Deep Learning applied for Spectrum Sensing using multiple and concurrent stages

Ouamna, Hamza; Madini, Zhour; Zouine, Y.

doi:10.21203/rs.3.rs-1821341/v1

Cited by 1 publication

(1 citation statement)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Though the SNR was minimized, the accuracy level was not improved by hybrid model. Energy Level detection was carried out in [11] with deep learning network to differentiate between presence of signal and noise. The AlexNet, SqueezeNet ResNet101 and LSTM neural networks were carried out for eliminating noise.…”

Section: Algorithms Like Multilayer Perceptrons (Mlps) Convolutional ...mentioning

confidence: 99%

Instance Weighting Domain Adaptive Deep Transfer Learning for Spectrum Sensing in Cognitive Radio Network

Muthukumaran,

Sharanya,

Natarajan

et al. 2023

Preprint

View full text Add to dashboard Cite

Cognitive radio allowed the licensed vacant spectrum sharing to unlicensed users depending on the radio scene analysis. Cognitive radio performed machine-to-machine communication. Radio scene analysis is carried out in spectrum sensing, interference and channel estimation. Cognitive radio increased the spectrum utilization in wireless communication. Spectrum sensing is an essential component of cognitive radio based on feature extraction of received signal at specified point. Many existing methods are discussed for performing efficient spectrum sensing. But, the spectrum sensing accuracy was not improved and time complexity was not minimized by existing methods. The transfer learning has an opportunity to increase spectrum sensing accuracy through cooperative spectrum sensing and to analyze the radio scene. In order to improve the accuracy and reduce the time complexity, Instance Weighting Domain Adaptive Deep Transfer Learning (IWDADTL) Method is introduced based on data features, patterns and temporal characteristics. IWDADTL Method comprised five layers, namely one input layer, three hidden layers and one output layer to identify the presence or absence of signal at specified location. The designed method has three stages in three hidden layers, namely data splitting, phase selection and fine tuning. In IWDADTL Method, data splitting stage is carried out to divide the data into source data and target data. The source data is used for training model where positive values are paramount. Target data is used for fine-tuning the negative value from − 20dB to − 2dB where majority values are negative. The data splitting is used to tackle unstable training of noisy signals to attain efficient performance accuracy. In phase selection stage, signal components (i.e., in-phase and quadrature-phase) are selected for training, validation and testing on source data. In second stage, an optimal structure is obtained through hyperparameter tuning. Then, IWDADTL Method is fine-tuned on target data to improve the accuracy results and to reduce time complexity. The results of IWDADTL Method show the improvement in spectrum sensing in terms of accuracy, precision and recall.

show abstract

Section: Algorithms Like Multilayer Perceptrons (Mlps) Convolutional ...mentioning

confidence: 99%