Study of lower sampling intervals on rainfall queue characteristics over Radio Links in South Africa.

Abstract

Rainfall attenuation in tropical and subtropical regions of the world has continued to attract great interest; as there is a urgent emphasis on proper spectrum management and sharing, particularly at microwave and millimeter bands above 10 GHz. To this end, there have been arguments pertaining to the need to improve the ‘sensing’ of rainfall events to enhance the opportunities provided by adaptive rain fade mitigation schemes, while conserving base station power requirements during rainy events. To implement this approach, an extensive understanding of rainfall time series via the available statistical tools is often required to properly harness the characteristics of rainfall behavior. To this end, a study was undertaken to examine the behavior of rainfall and its impact on radio links at 1-minute sampling time by using the Queueing Theory Technique (QTT). Interesting results were obtained in the process of the study, except that the effect of the sampling time on rainfall queues remained unknown. Therefore, this thesis presents the investigation of the sampling time effects on rainfall queues over radio links in Durban, South Africa. Rainfall measurements were collected at 30-second sampling time using the RD-80 Joss–Waldvogel (JW) distrometer in Durban (29o52’S, 30o58’E), the same location where the 1-minute data was previously collected. As before, the rainfall data is classified into four rainfall regimes, namely drizzle, widespread, shower and thunderstorm. The queue parameters required for rainfall traffic analysis such as inter-arrival time and service-time distribution are empirically determined to be Erlang-k distributed, whereas the overlap time is exponentially distributed. It is thus established that the queue discipline for rain spikes over radio waves is a non-Markovian process (Ek/Ek/s/∞/FCFS). Comparison between the 30-second rainfall queues results and previous results of 1-minute sampling time, shows that more rainfall spikes are revealed at 30-second sampling time. Furthermore, it is determined that there is a strong polynomial relationship between the 30-second and 1-minute sampling time data – hence some of the 1-minute data may be converted into 30-second data by using the polynomial function, with the appropriate polynomial coefficients according to rainfall queue parameters in each regime. The converted data is amalgamated with the actual 30-second data for the investigation of the rainfall long-term behavior. It is found that the rainfall long-term behavior resembles the behavior of the short-term data - hence implying that the rainfall process at 30-second sampling time in Durban has the attributes of a self-similar process. From rain attenuation investigation, it is determined that since more rain spikes are evident in the 30-second data, the former has higher rain attenuation exceedance values (R0.01) compared to the 1-minute data.