Periodic Fast Feedback Control over Low-power Wireless Multi-hop Networks
Closing feedback loops over wireless enables new technologies, such as autonomous driving or drone swarm coordination. Also for established fields, such as factory automation, wireless communication has benefits due to reduced installation costs and higher flexibility. Therefore, there have been many research efforts in the last years towards enabling feedback control over wireless. However, so far, none of these attempts achieved to control fast mechanical systems (that require update intervals of tens of milliseconds) over large distances. This is mainly due to some inherent challenges that come with the introduction of wireless technology:
- Information that is communicated over wireless networks is subject to delay.
- This delay is non-deterministic, the variation is known as jitter.
- Wireless communication is less reliable than cable-based solutions, i.e., data packets that are sent over wireless may be lost.
We tackle these challenges through a tight-co-design of control and communication: The design of the wireless embedded system aims to tame the network imperfections to the extent possible. While delays and message loss cannot be completely prevented, the resulting wireless embedded systems provides an extremely low jitter, which can be neglected from controls perspective, and high reliability, i.e., message loss is rare and the rare losses are independent and identically distributed. These remaining properties are then captured by a mathematical model that describes the physical and the wireless embedded system. Both properties are also reflected in the schematic provided below: Information that is sent at discrete time-step k is received one time-step later, i.e., at k+1. Message loss is represented by the variables phi and theta. Based on the model, a control strategy is designed that addresses transmission delays and message loss. Transmission delays and message loss are compensated through state predictions based on a mathematical model of the physical system. Further, the fact that losses are independent and identically distributed is exploited to come up with a stability proof for the overall system.
Most of classical control is based on periodic feedback. Communication of feedback signals is triggered by a clock, no matter whether these signals carry relevant information. In wireless cyber-physical systems, this wasteful use of communication resources is not desirable. Communication is costly in terms of energy. Since cyber-physical systems are often battery-powered, communication should only occur if necessary. Further, communication channels have a limited bandwidth. If all agents communicate at high, periodic rates, this can easily overload the channel and cause longer delays and more message loss.
To overcome this issue, we complement the integration at design time discussed above by an integration at design time using a self-triggered control approach. In self-triggered control, agents decide at the time they communicate about when to communicate the next next time and include this information in the data packet. The wireless system then takes this information and schedules communication for the next round. While we always assign highest priority to control tasks, i.e., every control system that signals communication needs will be assigned a slot, we schedule additional traffic (remote sensors, status information, see illustration below) in case not all control systems need to communicate. If after scheduling slots are still empty, these are left empty and radios are turned off to save energy.