This is the final installment of a guest series written by Walt Strickler of Anritsu Company.
As all engineers know, calibration is fundamental to making accurate measurements. Historically, calibration times have been relatively long, slowing throughput and increasing the cost of test. Knowing when measurement configuration changes require new calibrations is important to minimize test time. This can be especially tricky when making pulsed VNA measurements.
VNA advances and improved techniques, however, have helped to reduce calibration time and how often they must be made. This is evident with the Anritsu VectorStar™ MS4640B Vector Network Analyzer (VNA), which features a high-speed digitizer approach that combines improved measurement resolution and timing accuracy with the ability to independently adjust receiver timing to significantly reduce the need for recalibration and minimize uncertainty.
Reduce Calibration Time, Increase Throughput
PulseView is a new feature in the VectorStar MS464xB VNAs that allows point-in-pulse, pulse profiling, and pulse-to-pulse measurements to be performed in some calibrated sense beyond simple normalization (see my first column for more information on the pulsed measurements). While these are S-parameter measurements and all of the usual calibration algorithms and types still apply, there can be some setup and configuration differences of which one needs to be aware. Calibration implications are heavily dependent on the basic measurement configuration. For example, if a pulsed RF stimulus is not used, the calibration should be performed in a non-pulsed state.
If calibration is under non-pulsed conditions, the point-in-pulse situation is conceptually simple. A frequency sweep (or power sweep) is performed and appropriate error coefficients generated. Changing the measurement width or position after calibration will only affect relative levels of trace noise and dynamic range of the results. The frequency range and point count are constrained as in regular transmission/reflection mode, although interpolation is allowed. For pulse profile and pulse-to-pulse, the calibration is essentially a continuous wave (CW) process, since the same frequency and power are used for the measurements across an axis of time. The same error coefficients, within limits of trace noise, will be generated at each point and applied to the measurement result. Again, measurement widths and positions can be changed after the calibration with only relative changes in trace noise or dynamic range. The number of points, however, cannot be changed, as this would alter the indexing of the error coefficients.
If stimulus pulsing is needed, the situation can be slightly more complex. The central issue is the extent an engineer wants to calibrate out dynamics associated with the stimulus modulation process. One aspect to consider is how the reference signals are handled. Some VNA pulsed measurement configurations, such as those offered by Anritsu, have a post-modulator reference coupler that allows pulsed references to be used. Alternatively, a non-pulsed reference can be used by not connecting these paths. The resulting S-parameters are different in these two cases, as suggested in Figure 1. This non-calibrated measurement shows the behaviors of the ratio. Since the paths of the reference are different in the two cases, the absolute S21 value changes slightly. The central point is again the transition behavior.
Figure 1 reveals the heart of the matter in looking at the edge behavior. When using the pulsed reference, if there are transients associated with the modulation, they may cancel in the ratio process, assuming timing is aligned. The result is less distortion in the device under test (DUT) overshoot area. The pulsed reference also tends to reduce video ringing in the on-state of the pulse in a raw sense, so calibration has less work to do. The pulsed reference does, however, make it much harder to look at rising edge and falling edge details (particularly further down the edges), since the reference is changing rapidly at those times. The choice may depend on what part of the pulse is of the most interest.
Another reason to consider a pulsed reference is the question of stability in complex test setups. If pre-amplifiers, external modulators, or other external structures are used, stability could be somewhat impaired by the addition of the network of test accessories (figure 2) or by long runs of external cabling. By having the reference taken after these networks, which normally means it will be pulsed if a pulsed stimulus is used, much of the potential drift behavior will ratio out, thereby reducing the frequency of calibrations.
Calibration Approach Effects on Uncertainty
The overall uncertainty picture in a pulsed measurement is a complex interaction of traditional VNA uncertainty terms (residual source match and directivity, tracking, linearity, repeatability, etc) and pulse-specific terms (bandwidth and resolution limitations, timing errors, et al). Possible variations in calibration approaches are an additional complication in this scenario.
Consider a pulse profiling measurement with a non-pulsed reference, where the pulse is applied during calibration. The uncertainty can be evaluated in a Monte Carlo sense as an example case of a DUT with a fairly narrow overshoot, relatively good return loss, and not enough output power to cause system compression. The result of this example calculation is shown in Figure 3. The dotted line represents the nominal result, and the other colored traces represent possible outcomes, as various phases and error terms (such as repeatability) are randomized. The spread of results is fairly contained in the middle of the pulse, increases a little in the overshoot region, and expands more on the rising and falling edges.
The same calculation with a non-pulsed calibration is shown in Figure 4. In this case, the modulator overshoot is not corrected, so there is a deterministic error on the rising edge in addition to somewhat elevated scatter. There is also a bias and increased scatter on the falling edge, since those modulation characteristics are not taken into account. It should be pointed out that this is something of a worst-case example, since the pulse width is relatively narrow and the DUT had a narrow overshoot on the same scale as that of the modulation. The uncertainty change in the middle of the pulse or for other time epochs is insignificant.
Electrically Large Setups
Some additional calibration complications can be introduced in electrically large setups (e.g., long cable runs between parts of the measurement apparatus), and there is a distinction based on which parts of the setup the electrical length is added. In both cases, it is only when the electrical length change is significant to the pulse measurement resolution that action might be warranted. Thus, for 2.5 ns resolution measurements, length additions on the order of 1 m may become important.
Knowing when and why it may be necessary to re-calibrate your VNA for making pulsed measurements can help minimize test time. Calibration is simply one area that has been improved through new VNA designs. Our previous posts have outlined how today’s VNAs are better equipped to meet the requirements of conducting pulsed measurements and testing modern radar systems.
If you’d like to learn more about VNA advances and how they are helping engineers meet today’s high-speed design challenges, Anritsu has developed a VNA microsite with focuses on radar, signal integrity, on-wafer, and components.