Microwave Journal
www.microwavejournal.com/articles/35245-toward-augmented-ota-testing-bringing-full-wave-numerical-modeling-and-antenna-measurements-together

Toward Augmented OTA Testing: Bringing Full-Wave Numerical Modeling and Antenna Measurements Together

January 14, 2021

Evaluating the realistic performance of antennas integrated into complex setups has been a long-term challenge, currently relevant to automotive connectivity, for example. Numerical electromagnetic (EM) modeling offers a cost-effective alternative to device testing but is often limited by an incomplete or inaccurate knowledge of the device under test (DUT). While this shortcoming does not apply to test approaches, the implementation of adequate measurement systems, such as full-vehicle over-the-air (OTA) test setups, requires large investments and, possibly, a significant amount of space. We present a way to converge both worlds in an optimal cost-benefit technique by combining antenna measurements with full-wave simulation.

Three-dimensional EM modeling tools have seen extraordinary technical and commercial development since the early 90s. Now, there is almost no problem involving Maxwell’s equations which cannot be solved with the right software and computational resources. Nevertheless, the need for RF measurements continues to grow, and simulations are not foreseen as a substitute for testing soon. The main reason is that full-wave simulation, even performed by an expert, is only as good as the knowledge of the problem modeled. In most circumstances, this knowledge is incomplete, even for the device manufacturer. Taking the example of a mobile phone, some components come from suppliers, material dielectric properties are not fully characterized, manufacturing tolerances and operating uncertainties of the components are not completely accounted for, etc.

Measurements do not suffer from a similar lack of knowledge, as no simplifying assumptions are required about the DUT. However, the complexities of experimental setups, including realistic integration or usage conditions, must be considered. For example, for autonomous driving cars using cellular vehicle-to-everything (C-V2X) technology for data communication, the connectivity in cars is developing to unprecedented levels. Evaluating the accurate performance of all transceivers (i.e., cellular, WLAN, GNSS, etc.) necessitates including the impact of the complete car on all radiated emissions. To address these aspects, the 5GAA organization is currently standardizing the use of full-vehicle OTA testing systems using 10 m or larger anechoic chambers (see Figure 1).1 The corresponding costs are as substantial as the space requirements, and the realizable test scenarios are still limited. A close to real-world test solution would require unreasonable effort, time and cost for measuring the OTA performance of an integrated 5G mmWave antenna module, and no test chamber can assess the effects of various grounds on the measured metrics or an actual link budget in a car-to-car transmission.

Figure 1

Figure 1. R&S full-vehicle OTA testing system based on draft 5GAA requirements.

To overcome the limitations in both antenna simulation and measurement, the best of both numerical and experimental methods can be combined: measure the part of the problem which cannot be accurately modeled and compute the portion which is too costly or complex to measure.

MEASUREMENT + SIMULATION

Figure 2

Figure 2. Equivalent source calculated from complex voltage measurements over a sphere.

Figure 3

Figure 3 Steps in the measurement-simulation technique.

Combining measurement and simulation is a straightforward outcome of the fundamentals of wave physics. From the Huygens-Fresnel principle or the equivalence theorem, the six components of the electric (E) and magnetic (H) field vectors outside of a volume, Vs, encompassing a radiating source within a surface, Σs, can be exactly computed from only two of the EM field components over Σs or the associated surface electric and magnetic currents J and M (see Figure 2). Conversely, proven techniques can solve the inverse problem: calculating the currents J and M from two characterized components of E and/or H over a closed surface, Σm, containing Σs.2-5

Imagine that Σm is a sphere surrounding an antenna. A test probe scans Σm and delivers two voltages at each sampling point, which relate to the local, orthogonal, spherical coordinate phasor components Eθ and Eφ. Knowing the magnitude and phase of these voltages over the sphere is then sufficient to derive a model of the source, equivalent to the antenna under test, in the sense that the model generates the same E and H fields outside of Σs. This principle is not limited to a spherical surface; a sphere matches the results where the measurements are obtained using a spherical test range.



Once this new radiating object is created, there is no need to know the details of the DUT to model its operation and integrate it into a more complex environment. The new equivalent source can be imported in an EM simulator. Additional conditions can be introduced, with more quantities computed to simulate the influence of nearby scattering objects on the radiation pattern, the link budget with another antenna, the exposure of a person standing in the near-field and so on (see Figure 3).

Figure 4

Figure 4. Anechoic chamber with spherical scanner, measurement probe supported by the elevation arm, and DUT positioned at the azimuth pole.

Although the three steps of Figure 3 may be implemented in multiple ways,6 we detail one as an example. In this example, the measurement data are acquired (step 1) with a system comprising an anechoic mobile antenna measurement chamber and a vector network analyzer (VNA). The Rohde & Schwarz anechoic chamber shown in Figure 4 has tip-to-tip dimensions of 0.64 x 1.25 x 0.93 m between the absorbers and contains a distributed-axis positioner with a 360-degree rotating azimuth table to support the DUT. A gantry arm carries the measurement probe and provides elevation from 0 to 165 degrees. The mechanical precision of this spherical scanner is better than 0.03 degrees in both dimensions. The probe is a dual-polarized Vivaldi antenna, suitable for direct far-field and spherical near-field measurements from 12 to 90 GHz. The probe tip retains a separation of 50 cm from the center of the coordinate system, with the center accurately located using built-in lasers. The center of the DUT can be accurately positioned at this point using a crank, which enables fine DUT height adjustment.

In this test setup, three ports of the VNA are used: Port 1 feeds the RF signal to the DUT via a cable through the azimuth pole. Ports 2 and 3 collect the received signals simultaneously, connecting to a feedthrough at the chamber side and two cables running to the two polarization ports of the probe. To avoid cable movement, rotary joints for both azimuth and elevation are integrated into the chamber. A full measurement obtains S21 and S31 (i.e., magnitude and phase) for all desired angular positions.

Step 2 is performed using the Fast Irregular Antenna Field Transformation Algorithm (FIAFTA).2 The FIAFTA, which was developed at the Technical University of Munich, implements a generalized minimal residual equation solver for reconstructing the equivalent currents over a triangular mesh covering Σs from the measured S-parameters. The reference complex far-field of the test probe is injected in the algorithm to include a full probe correction. The resolution of the inverse problem is dependent on the applied domain boundary conditions. This study applies the “Huygens radiator” option, which assumes a fixed free-space impedance boundary condition at Σs. This approach is computationally efficient at delivering solutions for the emissions, predominantly radiating toward the outside of Σs. Subsequently, Σs was chosen to be a rectangular box closely encompassing the DUT.

In step 3, the box of equivalent currents is used as an EM source in the finite difference time domain (FDTD) software EMPIRE XPU by IMST GmbH.7 To correctly import the data, equivalent currents are interpolated so the triangular mesh is converted into a cube. The original triangulation must be fine enough to ensure the stability conditions of the FDTD are fulfilled. A sampling of λ/15, where λ is the free-space wavelength, meets this requirement.

FULL-VEHICLE APPLICATION

Considering the full-vehicle OTA case, the measurement system shown in Figure 1 can be used to evaluate the performance of a 5G mmWave module integrated into a car; however, it is costly and time-consuming. First, the cost of a large positioning system with a mechanical accuracy meeting the requirements of near-field testing at this frequency range are exorbitant. Second, the car is unlikely to be positioned on the turntable with the antenna module at the center of the coordinate system. With an offset and the small wavelength, the probe would have to scan the upper hemisphere -- at least a large portion of it -- with angular resolution of less than 1 degree. The test time would be extremely long, as distributed-axis positioners of this size have typical speeds of 6 to 12 degrees/s in azimuth and 1 to 2 degrees/s in elevation, respectively. On the other hand, only performing a full-wave simulation of the antenna integrated within the car is not completely suitable, as the simplifying assumptions would not guarantee a suitable model of the real-world installation.

Figure 5

Figure 5. mmWave antenna array integrated in the rear-view mirror.

To show the utility of combining measurement and simulation, we study a 5G mmWave transceiver integrated in the rear-view mirror behind the windshield of a car (see Figure 5). Measurements of the antenna module are used as an input to a full-wave computation, where the module is virtually integrated into a numerical model of the car. The approach uses the Rohde & Schwarz anechoic chamber measurement system with FIAFTA and EMPIRE XPU processing, previously described.

The DUT is an 8 x 8 antenna array with a mmWave front-end module operating at Ka-Band (see Figure 6).8 The array uses a 1:8 divider to distribute the signal from one waveguide interface to 16 beamforming ICs. Each IC feeds four, dual-port circular patch antennas, providing individual phase and amplitude control of every antenna element. Both physical and full numerical versions of the array were used. The numerical model, including all accessible knowledge of its architecture and key parameters, was used to design the array.

Figure 6

Figure 6 Antenna side of the 8 x 8 phased array (a). Simulation of the IC side of the array, showing the E-fields at the IC and antenna feeds (b).

Figure 7

Figure 7. 29 GHz azimuth patterns comparing the near-field to far-field processed results (red) with the pattern reconstructed using the combination technique (blue). φ = 0° plane (a); φ = 90° plane (b).

During the measurement step, the complete sphere around the DUT was scanned with 2 degree angular steps in both azimuth and elevation. To verify the results, two methods were used to compute directivity. With the first method, the acquired near-field data was processed using FIAFTA. The algorithm created the equivalent current box, which was then imported into EMPIRE XPU, where the fields were propagated via FDTD through the complete computational domain. With the second method, a far-field multi-level, fast multipole method translation, also supported in FIAFTA, was used to directly compute the far-field radiation pattern from the measurement data.9 The results of the two methods, compared in Figure 7, agree across the complete angular region. Observations of additional near-field distributions at various locations show complete consistency between the fields for Σs obtained only with FIAFTA and the combination with EMPIRE XPU.

In the next stage, the FDTD software was used to simulate the antenna integrated in the windshield of the car, shown in Figure 5. The computations were performed using two approaches: 100 percent numerical, where both the antenna and the vehicle were modeled, and hybrid, where the combination technique was employed, with the array represented in the EM simulation by its measurement-based equivalent source. Figure 8 compares the two, showing the near-field E-field magnitudes in a defined plane at 29 GHz. The field distributions agree with some differences, especially in the region between the array and the windshield. Figure 9 compares the far-field directivity patterns, showing reasonably consistent results when φ = 0 degrees and more deviations, especially in the sides lobes, when φ = 90 degrees.

These differences are expected. The FDTD model of the array does not account for production tolerances or tolerances in the operation of the electronics. Other error contributors include DUT modeling errors (e.g., imperfect knowledge of dielectric properties), measurement uncertainty and boundary conditions applied at Σs for deducing the radiating currents. A more systematic uncertainty evaluation is required, which will be the subject of future investigation.

At this point, neither of the two computational approaches delivers better results, i.e., closer to the real-world implementation. The hybrid or combined approach yields much higher computational efficiency than the 100 percent numerical analysis. Simulating the full EM model of the array is challenging due to the fine details in the multi-layer structure of the 5G module and the computational complexity of the scenario. Simulating the full model with 3600 million FDTD cells requires a computational time of 180 minutes and RAM usage of 108 GB. As these fine details disappear in the hybrid model, the simulation complexity drops to 320 million FDTD cells for the same calculation volume, requiring a calculation time of only 15 minutes and 10 GB of memory.

Figure 8

Figure 8. Near E-fields at 29 GHz using a 40 dB scale: cut-plane orientation with respect to computed model (a), FDTD-only calculations (b) and combination technique (c).

CONCLUSION

Figure 9

Figure 9. 29 GHz azimuth patterns comparing full-wave computation only (blue) with measurement–simulation (red). φ = 0° plane (a); φ = 90° plane (b).

As the need to improve antenna design and performance in real-world applications is constantly growing, as is the complexity of the scenarios, converging measurements with the power of numerical computation seems an unavoidable way forward. This article presented an augmented measurement or augmented simulation technique, exploring this approach and revealing multiple advantages. Numerical computation can be made more realistic, as no a priori and in-depth knowledge of the DUT, including manufacturing tolerances, is required. Calculations are also more efficient - by 12x in speed and 10in memory in the example described - as much of the complexity in modeling the radiating source is removed. Measurements can be more cost and time effective. In the example, an expensive, full-vehicle OTA facility can be replaced by a 1.3 m2 spherical scanning anechoic chamber, used just to measure the array module. Integration of the module in the car is obtained by processing. Future articles will demonstrate such techniques for other applications, involving measurements of active devices transmitting modulated signals and including more systematic evaluation of uncertainty contributors.

References:

  1. 5G Automotive Association, Web: 5gaa.org.
  2. T. Eibert et al., “Electromagnetic Field Transformations for Measurements and Simulations (invited paper),” Progress In Electromagnetics Research, Vol. 151, Dec. 2015, pp. 127–1250.
  3. R. A. M. Mauermayer, Y. Weitsch and T. F. Eibert, “Electromagnetic Field Synthesis by Hierarchical Plane Wave-based Field Transformation,” IEEE Transactions on Antennas and Propagation, Vol. 63, No. 12, Dec. 2015, pp. 5561–5572.
  4. O. Neitz, R. A. M. Mauermayer, Y. Weitsch and T. F. Eibert, “A Propagating Plane-Wave based Near-field Transmission Equation for Antenna Gain Determination from Irregular Measurement Samples,” IEEE Transactions on Antennas and Propagation, Vol. 65, No. 8, June 2017, pp. 4230–4238.
  5. T. F. Eibert, “A Diagonalized Multilevel Fast Multipole Method with Spherical Harmonics Expansion of the k-space Integrals,” IEEE Transactions on Antennas and Propagation, Vol. 5, No. 2, Feb. 2005, pp. 814–817.
  6. A. Cozza, B. Derat, N. Ribiere-Tharaud, “A Near SAR Assessment Procedure for Homogeneous and Heterogeneous Flat-phantoms Based on Near-field Free-space Measurements,” Annual Symposium Antenna Measurement Tech Assoc. (AMTA), Nov. 2007.
  7. A. Lauer, W. Simon, A. Wien, “XPU Technology for Fast and Efficient FDTD Simulations using Modern CPUs Cache Memory Bandwidth,” European Conference on Antennas and Propagation, Vol. 13, No. 15, Mar. 2015, pp. 2584–2589.
  8. W. Simon et al., “Highly Integrated Ka-Band Frontend Module for SATCOM and 5G,” Asia Pacific Microwave Conference, Dec. 2019.
  9. A. Tzoulis, T. F. Eibert, “Fast Computation of Electromagnetic Near-fields with the Multipole Method Combining Near-field and Far-field Translations,” Advances in Radio Science., Vol. 4, Sept. 2006, pp.111–115.