Closing the Loop: A Simple Distributed Method for Control over Wireless Networks ∗ Miroslav Pajic Shreyas Sundaram Dept. of Electrical & Systems Eng. Dept. of Electrical & Computer Eng. University of Pennsylvania University of Waterloo [email protected] [email protected] Jerome Le Ny George J. Pappas Dept. of Electrical & Systems Eng. Dept. of Electrical & Systems Eng. University of Pennsylvania University of Pennsylvania [email protected] [email protected] Rahul Mangharam Dept. of Electrical & Systems Eng. University of Pennsylvania [email protected] ABSTRACT Categories and Subject Descriptors We present a distributed scheme used for control over a net- C.2.1 [Computer-Communication Networks]: Network work of wireless nodes. As opposed to traditional networked Architecture and Design|Wireless communication; C.3 [Special- control schemes where the nodes simply route information to purpose and Application-based Systems]: Process con- and from a dedicated controller (perhaps performing some trol systems encoding along the way), our approach, Wireless Control Network (WCN), treats the network itself as the controller. In other words, the computation of the control law is done in General Terms a fully distributed way inside the network. We extend the Algorithms, Design, Theory, Performance, Reliability basic WCN strategy, where at each time-step, each node updates its internal state to be a linear combination of the states of the nodes in its neighborhood. This causes the en- Keywords tire network to behave as a linear dynamical system, with Cooperative control, decentralized control, networked con- sparsity constraints imposed by the network topology. We trol systems, wireless sensor networks demonstrate that with observer style updates, the WCN's robustness to link failures is substantially improved. Fur- thermore, we show how to design a WCN that can maintain 1. INTRODUCTION stability even in cases of node failures. We also address the Improvements in the capabilities and cost of wireless tech- problem of WCN synthesis with guaranteed optimal perfor- nologies have allowed multi-hop wireless networks to be used mance of the plant, with respect to standard cost functions. as a means of (open-loop) monitoring of large-scale indus- We extend the synthesis procedure to deal with continuous- trial plants [1, 2]. With this technology, sensor measure- time plants and demonstrate how the WCN can be used on a ments of plant variables can be transmitted to controllers, practical, industrial application, using a process-in-the-loop data centers and plant operators without the need for exces- setup with real hardware. sive wiring, thereby yielding gains in efficiency and flexibility for the operator. However, the use of multi-hop wireless net- works in closed-loop feedback control is in its infancy, and is ∗This work has been partially supported by NSF-CNS an active area of research [3, 4]. Wireless Networked Con- 0931239 and NSF-MRI 0923518 grants. It has also been trol Systems (WNCSs) fundamentally differ from standard funded by a grant from the Natural Sciences and Engineer- wired distributed systems in that the dynamics of the net- ing Research Council of Canada (NSERC). work (variable channel capacity, probabilistic connectivity, topological changes, node and link failures) can change the operating points and physical dynamics of the closed-loop system [3, 5]. The most important objective of WNCSs is Permission to make digital or hard copies of all or part of this work for to provide stability of the closed-loop system. An additional personal or classroom use is granted without fee provided that copies are requirement is optimality with respect to some appropriate not made or distributed for profit or commercial advantage and that copies cost function. It is therefore necessary for the network (along bear this notice and the full citation on the first page. To copy otherwise, to with its interfaces to sensors and actuators) to be able to republish, to post on servers or to redistribute to lists, requires prior specific provide some form of guarantee of the control system's sta- permission and/or a fee. bility in the face of the non-idealities of the wireless links IPSN’12, April 16–20, 2012, Beijing, China. Copyright 2012 ACM 978-1-4503-1227-1/12/04 ...$10.00. and the communication constraints of the wireless network. 25 WCN s1 s1 v1 v2 s1 v1 v2 a1 a1 a1 v5 v5 a2 a2 a2 v4 v3 v4 v3 s2 s2 s2 Plant Controller Plant Plant v6 v6 s3 s3 Controller s3 . ... ... v9 v9 s v8 v p sp v7 sp v7 8 v am am v 10 a v10 9 m v9 (a) Wired Network Control (b) Wireless Network Controlled System (c) Wireless Control Network Figure 1: Standard architectures for Networked Control Systems; (a) Wired system with a shared bus and dedicated controller; (b) Red links/nodes - routing data from the plant's sensors to the controller; Blue links/nodes - routing data from the controller to the plant's actuators; (c) A multi-hop wireless control network used as a distributed controller. The most common approach to incorporating WNCSs into order of several seconds to a few minutes and the control the feedback loop is to use it primarily as a communication network is expected to operate at rates of hundreds of mil- medium: the nodes in the network simply route informa- liseconds. While such plants may have as many as 80,000 tion to and from one or more dedicated controllers, which to 110,000 control loops, they are organized in a hierarchal are usually specialized CPUs capable of performing compu- manner such that networks span 10-20 wireless nodes (per tationally expensive procedures (see Fig. 1(b)). The use of gateway) for low-level control [6]. Therefore, in this work dedicated controllers imposes a routing requirement along we focus on the networks with up to a few tens of nodes. one or more fixed paths through the network, which must Furthermore, the networks might be shared among control meet the stability constraints, encapsulated by end-to-end loops (i.e., a node may be involved in several feedback loops), delay requirements [6, 7]. However, this assignment of routes and new feedback loops may be added at run-time. Adding is a static setup, which commonly requires global reorganiza- new communication loops in a standard WNCS could affect tion for changes in the underlying topology, node population the performance of the existing loops, and the system must and wireless link capacities. be analyzed as a whole. Although techniques have been de- Routing couples the communication, computation and con- veloped for compositional analysis of WNCS (e.g., [4]), their trol problems [4, 8, 9]. Therefore, when a new route is re- complexity limits their use in these applications. Therefore, quired due to topological changes, the computation and con- it is necessary to derive a composable control scheme, where trol configurations must also be recalculated. Merely insert- control loops can be easily added and a simple compositional ing a WNCS into the standard network architecture \sensor analysis can be performed at run-time, to ensure that one ! channel ! controller/estimator ! channel ! actuator" loop does not affect performance of other loops. requires the addition of significant software support [7, 10], Finally, with the use of asynchronous event triggered net- as the overhead of completely recomputing the computa- work substrates, it is difficult to design, model and analyze tion and control configurations, due to topological changes control networks and also hard, if not impossible, to ver- or packet drops, is too expensive and does not scale. ify performance of a system that consists of several event- triggered loops. On the other hand, full network synchro- 1.1 Wireless Control Design Challenge nization allows the use of Time-Triggered Architectures (TTAs) It should be noted that providing closed-loop stability where communication and computation are scheduled at and performance guarantees for wireless control networks particular instances of time (i.e., time slots) [11]. This sim- is a challenging problem. On one hand, the control sys- plifies modeling of the closed-loop system, which now con- tems community typically abstracts away the systems de- sists of continuous-time physical dynamics and a communi- tails and solves the problem for semi-idealized networks with cation network. With TTA, the closed-loop system can be approximated noise distributions and link perturbations [3, modeled as a switched control system [4], allowing the use 5]. While this approach provides mathematical certainty of existing techniques for switched-system analysis. There- of the properties of the network, it fails to provide a sys- fore, when communication and computation schedules are tematic path to real-world network design. On the other derived, it is possible to determine if the closed-loop system hand, the network systems community uses hardware and is stable (i.e., asymptotically stable or mean square stable) software approaches to address open-loop issues, but these for channel errors, with and/or without permanent link or fail to provide any guarantees to maintaining stability and node failures. performance of closed-loop control. We propose a control scheme over wireless networks that provides closed-loop sta- 1.2 Contributions bility or optimality with respect to standard metrics, while In this effort, we build on the work from [12], in which maintaining ease of implementation in real-world networks. we introduced the Wireless Control Network (WCN), a fun- The applications of interest in this work are industrial damentally new concept where the network itself acts as process control systems (such as natural gas refineries and the controller. We proposed a basic scheme that can be paper pulp manufacturing plants) and building automation used to guarantee closed-loop system stability. We hence- systems.