Channel state information estimation for 5G wireless communication systems: recurrent neural networks approach

Area of Science:

• Electrical Engineering
• Computer Science
• Signal Processing

Background:

• Accurate channel state information (CSI) is crucial for efficient wireless communication systems.
• Traditional CSI estimation methods face challenges with complex channel statistics and limited pilot data.
• Deep learning offers potential for advanced signal processing in modern communication.

Purpose of the Study:

• To propose a deep learning-based channel state information estimator for 5G orthogonal frequency-division multiplexing (OFDM) systems.
• To develop a pilot-dependent estimator utilizing a bidirectional long short-term memory (BiLSTM) recurrent neural network.
• To evaluate the estimator's performance under conditions with limited pilot availability and uncertain channel statistics.

Main Methods:

• Implementation of a pilot-dependent BiLSTM recurrent neural network for CSI estimation.
• Utilizing an online learning approach for training and an offline approach for practical implementation.
• Comparative analysis using three classification layers with different loss functions (MAE, cross-entropy, SSE) and optimization algorithms (Adam, RMSProp, SGdm, Adadelta).

Main Results:

• The proposed BiLSTM estimator demonstrated superior performance compared to LSTM, least squares (LS), and minimum mean square error (MMSE) estimators.
• Performance was evaluated using symbol error rate (SER) and accuracy metrics across various simulation conditions.
• Computational and training time complexities of BiLSTM and LSTM estimators were analyzed.

Conclusions:

• The deep learning BiLSTM-based CSI estimator shows significant promise for 5G and future communication systems.
• The approach effectively analyzes massive data, recognizes statistical dependencies, and generalizes knowledge to new datasets.
• The estimator achieves high accuracy and low SER, particularly with limited pilot data.