Differential-privacy federated learning (DP-FL) has emerged as a promising approach to mitigate the inherent risks associated with traditional FL architectures, which are susceptible to inferential attacks due to the continuous sharing and updating of model parameters. However, existing DP-FL frameworks typically assume that the perturbations introduced by clients remain constant throughout the FL process, overlooking the dynamic influence of these perturbations on model performance across different communication rounds. In this paper, we present a long-term privacy-preserving FL framework designed to address issues of optimal incentive design, considering the dynamic influence of perturbations on model performance. Specifically, we first analyze the effect of local perturbations on the model’s convergence performance during various communication rounds, elucidating the balance between learning performance and privacy loss. Then, to harmonize learning performance with privacy loss, we develop a long-term privacy-preserving incentive scheme, where the interactions between clients and the FL server throughout the FL process are modeled as a multi-stage privacy-preserving game. Furthermore, we utilize contract theory to derive the equilibrium of this game. Finally, simulations show that our scheme can incentivize clients to contribute high-quality models, thereby enhancing the accuracy of the global model, as compared to benchmarks.

Crowd computing in large-scale cellular networks typically involves a significant number of participants, leading to high spectrum interference, reduced communication efficiency, and low user trustworthiness. To overcome these challenges, this paper proposes trustworthy and spectrum-efficient crowd computing scheme in massive multiple-input multiple-output (MIMO) networks based on deep neural networks (DNNs). Existing machine learning-aided multiple antenna technologies usually ignore the pilot contamination, which reduces spectrum efficiency. Here, we leverage the DNN to devise detection and precoding algorithms by inputting an imperfect channel state information (CSI) big data and provides detection and precoding matrices as outputs, where the online-to-offline learning framework offloads the training task to servers to reduce the overhead of base station (BS). With the well-trained DNNs, the BS can generate the detection and precoding matrices with low computation overheads. Especially, minimum-mean-square-error (MMSE) triggers between received signals and sources considering channel estimation error are seen as labels to improve spectrum efficiency. Besides, a multi-factor trust model is designed to enhance user authentication security. The simulations and numerical analysis show the proposed scheme can provide higher spectral efficiency, compared to conventional methods.

Device-free based object tracking provides a promising solution for many localization and tracking systems to monitor non-cooperative objects which do not carry any transceiver such as intruders. However, existing device-free solutions mainly use sensors and active RFID tags, which are much more expensive compared to passive tags. In this paper, we propose a novel motion detection and tracking method using passive RFID tags, named Twins. The method leverages a phenomenon called critical state caused by interference among passive tags. We theoretically explain this phenomenon via an interference model and conduct extensive experiment to validate it. We design a practical Twins based intrusion detection system and implement a real prototype with commercial off-the-shelf reader and tags. Experimental results show that Twins is effective in detecting the moving object, with low location errors of 0.75m in average.