The 2D21 Thyratron: From Historical Curves to SPICE Models
18. From SPICE macromodels to experimental checks
A simulation can reproduce the shape of a pulse at the load, the grid potential or the energy loss in the switch. Each comparison tests a different part of the model. The following readings reconstruct which quantity the authors actually compared.
Sources examined on this page
- Barnes and Wait (1993). An equivalent network that turns on in stages
- Barnes and Wait (1994). The comparison moves to the magnetic field
- Zabb Csaba. What the note and the two simulation files show
- Giux-Lab. A model identified on the GE conduction curves
- Korolev and coworkers (2018). The auxiliary discharge before the command
- Kruglov and coworkers (2021). Where energy is dissipated during opening
- Kim and coworkers (2023). Measuring the jitter of the tube–drive assembly
- Han and coworkers (2023). A flatter pulse can have a slower edge
Barnes and Wait (1993). An equivalent network that turns on in stages
In the CX1171 the whole voltage is distributed across several gaps. When the first becomes conducting, the gaps still off also change voltage. Their capacitances allow displacement current to pass towards the external circuit. The model must therefore describe both how the impedance of each gap decreases and how the electrodes are coupled to each other. In figure 1 a controlled source generates the drop of a branch with variable effective resistance. In figure 3 the drift spaces are lumped with adjacent gaps.
One check consists in changing the capacitance to ground and observing how the predicted cathode current changes, as in figure 4. It is a prediction sensitive to the topology of the model. The contribution to the history of the 2D21 is methodological: a terminal current during commutation can include charge displaced in the electrodes even before the main path has been established. The number of gaps and the parameters of this work belong to the multigap tube.
A Mathematical Model of a Three-Gap Thyratron Simulating Turn-On, figures 1–5
Barnes and Wait (1994). The comparison moves to the magnetic field
The same authors move from the subcircuit to the prototype transmission-line magnet. Their criterion concerns the time needed for the field to go from 1% to 99% of its final value. Even a small disturbance before the main edge can bring forward the first crossing and lengthen the measured time. The precursor arises from the displacement current predicted when the three gaps of the thyratron turn on in sequence.
In the ten-cell prototype they compare measurements and PSpice calculations with different saturable ferrite geometries and a network that modifies the edge. The final quantity is the field useful to the beam. It depends on the switch, the wiring and the magnet. Agreement at this level proves the model’s ability to describe a system effect, while the verification of the internal laws of the single tube remains distinct.
EPAC 1994, Effect of Saturating Ferrite on the Field in a Prototype Kicker Magnet, pp. 2544–2546
Zabb Csaba. What the note and the two simulation files show
The note shows two uses, a relaxation oscillator and a time base. The author’s catalog identifies the model distributed for the PL21, 2D21 and 21TE31. To understand which tests can actually be run, however, the archive is needed. The netlist exposes A, G and K, adds capacitances between the terminals and represents the ON state and recovery through controlled sources, limiters and an RC memory. The shield is tied internally to the cathode. The distributed version therefore lacks a terminal to which the different G2 voltages of the factory plates could be applied.
The Bg source contributes to the grid current with a piecewise law. A second branch depends on an internal state that meets a limit. The quantitative reading on page 10 shows why the dependence of that contribution on anode current becomes weak already at a few milliamperes. This detail can be checked in the code even when the external waveform of the oscillator looks convincing. The note demonstrates that the examples work. The files allow the reader to repeat the simulation and to establish which measured families it can be tested against.
Thyratron Macro Model, the author’s examples · Catalog of SPICE models · Thyratron.zip archive with .CIR and .TSM
Giux-Lab. A model identified on the GE conduction curves
The work starts from repeated digitizations of the same plate. For each anode current it derives a consensus trace and an estimate of spread, locates the zero of the grid current and treats the two branches separately. Vertical normalization by the anode current is not enough for all the curves. The voltage scales and the position of the zero provide degrees of freedom that can be checked directly against the starting figure.
Two families illustrate why the deviations should be published both in amperes and relative to the uncertainty. In the negative branch at 300 mA the absolute deviation is large, but the printed curve is also less well determined. At 100 mA a smaller absolute deviation weighs more relative to the spread of the readings. The test on sampling density checks the numerical stability of the digitization. The ability to predict a family never used in the identification and agreement with measurements on other tubes remain two further checks.
Compact model of the grid current, data and quantitative verification
Korolev and coworkers (2018). The auxiliary discharge before the command
In the cold-cathode thyratron TPI1-10k/50 the sequence starts from a weak auxiliary discharge already present in the deuterium. The command and the main commutation can change its regime. The authors record the current–voltage characteristic of the auxiliary unit and superimpose waveforms of command, anode voltage and anode current. Figure 5 links different traces to different initial conditions. Figure 6 makes the time spread of the breakdown visible.
Around 40 kV the total delay reported is about 100 ns, with variations within 15 ns. The fact that the spread depends on the starting regime gives a physical meaning to what would otherwise appear as timing noise. For the 2D21 the question remains valid of how to prepare the initial state and define its repeatability. The specific mechanism of the deuterium device includes an auxiliary unit that the small RCA tube does not have.
Plasma Physics Reports 44, 110–117, figures 4–6
Kruglov and coworkers (2021). Where energy is dissipated during opening
The device has a sector-shaped hole in the shield and works as an opening switch in an inductive-storage circuit. The authors approximate the current and voltage traces during opening. The instantaneous product of the two quantities is the power dissipated at the terminals, provided the passive sign convention is kept. Integrating over distinct intervals they compare the losses accumulated in conduction and those concentrated in the commutation. Figure 5 compares the approximations with the oscillograms from which they derive.
Under the conditions of the work the split reverses as the peak voltage increases: from about 30 kV upwards the authors report switching losses greater than conduction losses. The threshold is specific to this configuration. The comparison is useful to understand why a current-only waveform, however well reproduced, can leave a considerable error in the energy estimate.
Kim and coworkers (2023). Measuring the jitter of the tube–drive assembly
The measured time runs from the generator command to the start of the current rise in the L4888. The bench includes three separate supplies: trigger, cathode heater and reservoir heater. The authors check the supplies separately and then observe how the spread of the responses changes as the drive is varied. In the conclusions they report about 341 ps in one of the configurations with the new heaters and 152 ps in the final test. Figure 12 specifies that the indicator is a standard deviation.
The time reference passes through several devices. Attributing a share of the jitter to the tube alone would also require knowing the spread of the command at the electrode. The difference from Korolev concerns both the architecture and the statistic used: an interval within which the deviations fall and a standard deviation require different information to be compared.
IEEE Transactions on Plasma Science, figures 9–12
Han and coworkers (2023). A flatter pulse can have a slower edge
The authors’ problem is to obtain a trapezoidal pulse at the load with a realizable pulse-forming network. They compare a third-order series-resonant network and an antiresonant one, transform the component values and verify in simulation the correspondence of the waveforms. They then vary the harmonic coefficients to make the top of the pulse flatter. Figures 12 and 13 show how the family of waveforms changes as those coefficients vary.
Figure 14 also documents the cost of the improvement. After optimization the plateau is flatter, but the rise is slower and the requirement on the edge is not met. The authors postpone the physical verification of the components to the next phase. It is a significant example of how two criteria on the same pulse can lead to different judgments. To use a comparison of this kind in evaluating a macromodel, one must specify which part of the waveform was actually optimized.