|Abstract:||By converging the cyber world with the physical world, Cyber-Physical Systems (CPS) is expected to create a great impact on computer science and the society. This thesis proposes several real-time building blocks for CPS infrastructures.
As for real-time wireless LAN, I compare various wireless communication paradigms and show that by deploying largest processing gain possible (i.e. lowest bit rate), DSSS-CDMA cell phone paradigm (i.e. each control loop occupies a CDMA channel between the remote station and the base station) achieves much higher reliability/robustness than main stream alternatives, such as IEEE 802.11, IEEE 802.15.4, and convolutional coding. For example, if a control loop sends one packet per second with a packet size of 152 bits, a DSSS-CDMA communication link using largest processing gain possible takes about 20dB more power to jam compared to an IEEE 802.11b link using packet retransmission. If the control loop sends ten packets per second, a DSSS-CDMA communication link is about 10dB more robust than an IEEE 802.11b link.
As for real-time wired network, I propose a real-time crossbar switch design, which is not only compatible to, but also simpler than the main stream iSLIP switch design. Specifically, since real-time flows are predetermined and run consistently for long time, an iSLIP switch can be easily revised into a real-time switch: we only need to grant according to a deterministic schedule; and eliminate the request and accept steps. The schedule can be represented as an N by M matrix S, where N is the number of input(output) ports of the crossbar switch, and M cell-time is the period of the schedule. Let S(i,k) denote the element at the ith row and kth column of S. Then at the kth cell-time of each period, the ith input port forwards a cell to the S(i,k)th output port. Basically, this means the switch serves each real-time flow with clock-driven time-slicing. A flow of period P cell-time and message size of C cells are over-provisioned with a server that forwards ceil(C/floor(P/M)) cells every M cell-time. By mapping each real-time flow to a server, we derive an N by N demand matrix D. The element at the ith row and jth column of D, denoted as D(i,j), means every M cell-time, input port i has to forward D(i,j) cells to output j. We prove that as long as each input (output) is required to forward (receive) no more than M cells per M cell-time period, demand matrix D can always be scheduled within polynomial time (O(power(N,4))).
I also designed a hard real-time, fast, and lightweight acoustic event localization protocol, the Lightning Protocol, for wireless sensor networks. Basically, wireless sensors are deployed in a square grid pattern. Every sensor is colored i (i = 1, 2, 3, 4), so that for any point on the plane, the enclosing four sensors have distinct colors. A sensor is either in RF listening or broadcasting mode (never both). Whenever a free (i.e. its state is neither winner nor loser) sensor hears the acoustic event, it broadcasts a noise on the wireless carrier for i time units. After that, if it does not hear any other noise on the wireless carrier, it wins the election; otherwise it loses the election. Whenever a free sensor hears a noise on the wireless medium, it loses the election. I prove this protocol elects the closest sensor with only O(1) RF broadcast within O(1) time. Energy Efficient Lightning Protocol is also designed, which only turns on RF module during localization period. Experiments using U. C. Berkeley Mica Motes show the feasibility of the protocol in lab environments.