We address physical layer security in distributed antenna (DA) systems, where eavesdroppers (Eves) can intercept the information transmitted for the intended receiver (IR). To realize a user-centric, powerefficient and physical layer security-addressing system, we aim at minimizing total power consumption by jointly designing DA selection and secure precoding. Different from the conventional artificial noise (AN)-aided secure transmission, where AN is treated as an undesired element for the IR, we design AN such that it is constructive to the IR while keeping destructive to the Eves. Importantly, we investigate two practical scenarios, where the IR and Eves' channel state information (CSI) is imperfectly obtained or the Eves' CSI is completely unknown. To handle the CSI uncertainties, we solve the problems in probabilistic and deterministic robust optimization respectively, both satisfying the IR' signal-tointerference-and-ratio (SINR) requirement by use of constructive AN and addressing security against the Eves. Simulation results demonstrate our algorithms consume much less power compared to the centralized antenna (CA) systems with/without antenna selection, as well as the DA systems with conventional AN processing. Last but not least, by the proposed algorithms, the activation of DAs closely relates to users' locations and quality-of-service (QoS) requirements, featuring a user-centric and on-demand structure. 2 I. INTRODUCTION Wireless Communications for the future Internet of Things (IoT) and Industry 4.0 are required to provide power-efficient transmission together with high security level [1]. In the last decade, centralized multiple-input multiple-output (MIMO) has been considered as a potential technique due to its high throughput [2] and additional spatial diversity for enhancing physical layer security [3]. However, it requires extremely high power consumption caused by the fully activated antennas, and centralized MIMO often suffers from an equal level of path loss (PL) from the antenna array to one user caused by the co-located antenna (CA) deployment [4]. Besides, edge users in CA deployment may not be well served due to the severe propagation attenuation, otherwise significant transmission power is required for compensating the propagation loss. To create a user-centric and power efficient structure necessary for the IoT and Industry 4.0, distributed antenna (DA) systems have attracted much attention [5]. By geographically distributing the antennas and hence placing them closer to users, DA systems can reduce the PL impact and obtain blockage-free effect and also facilitate an on-demand network structure by activating those DAs contributing the most. The concepts of user-centric DA are particularly suited for communications in industrial environments, where antennas can be carefully planned and distributed in the ceilings of large factories to effectively extend network coverage without crucial increment of power consumption, and are therefore a key contender for industry IoT and Industry 4.0 ...