Delay-reliability tradeoffs in wireless networked control systems

Abstract

Networked control systems (NCS) require the data to be communicated timely and reliably. However, random transmission errors incurred by wireless channels make it difficult to achieve these qualities simultaneously. Therefore, a tradeoff between reliability and latency exists in NCSs with wireless channels. Previous work on NCSs usually assumed a fixed transmission delay, which implied that the failed packet will be discarded without retransmission, and thus reliability is reduced. When the channel errors are severe, the NCS cannot afford the resulting packet loss. In this paper, a delay-bounded (DB) packet-dropping strategy is associated with automatic repeat request (ARQ) at the link layer such that the failed packets will be retransmitted unless they are out-of-date. On the one hand, the packet delay is controlled to be below the predetermined delay bound. On the other hand, reliability will be improved because failed packets are given more retransmission opportunities. This paper investigates the tradeoff between packet delay and packet loss rate with DB strategies. Due to the multihop topology of the NCS, medium access control (MAC) schemes are needed to schedule the transmission of multiple nodes. Moreover, spatial reuse should be taken into account to improve the network throughput. In this paper, two MAC schemes are considered, m-phase time division multiple access (TDMA) and slotted ALOHA. They are compared for different sampling rates and delay bounds. TDMA outperforms ALOHA in terms of both end-to-end (e2e) delay and loss rate when the channel reception probability is above 0.5 and/or traffic is heavy. However, ALOHA shows a self-regulating ability in that the node effective transmit probability depends only on the sampling rate and channel reception probability, but is essentially independent of the ALOHA-dependent parameters. Then, for light traffic, a simple ALOHA with transmit probability 1 is preferred over TDMA in NCSs. The derived relationship between the sampling rate, the e2e delay (or delay bound), and the packet loss rate is accurate and realistic and can be used in NCSs for more accurate performance analyses.