Methodology based on Vector and Scalar Measurement of Traffic Channel Power Levels to Assess Maximum Exposure to Electromagnetic Radiation generated by 5G NR Systems

Maximum-Power Extrapolation (MPE) for mobile telecommunication sources follows an established paradigm based on the identification and measurement of a channel that acts as a power reference. Prior to the 5G era, the role of reference channel has been played by always-on broadcast signals since they had the great advantage of being always transmitted at the maximum power level allowed for a generic signal channel. However, the beamforming implemented by 5G sources obliges to rethink this approach. In fact, with beamforming the 5G source can transmit data traffic streams through a beam characterized by a much higher gain than the broadcast one. This implies that the detected power for traffic beams could be much higher than the corresponding power of broadcast beams. In this paper, a novel approach for 5G MPE procedure is presented. For the very first time, the direct measurement of the received power of a traffic beam is used to assess the maximum exposure generated by a 5G system. An innovative specific experimental setup is also proposed, with the use of a User Equipment (UE) with the aim of forcing the traffic beam toward the measurement positions. In this way, it is possible to measure directly the power of each Resource Element (RE) transmitted by the traffic beam. As opposed to other MPE proposals for 5G, the discussed technique does not require any correction of the measured data since it relies only on the traffic beam pointing toward the measurement position, simplifying the overall MPE procedure and thus reducing the uncertainty of the MPE estimated field strength.


I. INTRODUCTION
5G represents a major technological leap over previous generations of cellular systems. The stringent specifications of 5G, that must meet the requirements not only of human-to-human communication but also those of human- 5 to-machine and machine-to-machine, required the introduction of numerous technological innovations. However, such innovations, like the beamforming technique inherent to 5G New Radio (NR) antenna systems, have raised new questions about the assessment and measurement of human 10 exposure to the electromagnetic radiation generated by the new radiation paradigm. Also, the characterization of Power Control systems that limit the radiated power when specific time-averaged exposure limits are being approached is a topic of interest when a multiple-beam coverage is used 15 like in a 5G NR [1].
To clarify the new challenges that 5G presents in this field, we recall that the estimation of the human exposure levels to electromagnetic fields essentially requires two steps. In the first phase, a special procedure, named 20 Maximum-Power Extrapolation (MPE), allows to estimate the maximum expected field level radiated by the Next Generation NodeB (gNB) at the measurement point. It should be emphasized that the value obtained by the MPE procedure is an unrealistic upper limit, since it assumes that 25 all the resources of the communication system are assigned to a single user. In the second phase, the MPE value is multiplied by an appropriate correction factor that takes into account the stochastic nature of the problem in order to obtain a realistic value [2]- [10]. 30 With reference to the first phase, the standard MPE approach up to now has been based on the identification of a constant power reference signal that is always present (i.e., "always on"), under the condition that all these signals are transmitted with constant and maximum power at all 35 times.
This condition is not totally fulfilled in 5G, because of a different approach to the managing of "always on" signals and because of beamforming. In particular, 5G uses an energy-efficient signaling strategy, limiting "always on" 40 signals as much as possible. In fact, 5G contains all the fundamental information necessary to access the network in a signal structure highly concentrated in frequency, time and possibly space, called SS/PBCH Block (SSB).
In addition, 5G systems implement antenna solutions able 45 to transmit user data on a different beam (named traffic beam) than the beam used to transmit the SSB channel (named broadcast beam). This approach is used both in sub-6 GHz active antennas and in mm-Wave band [11]. Since the traffic beam usually has a greater directivity than the 50 broadcast one, there is a power increase factor with respect to the power level measured by SSB which must be taken into account in MPE. The innovations introduced by 5G technology oblige to rethink and innovate the established paradigm of MPE pro-55 cedure. Due to beamforming, 5G systems are characterized by dynamic, high-gain traffic beams focused towards the UE resulting in received power levels that may be much higher than the corresponding power of broadcast beams. For this reason, the definition of an effective MPE procedure for 60 active antenna systems implementing beamforming is currently ongoing in International Electrotechnical Commission (IEC) TC106X, that is revising IEC 62232 standard [12] in order to define a method to take into account this power increase. The complexity of the experimental scenario sug-65 gests to shift towards a novel approach for MPE, based for the very first time on the usage of traffic Physical Downlink Shared CHannel (PDSCH) as the maximum power reference for 5G systems.
This paper presents an innovative proposal for 5G MPE 70 procedure based on direct measurement of the PDSCH power level using both Vector Signal Analyzer (VSA) and Spectrum Analyzer (SA). In addition, to address the issues introduced by beamforming, the procedure requires an original experimental setup based on the use of a UE, with the 75 aim of forcing the 5G system to focus a high-gain traffic beam towards the point of interest, throughout the whole duration of the PDSCH power measurement. Experimental results collected at several 5G sites confirm the validity of the proposed method, providing an accurate validation of its 80 effectiveness.
The paper is organized as follows: in Sect. II a review of the MPE procedure is given, while in Sect. III the measurement setup is presented. Experimental results are presented and discussed in Sect. IV: specific subsections 85 are provided for each tested methodology, along with an overall comparison. Measurement procedures are thoroughly discussed in Appendix A.

90
Although the experimental procedure for the assessment of human exposure to radio frequency (RF) electromagnetic fields is strongly dependent on the specific characteristics of the radio system, the methodologies adopted for getting the maximum received power in a measurement point -independently by the technology -are based on the identification of a reference channel. In order to be eligible to serve as a reference for the received power, a channel requires a timeconstant transmitted power representing an upper bound for the allowed transmitted power of a generic channel of the radio system. Once the reference channel received power P ref is measured, the extrapolated maximum power P max that can be received in the measurement point can be obtained by eq. (1): where K is a technology-dependent constant. Prior to 5G technology, the role of reference channel has been played by system-specific broadcast control channels -Broadcast Control CHannel (BCCH) for 2G [13,14], Primary Common PIlot CHannel (P-CPICH) for 3G and 95 Cell specific Reference Signal (CRS) for 4G -which had the great advantage to be always on and moderately easy to detect. For this reason, many early works about 5G electromagnetic field (EMF) assessment were based on the measurement of the power levels associated with the SSB 100 [11,15]- [21]. Nevertheless, it soon became evident that the SSB power could not be representative of the maximum power level of data channel due to the beamforming mechanism.
The peculiar ability of the active antenna systems -105 widely used for 5G technology -to focus narrow, highgain traffic beams to specific end-user directions implies that the PDSCH is received with a power level that could be several order of magnitude higher than the broadcast control channels. This behavior implies, for signals generated by 110 active antenna systems, that PDSCH power per Resource Element (RE) P PDSCH-RE can be assumed as an effective candidate for P ref for maximum power extrapolation just as long as the measurement is carried out while a traffic beam is pointing to the receiver antenna. This condition can be 115 achieved by forcing a connection between the radio base station and a UE placed in the measurement point ( Fig. 1). Under this assumption, eq. (1) for 5G signals can be rewritten as: where: is the total number of sub-carriers, which depends on both the numerology µ and channel band-120 width B of the incoming signal; • F TDC is the Time-Division Duplexing (TDD) dutycycle scale factor defined in [12]; • P PDSCH-RE,max is the maximum PDSCH received power per RE, according to the measurement scenario shown 125 in Fig. 1. Strictly speaking, the key step of the experimental methodology for the assessment of the maximum 5G received power is the on-field measurement of P PDSCH-RE,max . In the following, two alternative methods for the assessment of 130 P PDSCH-RE,max will be presented, based on code-domain and Zero Span (ZS) measurements.

III. MEASUREMENT SET-UP AND PROCEDURE
Two types of experimental campaigns have been conducted, the first one under controlled laboratory conditions and the 135 second one consisting of on-site measurements.

A. CONTROLLED-ENVIRONMENT MEASUREMENT SETUP
In the controlled-environment experimental session, the following setup has been used:  The 5G VSG was connected directly to the VSA through a coaxial cable. The main characteristics of the generated signals are summarized in Tab. 1.

B. ON-SITE MEASUREMENT SETUP
For the on-site measurements, three operation sites located 150 in Ivrea, Rome and Turin were chosen where 5G NR gNBs by two different vendors were active. Fig. 2 shows that each site allowed a full Line-of-Sight (LOS) propagation. The configuration of the signal sources is shown in Tab      connected to an R&S HL050 log-periodic antenna through a coaxial cable. Both vector analysis of the demodulated 5G frame and ZS measurements were carried out. In addition, Channel Power (CP) measurements were used as a reference value for maximum 175 power.
The instruments configuration for scalar measurements is reported in Tab. 3. 180 Preliminary measurements in a controlled environment were run to refine the procedure to be used during on-site measurements. Namely, the focus was to determine the effect of the modulation of the PDSCH on the demodulated power, considering both constant (QPSK) and non-constant 185 (256QAM) amplitude modulations. Spectrograms in Fig. 3 show a full 10 ms frame when the two modulations are applied.  At first, let us consider the PDSCH signal with a QPSK modulation. As we can see in the top strip of Fig. 4, 190 the effect is that there is a constant-power signal at the receiver input. As a matter of fact, the experimental pdf of the normalized power is very dense about the mean, and symmetric.

A. CONTROLLED-ENVIRONMENT MEASUREMENTS
Then, let us consider the PDSCH with a 256QAM mod-195 ulation. The center and bottom strips in Fig. 4 show the experimental and theoretical pdfs of the normalized power of the REs of that channel at the receiver input. As expected, the figure shows a large power spread, that is typical of the high-cardinality constellations adopted for payload data 200 transmission, where many symbols of different amplitudes are used.
The observations above suggest a statistical approach to the evaluation of the power associated to the PDSCH, whose potential estimates are the the medianx and mean 205 x of the PDSCH REs received power. In the following, both estimates will be considered, and it will be shown that, in all the experimental cases considered for on-site measurements, they return compatible results when used in the MPE procedure.

210
Tab. 4 lists the values ofx andx from the actual QPSKand 256QAM-modulated signals generated by the VSG and for the normalized theoretical 256QAM-modulated signal. We have also reported the difference ∆(x,x) in dB of the two quantities. As expected, the QPSK modulation returns 215 VOLUME 4, 2016 a substantially equal mean and median (∆(x,x) ≃ 0), while the 256QAM modulation results in distance between the two statistics that is almost the same for the actual and theoretical signal, which is consistent with the strong overlap of the two distributions in Fig. 4.

B. ON-SITE MEASUREMENTS: VECTOR APPROACH
The vector measurement approach for the evaluation of P PDSCH-RE,max has been applied to the measurement sites located in Rome and Turin (Fig. 2). For both sites, the 5G radio frame has been demodu-225 lated to obtain a complete symbol per carrier grid with the received power of each RE. Measurements were run according to the experimental setup shown in Fig. 1, i.e., forcing the traffic with a UE to obtain a full-loaded frame. The radio frames acquired are shown by the spectrograms 230 in Fig. 5, confirming that the connection between the UE and the gNB was very effective in inducing a complete PDSCH allocation within the frame at all measurement sites, although a fluctuation of the average power of the REs along the frequency axis can be observed. Measurements in Turin 235 were run both in the morning and afternoon to check for repeatability, i.e., to make sure that no significant variation had occurred in either the source or the propagation channel over time.
As discussed in Sect. IV-A, the usage of 256QAM mod-240 ulation for the PDSCH implies that the expected received power distribution of the associated REs is characterized by a large spread, due to the high crest factor of such a highorder modulation, unlike a constant-amplitude modulation like QPSK. Fig. 6 shows the pdf of the power of the PDSCH 245 REs for the radio frames acquired both in Rome and Turin, with vertical dashed lines representing the mean (red) and median (green) value for each distribution. The presence of a large spread is confirmed, with the pdf typically spanning slightly more than the 30 dB interval that characterizes the 250 256QAM modulation in Fig. 4. The most relevant statistics of the distributions shown in Fig. 6 are listed in Tab. 5, wherex is the median andx is the mean. Focusing on the distance ∆(x,x) between the dBm values ofx and x, we can notice that its value is virtually that same for 255 both measurements in Turin, possibly implying that the propagation channel has kept pretty stable during the day. Also, we can see that, on average, Turin has a smaller ∆(x,x) than Rome.
The extrapolated maximum power P max has been assessed 260 according to eq. (2) using both the median and mean as estimates for P PDSCH-RE,max . A CP measurement (Fig. 7) has also been performed during the full-loaded frame condition as a reference for the expected P max value. An overall comparison between the extrapolated and reference P max 265 value is presented in Tab. 6 with 95% (k = 2) expanded uncertainty. Results show that in each measurement sitex andx return extrapolated values that are compatible with the P CP reference value, confirming the results reported in [23]   regarding the effectiveness of the proposed vector approach.

C. ON-SITE MEASUREMENTS: SCALAR APPROACH
MPE from scalar measurements was obtained from ZS acquisitions in AVG mode and RMS detector, each consisting of the average of n = 100 traces.

275
The reason for using the AVG mode is that while MH-  mode acquisitions easily capture the full-frame condition required for the assessment of P PDSCH-RE,max , they also detect the peak value of the power of the PDSCH REs P PDSCH-RE (i.e., the right tail of the distributions in Fig. 6) that could 280 be not fully representative for MPE due to modulation and propagation conditions. A better representation of P PDSCH-RE can be obtained by using acquisitions averaged over a sufficiently extended time interval, thus minimizing the possibility of power overestimation.

285
A sample of a ZS AVG trace is shown in Fig. 8a, where the red dot indicates the position of the SSB. After converting from dBm to nW, the power kernel density estimate was obtained through the density() function available in the stats library of the R language [24], using a rectangular 290 kernel with a binwidth equal to 0.05 (see Fig. 8b). From the experimental density, the mode x mo of the traffic slots (indicated by the shadowed dotted line in Fig. 8b and, in log units, by the horizontal dashed blue line in Fig. 8a) was extracted. The use of x mo makes the procedure robust 295 against transient power reductions, under the hypothesis that the most frequent value and the maximum received power are the same.
To obtain the estimate of the MPE power P traffic from x mo , the following formula can been applied: where ∆f = 30 kHz is the sub-carrier frequency spacing for µ = 1, and NBW is the noise bandwidth. Various 300 configurations of the measurement equipment were tested at the sites shown in Fig. 2 Fig. 9, where dots represent the measured value and vertical segments the expanded uncertainty (k = 2). The horizontal dashed gray line represents VOLUME 4, 2016  value to assess the significance of the proposed scalar MPE procedure. The expanded uncertainty interval (k = 2) is represented by the light gray strip. A summary of ZS measurement is shown in Tab. 7, where N p is the number of points acquired in a trace, N T is 315 the number of averaged traces for each acquisition in AVG mode, n is the number of repeated measurements,P traffic is the average value of P traffic with the associated uncertainty. Where repeated measurements were not available, only the u b contribution to the uncertainty given by the instrument 320 could be evaluated resulting in a tighter uncertainty interval. Fig. 9 shows generally a very good agreement between the scalar MPE values and the reference CP measurements: most uncertainty intervals of scalar and CP measurements overlap, even if in some cases (see for example the set of 325 measurements for the Ivrea site) scalar measurements are at the lower bound of the CP uncertainty interval.

D. A COMPARISON BETWEEN VECTOR AND SCALAR MEASUREMENT
As a last step, it is of interest to compare the vector 330 and scalar methods. For a significant comparison, the two measurements methods had to be applied at the same time so that they operate on the same input signal. Therefore, in the measurement session in Turin a power splitter was inserted in the measurement chain to connect the same antenna to 335 both the R&S FSW and the Keysight N9020 analyzers. In the previous Sections it was showed that the parameters setting tested in the various measurement sessions give results within the uncertainty range of the measurement. For the comparison, the scalar method was applied with 340 the following experimental configuration: • SWT = 20 ms, equal to a multiple of the frame duration and to the SSB repetition period; • number of points = 606, so that each time interval spanned by a pixel on the screen of the SA is very 345 close to the symbol duration; • RBW = 5 MHz, chosen because it is the largest value, that guarantees that a sufficiently large statistical sample of traffic REs is acquired, while for the vector measurement the medianx of the 350 measurement session was chosen. Furthermore, since the data forcing procedure allowed for a full use of the frame, CP measurements were also collected and used as direct estimation of the maximum power.
The results shown in Fig. 10 confirm that scalar and 355 vector approach to MPE return comparable results (a t-test with t = 0.14 proves that they do not differ significantly, with p < 0.01), and that they are also in full agreement with the reference value provided by CP measurements.
With regards to the uncertainty associated with each 360 method, the figure shows that it is lower for vector measurement than for the scalar and CP measurement. This is explained by the fact that the vector measurement is more precise because the analysis algorithm assures a perfect synchronization with the signal, which also relax 365 the requirement for repeated measurement to assess the u a contribution to the uncertainty budget. On the contrary, repeatability is a significant contribution to uncertainty in the scalar approach, where synchronization is much less precise, and it is not possible to directly measure P PDSCH-RE,max 370 (which is obtained by post-processing traces as shown in Sect. IV-C). For the CP measurement, uncertainty has been evaluated also taking into account the variation of the In this paper, a novel MPE procedure for 5G systems, based on the direct measurement of the received power of traffic beams, is presented and experimentally validated. The technique takes advantage of a user terminal to force the traffic beam toward the measurement positions. In this 380 way it is possible to measure directly the power of the REs assigned to PDSCH.
The proposed procedure has a number of advantages: 1) it can be applied to both scalar or vector measurements; 4) the multiple-acquisition procedure required by scalar measurements can be easily automatized. Regarding the limitations, the ZS procedure proposed in this paper requires signal acquisition in average mode, therefore a uniformly full loaded frame is required during 400 the acquisition time. This problem will be subject to further investigation. . • Spectral occupation and center frequency of the 5G NR signal: each spectrum band has to be investigated 430 reducing progressively the span. A MH trace setting with a RBW and VBW of some hundreds of kHz allows to properly display the Orthogonal Frequency-Division Multiplexing (OFDM) spectrum of the signal and thus identify the spectral occupation and center 435 frequency;

APPENDIX A MEASUREMENT PROCEDURES
• Center frequency of the SSB: modern VSA can perform a real-time spectrum analysis (also known as waterfall spectrogram shown in Fig. 11). This analysis allows to highlight the region of the frequency 440 spectrum which is characterized by the presence of a persistent, always-on SSB. Starting from a waterfall plot, it is easy to identify the offset between the 5G carrier frequency and the SSB center frequency; • Cell ID: once all the parameters discussed above are properly set, the VSA should be able to resolve the Cell IDs associated to the signal under investigation. Although prior knowledge of the TDD configuration imple-460 mented by the 5G signal is not strictly required to perform a vector analysis of a 5G signal, the value of F TDC parameter -i.e., the fraction of the signal frame reserved to downlink transmission [25] -is still required in order to apply eq. (2). A detailed description of the experimental procedure for the 465 estimate of F TDC can be found in [18].
Once the VSA shows a condition of full synchronization with the 5G signal, it is possible to force the traffic with a UE to obtain a full-loaded frame. Finally, a detailed analysis of the received power level per RE allows to infer the value 470 of P PDSCH-RE,max which could be used in eq. (2).

B. SCALAR MEASUREMENTS
The measurement method developed on VSA is based on the following considerations: • for the goals of this work, the frequency setting for the 475 analysis in the time domain (ZS functionality of the analyzer) was chosen equal to that of the SSB center band; • to verify the duplexing and measure the level of the SSBs and/or of traffic slots, it is necessary to syn-480 chronize the analysis of the trace to the duration of VOLUME 4, 2016 Figure 11. Determination of center frequency and bandwidth extension of the SSB by a waterfall spectrogram.
the frame, and in particular to the repetition period of the SSBs. This synchronization was obtained using a periodic trigger, with a period equal to a multiple of the duration of the frame (10 ms), and a scan time 485 set in such a way that the duration of the analysis of each pixel on the screen was close to the duration of a symbol (this indication derives from the provisions for 4G in [12]); • in case of OFDM type signals with complex modula-490 tions, it is necessary to work with an RMS detector to avoid overestimating the read power (due to the high crest factor); • the opening of the RBW filter was chosen lower than the SSB bandwidth occupation, in order to guarantee 495 the proper detection both of the power level associated only with the control channels (not with any simultaneous and adjacent traffic channels in frequency), and the power level associated to traffic slots. In the measurement sessions of this work, depending on the 500 characteristics of the power distribution in each frame, we could observe that greater repeatability is obtained when using values of RBW greater than 3 MHz, due to the greater number of samples from which the power level is measured, whereas greater variability can be 505 seen when using an RBW of 1 MHz; • again due to the OFDM signal with complex modulations, the VBW value was chosen greater than 3 times the width of the RBW filter; • to obtain traffic slots amplitude, traces have to be 510 acquired in AVG mode. In this work, an average on 100 traces was chosen. For this type of measurement, it is important to assure, through adequate traffic-forcing, that the full-frame condition is maintained during the acquisition of the entire set of 100 traces.

515
After acquiring the traces obtained with the above described criteria, the traffic slots power level can be determined by analyzing the density function of the trace pixels, by calculating the mode of the distribution. Eventually, the power per RE has to be calculated, by 520 correcting the obtained power value for the ratio between sub-carrier spacing and noise bandwidth.