ESTIMATING FF RF PERFORMANCE FROM NF PVT CHAMBER MEASUREMENTS

Converting NF measurements to FF is studied by many^5-8, without a satisfactory solution for general purpose, DVT-type measurements. Here, the conversion problem is restricted to a PVT chamber custom-designed for a specific antenna design. A golden unit of the design is available as a reference. First, using the golden unit, a model is built reflecting the NF and FF data. This model is used to estimate the FF performance of a product from its NF measurements in the PVT chamber. The rest of this section will use boresight EIRP estimation as an example to illustrate the algorithm. Other test cases can be carried out similarly.

DVT Chamber with AUT Temperature Control

To characterize mmWave mMIMO radios, Jabil constructed a CATR chamber with AUT ambient temperature control. Figure 3 shows the chamber with a n258 5G mMIMO radio on its positioner. The quiet zone of the CATR is a 700 × 700 mm cylinder around the polarization axis of the positioner. The positioner has five degrees of freedom along with: azimuth: ± 180 degrees, elevation: ± 30 degrees, polarization: ± 180 degrees, x-axis: ± 150 mm and y-axis: ± 75 mm. When the AUT is under thermal cover, a remote thermal system controls the ambient temperature inside the thermal cover. Its temperature can be set from -40°C to +70°C with a 300 W load. The insertion loss of the cover is less than 0.2 dB. Figure 4 shows the CATR with its thermal cover installed.

Figure 3 Radio in CATR DVT chamber.

Figure 4 Thermal cover on AUT in CATR chamber.

NF-FF Conversion Modeling

The first step is characterizing the golden unit behavior in the DVT chamber to build a model for NF-FF conversion. Characterization is done by collecting IQ data from the CATR chamber while the golden unit transmits at a predefined setup, e.g., beamforming into the boresight direction. To capture the temperature dependency of the golden unit, data will be collected at a temperature range (T1~TN), that covers the internal working temperature of the PVT chamber. The IQ signal of the baseband source, X, is predefined by the test case, e.g., two frames of the TM3.1-100 MHz NR waveform. The golden unit data from the DVT chamber are denoted by {Y_T1,Y_T2,...Y_TN}, where {T₁,T₂,..T_N}, are temperatures where data, Y, is collected. All collected data are time-aligned with X.

Next, the golden unit is placed in the PVT chamber and IQ data is collected from all the probes while the golden unit transmits under the same condition as in the DVT chamber. Temperature will be recorded while this data is collected. Let M denote the number of probes in the chamber. For each transmitting condition, there are M sets of data, {Z₁,Z₂,..Z_M}. Each Z_i has the same dimension as Y. All collected data are time-aligned with X.

For a fixed transmitting condition, regardless of whether the golden unit is in a DVT or PVT chamber, the electromagnetic field generated around it should be the same at the same temperature. The propagation and reflection properties of the two chambers mean that the measurements from the probes are different but correlated. This correlation can be modeled by a linear time-invariant filter bank, i.e., the FF data from the DVT chamber is seen as the summation of filter outputs driven by PVT probe measurements. This can be shown in Equation 1:

Where:

* denotes filter operation, e.g., convolution of time sequence in Zi with impulse response of F_i

Letting P_i denote the Toeplitz matrix formed by time sequence in Z_i and F_i be coefficients of FIR filters, allows Equation (1) to be rewritten in matrix form as Equation 2:

Or simplified to Equation 3:

Where:

P=[P₁,P₂,…P_M] and F=[F₁^T,F₂^T,..F_M^T ]^T

Assuming that the PVT data are collected at a temperature T_n, the filter bank coefficients for NF-FF conversion are obtained by solving the least square problem as shown in Equation 4:

To estimate a product’s FF performance, the NF IQ data is collected from the PVT chamber when the product under test is transmitting under the specified setup. Then, Equation 1 is used to convert the data to FF. Test KPIs such as equivalent isotropic radiated power (EIRP) are computed from the FF IQ data.

Data shows that even though mmWave antenna array performance is quite sensitive to temperature change, the NF-FF conversion model does not change as much as the antenna. NF-FF conversion modeling can be done at a much lower temperature resolution.

DVT and PVT Test Data Comparison

Extensive validation has been carried out to verify the accuracy of KPI estimates based on PVT chamber measurements. Table 1 shows test data from DVT and PVT chambers for 10 units. The test frequency was 25.225 GHz and the test ambient temperature was 29°C. The results show that EIRP, error vector magnitude (EVM) and ACP estimation errors from the PVT chamber are within 0.19 dB, 0.21 percent and 0.38 dB, respectively, of the DVT measurements.

PRODUCTION TEST SOFTWARE

Cloud-Edge-Frontend

Figure 5 shows the architecture of Jabil’s mMIMO PVT system. It has three layers of computation and associated software, cloud capabilities, Edge Data Box (EDB) and front-end instrumentation. The master software, running on the EDB, controls test execution and all equipment. It also calculates test KPIs, classifies test results and faults and ports data to the cloud. The front-end instrument does data capture and injection. Data analytics and fault modeling are done on the cloud (Azure), exploiting its AI and machine learning tools. The three layers of software run asynchronously. This allows time sharing of expensive front-end instruments between multiple chambers and KPI calculation during AUT handling.

Figure 5 Jabil production OTA test system with a cloud-edge, front-end hierarchical architecture.

Web App for Data Visualization and Analytics

This system uses J-CloudView, a web application, to assist in data visualization and analytics. It is currently implemented on Microsoft Azure with Jabil’s proprietary mathematics library and signal processing library. Standard APIs are defined to facilitate porting to customer-specified cloud services. In addition to production-related statistics such as yield, Cpk and distribution of product performance measurements, J-CloudView provides visualization tools specifically for wireless products, such as spectrum and complementary cumulative distribution function (CCDF) of IQ data. It also has 3D plotting tools to show key performance across different test conditions for a group of products. For example, Figure 6 shows EVM tests of 100 units at five test frequencies. The plot indicates that the EVM of these products is relatively high at 26.5 GHz. These visualizations can demonstrate a design weakness of the product and enable effective engineering changes.

Figure 6 A 3-D plot of mMIMO TX EVM.

CONCLUSION

High volume and low-cost production OTA test of mmWave mMIMO radios for 5G/6G is a challenge to electronics manufacturers. This paper describes a compact chamber that has been custom-designed for each production line and how each product can be tested and calibrated using data from multiple probes installed in the near field. NF-FF data conversion is modeled using a golden unit. Factory production data shows that this approach is valid and efficient. A cloud-edge, front-end hierarchical software architecture is shown to be suitable for test automation, test cost reduction and production scaling.

ACKNOWLEDGMENT

The authors would like to thank Jabil OTA Test Solution team members for their contributions to the development of this solution.

References

1. T. Takahashi, H. Miyashita, Y. Konishi and S. Makino, “Theoretical study on measurement accuracy of rotating element electric field vector (REV) method,” Electron. Commun. Jpn., Vol. 89, No. 1, pp. 22–23, Jan. 2006
2. R. Sorace, “Phased array calibration,” IEEE Transactions on Antennas and Propagation, Vol. 49, 2001.
3. R. Long, J. Ouyang, F. Yang, W. Han and L. Zhou, “Multi-Element Phased Array Calibration Method by Solving Linear Equations,” IEEE Transactions on Antennas and Propagation, Vol. 65, No. 6, pp. 2931–2939, June 2017.
4. L. Lin, K. Loughran and J. A. Wildt, “Compact anechoic chamber for active and passive antenna over-the-air testing,” U.S. Patent number: 11280821, Filed: Mar 22, 2019, Assignee: JABIL INC. (St. Petersburg, Fla.).
5. M. H. Francis and R. C. Wittmann, Modern Antenna handbook, Chapter 19, Near-Field Scanning Measurements: Theory and Practice, John Wiley & Sons, 2008.
6. M. Laabs, D. Plettemeier, T. Deckert, V. Kotzsch and M. Vanden Bossche, “A novel OTA near-field measurement approach suitable for 5G millimeter wave validation and test,” 2021 51st European Microwave Conference (EuMC), 2022, pp. 564–567.
7. M. Shafiee and S. Ozev, “Contact-less near-field test of active integrated RF phased array antennas,” Journal of Electronic Testing, Vol. 35(3), pp. 335–347, 2019.
8. D. Janse van Rensburg, “Near-field test challenges of high frequency digital phased array antennas,” 2020 International Symposium on Antennas and Propagation (ISAP), pp. 327–328.
9. L. Lin and G. Rebeiz, “Fast Beamforming Calibration of mMIMO Radios – an Information Theory Perspective”, accepted for publication in Proceedings of EuMW2024.