I wanted to evaluate the typical error of common no-name SOLT calibration kits on the market, I purchased one set of self-characterized kits from a vendor and several sets of no-brand kits. Initially, I just wanted to do a simple comparison and publish the calibrated parameters online to help users of these no-brand kits calibrate their network analyzers.
-
During the first experiment, I found the data was inconsistent. I eventually traced it to poor contact on the VNA port - possibly non-compliant or damaged. So, I completely disassembled the VNA and replaced the connectors to continue testing.
-
However, since the VNA had been disassembled, hidden faults might now affect data reliability. Also, its frequency limit of 4 GHz couldn't cover the entire sub-6 GHz range I needed. I bought a 6 GHz analyzer for the experiment.
-
Realizing that non-compliant connectors might damage equipment (likely the cause of the previous instrument failure), I purchased a connector interface gauge. I studied IEEE, IEC, MIL-STD, HP, and Maury standards, compared tolerances across 8 standards, identified “safe” parameter values, and posted a tech memo.
-
During the experiments, I discovered a potential algorithm called SDDL that can derive SOLT parameters without a full SOLT calibration - allowing open standard self-calibration. To verify its accuracy, I bought a used Agilent 85033E for comparison.
-
To simplify parameter extraction from no-brand calibration kits, I developed a tool called EasySOLT that uses equivalent circuit fitting, allowing users to characterize unknown SOLT kits using a calibrated VNA.
-
After fitting experimental data with EasySOLT, I realized equivalent capacitance and electrical length were nearly indistinguishable, producing many physically meaningless fits. For better rigor and physical interpretability, I decided to fix the reactance based on theoretical capacitance values and only fit the electrical length.
-
While reviewing literature, I found conflicting methods for calculating the equivalent short standard. After comparing four papers and trial-and-error, I finally identified the correct method.
-
I then looked for ways to calculate the parasitic capacitance of the open standard. I initially spent three days struggling with a classic paper that used a Bessel-Neumann function in a numerical solution - only to discover I had misunderstood a math table. Eventually, I got usable results, but discrepancies remained. I switched to a simpler polynomial fit from another paper, but results were still incorrect. After reviewing more papers by the same author, I found a printing error - one parameter was off by a decimal. Fixing that gave me the correct theoretical value.
-
In the process, I realized that some “no-brand” open standards with lengthened shield body were not just arbitrarily made - they were actually based on classical electromagnetic theory of “coaxial-to-below-cutoff circular waveguide transitions” (from MIT Rad Lab), which once served as a 7 mm metrology standard and still studied in material science today as a dielectric measurement probe. This greatly boosted my confidence.
-
I added the open and short models to EasySOLT and was ready to wrap up - until I noticed the model failed to fit. The theoretical and experimental open capacitance values were wildly inconsistent.
-
Comparing various other papers confirmed my theoretical value was correct. So I turned my attention to the possibility that real-world open standards include non-ideal physical features (like hollowness, sharp pins, or step changes) not present in ideal coaxial opens.
-
I wrote my own simplified connector simulation script based on openEMS for full-wave simulation.
-
But soon, I encountered anomalous results. After a week of debugging, I found a bug in openEMS. As one of its core contributors, I fixed the bug and continued modeling.
-
The simulation results varied inexplicably with different input parameters. I suspected I had misread the mechanical drawings in the standard. So I decided to verify with real hardware. Since the end of microwave engineering is always mechanical engineering, I bought a vise and destructively disassembled the cheapest SMA through-connector I could find. I also bought a contact micrometer and 1-micron pin gauges to measure its mechanical dimensions. It turned out my modeling was correct, and I had read the drawings right. Incidentally, I also discovered why these cheap through-connectors often jam and damage ports. While I had long distrusted these flange-less, torque-wrench-incompatible, easy-to-jam adapters, I finally learned the root cause: one end of the adapter was 0.05 mm below the standard diameter, in violating to spec - so much so that even the connector interface gauge couldn’t enter. Only a pin gauge could measure it. This was likely intentional, increasing friction to prevent both ends from rotating simultaneously.
-
Because I had been using trial and error, it was hard to identify causes. So I redesigned the experiment systematically, ran dozens of simulations to compare parameter effects, and confirmed the simulation was accurate and close to literature values. The likely problem was parasitic effects from connectors.
-
The new direction was to independently determine the open capacitance using an S21 resonance-based method found in literature, which simulation also validated.
-
Wanting a reliable, independently verifiable method, I spent a week writing and debugging a Monte Carlo simulation script to calculate the error bounds. To my surprise, the method was highly sensitive to uncertainty - under worst-case assumptions, the experimental results were meaningless.
-
I purchased a large variety of coaxial cables, connectors, and adapters to assess real-world error contributions. If results were consistent across different setups, it would mean real-world errors weren't that significant.
-
While evaluating S21, I used this chance to apply the SDDL algorithm to measure open capacitance and reassess SDDL. The results were completely invalid - but only for open capacitance. Measuring DUTs, including calibration kits, worked fine. This strongly suggests that connected vs. unconnected states of the port introduce different parasitic effects.
-
I then replaced the cheap SMA adapters with 3.5 mm ones and re-ran the initial experiment. Sure enough, results shifted significantly - much closer to theoretical values, but still off by as much as -40%. This further confirmed the critical role of connector parasitics.
-
The research now shifted toward connector parasitic parameters. I placed a high-priced order for a Precision Slotless Connector (PSC) version of a 3.5 mm connector to continue experiments - still waiting for delivery.
-
At the same time, I'm preparing to construct a simple air-line structure to test whether inserting the center pin causes mechanical and parasitic changes. This is to validate the hypothesis that “the parasitic parameters of a connected port (with pin) vs. a disconnected port (without pin) are not time-invariant.”
-
By this point, I’ve read dozens of papers published on this topic, from WW2 to the 21st century, including both standard results and little-known papers.
What started as a simple personal project to measure and publish some parameters has ballooned in complexity, budget, and workload. So far, I've spent over $3000 - and may now be one of the world’s top experts on the seemingly utterly useless subject of “a coax connected to nothing" in the amateur radio community.
Ironically, I never solved the original “simple” question. Instead, I’ve concluded: no strict characterization or analysis can be performed until the center pin insertion effect is fully understood. This might also explain why no commercial 3.5 mm or SMA open standards use pinless circular waveguides. (I even found a suspected early Maury Microwave open standard on eBay, 1980s [?] which required manual pin insertion - possibly to avoid operation without a center pin)
