Attention-based deep convolutional neural network for spectral efficiency optimization in MIMO systems

Abstract

Spectral efficiency (SE) optimization in massive multiple input multiple output (MIMO) antenna cognitive systems is a challenge originated from the coexistence restrictions. Traditional power allocation can optimize the SE; however, involving deep learning can meet real-time and fairness processing requirements. In unfair allocation problem, all power is possibly assigned to one or few antennas of a particular user. In this paper, we build a mathematical optimization model considering the fairness problem such that SE is optimized for all users. To implement the model, we propose an attentionbased convolutional neural network (Att-CNN), where h 0 and h k (i.e., cross-interference and direct channels) attention mechanisms are used to improve the SE. The convolutional neural network is applied to decrease the floating point operations (FLOPs) and number of network parameters. We conducted experiments from these aspects: Fair antenna power allocation, power allocation performance and computational performance. Heat maps with different interference thresholds show the fair allocation for all users. Analyses of SE validate the superiority of the Att-CNN compared with the equal power allocation and fully connected neural network (FNN) schemes. The analyses of the FLOPs and number of parameters show the superiority of the Att-CNN over the FNN.

Ensemble learning‐based channel estimation and hybrid precoding for millimeter‐wave massive multiple input multiple output system

“…The efficiency of the suggested method was compared with the different deep learning models and algorithms regarding various measures. The suggested model was compared with various meat‐heuristic algorithms like Dragonfly Algorithm (DA), 35 Harris Hawks Optimization (HHO), 36 Grey Wolf Optimization (GWO), 37 Forest Optimization Algorithm (FOA), 38 Tunicate Swarm Algorithm (TSA), 39 GSO, 27 F‐TSA 40 and compared with other existing models like Orthogonal Matching Pursuit (OMP), 41 Approximate Message Passing (AMP), 42 DLCS, 20 Quantized Angle Linear Search (QALS), 43 DLQP, 20 Cat Swarm Optimization (CSO) 44 and compared with other deep learning models like CNN, 28 RNN, 29 Adaboost, 34 CNN + RNN, 45 CNN + Adaboost 46 . In predictive analytics, the learning rate is a tuning parameter in an optimization algorithm for each iteration through moves toward the lowest amount of a loss function.…”

“…The enhanced CNN‐RNN approaches estimate the channels from the original signals. The gathered benchmark dataset is trained using the fully connected layer of CNN, the extracted features from CNN 28 are inputted to RNN, 29 and the channel is estimated using RNN. The maximum epoch of CNN, the learning rate of CNN and hidden neuron of CNN, 30 and the hidden neuron count of RNN are optimized using the SAS‐GSO algorithm.…”

Throughput Maximization Using Deep Complex Networks for Industrial Internet of Things

“…The input data are directly fed into three consecutive complex dense layers, then the output will be flattened and fed into four complex dense layers with the complex activation functions: , , , and , respectively, to generate the final result. The AttCNN is a real-valued attention-based power allocation network, which was proposed in [ 20 ]. The AttCVNN is defined in Section 4 , which realizes the complex-valued layers and complex-valued attention mechanism.…”

“…The authors of [ 7 , 18 ] used a fully connected neural network (FNN) to estimate the best power allocation solution to maximize SE, and Lee et al [ 19 ] proposed a convolutional neural network for power control; however, their method cannot strictly control constraints, and the FNN has the problem of unfair power allocation. Hence, Sun et al [ 20 ] proposed the attention-based deep convolutional neural network, which has also a better time and storage space complexity. However, they all utilized real-valued neural networks to process the complex-valued channel data, which generally take the complex-valued input data as two separate parts of real-valued data.…”

“… Four power allocation benchmarks are implemented to show the superiority of our model. They are the equal power allocation method (EPM), the real-valued fully connected network (FNN) [ 7 ], the complex-valued fully connected network (CVFNN), and the real-valued convolutional network (AttCNN) [ 20 ]. …”