چكيده به لاتين
In recent years, many uses have been made on the Internet of Things (IOT), one of the most promising applications associated with industrial areas. The most prominent standard in the field of technology for low-power wireless nodes is the IEEE 802.15.4 standard. However, this standard faces limitations such as low reliability and unlimited latency. IEEE 802.15.4e has recently been updated to overcome these limitations. Part of this emerging standard defined in the Media Access Control Layer Protocol (MAC). The TSCH protocol meets the requirements for very low power consumption, high reliability and definite delay. In the TSCH IEEE 802.15.4e Media Access Control Protocol, sending packets between network nodes is coordinated through a scheduler. This standard specifies how each node should execute scheduling, but the way of calculating, constructing, and maintaining TSCH scheduling is not raised. Previous approaches to determining the scheduler policy to ensure the optimal throughput are reliant on the status of the channel information (CSI) in an Instantaneous and accurate manner. The assumption does not materialize in many applied scenarios and, as a result, makes existing solutions unrealistic for csi entific applications.
In this thesis, for assigning slot frames, first, assuming that the channel statistical knowledge (and not the exact moment values) is available, an optimal scheduling method (in terms of the mean value of the total network) is presented. In fact, under the assumption of statistical CSI, the average number of packets sent by each link in each time slot can be achieved in advance then the scheduling was optimized based on the average link rate. But in more realistic circumstances, it should be kept in mind that the CSI changes with time and it is not possible to modeling the exact probabilistic structure of that. The second proposed strategy does not rely on CSI statistical knowledge, and instead, using a machine learning approach to optimize the scheduling of sending packets without model.
Compared with previous work in this area, the proposed method, due to the operation of the machine learning technique, provides a higher degree of integrity in terms of its capability to operate in a wider range of operational environments. The final simulation results also show that the average throughput of the proposed second strategy (Despite the free of model) can be converted to the efficiency of the proposed1 (model-based) during the run-time.
