In many real-time networked control and communication systems such as sensor networks, smart grids, transportation networks etc., sequential data transmission takes place. The transmitters are often battery-powered devices that transmit over a packet-switched network. Turning the radio on for transmission is significantly more costly compared to the size of data-packet. Hence the transmission is intermittent but when there is a transmission, the entire data-packet gets transmitted.
In this talk, we characterize the trade-offs between transmission cost and estimation accuracy arising in remote state estimation of an autoregressive scalar Markov process. The remote estimation system consists of a sensor and an estimator along with an erasure channel with known i.i.d packet drops. The sensor observes a discrete-time autoregressive Markov process driven by a symmetric and unimodal innovations process. At each time, the sensor either transmits the current state of the Markov process or does not transmit at all. The estimator estimates the Markov process based on the transmitted observations. In this context, we address the following two fundamental trade-offs, (i) when the communication is costly, find the minimum estimation error plus the cost for the total number of transmissions, and (ii) when there is a constraint on the number of transmissions, find the minimum estimation error. In the infinite horizon discounted cost and average cost setups, we identify the optimal transmission and estimation strategies and characterize the optimal performances.