Climate, mdt and machine learning the clima variables in the city

Climate, mdt and machine learning the clima variables in the city
 

Climate, mdt and machine learning the clima variables in the city

The privileged view offered by the MDT mobile radio networks measurements represents an added value also for the study of the environmental impacts and of the related effects on social behavior.
This research focuses on the correlation between rainfall weather events and variations both in the distribution of people in the territory and in the radio measurements performed by mobile terminals.
The research has been carried out analyzing the MDT 2019 measurements of the region Emilia Romagna in 2019 and of the region Veneto in 2018, by comparing climatically different days and moments. This document describes a way to measure both types of climate-induced variations, the behavioral variation of users, through the measurement of positional entropy, and the variation of physical parameters, through the measurement of the RSRP (4G LTE Reference Signal Received Power).
The document introduces also the opportunity, offered by the MDT innovation, that these types of analysis provide both for the study of the effects of climate changes in our cities and for the evolution of the mobile operators' business.

 

Introduction

The 3GPP MDT (Minimization of Drive Test) standard [1-4] offers the possibility to collect radio measurements that the mobile terminal performs both when it is using a connection (Connected Mode) and when it is in waiting state (Logged Mode), all linked to the GPS coordinates, if available, of the position in which the single measurement takes place.
The distribution across the territory of mobile phones and the high number of the anonymized MDT measurements allows to build a database with a sufficient statistical validity, which forms a relevant aspect for the study of the influence of atmospheric rainfall on the traffic distribution and on the signal strength level received from mobile terminals. Typically, the effect of rainfall on mobile communications is considered negligible, since the attenuation of rain it is very low for the usual frequencies between 800 MHz and 2.7 GHz, as reported in the ITU recommendations [5-8]. However, compared to point-to-point connections, the mobile telephony, especially within cities, is characterized by the presence of multiple paths of radio signals [9].
This affects the propagation of electromagnetic waves and, when the environment becomes wet, the characteristics of the dielectric surface of the soil, walls, roof buildings, foliage, cars and in general of all surfaces are altered. Other authors have made considerations on the effect of rain. For example, Rogers et al. [10], highlight how the dielectric properties of the trees vary from season to season also depending on the salinity of the water further to the atmospheric precipitations. Li et al. [11] show how the equivalent electrical permittivity of rain depends on both the working frequency and the rainfall rate.
S. Helhel et al. [12] report a study with measurements at 900 MHz and 1800 MHz showing how the surrounding wet environment formed by soil and trees reduces globally the value of the radio signal received by a few dB.
The novelty of the research presented in this article is the use of a large amount of measurements generated by mobile terminals. The mobile terminals are in fact instruments of measurement of the electromagnetic field and the MDT technology makes available a high number of geolocalized measurements, enabling the application of statistical analysis models to isolate the single hypothesized effects and identify the searched feedback, in real traffic scenarios.
The article shows that the weather conditions can already affect the carrier frequencies currently used in 2G, 3G and 4G mobile networks (800-1800-2600 MHz), but some of the effects highlighted in this study will be further observable on 5G networks, that is, when higher carrier frequencies (millimeter waves) will be used, frequencies more influenced by the presence of water. Therefore, the evolution towards 5G cannot ignore the development of SON (Self-Organizing Network) solutions, to make the infrastructures able to continuously measure, and therefore dynamically adapt to, the variations due to meteorological events, to maintain the quality of the radio network at the desired level.
The article is divided into four main sections. In the first section we analyze weather data of the Emilia Romagna (Bologna), in relation to the issue of positional entropy. In the second section we analyze weather data of the region Veneto (Padua) in relation to the topic [9] of the received signal level. The third section discusses the potential that MDT-based studies also project for analyzes on the life and evolution of the city. The MDT data covered in this article have been collected with the GeoSynthesis system developed by Nokia.
The fourth section reflects on how the technical evolution of the networks and the analysis tools available today end up influencing the role itself of the mobile operators.

Emilia Weather and Positional Entropy 

The month of May 2019 was a climatically exceptional period for Italy, with several episodes of bad weather all over the peninsula, often during the weekend. And it is precisely a weekend that is examined here, in particular Sunday 12th May 2019, with very intense weather phenomena that affected the region Emilia Romagna, as can be seen also from the frame of Figure 1 (processing of a radar image of the Civil Protection), in a meteorological context that has presented rapid succession of rainy moments and pauses all over this region.

 

Figure 1 - Frame taken from radar images published by the Civil Protection. Shortly before 18:00 (16:50 UTC + 2H, to obtain the Italian time) a strong thunderstorm core (the red spot, highlighted by the arrow) was formed roughly at the intersection across the provinces of Bologna, Ferrara and Ravenna. That strong stormy core then quickly moved towards the west, before losing intensity

In the same days a measurement campaign was also active MDT 4G of the cluster (set of 4G LTE Cells) of Bologna, for which it was possible to study the effects of that stormy core on the distribution of telephone traffic in the area. The importance of the study of these variations depends on the fact that the radioelectric coverage of a territory is only ideally considered uniform in all positions, but in the real world the radioelectric signal has specific emission points (the antennas) and is also subjected to attenuations everywhere, reflections and refractions that alter the signal level in the single points of a territory. The quality of the services therefore also varies depending on where the services are used, and, in turn, it can be influenced by the environment and by its variations.
It is understandable that in situations of hot and stable weather a public space can be occupied in a different way compared to a rainy occasion, especially if it is very intense, during which the occupation of individual places more sheltered from the weather can suddenly become a position used more than other ones.
As the possible distributions of positions along a territory are theoretically many, it becomes relevant to identify a mode of measurement of the distributions which is objective, rapidly obtainable from the radio data and still indicative of the different (overall) scenarios of the services’ usage.
It meets this summary need for complex phenomena the concept of entropy, a concept that emerged in physics to summarize the state of greater or lesser order of a system composed of many components, that in our case will not be the physical particles but the swarm of terminals that access and use mobile radio services.
We will therefore talk about positional entropy, a simple numerical representation of how long the users of the mobile radio services tend to be "chaotically dispersed" (in all possible positions) or, on the contrary, tend to be "neatly collected" around some specific locations.
Moving to an illustrative example of the Emilia area in question, a rectangle (with a diagonal of 5 km) of the territory surrounding the Meraville shopping center in Bologna (Viale Tito Carnacini) is analyzed. In fact, this type of area represents a place that is sometimes enough frequented (during opening hours), in order to be able to have a statistically significant measurements basis, but also a very accessible place with a few positional constraints, as it is a shopping center typically equipped with large parking lots and various access and runoff roads.
The rectangle of the territory above was therefore ideally divided into 48 identical tiles, each of which can be associated to the number of radio measurements generated therein, within a specific time interval (as e.g. 5 minutes).
Thus the distribution of radio measurements in the area can be transformed (see Figure 2) into a sequence of "pictures" that represent the various tiles, each more or less populated, in a specific time.

 

Figure 2 - Example of rasterization of measurements in the area of the Meraville Shopping Center (Bologna). Each box represents a time interval of 5 minutes, and the different colors the density of measurements in a tile (the darker color indicates a greater density, the lighter one indicates a lower density)

The subdivision (rasterization) of presences within the tiles can be finally converted into a vector (a list of numbers), where each position of the vector represents one of the various tiles, and in each tile (position of the vector) shows the count of radio measurements in that specific tile (at that time).
In this example, only the radio measurements that can be associated with low-speed movements (less than 5 km / h) have been counted, with the objective to isolate pedestrian-type movements (which are free to occupy any portion of territory) than road ones (more constrained by the roads and to the forced directions).
It is then possible to measure (using the tools of Information Theory) exactly how much information each vector (which represents the tiles within a single time interval) contains, for example, how many informational bits would be needed to describe its specific composition. Clearly the case of a single populated tile would be a situation that would intuitively require fewer bits, to be described, than in the case of many populated tiles, and all populated differently between them. In terms of entropy we would say that in the first case the system is quite orderly (low entropy), while in the second case we would affirm the exact opposite, that is that the system is quite disjointed (high entropy). In Figure 3 an example of distribution of MDT samples in the area is reported.

 

Figure 3 - Graphic representation of an example of geographical distribution of samples in the area under investigation

We can therefore compare two different days, i.e. Saturday 11th May, which did not present particular meteorological criticalities, with the following day, Sunday 12th May, which towards the late afternoon showed a rapid increase in rainfall.
What is expected is that, when moments of intense rainfall begin, the area hit by those phenomena tends to decrease its positional entropy, because the positions of use of the services will tend to decrease as many more users will tend to group themselves in some specific areas (the more sheltered ones) and few users will remain scattered in the positions that before (when there was no rain) were accessible without problems. Figure 4 confirms this hypothesis showing that when the period of intense rainfall begins (after 18:00) the trend of the measured entropy tends to be reduced, while in the same period of the previous day (less rainy) this phenomenon does not occur.

 

Figure 4 - Variations of Entropy (positional) in the vicinity of the Shopping Center Meraville (Bologna) on May 11th (less rainy weather, in the area) and May 12th (rainy afternoon phenomena, sharp from late afternoon as evidenced by the rain symbol). To allow the comparison between the two days in the graph, the mean value ("Mean") of entropy is reported, referring to the two compared days

Generalizing the single example described here we can guess the potential deriving from the use of MDT as a tool to monitor places, times and situations in urban contexts, leveraging the potential that the concept of positional entropy makes available to us.
Numerous studies [15-18] based on the analysis of a considerable number of space-time traces generated by mobile devices have outlined the possibility of identifying different levels of "order" (e.g. spatial order, social order, spontaneous self-organized order), in the apparently casual distribution of citizens in the urban area.

Veneto Weather and Radio Signal

If the study of meteorological phenomena through MDT brings out different behaviors for the use of mobile radio services, the same meteorological phenomena can also modify the characteristics of the mobile radio service. The signal strength received from an antenna is one of the main parameters that our phones constantly monitor and communicate to the cellular network infrastructure so that, from time to time, we can use the antenna (the Cell) that can best serve, in that moment, in that position, in that situation.
While in the case of positional entropy the cause (the weather) and the effect (the change in the use scenario) are directly correlated, in the case of the variations of the radio characteristics this happens in a particularly complex way, due to the multiple causes that contribute to the received signal level (e.g. the fact that people hold the phone in our hands or that we put it in a pocket or bag).
The mobile radio frequencies used in the 2G, 3G and 4G networks have physical characteristics that make them very robust from a time point of view (see Figure 5), however, the signal level received from mobile phones also depends on reflections (e.g. from soil), refractions (e.g. from buildings) and absorptions (e.g. walls or trees), as well as how direct signals and reflected signals are recombined (interference phenomenon).

 

Figure 5 - Relationship between rainfall intensity and the specific attenuation at different frequencies [13]. Please note that all the main frequency bands of networks in 2G, 3G and 4G technology, falling within 4 GHz, present low specific attenuations (dB / km) even for very intense rainfall intensity (mm/h)

The climate therefore, modifying the entire territory hit by humidity and rain, affects the mobile radio scenario by introducing a further level of complexity (see Figure 6).
To better understand this complexity, Figure 6 shows the result of the simulation of electromagnetic propagation through a 5 mm thick flat glass in the absence and in the presence of a thin layer of 0.3 mm thick overlapped water.
This scenario tries to represent a car glass or a window in a simplified way in the event of heavy rain on the glass.
The simulations were carried out considering the dielectric characteristics of the glass indicated by the ITU recommendations (ITU-R P.2040-1) [19] and water [20], considering a model of reflection and transmission of a plain wave through a multilayer structure as shown in Figure 7. The mathematical model of electromagnetic propagation through a structure of multilayer materials was implemented in Matlab for the calculation of the reflection coefficients (RC) in dB, of the transmission coefficient (TC) in dB and of the factor of loss (LF) in percentage [21].
From the graphs of Figure 6, it is observed that the presence of water on the surfaces favors the reflection of the incident electromagnetic field. Furthermore, it follows that the electromagnetic field transmitted through the glass + water structure is reduced by a few dB and even if not shown in the figure it increases its overall dissipation.
The parameters of the structure influencing these effects are the dielectric characteristics of the materials and thicknesses of each material.

 

Figure 6 - Simulation of the Coefficients of reflection (RC), transmission (TC), of a layer of glass 5 mm thick in the absence and in the presence of a superimposed layer of water of 0.3 mm thickness towards which the electromagnetic field affects for angles of incidence between 0 and 80 °. In the case of oblique incidence of the plane wave of the electromagnetic field on the surface of the materials, the cases of electric transversal (TE) and transverse magnetic (TM) are distinguished, depending on whether the tangent component of the electromagnetic field incident to the multilayer air interface is the electric field or the magnetic field. Only the graphs for the TE type incidence (Electric Transversal) are reported here for simplicity

This simple example suggests that in the real world the scenario of real electromagnetic propagation can be modified in the event of rain even for the radio frequencies used today. Therefore, that the presence of water on walls, soil, roads and trees modifies overall electromagnetic propagation in the environment producing in some cases greater reflection and in others a greater attenuation of electromagnetic fields.
In literature there are other studies that confirm this direction as already highlighted in the introduction.

 

Figura 7 - Electromagnetic propagation model through a multilayer sandwich of materials

To isolate the specific effects of the weather on the signal level available for the devices it is therefore necessary to operate in a very selective manner on the MDT measurements to be analyzed, both in geographical terms, appropriately cutting out the area to be placed from time to time, both in temporal terms, to be able to correlate the variations of the weather parameters with the variations of the mobile radio parameters measured by the devices. Only with this approach is it possible to isolate, study, and therefore measure, those specific effects that the climate induces in individual areas [note 1].
The magnificent piazza of Padua Prato della Valle, one of the largest squares in Europe, is a good example of an urban area suitable for studying the effects of rain on the mobile radio signal. First, the size and continuous attendance of the square provide a good statistical basis for the analyzes. The meteorological survey station is then in proximity of the square, therefore the hourly rainfall data collected by that station are directly usable, without the need to interpolate data from several weather stations (this is possible but would lower the reliability of the weather data in the area in analysis). Finally the position of the main cell serving that area is placed at an optimal distance from the square (about 1 km) for an urban study, a study in which we want to analyze both effects on the direct lighting component but also possible effects due to reflections and signal refractions.
The results obtained (see Figure 8) show an effective reduction, on average, of the power level (RSRP) received from the devices that were in Prato Della Valle during the rainy time slot (RSRP -87.5 dBm between 16.00 and 18.00 ), while in the absence of rainfall in the morning and early afternoon the RSRP averages were respectively -84.4 dBm and -82 dBm. Similar received power level, -84.5 dBm, was then recorded in the evening period, with the weather having meanwhile returned clear.
It was therefore possible to measure a case of influence of the meteorological context on the provision of mobile radio services that was also little influenced by behavioral reasons of the users, so as to be able to isolate this specific physical effect on the signal received from the other possible contributing factors.
The signal power level received by the devices in a specific area is one of the most significant parameters for the quality of the services offered in that area.

 

Figure 8 - Main positions (maps from www.OpenStreetMap.org elaborated with R Study [14]) from which Power MDT measurements of the LTE Reference Signal 4G were carried out in Prato Della Valle (Padua) on 7/6/2018, in four different time slots (9-12, 12-15, 15-18, 18-21), of which the afternoon session (155-18) coincided with a rainfall sequence (about 17 mm of rain in 3 h). The four frames are quite homogeneous between them, thus making it possible to compare the specific variations in power occurring within each time slot, reducing the possible distortions due to other effects (eg if there were extreme rainfall phenomena on 7/6) that area, this would have involved measures not homogeneously distributed, or even completely absent in the bundles

Measure cities with MDT

It is known that climate influences the way people live their own territory. And it is also now part of the common feeling that the climatic variations, which have always been able to generate changes to the livability of entire areas of the planet, could still have a profound effect on it due to global warming.
It is therefore always necessary to have a double view of the climatic phenomena we undergo, a physical view and a behavioral, human view, that is linked to the reflections on our lives of the physical changes of the ecosystem in which we are immersed.
To this end the technological innovation brought by MDT, being associated with very frequent measurements (radio), to guarantee full continuity of operation to the mobile radio system (moreover used in an increasingly intense way), and being MDT also combined with the main instrument of our times, that is, the mobile phone that we constantly keep with us, is increasingly showing itself capable of forging instruments capable of offering accurate views of the scenarios in which we live, offering us methods of investigation capable of objectively measuring the emerging trends, capable of comparing similar situations in different scenarios, capable of monitoring the effects of interventions aimed at improving a specific situation, capable of evaluating the effort made to improve, for example, the mobility of an area, and ultimately capable to privilege those actions that are really more effective in that specific context, urban or non-urban.
The possibility of inferring the behavioral models of cities (mobility, entropy, density of citizens in particular social gathering points, etc.) through aggregated MDT data and Machine Learning tools, also in relation to the different local weather conditions, represents for Mobile Operators an opportunity to develop new services to be positioned on the market in a distinctive manner compared to other ICT players.

 

Figure 9 - Left: geographical location of the Cell whose power is measured by telephones, the survey point (Botanical Garden) of the weather data of the period under investigation (courtesy of Radarmeteo) and the magnificent square of Padua, Prato della Valle, from which power measurements are taken. On the lower side: the average power loss (RSRP) measured by mobile phones in the square during the three hours of rain, compared to the rest of the period in question, with no precipitation

The role of the Operator in the future

It is easy to think of the Operator as a mere provider of connectivity, but the wealth of knowledge that houses in an operator now becomes only partly linked to this fundamental professionalism.
Disciplines once distant from the daily practice of a TLC Operator, like the Machine Learning, nowadays become more and more a daily practice, necessary to face a technological evolution that shows increasing complexity and increasingly sophisticated automation.
This trend joins new information  (computer science, mathematics, etc.) to the consolidated knowledge (propagation, connections, protocols, communications market, etc.), taking place all around the use of tools, such as smartphones, which now accompany in every moment our lives, is a trend able to free new opportunities for the Operator...
With the evolution of networks towards 5G, the volume of data generated by the network itself will grow exponentially.
This will be partly due to the innovative features of the 5G that will enable greater flexibility thanks to the peculiarities of active antennas and beamforming, or slicing. It will also be due to the increasing diversity of objects connected and managed by mobile radio networks that will support not only traditional smartphones, but a growing number of connected objects, both passive (sensors) and active (robotics, drones, vehicles).
The increasing complexity in planning and managing the topology of networks and the diversity of use cases will be possible only thanks to a greater intelligence that will be fed by data.
Therefore, the network infrastructure will have to be engineered to be AI-ready, and to produce and make available data that will enable its management. This same data, as can be seen from this article, can be analyzed and shared with other data, to produce analyzes and perspectives so far unimaginable.
The potential is very high and quite interesting as well, but it will require an innovative approach that is more oriented towards innovation and the creation of an ecosystem.
In fact, it will be necessary to imagine the networks as platforms that, via API and Exposure, enable the provision of network data, in respect of privacy and security.

Conclusions

The analysis conducted in this article, although susceptible to refinement and validation on a significant number of experimental cases, illustrates the way to move from large sets of MDT measures to synthetic numerical descriptions of the places and related service usage scenarios, to varying weather conditions.
Synthetic numerical descriptions are very useful for the application of data analysis techniques and for the development of predictive models, whose training requires coding the aspects that the algorithm must learn.
Automatically recognizing a usage scenario and the context in which it is immersed then lends itself to very different applications. These applications are evident if the learning process is grafted into the SON (Self Organizing Network) evolution in networks and continuous quality improvement.
But the field of application is not exclusively technical. It is easy to understand how this type of study can also benefit the analysis of the ways in which the city is experienced from moment to moment, and how the onset of this new value ends up influencing the very way in which the Operators they will reach the market, increasingly expanding the role of connectivity providers.

 

Nota

  1. The different effects, direct and indirect, that the climate induces on the power received by the devices would require to illustrate multiple types of case studies, within which they can from time to time become predominant depressive effects (absorption) on the signal level or, at On the contrary, it has a more expansive effect (reflections on the level of the signal received. The set of cases is not adequately treatable in a single article that also focuses on indirect effects, such as behavioral, on the signal level. In this article however, a single example of a significant weather situation is reported.
 

Bibliografia

  1. "La mobilità di breve e lungo raggio con le innovative misure radiomobili e l’Intelligenza Artificiale" - Notiziario Tecnico n.3-2018
  2. "Big Data georeferenziati MDT per servizi digitali nelle Smart Cities" - Notiziario Tecnico n.1-2018
  3. 3rd Generation Partnership Project; Technical Specification Group Radio Access Network; Universal Terrestrial Radio Access (UTRA) and Evolved Universal Terrestrial Radio Access (E-UTRA); Radio measurement collection for Minimization of Drive Tests (MDT)”; Overall description; Stage 2 3GPP TS 37.320 V14.0.0 (2017-03).
  4. Chiara Mizzi, Alessandro Fabbri, Sandro Rambaldi, Flavio Bertini, Nico Curti, Stefano Sinigardi, Rachele Luzi, Giulia Venturi, Micheli Davide, Giuliano Muratore, Aldo Vannelli, Armando Bazzani. “Unraveling pedestrian mobility on a road network using ICTs data during great tourist events”, EPJ Data Sci. (2018) 7: 44,  https://doi.org/10.1140/epjds/s13688-018-0168-2.
  5. Rec. ITU-R P.838 “Specific attenuation model for rain for use in prediction methods
  6. Rec. ITU-R P.530 “Propagation data and prediction methods required for the design of terrestrial-line-of-sight systems”      
  7. Rec. ITU-R P.618 “Propagation data and prediction methods required for the design of trans-horizon radio-relay systems
  8. Rec. ITU-R P.676 “Attenuation by atmospheric gases
  9. Andrea Scaloni, Pasquale Cirella, Mauro Sgheiz, Riccardo Diamanti, Davide Micheli, “Multipath and Doppler Characterization of an Electromagnetic Environment by Massive MDT Measurements From 3G and 4G Mobile Terminals”, Published in: IEEE Access ( Volume: 7 ), 21 January 2019, Page(s): 13024 – 13034, DOI: 10.1109/ACCESS.2019.2892864
  10. Rogers, N. C., A. Seville, J. Richter, D. Ndzi, N. Savage, R. F. S. Caldeirinha, A. K. Shukla, M. O. Al-Nuaimi, K. Craig, E. Vilar, and J. Austin, “A generic model of 1–60 GHz radio propagation through vegetation” Final Report, May 2002.
  11. Li, Y. and P. Yang, “The permittivity based on electromagnetic wave attenuation for rain medium and its applications” J. of Electromagnetic Waves and Appl., Vol. 20, No. 15, 2231–2238, 2006.
  12. S. Helhel, S¸. ¨ Ozen, and H. G¨oksu. “INVESTIGATION OF GSM SIGNAL VARIATION DRY AND WET EARTH EFFECTS”, Progress In Electromagnetics Research B, Vol. 1, 147–157, 2008.
  13. M.Cubaiu, L.Cerea, R.Nebuloni, G.L.Solazzi, M.Oldoni  “Caratterizzazione dei segnali di link a microonde per scopi di modellizzazione e rilevamento pioggia” 05/06/2018 Progetto MOPRAM.
  14. R is a project which is attempting to provide a modern piece of statistical software for the GNU suite of software. The current R is the result of a collaborative effort with contributions from all over the world. R was initially written by Robert Gentleman and Ross Ihaka—also known as "R & R" of the Statistics Department of the University of Auckland. R Core Team (2017). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/.
  15. Entropy and its Application to Urban Systems - Ben Purvis et al. – MDPI, 2019
  16. Spatiotemporal Analysis of Urban Mobility Using Aggregate Mobile Phone – Gariazzo et al. – MDPI, 2019
  17. An Entropy-Based Model to Infer Social Strength from Spatiotemporal Data – Pham et al. - Univ. of Southern California, 2013
  18. Entropy and order in urban street networks - A. Gudmundsson and N. Mohajeri - Nature Scientific Reports, vol. 3, p. Article number: 3324, 2013
  19. Recommendation P.2040-1 (07/2015). Effects of building materials and structures on radiowave propagation above about 100 MHz.
  20. Valori della costante dielettrica (relativa) e del fattore di perdita (relativo) dell’acqua a diverse frequenze e a due diverse temperature”, Proprietà dielettriche dei materiali, 2006.  https://studylibit.com/doc/5634917/proprieta--dielettriche-dei-materiali.
  21. Davide Micheli et al. - “Broadband Electromagnetic Absorbers Using Carbon Nanostructure-Based Composites,” Published in: IEEE Transactions on Microwave Theory and Techniques ( Volume: 59 , Issue: 10 , Oct. 2011 ), 29 July 2011, Page(s): 2633 – 2646, DOI: 10.1109/TMTT.2011.2160198.