This paper proposes a novel indoor optical wireless positioning (IOWP) framework that aims to enhance localization precision and robustness through an advanced machine learning (ML)-driven fusion technique. Unlike traditional single-model approaches, the proposed framework uses received signal strength (RSS) data to intelligently combine multiple lightweight ML algorithms, including K-Nearest Neighbors (KNN), Random Forest (RF), and Gaussian Process Regression (GPR). In the training phase, our system utilizes a performance-optimized weight allocation strategy to identify the optimal weights, harnessing the complementary strengths of individual models while mitigating their limitations to achieve exceptional generalization in complex indoor environments. A comprehensive evaluation is conducted under a realistic ray-traced channel model that incorporates typical light distributions, high-order multipath reflections from walls and objects, and mixed diffuse-reflective surface interactions. Performance is assessed in terms of mean positioning error (MPE), 90th percentile (P90) error, and computational complexity. Results demonstrate that the proposed method achieves an MPE of 0.5 cm and a P90 error below 1 cm, offering a practical and scalable solution for next-generation IOWP applications in smart environments.

The practical deployment of visible light communication (VLC)-enabled flying networks could be accelerated via the simultaneous services offered by mixed carrier communication (MCC). The MCC is a unified physical-layer waveform that uniquely integrates various communication streams for data/control and signaling to support localization and dimming. For such a composite waveform, the receiver must conduct successive decoding operations of the constituent streams; thus, optimum decoding of the individual streams is essential. Existing publications rely on a conventional way to design the MCC communication chain using a block-based standalone approach that provides suboptimal performance. However, a machine learning (ML) framework has the potential for better optimization of the overall MCC system’s performance. This article proposes a novel end-to-end learning framework toward a unified physical-layer waveform for VLC-enabled indoor flying networks. The obtained symbol error rate (SER) performance results confirm that the proposed learning framework outperforms the existing work by ensuring perfect decoding of the control stream and detecting signals for localization. A 12 to 15-dB gain in signal-to-noise ratio (SNR) is demonstrated at a target of 10−3 SER. The results also indicate that a wide range of dimming is supported without significantly altering SER performance. The topics of convergence, spectral efficiency, and complexity analysis of the proposed framework are discussed in detail.

To enable intelligent network management and various 6G smart services, the precise estimation of user location and device orientation is required. Light fidelity (LiFi) based on vertical cavity surface emitting lasers (VCSELs) can not only respond to the needs of 6G communication networks in terms of ultra-high data rate, connection density and area capacity, but also enable high precision position and orientation estimation. However, this problem of joint position and orientation estimation is a non-convex optimization problem. Therefore, in this paper, we design deep neural networks (DNNs) for joint position and orientation estimation of user devices in a VCSEL-based LiFi access network. Simulation results demonstrate that the proposed framework outperforms state-of-the-art methods by significantly reducing position and orientation estimation errors while maintaining a lower complexity. We illustrate the effectiveness of the proposed DNN solution by considering two types of network deployment including distributed VCSELs and collocated VCSELs. In addition, we present the convergence and complexity analysis for the proposed learning framework. It is shown that the proposed DNN provides at least 69% and 27.9% improvements in the mean estimation error for position and orientation, respectively, over the baseline method.

The Light Fidelity (LiFi) and Wireless Fidelity (WiFi) technologies operate in non-overlapping spectrum, thus, a user is capable of receiving data concurrently from both LiFi and WiFi Access points (APs). Therefore, facilitating a link aggregation (LA) enabled heterogeneous LiFi-WiFi network (HLWN). In this paper, we have analysed the performance of LA enabled HLWN by utilizing an LA based on SINR (LA-SINR) algorithm and compared it with the conventional indoor access networks like the hybrid LiFi-WiFi, standalone WiFi and standalone LiFi. Additionally, an intuitive LA for enhancement of QoS (LA-EQoS) algorithm has been proposed to further improve the quality of service (QoS) and average data rate performance of LA enabled HLWN. The simulation results validate that for a predefined QoS requirement, proposed LA-EQoS provides 80% to 100% coverage for more number of users as compared to other networks. Moreover, for a given outage constraint, the maximum QoS offered by LA-EQoS is 1.5 times higher than other networks. 16

Light fidelity (LiFi) is an emerging communication technology that utilizes light intensity modulation in order to transfer data from light-emitting diode (LED) to users. Due to the vast visible light spectrum, LiFi can support high data rates; however, its coverage is limited. In contrast to LiFi, WiFi works in radio frequency and is capable of providing ubiquitous coverage with limited data rates. Since the spectrum of LiFi does not overlap with WiFi, both can co-exist to form a hybrid LiFi and WiFi network. The advantage of hybrid LiFi and WiFi network is that it provides high data rates and better connectivity. The performance of a hybrid LiFi and WiFi network significantly depends upon the load balancing strategies. Therefore, in this paper, gradient descent-based reinforcement learning (RL) has been proposed to determine an optimal access point (AP) assignment policy that aims to maximize the average network throughput while ensuring user's satisfaction. The performance of the proposed method is then compared against conventional signal strength strategy (SSS); the results are presented in terms of the average network throughput, user satisfaction, and outage probability. Based on the results, it was observed that the proposed RL method provides a significant improvement in all the performance metrics over the SSS based method.