Efficient Spectrum Occupancy Prediction Exploiting Multidimensional Correlations through Composite 2D-LSTM Models

Area of Science:

• Electrical Engineering
• Computer Science
• Telecommunications

Background:

• Efficient spectrum utilization is crucial for cognitive radio systems.
• Spectrum occupancy prediction leverages historical data to identify available frequencies.
• Multidimensional correlations in time, frequency, and space are key to spectrum usage.

Purpose of the Study:

• To address the computational complexity and retraining issues of existing tensor-based spectrum prediction methods.
• To propose a novel approach for exploiting multidimensional spectrum correlations more efficiently.
• To enhance the robustness and reduce the complexity of spectrum opportunity identification.

Main Methods:

• Decomposition of the multidimensional correlation problem into smaller sub-problems.
• Application of composite two-dimensional (2D)-Long Short-Term Memory (LSTM) models.
• Validation using extensive experimental results and real-world mobile network operator data.

Main Results:

• The proposed method demonstrates high spectrum detection performance.
• Achieved greater robustness in spectrum occupancy prediction.
• Significantly reduced computational complexity compared to existing tensor-based methods.
• Validated by real-world measurements from a leading Turkish mobile network operator.

Conclusions:

• The composite 2D-LSTM approach offers a more efficient and robust solution for spectrum occupancy prediction.
• This method effectively overcomes the limitations of traditional tensor-based techniques.
• The findings have practical implications for improving cognitive radio system performance.