Report on the development of a standardised measurement procedure for the quantification of RF exposure in terms of SAR and APD measurements of 5G new radio mobile phones with evidence of its submission to standards bodies, for example, CENELEC CLC/TC 106X TC and IEC TC106 JWG12 technical committees for consideration for inclusion in standards documents/technical specifications

Confidentiality Status: PU - Public, fully open (remember to deposit public deliverables in a trusted repository) Deliverable Cover Sheet Funded by the European Union. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or EURAMET. Neither the European Union nor the granting authority can be held responsible for them. The project has received funding from the European Partnership on Metrology, co-financed from the European Union’s Horizon Europe Research and Innovation Programme and by the Participating States. 1 of 33 21NRM03 MEWS D7: Report on the development of a standardised measurement procedure for the quantification of RF exposure in terms of SAR and APD measurements of 5G new radio mobile phones with evidence of its submission to standards bodies, for example, CENELEC CLC/TC 106X TC and IEC TC106 JWG12 technical committees for consideration for inclusion in standards documents/technical specifications Organisation name of the lead participant for the deliverable: IMTelecom Due date of the deliverable: 30/09/2025 Actual submission date of the deliverable: 30/09/2025 2 of 33 TABLE OF CONTENTS 21NRM03 MEWS ............................................................................................................................ 1 1 Executive Summary ..................................................................................................................... 3 2 Introduction .................................................................................................................................. 5 3 Objectives .................................................................................................................................... 6 3.1 Activity-level objectives (A3.4.1–A3.4.6) ................................................................................ 6 4 Results of Research Activities ...................................................................................................... 7 4.1 SAR and APD Maximum and Statistically Averaged RF Exposures of 5G User-Equipment using Inverse Methods [A3.4.1, A3.4.2] ....................................................................................... 7 4.1.1 Introduction .................................................................................................................................. 7 4.1.2 Non-Invasive measurements ....................................................................................................... 7 4.1.3 Conclusion ................................................................................................................................. 15 4.2 Linearisation for 5G NR SAR/APD Near-Field Probes [A3.4.3] ............................................ 16 4.3 Traceable APD Probe Calibration [A3.4.4] .......................................................................... 19 4.4 5G NR Reference Sources for APD System Validation [A3.4.5] .......................................... 22 4.5 Standardised Measurement Procedure Synthesis & Recommendations [A3.4.6] ................ 25 5 Conclusions ............................................................................................................................... 26 6 References & Publications ......................................................................................................... 27 Annex Emails sent as Evidence of Submission to Standards Bodies ............................................ 29 21NRM03 MEWS 3 of 33 1 Executive Summary This task develops a measurement methodology for quantifying public RF exposure from 5G new radio (NR) mobile phones in terms of Specific Absorption Rate (SAR, W·kg⁻¹) for sub-6 GHz and Absorbed Power Density (APD, W·m⁻²) for mm-wave bands above 6 GHz. This report is shared in different parts: The first one (A3.4.1, A3.4.2) is dedicated investigate the feasibility of a non-invasive electric field assessment method, the second one (A3.4.3) deals with Linearisation for 5G NR SAR/APD Near-Field Probes, the third one (A3.4.4) deals with traceable APD Probe Calibration and the fourth one (A3.4.5) develop NR Reference Sources for APD System Validation. The last part A3.4.6 provides the Standardised Measurement Procedure Synthesis & Recommendations. In the first part (A3.4.1, A3.4.2) the feasibility of a non-invasive electric field assessment is investigated. Specific absorption rate (SAR) measurement techniques are classified into direct (invasive) and indirect (non-invasive) measurements. In the conventional direct approach recommended by International Electrotechnical Commission (IEC) the tissue simulating liquid is placed inside a human-shaped tank which is exposed to the device under test (DUT), a smart phone for instance. The electric field is assessed using probe attached to a robotic arm or using an array of probes inserted inside the liquid phantom. These methods are accurate but presents some disadvantages linked to the invasive approach. In order to overcome the limitations of the conventional SAR measurement methods, a non-invasive electric field reconstruction method based on boundary element method was proposed. Evaluating SAR from non-invasive measurements on a surface enclosing DUT and phantom is challenging due to SAR reconstruction errors which arise due to DUT/phantom coupling The main research goal of this activity was to resolve the problem of SAR reconstruction errors due to the coupling between DUT and phantom in non-invasive approach for SAR measurement. We used spherical mode approach to address this research problem. In the first attempt, we performed low pass filtering of the spherical modes based on the minimum sphere enclosing the DUT in order to get a first estimate of radiated field of DUT which takes into account the coupling with the phantom. However, in this case, the filtered out lower order modes were also contaminated by the coupling from the phantom and we were not able to improve the SAR reconstruction errors. In a second attempt, we used Pogorzelski algorithm from literature in order to isolate the radiation of DUT (including coupling) from the total noninvasively measured spherical near field. However, this algorithm in literature was only applied to the cases where both radiators were in the far field and in our case DUT and the phantom were in the near field and because of this we were not able to improve the SAR reconstruction errors using this spherical mode approach. Another limitation of this approach was that the minimum spheres enclosing DUT and phantom must not intersect. This puts a limit on the minimum distance between DUT and phantom for the applicability of this algorithm. In conclusion, based on the above-mentioned research and despite the significant efforts made, we do not identify a clear solution to improve SAR reconstruction errors due to DUT/phantom coupling in context of non-invasive SAR measurements. Indeed, estimate coupled fields from external measurements is complex especially when the distance between DUT and phantom is electrically small. The second part of this reports (A3.4.3) deals with linearisation for 5G NR SAR/APD Near-Field Probes. Traditional signal-specific linearization methods have become impractical with 5G’s thousands of modulations and would require new measurements whenever a signal changes. The objective of this task has been to replace the signal-specific linearization of short-dipole diode probes with a physics-informed neural network (PINN). The resistively loaded short-dipole probe was modelled with an ordinary differential equation (ODE) and an innovative approach was developed to accelerate sensor model simulations. Then Artificial Intelligence, using a physicsinformed neural network (PINN), was trained to predicts the coefficients which are used to linearize a measure voltage response. The results of this work have been published and commercially implemented in SPEAG’s DASY8 Module SAR V17.0 and released on September 9th, 2025. The third part (A3.4.4) establishes a traceable, tested, and validated calibration for Absorbed Power Density (APD) probes operating over the 10– 30 GHz frequency range. It includes the development of a composite skin phantom. The APD measurement concept measures the complex electric field (E-field) inside a skin-simulating liquid (SSL) and reconstructs APD at the phantom surface. The calibration concept was initially implemented and validated for the frequency range of 24–30 GHz, and subsequently extended to cover the frequency range of 10–45 GHz. The developed calibration system and phantom were published in a peer-reviewed journal. In the fourth part (A3.4.5), 5G NR Reference Sources for APD System Validation has been analysed. We developed and characterized two complementary 5G New Radio (NR) mmWave reference radiation sources for validating the Absorbed Power Density (APD) in a measurement technology agnostic manner. The developed validation system was published in a peer-reviewed journal publication. The last part describes the work carried out to disseminate our result within CENELEC and IEC Technical committees, as well as IEEE, a technical report and 21NRM03 MEWS 4 of 33 international standards have been drafted and provided to IEC/IEEE P63195-3 for the compliance testing of wireless products with APD limits . The APD validation sources developed in this project became the reference sources in an APD measurement round-robin conducted by IEC/IEEE TC106. 21NRM03 MEWS 5 of 33 2 Introduction With the increasing use of wireless communication systems, constant advances in technology and an increasingly wide use of the frequency spectrum, it is fundamental to adapt the methods of assessing exposure to EM fields in order to quantify human exposure and check the compliance with existing limits. This task develops a measurement methodology for quantifying public RF exposure from 5G new radio (NR) mobile phones in terms of Specific Absorption Rate (SAR, W·kg⁻¹) for sub-6 GHz and Absorbed Power Density (APD, W·m⁻²) for mm-wave bands above 6 GHz. In this task the feasibility of a non-invasive electric field assessment is investigated respond to some disadvantages linked to the invasive approach. This report describes the works carried out and the limits linked to the estimate coupled fields from external measurements that is complex especially when the distance between DUT and phantom is electrically small. This document also addresses the challenges linked with 5G NR SAR/APD Near-Field Probes. The use of Artificial Intelligence using a physics-informed neural network (PINN) and short dipole probe model are carried out to predicts the coefficients which are used to linearize a measure voltage response. Traceable, tested, and validated calibration for Absorbed Power Density (APD) probes operating over the 10–30 GHz frequency range are established and 5G NR Reference Sources for APD System Validation has been analysed and tested. As described in section 4.5 a technical report and international standards have been drafted and provided to IEC/IEEE P63195-3, moreover the APD validation sources developed in this project have been disseminated in IEXC/IEEE and became the reference sources in an APD measurement round-robin conducted by IEC/IEEE TC106. 21NRM03 MEWS 6 of 33 3 Objectives This task develops a measurement methodology for quantifying public RF exposure from 5G new radio (NR) mobile phones in terms of Specific Absorption Rate (SAR, W·kg⁻¹) for sub-6 GHz and Absorbed Power Density (APD, W·m⁻²) for mm-wave bands above 6 GHz. It includes collecting datasets in both regimes, developing and applying advanced evaluation schemes (e.g., machine-learning and statistical approaches), and focusing on two APD measurement approaches: (i) traceably measuring the electric field inside a skin-emulating phantom and reconstructing APD, and (ii) measuring the free-space electric field outside the device under test and using field-reconstruction techniques to quantify incident and reflected fields and the resulting APD. 3.1 Activity-level objectives (A3.4.1–A3.4.6) A3.4.1 (M22) — Perform SAR and APD measurements on at least one multi-antenna 5G NR phone to generate analysis datasets; investigate inverse, over-the-air (external-to-phantom) methods and determine associated uncertainties (target 30%); propose and apply at least one advanced evaluation scheme A3.4.2 (M34) — Provide ≥2 independent approaches for SAR/APD assessment based on free-space near-source electric-field measurements; deliver written recommendations grounded in ICNIRP guidance for CENELEC TC106X, IEC TC106, equipment manufacturers, national authorities, and handset manufacturers. A3.4.3 (M35) — Develop and validate near-field probe linearisation methods for 5G NR signals for SAR and APD probes, including ML-based signal quantification; compile an uncertainty budget and verify results against at least two methods from A3.4.2. A3.4.4 (M18) — Establish a traceable, tested, and validated calibration for APD probes (10–30 GHz), including development of an APD test phantom simulating the human body, commissioning of the calibration setup, definition of the calibration methodology, and determination of uncertainties (target < 1.6 dB). A3.4.5 (M24) — Develop and verify two 5G NR reference radiation sources to validate the APD systems from A3.4.4; collect datasets in FR2 mm-wave (APD, W·m⁻²). A3.4.6 (M36) — Synthesize outcomes from A3.4.1–A3.4.5 into a standardised measurement procedure for SAR/APD of 5G NR phones; disseminate to international bodies (e.g., CENELEC CLC/TC 106X, IEC TC106 MT3/JWG12, ITU-T SG5, IEEE ICES TC95); obtain letters confirming submission/consideration. 21NRM03 MEWS 7 of 33 4 Results of Research Activities 4.1 SAR and APD Maximum and Statistically Averaged RF Exposures of 5G UserEquipment using Inverse Methods [A3.4.1, A3.4.2] 4.1.1 Introduction Specific absorption rate (SAR) measurement techniques can be classified into direct (invasive) and indirect (non-invasive) measurements. In the conventional direct approach recommended by International Electrotechnical Commission (IEC)[1], the tissue simulating liquid is placed inside a human-shaped tank which is exposed to the device under test (DUT) i.e., RF appliances or wireless communication devices such as smart phones. The electric field probe attached to a robotic arm is invasively inserted inside the liquid phantom to sequentially measure the induced electric field in three dimensions. SAR is then evaluated by know value of mass density and conductivity of the phantom. This method, although straightforward and accurate, presents several disadvantages. First, this procedure as recommended by IEC standards, is very time consuming since the volumetric (3D) electric field measurements have to be performed within the body simulating mannequin (phantom) for multiple frequency bands and modulations supported by modern smart phones. A typical smart phone may incorporate multiple sub-6-GHz frequency bands such as 4G long-term evolution (LTE) bands [LTE700 (704–803 MHz), GSM850 (824–894 MHz), GSM900 (880–960 MHz), GSM1800 (1710–1880 MHz), GSM1900 (1850–1990 MHz), UMTS (1920– 2170 MHz), LTE2300 (2300–2400 MHz), and LTE2500 (2500–2690 MHz)] and 5G new radio (NR) bands [N77 (3300–4200 MHz), N78 (3300–3800 MHz), and N79 (4400–5000 MHz)], and so on [2]. Second, in order to allow the easy movement of the probe, the phantom filler has to be a fluid. This requirement prevents the use of inhomogeneous and anisotropic solid phantoms which have not only more stable electromagnetic properties but also can model human body more accurately. Lastly, the invasive internal probing significantly perturbs the electric field distribution leading to measurement errors. The time for the direct measurements can be reduced by either using multiple probe systems which perform parallel measurements on a given plane [3]-[4] or by using wave propagation techniques [5]-[8]. Optimal set of measurement samples have also been used to reduce SAR measurement time [9]. Although, the aforementioned techniques reduce the SAR measurement time, these techniques are not only invasive but also require measurements for each device orientation and separation distance with the phantom to represent various use cases. 4.1.2 Non-Invasive measurements The inherent disadvantages of the conventional direct measurements can be addressed by indirect (noninvasive) measurements in which the electric field within the phantom is reconstructed from measurements performed on a two-dimensional (2D) surface enclosing the phantom. This approach allows the use of stable solid phantoms since the measurements are performed non-invasively. Additionally, the SAR acquisition is accelerated due to surface measurements. 4.1.2.1 State of the Art In order to overcome the limitations of the conventional SAR measurement method, a non-invasive electric field reconstruction method based on boundary element method was proposed in [10]-[11]. The equivalent electric currents on the phantom surface were reconstructed from electric field observed on a surface surrounding only the phantom in [10] and only the DUT in [11]. Despite the accuracy of the aforementioned studies, the probe measurements were required to be performed in a narrow region between DUT and phantom. This made the measurements impractical and led to the problem of probe coupling with the phantom and DUT. In order to avoid the problem of probe coupling, the electric field measurements were performed on a surface enclosing both DUT and phantom to reconstruct the equivalent currents on the phantom in [12]-[13]. However, the accuracy degradation in SAR reconstruction occurred due to loss of signal inside lossy phantom and because of interruption of phantom reflection by DUT before reaching the measurement probe. The 21NRM03 MEWS 8 of 33 stability of the field reconstruction algorithm was improved in [14]. More recently, spherical near field measurements were used to generate a near field source which was subsequently imported in full wave electromagnetic simulation software to reconstruct SAR numerically in the phantom [15]-[18]. The main advantage of this method was that SAR can be reconstructed rapidly for various device orientations and separation distances with the phantom. However, since the near field source didn’t take into account coupling with phantom, the SAR reconstructions errors were up to 31% at a distance of 6.2 mm between phantom and DUT [15]. Recently, absorbed power density characterization was performed in a non-invasive way [19]-[20]. 4.1.2.2 Work carried out Evaluating SAR from non-invasive measurements on a surface enclosing DUT and phantom is challenging due to SAR reconstruction errors which arise due to DUT/phantom coupling. In this work, spherical mode filtering, back propagation and Pogorzelski algorithm were utilized in order to isolate radiation of the DUT (patch antenna) in the presence of the phantom at 6 GHz. The isolated spherical modes of DUT were used to evaluate equivalent currents in MVG INSIGHT, which were subsequently imported in FEKO for SAR reconstruction.  The rectangular patch antenna was designed to operate at 6 GHz. The length and width of the patch antenna were 23.2 mm and 15.6 mm. The dimensions of the ground plane were 41.8 mm x 28.2 mm. The height of the substrate was 1.47 mm and dielectric constant was 2.1.  In order to reduce the computational resources, the dimensions of the cuboidal phantom were selected to be 20 mm x 20 mm x 20 mm The relative permittivity and conductivity of the phantom were 35.1 and 5.48 S/m. The distance between the antenna and phantom was 3.1 cm as shown in Figure 1. Figure 1 Simulation setup 4.1.2.3 Numerical Simulation: As a preliminary step, the numerical simulations were substituted for spherical near field measurements. The numerical simulations were performed in FEKO which is based on method of moments. The input impedance and reflection coefficient of the patch antenna alone (49.4-5.9j, -24.5 dB) and patch antenna (58-4.35j, -21.5 dB) in the presence phantom changed. The spherical near field data was recorded on sphere with a radius of RM=70 mm. 4.1.2.4 Results In the first step, the spherical wave expansion was performed on the spherical near field data centered at origin O (0,0,0) as shown in Figure 2. 𝛹=∑ ∑ 𝑂𝑙𝑚𝛹 󰇍 󰇍 𝑙𝑚 (1)(𝑟𝑂,𝜃𝑂,𝜑𝑂), 𝑟𝑂>𝑅𝑀 𝑙𝑚=−𝑙 𝐿𝑂 𝑙=1 (1) Then, the spherical mode filtering was performed based on a minimum sphere with a radius of 30 mm enclosing the patch antenna as displayed in Figure 2. The filtering was performed based on the assumption that spherical modes with n>kr were heavily attenuated. 21NRM03 MEWS 9 of 33 Figure 2 Spherical modes The filtered spherical modes were back propagated to a radius of 30 mm. The resulting electric field was imported in MVG INSIGHT to generate a Huygen’s Box as shown in Figure 3. The equivalent current reconstruction was not accurate because lower order filtered modes were also contaminated due to coupling with phantom. Figure 3 Equivalent currents The Huygen’s box was subsequently imported in FEKO to reconstruct the SAR within the phantom as shown in Figure 4. The maximum SAR reconstruction errors were 72.6%. 21NRM03 MEWS 16 of 33 4.2 Linearisation for 5G NR SAR/APD Near-Field Probes [A3.4.3] Resistively loaded short-dipole diode sensors are precise, broadband, and low-scattering RMS detectors suited for near-field electromagnetic field evaluation; however, their linear region (~30 dB) is insufficient for typical near-field dynamics (>50 dB). Traditional signal-specific linearization methods (per-signal calibration or sensor-model calibration combined with numerical linearization, SMC [1]) have become impractical with 5G’s thousands of modulations and would require new measurements whenever a signal changes. Therefore, the objective of this activity was to replace the signal-specific linearization of short-dipole diode probes with a physics-informed neural network (PINN) that predicts the linearization parameters directly from a few readily available signal descriptors, achieving an error of ≤ 0.4 dB up to a peak SAR of> 200 W·kg⁻¹ and enabling on-site computation. We modeled the resistively loaded short-dipole probe (Figure 17) with an ordinary differential equation (ODE) derived from its equivalent circuit. 𝐶⋅𝑑𝑣𝑜(𝑡) 𝑑𝑡 +𝐼𝑠(𝑒α⋅𝑣𝑜(𝑡)−1)=𝐶𝑎⋅𝑑𝑣𝑖(𝑡) 𝑑𝑡 We first developed an innovative approach to accelerate sensor model simulations, enhancing accuracy. This acceleration, together with improved accuracy, made it feasible to generate a large training data set covering many calibrated sensor parameter sets and a wide range of 5G New Radio (NR) modulation characteristics. Figure 17 The equivalent circuit for the short‐dipole sensors modelling broadband near-field probes Then a physics-informed neural network (PINN) was trained that predicts the four linearization coefficients (A, B, C, D), which are used to linearize a measure voltage response (vresp): 𝑣comp=𝑣resp+𝐷 dcp (𝑣resp)2 𝑣𝑐𝑜𝑚𝑝,𝑑𝐵=10⋅log10(𝑣𝑐𝑜𝑚𝑝) vcorr,dB=vcomp,dB−A⋅exp[−(vcomp,dB−B C)2] vcorr=10 vcorr,dB/10 This prediction is done directly from readily available descriptors of each modulation: i) occupied bandwidth (95% power), ii) peak-to-average ratio (99.9th percentile divided by mean), and iii) duty cycle, see Figure 18. With that, on-the-fly linearization parameters with acceptable uncertainties were obtained across the relevant dynamic range. The training loss penalizes the mismatch between the corrected sensor response and its linear reference, so the model remains constrained by probe physics rather than relying on unconstrained pattern matching. 21NRM03 MEWS 17 of 33 Figure 18 Illustration of the physics‐informed training workflow for linearization. An NN is used to predict the linearization parameters Experimental validation used the Dosimetric Assessment System (DASY8) to position the probe in a highpower transverse electromagnetic (TEM) cell (Figure 19). Figure 19 Measurement setup. Probe calibration setup showing a DASY8 robot and the high‐power TEM cell used for the linearization and validation measurements. Modulated carriers were synthesized from in-phase/quadrature (IQ) data, amplified, and applied. At the same time, the incident power was monitored with a calibrated average-power sensor, and the probe’s direct-current output was recorded. Across unseen signals and sensors (not in training data set), the method achieved <0.4 dB (rect.) linearization error up to > 200 W·kg⁻¹ peak specific absorption rate (SAR) or 400 W·m-2 APD. Compared to simulation-based parameter extraction [1], the PINN determines linearization parameters more than 34,000 times faster and reduces storage requirements by more than 350,000 times, enabling on-site linearization computation within existing DASY8 measurement workflows. The results of this work have been published in the peer-reviewed journal publication [23] and commercially implemented in SPEAG’s DASY8 Module SAR V17.0 and released on September 9th, 2025. 21NRM03 MEWS 18 of 33 Figure 20 Experimental validation. Shown is the linear voltage correction obtained by the AI approach for the measured 5610 responses. The green line indicates the required accuracy of 200 W/kg SAR, equivalent to 200 W/m². 21NRM03 MEWS 19 of 33 4.3 Traceable APD Probe Calibration [A3.4.4] This activity establishes a traceable, tested, and validated calibration for Absorbed Power Density (APD) probes operating over the 10–30 GHz frequency range. It further includes the development of a composite skin phantom and implementation in DASY8 Module APD with a target expanded measurement uncertainty <1.6 dB. The APD measurement concept measures the complex electric field (E-field) inside a skin-simulating liquid (SSL) and reconstructs APD at the phantom surface, which is equivalent to APD at the human-skin surface for both propagating and relevant evanescent spectra. APD and its spatial average (sAPD over 1 cm² / 4 cm²) are defined via the orthogonal component of the Poynting vector; peak quantities (pAPD, psAPD) follow the same convention and are used to test compliance with limits [24]. A composite phantom (Figure 21) is employed to emulate the reflection and transmission coefficients of human skin under reactive near-field conditions while increasing penetration depth to permit in-liquid scanning with the APD probe. The design uses a lossless shell as an impedance-transformation layer above the SSL and a thin foam for mechanical stability; an optimized implementation for 24–30 GHz comprises 2 mm foam (εr≈1.05), 1.08 mm shell (εr≈11.9), and ≥20 mm SSL (εr≈6.9, σ≈1.95 S m⁻¹), and is constrained to capture essentially all practically relevant spatial frequencies up to kr/k0=2 (propagating + evanescent content) with SSL penetration depth ≥ 4 mm. Figure 21 Target 2-layer skin model (left) and corresponding composite APD phantom (right) approximating the skin model for propagating and evanescent waves in a specific frequency range for E-field measurements in the skinsimulating liquid (SSL). Analytical optimization against a two-layer skin model yields mean APD and reflection-coefficient errors typically ≤0.2 dB (TE) and ≤0.8 dB (TM) across the design band (Figure 22). A study on manufacturing tolerances of the composite phantom resulted in a manufacturing parameter related uncertainty of the APD of 0.32 dB (rect.) using realistic manufacturing spreads in shell thickness/permittivity and SSL permittivity/conductivity. 21NRM03 MEWS 20 of 33 Figure 22 Deviation of the APD in the composite phantom compared to the 2-layered skin model. The APD probe is calibrated in a traceable setup purpose-built for APD metrology based on the following concept:  traceable (SPEAG laboratory SCS0108) measurement of SSL dielectric parameter (uncertainty (k=2): 3.2% (ε) and 5.2% (σ))  traceable signal source frequency  traceable RF power measurement  analytically defined calibration source, relates the electric field in the SSL (Figure 23) to the input power and frequency of the source Figure 23 Calibration setup employing a frequency-traceable signal source, traceable power at input port A, and an analytically defined calibration source (right). 21NRM03 MEWS 21 of 33 The calibration concept was initially implemented and validated for the frequency range of 24–30 GHz, and subsequently extended to cover the frequency range of 10–45 GHz. With the setup, the APD probe can be calibrated in a traceable electric field induced in the SSL of the calibration source (Figure 24) by scanning the probe inside the SSL from the surface in the z-direction and determining the probe sensitivity factor by leastsquares fitting the measured probe voltage output to the calibration target E-field. Figure 24 Analytical, simulated, and measured E‐field profiles at 24 GHz. A detailed calibration uncertainty budget for the APD probe calibration was developed, and the expanded measurement uncertainty of the calibration setup was determined to be 0.74 dB (k = 2). The calibration system was implemented in the SPEAG near-field probe calibration laboratory and used for manufacturer calibrations of the APD probes delivered to customers. Towards the end of 2025, SPEAG will seek ISO 17025-compliant accreditation for its calibration setup and extend its calibration scope. The developed calibration system is used for calibrating APD E-field probes. Together with the developed APD phantom, they form key components of the DASY8 Module APD commercial APD test system from SPEAG, which was released in its first version in Q1/2024 and is now available in Version 1.2. The expanded APD measurement uncertainty of DASY8 Module APD is <1.5dB for the determination of the psAPD 1cm2 and 4cm2 The developed calibration system and phantom were published in a peer-reviewed journal publication in [26]. 21NRM03 MEWS 22 of 33 4.4 5G NR Reference Sources for APD System Validation [A3.4.5] In this activity, we developed and characterized two complementary 5G New Radio (NR) mmWave reference radiation sources for validating the Absorbed Power Density (APD) in a measurement technology agnostic manner. Typical antennas operating in the 24–30 GHz range show a spectral power of <0.05% of the total spectral power for kr/k0>2, for a distance to the phantom of r0 at 2 mm, which corresponds to ~0.2λ (Figure 25). More than 99% of the spectral power is captured by kr/k0≤2. Figure 25 Spectral distribution normalized to the total power of different sources operating in the 24–30 GHz range in air at 2 mm from the source, corresponding to 0.18λ to 0.2λ. The direction of propagation for all antennas is along the z-axis. The spectral distribution is plotted for the z‐normal plane for the dipole, WG, slotted horn, open horn, and CDA. For the endfire array, the spectral distribution is plotted for the y‐normal plane; due to the array design, the x‐normal plane contains negligible spectral power. Consequently, sources selected for APD system validation should cover a spectral domain in the same range as typical antennas in wireless devices. The developed system validation process follows a compare-to-target paradigm using numerically established reference values and acceptance based on expanded uncertainty and normalized error En criteria. |En|= | | psAPDmeas/psAPDsim−1 √umeas 2+usim 2 | | The sources selected for system validation, based on the above criteria, are horn slotted array antennas (HSA) and cavity-backed dipole arrays (CDA), which were initially introduced and well characterized for validating incident power density measurement systems [27] as well as an open-ended waveguide source (WG). These sources were characterized numerically at varying distances in front of a numerical skin phantom resembling the properties of the skin target model [25] with a numerical uncertainty ranging between 0.4 and 1.2 dB for psAPD, depending on the source type and distance to the phantom. The developed validation sources and process were subsequently used to validate both the DASY8 Module APD measurement system and the calibration method developed in Section A3.4.4. The measured and 21NRM03 MEWS 23 of 33 simulated APD absorption patterns from the validation sources were compared qualitatively, as shown in Figure 26 and then quantitatively using the normalized error En, as summarized in Table 1. Figure 26 Simulated (a) and measured (b) APD magnitude for the cavity‐fed dipole array with a CW signal at P in = 15dBm and d = 2 mm (top); horn slotted array with a CW signal at P in = 15dBm and d = 2 mm (middle); and the openended waveguide with a CW signal at P in = 7dBm and d = 2 mm (bottom). With normalized errors En <1, the developed APD measurement phantom and calibration methods can be considered validated and fit for the compliance testing of 5G FR2 wireless mobile devices. The developed validation system was published in a peer-reviewed journal publication in [26]. Results have been disseminated to the TC106 Committee of the IEC/IEEE, which has been drafting a technical report and now international standards for the compliance testing of wireless products with APD limits (IEC/IEEE P631953 [28]). The APD validation sources developed in this project became the reference sources in an APD measurement round-robin conducted by IEC/IEEE TC106. International regulators such as the FCC, ISED, and CTTL were updated on the developments through workshop presentations. Table 1 Validation results comparing the surface pAPD and psAPD determined from measurements with the numerical target values for sources at different distances from the skin model. The distance d is defined as relative to the skin surface. 21NRM03 MEWS 24 of 33 Source Signal f (GHz) P_in (dBm) d (mm) Target APD (W/m²) Measured APD (W/m²) Eₙ APD p ps1cm² ps4cm² p ps1cm² ps4cm² p ps1cm² ps4cm² Horn slot array CW 30 15 2 156.9 80.8 43.0 161.0 86.7 53.8 0.1 0.2 0.7 5 49.5 30.7 21.4 47.9 31.9 22.3 0.1 0.1 0.1 Cavity-fed dipole array CW 30 15 2 94.2 30.9 22.4 95.6 35.0 26.3 0.0 0.3 0.4 5 93.7 45.7 25.7 83.6 44.5 24.0 0.1 0.1 0.2 Openended waveguide CW 27 7 2 71.2 23.7 9.2 54.7 22.9 10.0 0.5 0.1 0.2 BPSK 71.2 23.7 9.2 55.5 23.3 9.9 0.5 0.1 0.2 21NRM03 MEWS 25 of 33 4.5 Standardised Measurement Procedure Synthesis & Recommendations [A3.4.6] CENELEC (European Committee for Electrotechnical Standardization) in Europe and the IEC (International Electrotechnical Commission) at the international level are responsible for developing standardized measurement procedures to assess the SAR or APD emitted by equipment placed on the market and verify their compliance with the limits. They have established technical committees (TC106x for CENELEC and TC106 for the IEC). The IEC and CENELEC collaborate, and the IEC develops the technical standards used by CENELEC. CENELEC, for its part, develops technical standards if no work is underway at the IEC level. CENELEC's main objective today is to develop harmonized standards. These standards are developed under European mandates to enable the placing on the market or putting into service of devices that comply with the fundamental requirements of the Radio Equipment Directive.It is therefore essential to use new methods in technical standards. With 5G and future wireless communication protocols, traditional signal-specific linearization methods would require new measurements for each signal variation. The work of this project was primarily disseminated to IEC/IEEE Technical Committee 106, which produced a technical report and, now, international standards for testing wireless products for compliance with APD limits (IEC/IEEE P63195-3). The APD validation sources developed in this project became the reference sources for a series of APD measurements conducted by IEC/IEEE Technical Committee 106. 21NRM03 MEWS 32 of 33 21NRM03 MEWS 33 of 33