Measurement based statistical channel modeling 231 15 0 20 30 20 40 30 Fig.8.7 The convergence of the spreads of the paths in the clusters versus the iteration index
Measurement based statistical channel modeling 231 0 10 20 30 40 50 0 5 10 15 0 10 20 30 40 50 0 5 10 15 0 10 20 30 40 50 0 10 20 30 0 10 20 30 40 50 0 10 20 30 Iteration index σθ [ ◦ ] σφ [ ◦ ] σν [Hz] στ [samples] Fig. 8.7 The convergence of the spreads of the paths in the clusters versus the iteration index
232 Measurement based statistical channel modeling 8.4 Data segment length selection In order to ext the statistics of channel characteristics needed.For exactin the information of macparmparatata saytheoftheatthe umberofbou collected in time.y wed出seuss the imp for determining the length of data segment in term of number of bursts. 8.4.1 Problem statement .The stationarity of the channel for analyzing the distribution of large-scale parameters and the small-scale .The necessity for keeping the results at a certain significance level Expe sugati ns show that this me nt groups consisting samples. .Channels from which the clusters are established have to be wide-sense-stationary e spread parameters In this sec ion ie how long the se ation of the data should be ir for data segment.The coh ence time can be computed here for one ossible value for the length of the data segment is three ursts in one
232 Measurement based statistical channel modeling 8.4 Data segment length selection In order to extract the statistics of channel characteristics, multiple observations of a wide-sense-stationary (WSS) channel are needed. For extracting the information of small-scale parameters, separate data segments each consisting of some consecutive bursts of received data are processed using the parametric estimation technique, e.g. the SAGE algorithm. The estimated paths are then grouped into clusters. In order to maintain the stationarity of the channel during each segment, it is necessary to determine the length of the data segment, i.e. the number of bursts in our case. This issue is solved in the third period. The measurement data considered in our analysis were collected in time-variant environments. The time-variability was caused by either the movement of the Tx or of the Rx, or the people walking in the measurement environment. It is necessary to maintain the stationarity of the channel in order to extract the distributions of model parameters which are applicable for describing the typical channel statistics in the same or similar environment. In this section, we discuss the impact of the number of bursts in one segment on the clusters’ behavior. We propose new approaches for determining the length of data segment in term of number of bursts. 8.4.1 Problem statement We always need to construct a control group of data for establishing the distributions. The length and the selection of the control group should be set for considering the stationarity of the channel from different perspectives: • The stationarity of the channel for analyzing the distribution of large-scale parameters and the small-scale parameters; • The necessity for keeping the results at a certain significance level. We have realized that when investigating the distribution of composite spread parameters, we should not use the group with a certain number of consecutive cycles. The distribution extracted from the control group with consecutive cycles does not apply to represent the distribution in the overall environment. Thus, a better choice is to select randomly the samples from all observations. Experimental investigations show that this method allows establishing the distributions of the parameters which are applicable for the treatment groups consisting of randomly selected samples. For the small-scale parameters, for example, the spread parameters of clusters, we need to consider at least three problems: • Channels from which the clusters are established have to be wide-sense-stationary. • The number of the samples used for establishing the clusters, should be beyond a certain a number, in order to keep a significance level, in order to specify the confidence interval for the values of the spread parameters. • The control group should consist of independent randomly selected multiple segments, in order to extract the distribution which is applicable for describing the characteristics of the overall channel. In this section, we try to solve the first question, i.e. how long the segmentation of the data should be in order to keep the channel wide-sense-stationary. We first use the coherence time to find out how many bursts should be considered for data segment. The coherence time can be computed as Tc = 0.423/fm, where fm denotes the maximum Doppler frequency fm = v/c · f0 with v denoting the relative speed, c the speed of light and f0 the carrier frequency. For the indoor measurements, the Tx was moving at the speed of 0.5m/s. Thus Tc in this case can be computed as 1 s. We know that the interval between two neighboring burst equals 311.6 ms, therefore, in the period of coherence time, the observations of the channel within the duration of 3 bursts can be considered as satisfying the WSS assumption. Therefore, one possible value for the length of the data segment is three. When applying 3 bursts as one data segment, we found that the values of the spread parameters estimated using the parametric method, are always smaller than those calculated using the non-parametric method. Furthermore, we found in some bursts, the number of paths is extremely, which results in low significance level for the estimated spread parameters per cluster. Due to the above two reasons, we try to find more bursts in one data segment, in order to have more accurate estimates. Thus, we need to investigate the approach for determining the number of bursts in one segment. Some indoor measurement data are used for this investigation. The measurement data collected in the scenario named “TxR9” is considered. The map of the environment is shown in Fig. 8.8