OVERVIEW

The emergence of pervasive networking technologies such as ZigBee and other low power radios, have opened up opportunities to apply wireless communication to several new applications. Control systems, earlier limited to wired star or bus topologies, may now use ad hoc wireless topologies where it facilitates deployment and maintenance. In this research we focus on a class of systems where a control operation is exercised over a system comprising of spatially distributed sensors and actuators that communicate through a wireless ad-hoc network. Such systems are emerging everywhere. A smart workspace that changes the lighting by sensing the user actions; A smart building that continuously reduces structural vibrations due to external disturbances; An intelligent irrigation system that micro-controls the soil conditions for precision agriculture. These are a few examples of a growing class of control applications.

A challenging issue for control over ad-hoc networks is the latency between sensor inputs and actuator outputs. Latency concerns can be addressed by operating the system in a distributed manner. However, an additional challenge then is to ensure coherent operation throughout the system without explicit central coordination. We design distributed ad hoc network algorithms to address these challenges.

A barrier to enabling end-to-end control applications is the difficulty in writing complex distributed embedded software for these systems. We provide a middleware service and a set of reusable components that abstract the low level timing and coordination details and provide an easy programming model for developers to quicly deploy their control applications.

Illustration of a distributed control system:

APPROACH

These ad hoc control systems encompass a diverse set of systems and applications. Hence, the design challenges and their solutions vary with the properties of the systems under consideration. We first attempt to classify the system based the spatial and temporal properties of the controller and the system components.

Spatial Temporal Properties
Actuator Influence Region
The actuator changes the state of the physical world or the plant. The spatial region of the plant where the state change occurs due to a particular actuator is defined as its influence region. The influence region is an important property in the design of the system because all the actuators whose influence regions overlap need to coordinate with one another in order to avoid any conflicts. The relationship between the actuator influence region and the communication range of the actuators has an impact on the coordination mechanism between them.

Spatial Extent of Controlled System
The spatial extent of the controlled system is the physical span of the plant under control. The relationship between the spatial extent and the communication range of the components has an impact on the choice of control approach employed.

Periodic Feedback Control
Periodic feedback control is a temporal property of the controller. A control system is continuously subjected to noise from various sources. The state of the physical world or the plant is constantly perturbed due to external disturbances. The operations of the sensors and the actuators also vary due to the presence of a zero mean thermal noise. If the system objective is to maintain the state of the physical world at a constant level, then there is a need for a periodic feedback control system to counter the effects of noise and make the system more robust. The frequency of the periodic loop is determined by the characteristics of the noise and the desired performance of the controller. The feedback loop in these systems is implemented by an ad-hoc network and therefore its frequency imposes timing constraints on data delivery. The emergent traffic patterns in the network are periodic.

Event-triggered Control
The event-triggered system reacts to discrete events in the physical world. The performance of the system is characterized by the response time of the system to the events. The event transport in these systems from the sensors to the controllers is done by the ad-hoc network. The desired response time imposes latency constraints on the delivery of events. Reliability of the event transport is another important metric.

The specific design challenges that emerge in all the classes of ad hoc control systems are:

• Meeting strict latency constraints imposed by the control system
• Techniques for coordinating multiple actuators in the control system
• Efficient techniques for programming these systems

We propose the implementation of control-centric middleware service and a programming framework to address these design challenges.

Control-centric Middleware Components
Traffic Management
Ad hoc control systems comprise of multiple kinds of concurrent traffic flows. A system comprising of event triggered and periodic controller will have periodic traffic flows interspersed with the sporadic events. The requirements on the latency and reliability of the different traffic flows depend upon the nature of the control system. The traffic management run-time component is derived from the cross layer optimization of the Medium Access, Routing and Packet Scheduling layers of the network stack in sensor nodes.

Actuator Coordination
As mentioned already, the region of actuator influence of multiple actuators can overlap. The objective of the actuator coordination middleware is to ensure conflict free operation of the actuators in the system. The conflicting actuators should not be permitted to actuate simultaneously in time. Therefore, we propose to implement co-ordination protocols that are somewhat similar to the medium access protocols used for the radios. The first step in this process is a protocol for the actuators to determine their actuation model and derive conflict relationships with one another.

Programming Framework
Programming of these systems is a challenging task because the various components of the system execute heterogeneous code. Nodes in the system perform different functionality depending upon whether they are controllers, sensors or actuators. The strict latency constraints introduce further constraints to programming. In such a scenario, it is difficult to program the system by programming every individual node in the system. There is a need for raising the level of abstraction and enable programming of the system as an aggregate as opposed to programming individual nodes. We propose to extend NesC as a wiring language between modules at the network level. Currently, there are no language features in NesC that specify the communication between two functional modules present in different nodes. Also, there is no API in NesC for specifying the timing of the different traffic flows in the system. The distributed application is specified as a component graph at the network level with the edges joining the components representing communication between them. The timing constraints are specified by annotating the edges with the constraints. The target operating system for the programming model is SOS.

FUTURE DIRECTIONS

Future work is going to focus on implementing the traffic management run-time components in SOS for a periodic controller with a large spatial extent. Specifically, we are going to implement a TDMA protocol and couple it with the Tiny Diffusion routing protocol.

SYSTEMS / EXPERIMENTS

We verified the impact of traffic management component by simulating a proportional periodic feedback controller over an ad-hoc network. A proportional controller produces a control output that is directly proportional to the error input. The error input is the difference between the current system output and the desired set point. Figure (1) shows the output of the perfect proportional controller that is sampled at a period of 5 seconds. The topology of the  system is shown in Figure (2). The network stack comprised of the Berkeley B-MAC protocol (CSMA + random backoff) and the Tiny Diffusion routing protocol. Figure (3) shows the output of the system in the absence of traffic management. The output of the system is superimposed on the delay values of the corresponding actuator command sample as it traveled from the controller to the actuator. As can be seen from the figure, the average delay is very high and it has a high jitter. Also, the packet losses lead to significant degradation in the system output. Figure (4) shows the output of the system in the presence of traffic management. The medium access protocol was changed from B-MAC to TDMA in order to prevent losses due to collisions. The packet scheduler in the Tiny Diffusion routing protocol was modified to give a higher priority to the data packets over interest packets. This was done to prevent loses due to congestion in the network. As can be seen from the figure, the output of the system closely tracks the output of the perfect proportional controller.

PEOPLE

FACULTY

Prof. Mani B Srivastava

GRADUATE STUDENTS

Ram Kumar