In massive multiple-input multiple-output (MIMO) wireless communication systems, power control and user association are crucial factors that affect the overall performance. In this article, we propose an efficient method for power control and user association using deep learning techniques. Our approach leverages the large scale fading (LSF) coefficients as input to a designed neural network (DNN) topology, which produces the desired power control (PC) and user association (UA) coefficients.
Convolutional Neural Networks
We employ convolutional neural networks (CNNs) to model the LSF coefficients. Each CNN block consists of convolutional layers, batch normalization layers, and rectified linear units (ReLU) activation functions. We use a modified residual dense network (ResNet) to fully utilize all features from the original LSF coefficients.
Deep Learning Topology
Our DNN topology is designed to guarantee the same dimension between the input and output of each convolutional layer. We set stride and zero padding to 1 to ensure clarity in expression. The number of kernels (d1) and their values are d2 = 64, and we omit the bias term from each layer to simplify the expression.
Power Control
Our proposed method captures the power control coefficients using a sigmoid function applied to the output of the DNN. The sigmoid function maps the output to a value between 0 and 1, which represents the transmitted power level.
User Association
We associate each user with the AP that maximizes the association coefficient. The association coefficient is computed using the output of the DNN. We use a modified version of the sum spectral efficiency (SE) algorithm to compute the SE for each user and associate them with the AP that provides the maximum SE.
Simulation Results
We evaluate our proposed method through simulations, comparing it to traditional power control and user association methods. Our results show that our proposed method outperforms traditional methods in terms of sum spectral efficiency and downlink max-min power control.
Conclusion
In conclusion, our proposed method for efficient power control and user association in massive MIMO downlink wireless communication systems using deep learning techniques significantly improves the overall performance compared to traditional methods. Our approach leverages the large scale fading coefficients as input to a designed neural network topology, which produces the desired PC and UA coefficients. The use of CNNs and ResNet allows us to model complex relationships between the LSF coefficients and the PC and UA coefficients. Our proposed method can be used in practical scenarios to improve the performance of massive MIMO downlink wireless communication systems.
