Research On Calibration Techniques For Multiport Vector Network Analyzer

The vector network analyzer (VNA) has, for almost six decades, found frequent applicationwithin microwave measurement. Calibration and error correction have always been acknowledged tobe one of the key technologies. It is well known that measurement devices can hardly perform ideally,especially for the VNA, which works from several MHz to tens of GHz, the perfect characterizationand excellent consistency can not be achieved within such a wide frequency range, hencemeasurement errors is absolutely inevitable. Therefore, it is not feasible to emphasize improvement ofhardware performance blindly without considering the design difficulty and product costs. Inconclusion, the practical solution is to calibrate the measurement error with proper method toapproach the accurate result.With the frequent application of multiport components in millimeter-wave engineering, thewhole network parameter matrix of the device under test (DUT) can be deduced after severalmeasurements by the two-port VNA, hence will result in calculation complexity and accumulatederror, while with the help of multiport VNA, the whole network parameter matrix can be achievedwith single measurement. Therefore the demand of the multiport VNA increases explosively. However,its hardware structure and error models are more complicated than two-port VNA, hence thecorresponding calibration and error correction is far more difficult. In this paper, calibration methodsare proposed and studied for n-port VNA (n≥2), accordingly the research works and innovations are asfollows.Two calibration techniques for two-port VNA with three measurement channels are studied. Thefirst one is based on the10-term error model using the concept of T-matrix. In the calibrationprocedure the system errors are not required and the length-unknown50transmission line can beused as a calibration standard. By this method the measurement cost is decreased and the calibrationprocess is simplified. The second one is based on8-term error model with the offset error of theswitch. Because the calibration results based on8-term error model only considering the reflectioneffect of the switch are obviously worse than those based on the12-term error model, the8-term errormodel is further modified from the physical situation and corresponding formulas are deduced forcalibration and error correction.Three calibration techniques for n-port VNA are proposed. The first one is based on4n-termerror model with the switch offset error term at each unstimulated port. Fully considering theimperfect switch effect, the new error model is closer to the actual situation. The error correctionequation is also simplified by the matrix operation. The second one is a calibration algorithm based ongeneral node equations. With the crosstalk error term in the error model for high precision, themathematic relationship between calibrated S-parameters and raw S-parameters can be obtained byonly three general node equations. Thus the formula can be easily deduced for error correction. Thethird one is proposed for n-port VNA calibration by the general6-term error model. Using the general6-term error model to describe3n~2-term error model for n-port VNA, the nodes related to unexcitedports are rearranged as the complex vectors and coefficients in signal paths are accordingly replaced by the complex matrixes. From the node equations deduced from the general6-term error model, thescattering parameters of n-port device can be solved by a unified formula.The calculation procedure for the uncertainty of multiport VNA S-parameter measurement due tonon-ideal calibration standards is discussed. Only considering measurement uncertainties come fromnon-ideal SOLT standards, type B standard uncertainty components can be estimated by theuncertainty information of SOLT standards. Based on general node equations, the mathematicrelationship between the deviations of the S-parameters of DUT from their true values and of theS-parameters of non-ideal calibration standards from their ideal values can be got. Finally, thesensitivity coefficients can be obtained for establishing each standard uncertainty component.