to utilize resources within the licensed spectrum as secondary users (SUs). To avoid interference with primary users (PUs), a vehicle needs to accurately sense the medium and detect white space. Given the dynamic nature of the VANET topology and varying volume of message traffic, machine-learning (ML) techniques, particularly those pursuing deep learning models, proved to be a viable option for spectrum sensing. However, the computational complexity of these techniques becomes an obstacle for vehicles. This paper opts to fill the technical gap by proposing a novel lightweight spectrum detector that employs spiking neural networks (SNN). The validation results using a field collected LTE-dataset clearly show that our proposed SNN model enables fast inference time, which ultimately expedites the spectrum sensing process and provides additional spectral resources for vehicular communication.

Optoacoustic communication enables an airborne unit to directly reach nodes deep underwater. To achieve high data rates in optoacoustic communications, implementing a multilevel modulation scheme is necessary where distinct acoustic signals can convey multiple symbols. However, demodulating these signals proves challenging amidst the complexities of underwater environments characterized by multipath propagation and resultant inter-symbol interferences. To overcome these challenges, this paper presents a novel demodulation scheme using a Spiking Neural Network (SNN). Our SNN model has undergone training using a laboratory-constructed dataset, comprising eight levels of optoacoustic signals recorded from three different underwater positions. Validation is conducted with a dataset deliberately designed to include severe interference from multipath-generated echoes and reverberations. The results indicate that our SNN-based demodulation scheme achieves an impressive accuracy of 90.16%, surpassing the 65.30% accuracy obtained through conventional peak detection-based techniques.

The major advances in wireless communication technology have led to increased adoption across almost all application domains. However, the massive growth has caused spectrum scarcity despite the fact that many of the frequency bands are not fully-utilized. Cognitive radios have emerged as a viable means to support dynamic spectrum access. Particularly, supporting opportunistic access through passive spectrum monitoring is of great interest. Existing techniques for detecting white space either require modification to commodity radio transceivers, or involve computationally complex models that do not suit resource-constrained devices. This paper opts to fill the technical gap by proposing a novel lightweight white space detector that employs spiking neural networks (SNN). SNN is a bio-inspired technique for creating data-driven models. The proposed design relies on the sensed energy in the medium to determine whether a primary user is active. The validation results using live LTE data demonstrate the effectiveness of our novel detector. Suitability for edge devices is confirmed through implementation on a Raspberry-PI platform.