A Kmeans-LSTM hybrid model for optimising spectrum sensing in wireless networks
| dc.contributor.advisor | Sibanda, K. | |
| dc.contributor.author | Tamuka, Nyashadzashe Everson | |
| dc.date.accessioned | 2026-09-07T10:08:12Z | |
| dc.date.available | 2026-09-07T10:08:12Z | |
| dc.date.issued | 2024-03 | |
| dc.description | PhD thesis | |
| dc.description.abstract | Due to the increasing demand for the wireless spectrum, cognitive radio (CR) technology has gained attention. Spectrum sensing, an essential task of CR, involves detecting vacant frequency bands for unlicensed users when licensed users are not fully utilising them. Traditional techniques like energy detection and matched-filter detection have limitations, including poor performance at low signal-to-noise ratios (SNR) and the need for prior knowledge of licensed user signal characteristics. Furthermore, most machine learning (ML) and hybrid techniques rely on simulated datasets, lacking real-world validation. This research proposed a novel hybrid Kmeans-LSTM model, robust at low SNR and independent of prior licensed user information. The model was validated with real-world spectrum datasets collected using an RTL-SDR dongle, unlike most of previous studies that depend on simulated datasets. Guided by design-science research methodology, the study involved iterative training, testing, and evaluation. The spectrum data was collected and labelled using K-means, followed by training the LSTM algorithm. Performance metrics, including accuracy, detection probability, false-alarm probability, precisionrecall curves, and ROC curves, showed that the Kmeans-LSTM model outperformed other models such as Support Vector Machine, Artificial Neural Network, and Random Forest. The real-world dataset provided a realistic assessment of the model’s performance, confirming its robustness and potential for optimizing spectrum sensing in wireless networks. This study demonstrated that the Kmeans-LSTM model can optimise spectrum sensing at low Signal-Noise-Ratio conditions without simulated and labelled spectrum datasets. | |
| dc.identifier.citation | Tamuka, N.E. (2024) A Kmeans-LSTM hybrid model for optimising spectrum sensing in wireless networks. PhD thesis. Alice, South Africa: University of Fort Hare. | |
| dc.identifier.uri | http://hdl.handle.net/20.500.11837/5093 | |
| dc.language.iso | en | |
| dc.publisher | University of Fort Hare | |
| dc.subject | Wireless communication systems | |
| dc.subject | Cognitive radio networks | |
| dc.subject | Machine learning | |
| dc.subject | Optimisation | |
| dc.title | A Kmeans-LSTM hybrid model for optimising spectrum sensing in wireless networks | |
| dc.type | Thesis | |
| person.identifier.orcid | 0000-0003-3391-7010 |