CHAPTER 8
• We designed and evaluated on-the-fly adaptive traffic mechanism, which achieves a highly superior performance compared to the WirelessHART system. The node relies on immediate updates whenever there is traffic change. These updates also make the solution robust, as traffic changes are immediately reflected in the forwarder node. Because updates are entirely piggybacked inside the packet (using single bit only), the solution is nearly overhead-free and thus energy-efficient.
• Moreover, we have looked at two different strategies for exploiting the sleep modes of the CC2420 transceiver and have highlighted that signif-icant savings can be achieved with only moderate increases in run-time complexity. We then proposed a practical and effective energy manage-ment scheme which exploits the multiple sleep states of a transceiver and utilizes them in efficient manner. We also adapted this approach for TDMA-based protocols. This approach is independent of the under-lying link scheduling algorithm, but a node uses it’s given schedule to determine the right sleep states.
• In this thesis we also evaluated and benchmarked our proposed approach with the state-of-the-art solution WielessHART. WirelessHART com-bines frequency hopping with a TDMA scheme utilizing a centralized a-priori slot allocation mechanism. The comparison provides a careful and balanced answer on the advantages and drawbacks of the Wire-lessHART and autonomous solutions for supporting periodic traffic. We evaluated both approaches in a range of scenarios using trace-based simu-lations, and we have shown that our approach, indeed, reaps the benefits of frequency hopping and also improves the adaptability to varying traf-fic, energy consumption and end delay over centralized system such as WirelessHART.
• Further, we studied sensitivity analysis which showed how the energy consumption depends on certain parameters such as the power consump-tion of the transceiver in different modes of operaconsump-tion (Tx, Rx, Listen-ing, SleepListen-ing, and turnover), as long as the length of learning phase and length of wakeup window factors. In this analysis we used the response surface methodology and showed the most influenced parameters to the energy consumption that one can optimize further.
• Based on our analysis of the WirelessHART we suggested several ideas to improve the WirelessHART standard performance in terms of energy and delay. One idea is to adapt the frequency of management traffic
by starting with a high frequency, and as soon as the network becomes somewhat stable, the management rate can be reduced. This can signif-icantly reduce energy saving at the expense of longer joining times and slower network update times resulting from topology changes. Another approach would be to use piggybacking more extensively, for example to use periodic data packets also for management purposes by piggybacking additional information (e.g. keep-alive and health reports).
• Our approach is extremely light in terms of signaling, as only ACK pack-ets need to carry few bits of information. We believe that our approach is an attractive alternative to WirelessHART and similar systems in lightly loaded networks with periodic traffic.
8.1 Future Works
In this section, we identified some future directions which we believe are worthwhile for farther future works for both autonomous framework and Wire-lessHART system.
8.1.1 Autonomous framework
The autonomous approach developed so far has some potential for optimization by considering the following ideas:
• One can enable traffic shaping mechanism for forwarder nodes to prevent large deviation in the periodicity and thus reduce the jitter.
• Light forms of signaling could be used, in which for example one node can piggyback local estimates onto data packets, helping an upstream node with estimating period and jitter.
• One can enhance the end-to-end delay by reducing the back-off con-tention interval dynamically. This is more useful in the case of moderate and low traffic rate.
• It would be also worthwhile to investigate different channel hopping pat-terns such as adaptive channel hopping in which a dynamic estimate of the channels is maintained with dynamic blacklisting.
8.1.2 WirelessHART
In the following we identified some future directions we believe are promising for WirelessHART system:
• Design and implement TDMA scheduling algorithms which explicitly take the presence of a local sleep scheduling algorithm and multiple sleep states into account by constructing schedules in which the slots of indi-vidual nodes have larger separations in time, so as to allow them to enter deeper sleep modes.
• Develop supporting features within the slot assignment algorithm to ploit the re-transmission slots and reuse them in efficient manner. For ex-ample the retransmission slots could be used to transmit buffered packets instead of wasting the retransmission slots in case of successful transmis-sion. Also assignment algorithm could improve both energy consumption and packet delay by assigning same slots for multiple nodes (shared slots) for re-transmission.
• Another important improvement could be the design of hybrid slot as-signment in which both centralized and distributed algorithms are to be selected based on the traffic load. As an example if the load is low then one can use distributed algorithm, otherwise use centralized algorithm.
This could be triggered based on the traffic information on the network manager or even could be done by examining the buffer size of the nodes.
Another idea would be to separate the two approaches; to use central-ized algorithm for control packet only and distributed algorithm for data packet exchange.
• We also believe that a further investigation of different channel hopping patterns such as adaptive channel hopping in which a dynamic estimate of the channels is maintained is needed for such centralized solutions.
• Another interesting enhancement would be to adapt our proposed ap-proach (on-the-fly adaptive traffic mechanism) to the WirelessHART standard. This would definitely improve adaptability of the WirelssHART and hence, improve the energy consumption and end-to end delay. Only one single bit should be reserved in the data or control packet. The network manager then could react faster to the new traffic demands and compute a new scheduler much faster than waiting for long time (about 15 minutes in light loaded network and much more in heavy loaded net-work).