Using analog beamforming in mmWave frequency bands we can focus the energy towards a receiver to achieve high throughput. However, this requires the network to quickly find the best downlink beam configuration in the face of non-IID data. We propose a personalized Federated Learning (FL) method to address this challenge, where we learn a mapping between uplink Sub-6GHz channel estimates and the best downlink beam in heterogeneous scenarios with non-IID characteristics. We also devise FedLion, a FL implementation of the Lion optimization algorithm. Our approach reduces the signalling overhead and provides superior performance, up to 33.6 % higher accuracy than a single FL model and 6 % higher than a local model.
Federated Learning (FL) is a promising framework for distributed learning when data is private and sensitive. However, the state-of-the-art solutions in this framework are not optimal when data is heterogeneous and non-Independent and Identically Distributed (non-IID). We propose a practical and robust approach to personalization in FL that adjusts to heterogeneous and non-IID data by balancing exploration and exploitation of several global models. To achieve our aim of personalization, we use a Mixture of Experts (MoE) that learns to group clients that are similar to each other, while using the global models more efficiently. We show that our approach achieves an accuracy up to 29.78 % and up to 4.38 % better compared to a local model in a pathological non-IID setting, even though we tune our approach in the IID setting. Adaptive Expert Models for Personalization in Federated Learning poster.
Machine Learning (ML) is an important enabler for optimizing, securing and managing mobile networks. This leads to increased collection and processing of data from network functions, which in turn may increase threats to sensitive end-user information. Consequently, mechanisms to reduce threats to end-user privacy are needed to take full advantage of ML. We seamlessly integrate Federated Learning (FL) into the 3GPP 5G Network Data Analytics (NWDA) architecture, and add a Multi-Party Computation (MPC) protocol for protecting the confidentiality of local updates. We evaluate the protocol and find that it has much lower overhead than previous work, without affecting ML performance. Secure Federated Learning in 5G Mobile Networks poster.
There is a rapid evolution in telecommunication with denser networks and systems operating on an increasing number of frequency bands. Also in the next generation 5G networks, even further densification is needed to be able to reach the tight requirements, implying more nodes on each carrier. Denser networks and more frequencies makes it challenging to ensure the best possible cell and frequency carrier assignment to a User Equipment (UE), without the UE needing to perform an excessive amount of inter-frequency measurements and reporting. In this paper, we propose a procedure of predicting the strongest cell of a secondary carrier, and the procedure is exemplified in a UE load-balancing use case. The prediction is based on only measurements on the primary carrier cells, avoiding costly inter-frequency measurements. Simulations of a realistic network deployment show that a UE selection based on the proposed secondary carrier prediction is significantly better than a random UE selection for load balancing.
A Wireless Sensor Network (WSN) of anisotropic magnetoresistor sensors offers a low-cost alter- native to other traffic measurement technologies. The WSNs offer better reliability than other solutions, they offer more information, can be deployed quickly and be reused. In this thesis the sensor algorithms used for detection, velocity estimation, queue detection and classification in such a network are evaluated based on simulated and measured data. A number of algorithms are evaluated and the results are compared. A new algorithm for speed estimation using two sensor nodes is proposed and evaluated. It is found to be much better than earlier algorithms, requiring a signal to noise ratio (SNR) of 20 dB less than the traditional algorithm.