This work studies learning-aided indoor fingerprint positioning (FP) for multiple-input multiple-output (MIMO) orthogonal frequency-division multiplexing (OFDM) systems, by exploiting their channel response matrices (CRMs). The existing learning methods of CRM-based FP mostly suffer from two main limitations: i) they require retraining when CRM dimensions change, which severely limits their practicality since the number of antennas and subcarriers may vary dynamically during practical implementation; ii) they overlook the temporal correlations in CRMs along the user trajectory, failing to exploit this information to improve positioning accuracy. Motivated by these limitations, we propose a novel deep learning-assisted FP method, named adaptive spatio-temporal neural network (A-STNN). This method consists of two key components: i) an adaptive mechanism that transforms spatial-frequency domain CRMs (SFCRMs) of arbitrary dimensions into truncated angle-delay domain CRMs (T-ADCRMs) with a unified dimension, enabling the neural network to handle varying antenna and subcarrier configurations; and ii) an STNN that exploits attention mechanisms to jointly extract spatial and temporal features embedded in T-ADCRM sequences along the user’s trajectory, thus improving positioning accuracy. Upon examination of simulation and measurement datasets, A-STNN achieves a prominent improvement in positioning accuracy over existing FP methods, in addition to its unique generalization capability across different antenna and subcarrier settings.

The ever growing number of antennas in massive multiple-input multiple-output (MIMO) systems have significantly burdened the demand for the feedback of channel state information (CSI). This drives the research on CSI compression for massive MIMO. Recent development of autoencoder-based methods has proven to break the limit of conventional data compression methods in terms of both accuracy and computational complexity. However, like conventional methods, those autoencoder-based methods are trained for certain channel scenarios (such as indoor and outdoor) and would require a dedicated model for each scenario, limiting the practicability. It is challenging to develop a unified model to compress CSI across different channel scenarios, as their properties are diverse. In this paper, such a model is proposed for the first time, which is named attention-infused autoencoder network (AiANet). A dual attention mechanism is developed to capture the spatial and channel features of distinctive CSI. Multi-resolution convolutions are also employed to enhance the ability of feature extraction. Simulation results demonstrate that AiANet can substantially outperform ACRNet, which is a state-of-the-art auto encoder model. In terms of normalized mean squared error (NMSE), the proposed method can achieve an improvement of up to 3.69 dB in indoor and 1.26 dB in outdoor.

Minimizing energy consumption is critical for the long-range (LoRa) Internet of Things (IoT) networks, to extend the battery lifetime of end devices (EDs) while reducing the maintenance cost. To overcome the excessive computational complexity required by the traditional optimization methods, learning approaches such as deep neural network (DNN) and reinforcement learning (RL) have been researched in wireless networks, where star topologies are widely employed. In contrast, LoRa usually deploys relays to assist the connection between the EDs and the gateway (GW), leading to a much more complex network topology. Consequently, DNN and RL would become less effective in LoRa, since those methods are difficult to learn the complex network topology constructed by LoRa. In this paper, we propose a learning method based on graph neural network (GNN), which is known for its prominent ability to capture and represent intricate graph-structural dependencies, to tackle the energy optimization problem for relay-assisted LoRa. Specifically, the multi-hop LoRa network is modeled as a directed graph, with channel state information (CSI) defined as node features, while the spreading factor and transmission power are deemed labels. A hierarchical message aggregation mechanism is proposed to effectively capture the multi-hop structural dependencies, followed by the process of inductive learning. Results show that against conventional optimization algorithms, the proposed method can achieve near-optimal energy consumption with a gap of 10% to 16%, while reducing the runtime by about six orders of magnitude. Compared to DNN, the GNN-based model can provide an energy saving of up to 32%, at a similar level of inference time.

Energy optimization is crucial for extending battery life and reducing maintenance costs in long-range (LoRa) Internet of Things (IoT) networks. Traditional optimization methods usually need excessive computational complexity, limiting the practicability. This drives the recent development of machine learning (ML)-based optimization, such as deep neural networks (DNNs) and reinforcement learning (RL), in wireless local area networks (WLANs). However, compared to WLANs, LoRa owns a more complicated network topology due to the engagement of multi-hop, which is difficult for the existing ML methods to learn. Motivated by this, we propose a topology-aware graph neural network (GNN) learning method, which is specially tailored to tackle the energy optimization problem in multi-hop LoRa networks. By leveraging each node’s topological position to adaptively determine the optimal message-passing depth, the model better integrates the neighborhood information with node feature representations, enhancing the prediction of transmission parameter and overall energy efficiency. Also, a closed-form model of collision probability is derived for the nodes in LoRa, to measure the energy consumption due to retransmissions. Simulation results show that against traditional optimization methods such as game theory, the proposed method can reduce the runtime by five orders of magnitude, with an energy optimization gap below 13%. Compared to existing GNN-based methods, it achieves up to 50% lower energy consumption 20% fewer outage probability, at a similar level of inference time.

Hybrid light fidelity (LiFi) and wireless fidelity (WiFi) networks (HLWNets) are an emerging indoor wireless communication paradigm, which combines the complementary advantages of LiFi and WiFi. Meanwhile, load balancing (LB) becomes an essential and critical challenge, due to the nature of hybrid networks. The existing LB methods are mostly network-centric, relying on a central unit to make a solution for the users all at once. Consequently, the solution needs to be updated for all users at the same pace, regardless of their moving status. This would affect the network performance in two aspects: 1) a lower update frequency would compromise the connectivity of fast-moving users; 2) a higher update frequency would cause unnecessary handovers as well as hefty feedback costs for slow-moving users. Motivated by this, we investigate user-centric LB so that users can update their solutions at different paces. The research is developed upon our previous work on adaptive target-condition neural network (ATCNN), which carries out LB for individual users in quasi-static channels. In this paper, a deep neural network (DNN) model is designed to enable an adaptive update interval for each individual user. This new model is termed as mobility-supporting neural network (MSNN). Associating MSNN with ATCNN, a user-centric LB framework named mobility-supporting ATCNN (MS-ATCNN) is proposed to handle resource management and mobility management simultaneously. Results show that at the same level of average update interval, MS-ATCNN can achieve a network throughput up to 215% higher than conventional LB methods such as game theory (GT), especially for a larger number of users. In addition, MS-ATCNN costs an ultra-low inference time in sub-milliseconds, which is two to three orders of magnitude lower than the GT baseline.

Load balancing (LB) is a key challenge in hybrid light fidelity (LiFi) and wireless fidelity (WiFi) networks (HLWNets), due to the nature of heterogeneous access points (APs). Machine learning has the potential to provide a complexity-friendly LB solution with near-optimal network performance, at the cost of a non-trivial training process. The state-of-the-art learning-aided LB methods require retraining when the network environment (particularly the user number) changes, significantly limiting their practicability. In this paper a novel deep neural network (DNN) structure, named adaptive target-condition neural network (A-TCNN), is proposed to tackle the LB issue for a varying number of users, without the need for retraining. Unlike the existing LB methods conducting AP selection for all users together, the new method performs AP selection for a single target user, upon the condition of other users. Also, A-TCNN involves an adaptive mechanism which maps any smaller number of users to a preset number by splitting the users’ data rate requirements, without affecting the AP selection result for the target user. Once trained, A-TCNN can be used for any user numbers not exceeding the maximum user number that the network can support. Results show that apart from the adaptiveness to a varying user number, A-TCNN provides a higher network throughput (up to 45%) than the conventional DNN in most cases, especially for a larger scale of network. In terms of computational complexity, A-TCNN can achieve a sub-millisecond level runtime, which is 2 orders of magnitude lower than fuzzy logic and 3 orders of magnitude lower than game theory.

Layered or hybrid optical orthogonal frequency division multiplexing (OFDM) has been proposed for use in optical communications due to its excellent spectral and power efficiencies, especially in visible light communications (VLC). However, most of the current works concentrate on transmitter and receiver design as well as the quality of service in communication networks. In this paper, we propose a spectrum-efficient dimmable triple-layer hybrid optical OFDM (DTH-OFDM) scheme to tackle the illumination requirements, considering different practical indoor VLC scenarios from low illumination to high illumination intensities. In the proposed DTH-OFDM scheme, the required dimming level is achieved by jointly adjusting the dimming factors and direct current bias. We investigate the comprehensive performance analysis of the proposed DTH-OFDM in detail, including probability density function, bit error rate (BER), spectral and energy efficiencies. In addition, a joint dimming control and optimal power allocation problem for DTH-OFDM is formulated and solved using convex optimization under the constraints of light emitting diode (LED) nonlinearity, dimming target and communications reliability. Numerical results show that, the proposed DTH-OFDM can offer continuous and arbitrary dimming target with higher spectral efficiency and lower BER compared with its counterparts, as well as an enhanced tolerance to the LED nonlinearity.