After depicting the deployed testbed, this section discusses some of the main results that have been obtained by means of this measurement framework, in terms of both E-field characterization and exposure index computation within the Santander downtown area. First,
Section 4.1 describes the methodology that has been used to compute the
from both the E-field measurements and life-segmentation statistics. Then, a thorough evaluation of the E-field, from information gathered during more than one year within the target area, will be discussed in
Section 4.2. Special attention is paid to the suitability of these measurements for the
calculation. Finally, the downlink component of the EI will be also evaluated.
4.1. EI Computation Methodology
As was mentioned earlier, the
definition considers not only electro-magnetic field values, but also a range of data elements that are connected to life segmentation, use patterns of communications devices and SAR values. This information has been calculated for a typical reference scenario, which was defined within the LEXNET project and whose main characteristics are shown in
Table 3. All of them are based on measurements and simulation-based studies that were carried on during the project lifetime.
As can be seen, the population is assumed to belong to four different segments, each of them with four traffic profiles: none, light, moderate and heavy. It is worth highlighting that users within the none-traffic profile are also considered, since they would also be impacted by the E-field exposure (passive receivers). The traffic profiles are different for every population segment and radio access technology. The reader may refer to [
5] for a succinct discussion of these values and the assumptions that were made to obtain them.
According to the
definition that was described in
Section 2, the corresponding contribution from each configuration can be defined as a scaled value of either the mean transmission power (
) or the mean received power density (
) for uplink and downlink exposure, respectively. In this sense, we can simplify the
as follows:
where
are constant exposure coefficients for a specific configuration and are calculated by scaling a reference Whole Body averaged SAR (WBSAR) to account for ICT usage and technology. The value of the WBSAR is typically used as a basic constraint for human exposure to radio frequency [
21]. This coefficient-based simplification was proposed in [
5] to facilitate the
computation. The reference values of the WBSAR were estimated by means of studies conducted in different European countries, assuming a transmission power and a power density of 1W and 1W/m
2, for uplink and downlink, respectively. Besides, WBSAR reference values were calculated for different services (e.g., voice, data, etc.) and both in outdoor and indoor conditions. Finally, the coefficients to calculate the
were obtained by modulating the WBSAR reference values with ICT usage statistical information.
For instance, the coefficient of the
uplink contribution from indoor voice calls during morning periods using 3G networks, and assuming a reference time interval of one day, would be thus defined as:
where the factors
represent reference values for this particular configuration. In addition, the use of this formulation to calculate the
from the values provided by the low-complexity dosimeter platform requires some further simplifications to be made. As discussed before, the measurements gathered by the dosimeters do not provide meaningful values for the uplink exposure, and we should just thus consider downlink
components. Besides, since the low complexity dosimeter is not able to differentiate between technologies within the same frequency band, the average whole band WBSAR has been estimated, by using typical usage values (i.e., how the band is used by the various cellular technologies). Finally, we will focus on the outdoors EMF exposure, since we did not deploy dosimeters in indoor environments. All in all,
Table 4 shows the values of these parameters, for the different technologies and population segments in the considered urban scenario [
5].
Afterwards, and before the calculation can be carried out, the raw information gathered by the low-complexity dosimeters should be processed, to take into account uncertainty sources and particular scenario characteristics.
As was already mentioned, the low complexity dosimeter has a mono-axial probe, and correction factors need therefore to be applied to yield an accuracy level comparable to that from an isotropic device. Furthermore, the low-complexity dosimeters take wide-band measurements, not being able to distinguish the different technologies that are deployed by the existing operators within the same frequency band. Last, but not least, the installation of devices might also cause some inaccuracy, due to the existing urban furniture, which has to be considered during the data process.
Figure 9 illustrates the whole data processing methodology. Once the E-field values are obtained, they are scaled with the mono-axial to isotropic and installation correction factors, whose values can be found in
Table 5.
These values were obtained by carrying out on-field measurements, in Santander city, with more capable devices (advanced spectrum analyzers). This campaign embraced static measurements at different locations, so as to compare them with the E-field values gathered by the low-complexity devices. We covered the whole downtown of the city (area where the platform is deployed) to validate the overall methodology and the correct values provided by the dosimeters. The reader may refer to [
22] for a more thorough description of the measurement campaign.
Afterwards, the contribution of the different technologies within each of the frequency bands is estimated, based on ICT statistics. We have to recall here that the usage and population factors that are applied in the computation is not based on frequency bands, but on the technologies. Once all of the correction operations are applied, the E-field values are transformed into power density, and the is finally calculated by applying the factors shown above.
4.2. E-Field Characterization and Exposure Index
This section discusses some of the results that have been obtained by using the measurement platform and methodology that have been previously described. As has been already said, their main target is to carry out the EMF exposure characterization within large areas during long time periods. We will first study the evolution of E-field measurements, discussing how this parameter evolves. Afterwards, we compute the downlink component of the exposure. In order to do so, we have used all of the data gathered for a complete year (2016), focusing on the outdoors E-field and leaving aside the ISM 2.4-GHz band, since the obtained values were much lower than for the cellular technology bands.
For the first group of results, we select a single dosimeter, and we study the temporal variation of the E-field values. This analysis aims to validate the 24-h usage, as the reference time period
T for the calculation of the
.
Figure 10 illustrates such temporal E-field evolution. We select a particular dosimeter and three days: Wednesday, Friday and Sunday. For a single day, we have datasets of 288 samples, since the E-Field is measured every 5 min. Then, we take the 52 datasets during the year (one per week), and we plot both the average value, as well as the 95% confidence interval. Each of the nine figures is thus obtained by processing ≈15,000 samples. We can see that there is a non-negligible variance of the results, although the trend is quite stable. The E-field values for the 1800-MHz band are the lowest ones, which corresponds to the fact that this is the least used band. We can also see that the 3G band (2100 MHz) shows a more relevant variation on the E-field, which is due to the stronger impact of traffic patterns over CDMA-based technologies. Note that this is the unique band without GSM technology. Another interesting conclusion can be derived by comparing the evolution of different days. In this sense, the E-field on Saturday nights (left part of the lower figures) is clearly higher than in the other days. This reflects the fact that the downtown has several restaurants and pubs and people usually go out on Saturday nights.
We have afterwards studied the statistical behavior of the E-field during different day periods. In particular, we have defined five time intervals, according to different activity levels and habits in Spain:
Standard Working Time (SWT): comprises the time periods in the ranges 9:00–13:00 (a.m.) and 16:00–19:00 (p.m.)
Rest Period (RP): is assumed to be in the ranges 14:00–16:00 (a.m.) and 19:00–22:00 (p.m.)
Night: comprises the time interval when communications are less likely, comprising from 23:00 (p.m.) to 7:00 (a.m.)
Based on such classification,
Figure 11 depicts the statistical distribution of the E-field during different day periods for one particular dosimeter, using a box-plot diagram. For each period of the day and frequency band, the boxes indicate the 25 (
), 50 (
or median) and 75 (
) percentiles. Furthermore, based on
and
, we establish the interquartile range (
), as
, so that the bottom/top lines in the graphs are defined as
and
, respectively.
As was also observed before, these results yield a rather high variability for all frequency bands. Furthermore, there is not a clear impact of the day time over the E-field values, but for a slight increase during the evening rest period. Besides, the results also show a more noticeable E-field reduction during night hours.
Once we have seen how the E-field varies during a day,
Figure 12 shows the distribution of values taken by a single dosimeter, for the seven days of a week. The statistical distribution is represented by using box-plot diagrams, whose values have the same meaning as in
Figure 11.
As was the case of the previous results, we are considering the values taken during a whole year, 2016. We can again see that there is not a clear difference between the E-field measured for each day. The graphs yield a non-negligible increase for the 900-MHz band during the weekend. However, for the 1800-MHz band, the E-field on Sundays gets lower, being rather constant for the rest of the week. This is due to the fact that during most of the year, this band has been exploited by the operators to carry 4G traffic (refarming), as was previously discussed. We cannot see any clear pattern for the 2100-MHz band, which is mostly devoted to 3G technology.
Hence, we can conclude that the measurement variability at a particular point is rather low, validating our initial assumption of using a 24-h interval as the reference period for calculation. As can been seen, we ensure not overlooking EMF peaks and valleys during the day, for the different frequency bands. It is worth pointing out that the same behavior was indeed observed in the remaining dosimeters, and the conclusion could be therefore generalized to the whole set of points of the deployment.
Once we have seen how the E-field varies at a single measurement point,
Figure 13 shows the correlation between the values provided by close-by dosimeters. We sort them according to their longitude, from the most western dosimeter (the one which is more at the left on
Figure 6a) towards the most eastern one. We plot, for each of the 40 devices, three percentile values of the E-field (
), which represent the value below which there is an
r probability of the E-field. These results yield interesting insights on how the E-field varies within the scenario under study. On the one hand, the measurements taken by a dosimeter do not show a high variability, yielding a tight distribution, where the highest and lowest percentile values are quite close to each other, thus confirming what was previously seen. On the other hand, although there is a clear correlation between some of the close-by devices, there are some cases where the measurements exhibit larger differences. This somehow corroborates the initial assumption that the density of probes needs to be high enough to leverage an accurate characterization. Both aspects are observed for the three frequency bands. This is in fact confirming the simulation-based analysis presented in
Section 3.3.
In order to derive the
from the E-field values provided by the dosimeters, we will use the methodology that was previously discussed. As a first step, we need to identify the technologies that are used in each frequency band, so as to apply the appropriate factors; see
Table 4. This is complicated by the fact that operators exploit refarming techniques and make use of some bands to deploy technologies for which they were not initially conceived. Hence, we need to take some assumptions regarding the spectrum usage within the scenario. In this sense, based on the outcome of a more accurate measurement campaign, carried out with a Received Signal Strength Indicator (RSSI) scanner, we extracted the following technology distribution: 2G technologies (i.e., GSM) is using both 900- and 1800-MHz bands; 3G networks are allocated in the 2100-MHz band; and 4G is only deployed in the 1800-MHz band. Hence, there are two different technologies (2G and 4G) coexisting in the same band (1800 MHz). Based on field measurements, (see Appendix 2.8 in [
22]), we assume that the E-field induced by 2G is
-times higher than that caused by LTE. As a consequence, we establish the corresponding correction factors for GSM and LTE:
and
, respectively.
Figure 14 shows the temporal evolution of the downlink
, considering the contribution coming from the different technologies. It can be clearly seen that the strongest contribution is the one caused by 2G technology, which is almost 10 dB higher than the one observed for 3G and 4G technologies. Besides, we can also observe, for the two 2G bands, a smooth decrease of the
, being slightly more relevant for the 1800-MHz case. This might be caused by the fact that during the last few months, cellular operators have deployed 4G networks using white-spaces left by TV broadcasting services (800 MHz). In this sense, it is sensible to assume that the E-field values can be affected by adjacent bands due to the sensitivity of the dosimeters. On the other hand, the values observed for the 2100-MHz band, uniquely devoted to 3G, shows a rather constant level during the whole year.