This paper discusses the stability of TCP. More precisely, the stationary behavior of the congestion window process is analyzed based on the measured real Internet traffic (bulk data transfer, FTP of large files). Appropriate algorithm approximating the actual congestion window is presented, and statistical methods to analyze the data are discussed in detail: change-point detection, distribution testing and parameter estimation. It has been shown that the distribution of the congestion window process during stable periods has a bell-like shape and can be approximated by a normal distribution. A number of interesting observations during the study are depicted and explained. Finally, possibilities for building reliable and robust models for the TCP protocol in the future, are discussed.

The error floor of minimum shift keying (MSK) signal transmission over a mobile radio channel with small time dispersion is analyzed. In contrast to previous investigations, no restrictive assumptions about the impulse response of the channel and/or sampling time are made, and closed-form analytical expressions for the error floor and optimum sampling time are derived. It is found that, generally, the error probability depends on the variance of relative delay (which conforms to the conventional RMS delay spread) as well as on the product of mean relative delays of echoes advancing and succeeding the chosen sampling time, weighted by corresponding mean powers. This implies that (elsewhere widely accepted) simple determination of the error floor by RMS delay spread only can now be seen as a special case when sampling is at mean delay and distribution of the delay profile's power around it is balanced. Furthermore, it is shown that the novel formula reflects the impact of delay profile shape on the error performance (that has been usually neglected, although not always justifiably), and that in general case, the optimum sampling instant is not at the mean delay. The theoretical results are verified by comparison with Monte Carlo (MC) simulations' results; achieved agreement lies within the accuracy of the simulations.

Taking advantage of GSM's elaborated mobility managementand its intemational roaming capability, the IntemationalUnion of European Railroad Operators (UJC) has decidedto build a supra-national rail radio network based on enhancedGSM technology.In this pape1; some implemmtation aspects specific to railroadradio environment resulting from the effects of vehiclespeed - in particular Doppler shift - are discussed, and an'off-line' algorithm for the traffic handover is proposed, whichoffers bel/er radio link reliability and less critical timing constraintsthan conventional handover schemes. More reliablehandover is of vital significance in the case of high-speed trainssince the radio link is supposed to cany the automatic controlsignals (ATC), too.

In radio network planning, the goal is to specify optimalbase station locations, service areas, antenna patterns and handoverstrategies for a given mobile radio system. The system performanceis fixed and the radio channel is subject to optimisation.Network planning requires channel characteristics thatprovide information about the expected sendee quality, especiallythe outage probability. The channel description must onlyperform as a qualitative measure of an actual receiving area.Power delay profiles are a convenient and very common descriptionof channel time dispersion, which can be easily physicallyunderstood as footprints of individual reflected or scatteredpaths, and provide a capability to a network planning engineerto discover areas of heavy time dispersion and importantscattering regions on the terrain. This in turn enables better assessmentof base station sites, antenna pattern selection, sen,iceregion (cell) shaping (handover criteria) and solving networkproblems.Time dispersion is mostly found to be the cause of poor coverageby excluding other possible causes. In cases where fieldstrength coverage does not overlap and base station sites cannotbe moved, shaping antenna directivity will be the only way toeliminate excessive time dispersion.

In this paper, the performance of MSK signal transmission over a mobile radio channel with small time dispersion is analyzed. In contrast to existing investigations, we made no restrictive assumptions about the impulse response of the channel or the sampling time and derive analytical expressions for the error floor and the optimum sampling instant. We found that the error probability depends not only on the delay spread, but also on the squared mean delays of echoes that are advanced or delayed with respect to the sampling point and on the power balance between them. Furthermore, we show that in the general case, the optimum sampling instant is not identical to the mean delay. Comparison of our results with Monte Carlo simulations show very good agreement.

The satisfactory estimation of speech autocorrelation by means of generalised zero-crossings indicates that they can be used for efficient feature extraction in speech recognition. In addition, high consistency between the Itakura-Saito distances, calculated before and after clipping, allowed for only a mode-rate degradation of the related recognition performance, which was compensated by including the excitation distortion into the distance measure.