Driving the Dee: RF Coupling

Getting kilovolts of RF onto a dee is an impedance-matching problem: a modest transmitter must drive a high-Q resonant load whose tuning moves as the machine warms up. This page covers the architecture choice (self-excited oscillator or driven amplifier), the coupling choice (inductive loop or capacitive tap), the discharges that keep the voltage from rising, and how to measure what the dee is actually doing. The matching arithmetic itself lives in the dee capacitance & matching calculator; RF power-amplifier construction is its own well-documented amateur-radio discipline, treated here as a protected 50 Ω source with pointers in Go deeper. The hazards of an energized RF system — exposure, burns, stored energy — are covered in Safety, which belongs before the first watt, not after.

The dee is a load only a resonator can drive

Electrically, a dee is a capacitor: a metal electrode facing grounded liners across small gaps, plus a stem carrying it through the chamber wall. Wouters' design memo for small cyclotrons puts it flatly: “the dee-to-ground capacity appears as the major portion of the capacitance in the tank circuit, which must be calculated taking this into account” (UCRL-476, p. 8). Typical small-machine values are tens of picofarads — the Rutgers 12-inch dee measured 78 pF against a parallel-plate estimate of 77.5 pF (Koeth 2005) — and an inductor across that capacitance completes a tank circuit resonant at the cyclotron frequency. For protons that frequency is 15.2 MHz per tesla, so iron-magnet machines (0.3–1.3 T) live between roughly 5 and 20 MHz.

Why resonate at all — why not connect the transmitter straight to the dee? Because the dee needs kilovolts and the transmitter has watts. The Oak Ridge ORIC designers state the principle: “The high dee voltage required in cyclotrons can be achieved for practical driving power only by using a high-Q resonant circuit” (ORNL-2648, p. 17). At resonance the tank recirculates energy, and the voltage it rings up to is set by the drive power against the tank's losses:

where P is forward power at match, L and C the tank inductance and capacitance, and R_s the effective series resistance of the whole RF path. In practice: voltage grows only as the square root of power, and every milliohm of loss is paid in volts. Koeth derived and then tested this on the Rutgers machine: dee voltage followed the square-root-of-power law over every range measured, 5 W to 1300 W (Koeth 2005, pp. 2–3). The formula also predicts the machine's measured record honestly — with the Rutgers values (L = 1.1 µH, C = 78 pF, R_s = 0.8 Ω), 2 kW forward gives 8.4 kV peak, exactly the operating record reported (Koeth et al. 2010, p. 19).

Two warnings hide in that formula. First, R_s is a property of the whole system, and it is much worse than the coil alone: the Rutgers tank coil computed 50 mΩ from handbook copper-tube values, but the assembled system measured 800 mΩ — a factor of 16 — because the return path ran through a stainless chamber wall, a stainless dee-stem support, and feedthroughs (Koeth 2005, pp. 2–3). Every lossy joint is paid for in amplifier watts. Second, at a fixed resonant frequency the only levers on voltage are lowering R_s or raising L/C — which means a physically smaller dee capacitance, since f₀ = 1/2π√(LC) is not negotiable.

What Q to expect: a clean copper tank on the bench measures an unloaded Q₀ near 900–1000 at ~15 MHz — the Rutgers team got 920–968 on a bench replica of their tank, where R_s was 0.107 Ω (Koeth 2005, pp. 6–8) — and a well-made small dee circuit has measured 1600 (Chun 2003). The installed system does worse: with the full-machine R_s of 0.8 Ω above, the same Q = ωL/R_s arithmetic gives about 130. An antenna-tuner chain does far worse still — Q of 16–22 at Houghton College (Haas 2009, p. 64), convenient to tune at a real cost in voltage per watt. Larger machines escape lumped elements entirely and build the resonator into the structure: dee and stem as a quarter-wave line foreshortened by the dee capacitance, tuned by trimmer capacitance, stem length, or stem impedance (ORNL-2648, p. 17).

All rows are measured operating points, and the power figures mean different things — 50 Ω forward power for the driven rows, oscillator or system power for the tube rows — so compare within a row's own machine, not across rows. Within a system the √P law is the reference: Houghton's two points follow it roughly, and the Rutgers record matches the formula above.

Self-excited or driven: the choice is who sets the frequency

Every cyclotron RF system answers one question first: does the dee resonator set the operating frequency, or does an external source? The classical answer is the self-excited oscillator — a tube whose feedback comes from the dee tank itself, so the system oscillates at whatever frequency the tank presents. The 184-inch report calls the grounded-grid version (grid held at RF ground, cathode driven) “an almost foregone conclusion” and couples it “tightly…to the single-dee resonant circuit so that the oscillator follows the frequency of the dee” (UCRL-64, abstract and p. 13). Livingston and Blewett's survey of classic circuits lands on the same virtue: “The most significant advantage of this circuit is its simplicity and compactness along with the freedom from delicate tuning requirements or precise construction” (Particle Accelerators, pp. 190–193). When the dee warps with heat or the chamber is reassembled a millimeter differently, a self-excited oscillator simply moves with it. One caution: following the tank is not the same as following the beam. The resonance condition ties the useful frequency to the magnetic field, and a classical machine tolerates only a percent or two of error, so tank drift still has to be trimmed back — classically with a remotely adjustable trimmer capacitor good for about 1% of frequency (Livingston & Blewett, Particle Accelerators, p. 188). For small machines Wouters recommends the grounded-grid Hartley specifically because it “confines RF currents to intended paths” better than most circuits (UCRL-476, p. 8).

The price of self-excitation is that everything couples to everything. The oscillator's own plate-loop circuit is a second resonance riding on the system, and if it lands on a harmonic of the dee frequency the dee amplitude dips — the 37-inch team measured the dip, predicted “most of the ions can be lost” if it falls where the ions reach final radius, and moved the loop resonance with 15 pF of added capacity (MDDC-1045, pp. 9–10). Grounded-grid circuits also need explicit phase correction — the shift between plate and filament RF voltage “can easily reach 20° or more if no correction is made,” fixed on the 37-inch with an adjustable bypass capacitor trimmed for minimum plate current (MDDC-1045, p. 7). And the grid-leak RC must charge fast relative to the resonant system or the oscillator stutters into blocking oscillations — starting and dying over and over instead of running (MDDC-1045, p. 7). None of this is exotic, but all of it must be gotten right, and the classic literature is the manual.

The modern alternative is the driven system: a master oscillator or signal generator sets the frequency, a 50 Ω solid-state amplifier provides the power, and a matching network transforms the dee tank to 50 Ω. Its appeals are real. Frequency becomes a number on a dial rather than a property of copper; commercial amplifiers, wattmeters, and SWR bridges all speak 50 Ω; and choosing the 13.56 MHz ISM frequency (0.889 T for protons) opens the market of industrial RF generators (Chun 2003, pp. 10–11). The cost is the mirror image of the self-excited system's virtue: the resonator no longer follows anything. Warm-up drift of the dee, chamber, and tank coil detunes the system enough to need continual retuning, and the working fix is a servo — compare the phase of the drive against a dee pickup and let the error signal drive a motorized trim capacitor (Koeth et al. 2010, p. 21). Houghton's automatic antenna tuner is the low-Q version of the same surrender: let a lossy, broad match absorb the drift and pay for it in dee voltage (Haas 2009, p. 64).

One behavioral difference matters enough to decide the choice for some machines: what happens when the dee is suddenly loaded down — by a discharge, an arc, or heavy ion loading. A self-excited oscillator derives its excitation from the load, so when a discharge clamps the dee voltage the feedback collapses with it and the oscillator can sit trapped below the discharge's extinction voltage. The 37-inch crew's fix was a “tickler” oscillator: an auxiliary drive that “does not derive its excitation from the load” and can push the main oscillator over the critical voltage (MDDC-1045, p. 11). A driven amplifier can play the tickler's role because its excitation likewise does not come from the load — but driving into a discharging, mismatched dee is abuse the drive chain must be built to survive. The 37-inch put a 20 kW series triode in the oscillator supply as an emission limiter to protect the tubes when tank discharges occurred (MDDC-1045, p. 11); the modern solid-state equivalent is hardware monitoring of reflected power and temperature with SWR lockout, which the QST amplifier treats as non-optional (QST LDMOS amplifier).

The literature disagrees about this, and the disagreement is worth seeing rather than averaging. Oak Ridge argued for a master-oscillator power-amplifier chain — a separate oscillator driving an amplifier — because on a machine that must produce more than one frequency, a self-excited driver makes harmonic amplitude and phase interdependent and awkward to set (dg-724). Berkeley took the other side on the 88-inch: run the resonator as the frequency-determining element, and buy back the amplifier chain's advantages piecewise — frequency accuracy from a servo trimmer and AFC, amplitude stability from a modulator regulating dee voltage, mode integrity from the coupling design itself (dg-1367).

Both machines worked. The choice is not which architecture is better but which failure you would rather engineer around: a self-excited system follows its tank wherever thermal drift takes it and needs help to be accurate, while a driven system is accurate by construction and needs help to stay matched as the tank moves. For a single-frequency tabletop machine the self-excited answer is the classical one and the simpler one, which is why every small-machine source above recommends it — but it is a recommendation with a documented counterargument behind it, not a consensus.

Loop or probe: two ways into the tank

Once an architecture is chosen, power still has to cross into the resonator. The two idioms are magnetic and electric: a coupling loop sharing flux with the tank inductance, or a capacitive tap — a series capacitor or probe plate — onto the dee stem. Both are impedance transformers, and on a given low-loss tank they are equivalent at match: every coupling geometry Koeth tested on the Rutgers resonator that presented 50+j0 Ω at resonance produced the same dee voltage for the same forward power, so the choice can be made on mechanical grounds (Koeth 2005, p. 7). The equivalence assumes the coupler itself stays low-loss; a dissipative matching chain buys its match with dee voltage, as the Houghton numbers above show.

The loop is the classic choice, and the 184-inch and 37-inch systems show its manners. Coupling strength is geometric — loop area, spacing, rotation — so it can be adjusted without touching the resonator electrically; the Berkeley loops entered slots cut in the transmission-line conductor to get “tight coupling” while keeping “sufficient clearance to prevent sparking,” with the caveat that pushing a loop deeper rapidly raises its self-inductance and stops helping (MDDC-1045, p. 7). Mutual inductance is directly measurable — connect loop and coil in series aiding, then series opposing; the difference of the two inductance readings is 4M (Koeth 2005, pp. 5–6). And the loop leaves no wired path between amplifier and resonator, which keeps amplifier-side DC and the dee's own bias circuits (below) cleanly separated.

The capacitive tap puts a series capacitance between the feed and a point on the stem or tank. Where the loop's adjustment is geometric, the tap's is electrical: change the capacitor, the probe gap, or the tap height — the impedance seen rises as the tap moves toward the dee (the voltage antinode), exactly like tapping a plate lead down a Hartley coil for step-up (UCRL-476, p. 8). In the lumped L-network form (series C_s, shunt L_p) it is the topology of choice for automated matching: within one L-network configuration each matchable load corresponds to a single L-C setting, and two complementary configurations cover the rest of the impedance plane — the property auto-tuners exploit (Blodgett 2012, p. 11). Its vulnerabilities are its own: the series capacitor and probe gap sit directly in the high-field path, so they are the components that arc, and the coax behind them is an arc-energy reservoir — 20 feet of HV cable at 56 pF/ft holds about half a joule charged to 30 kV (½CV²; the build log's own estimate was ~0.4 J), and one machine's persistent arcing damage stopped only when the run was cut to 5 feet (Ponter 2010, pp. 43–49).

Either way, plan the dee's DC circuit deliberately. In both figures the tank coil (or the shorted quarter-wave stem) holds the dee at DC ground; a biasable dee requires breaking that path with insulation designed in from the start. ORIC's designers noted that if multipactor blocks turn-on the options are biased dees or a more complex drive scheme (ORNL-2648, p. 99) — a problem to settle on the drawing board. The 86-inch avoided oscillator starting trouble from ion loading with insulated, negatively biased dees (ORNL-1196, pp. 7, 47), and the 184-inch went as far as feeding dee cooling water through low-conductivity lines so the bias would not leak away through the plumbing (UCRL-64, p. 10).

When the voltage refuses to rise: multipactor and its cures

Sooner or later most builders meet the same failure: RF drive goes up, dee voltage does not. The match checks out, the amplifier is delivering power, and the dee sits at a few hundred volts glowing blue. This is multipactoring and its relatives — a family of electron discharges that live in a low-voltage band and absorb everything the drive can deliver. (Arcs at high voltage are the separate ceiling treated below.) Livingston and Blewett describe the trap: “This loading of the D circuit by discharge currents holds the D potentials down to a few hundred volts… Unless the loading is removed, the chamber will continue to operate in the low-voltage, blue-glow discharge condition indefinitely” (Particle Accelerators, p. 188). The 37-inch team measured the mechanism: electrons oscillating in the RF field reach ionizing energies at gap voltages below an extinction threshold near 500 V, the discharge persists even at 10⁻⁵ mm Hg, and it “prevents the oscillator from driving the dee to high r.f. voltages where the discharge does not occur” (MDDC-1045, p. 11).

The cure catalog is nearly eighty years old and still the working playbook. The 184-inch commissioning is the canonical entry: “All discharges were eliminated by cutting down the available volume by means of perforated shields around the sides of the dee, by adding a grounded ‘dummy’ dee and by applying a negative bias to the dee” (UCRL-64, p. 20). Unpacked:

• Deny the electrons room. Oscillating electrons need path length to reach ionizing energy — about 30 eV in 5 cm at 20 MHz. Around the 37-inch dee, few field-aligned paths exceeded that and no trouble occurred; in the roomy rotary-condenser volume, paths ran ~20 cm and the discharge lived there (MDDC-1045, p. 11). Shields and a dummy dee shrink the available volume.
• Bias the dee. A DC sweeping field breaks the electron resonance. The 37-inch used a few hundred volts positive on the dee and line — which also, unexpectedly, doubled the beam (MDDC-1045, p. 11). The 184-inch found negative bias imperative where magnetic field reaches the RF volume, because a positively biased electrode in a magnetic field sustains a Philips-gauge (Penning) discharge (UCRL-64, p. 20). The 86-inch settled on insulated dees biased a few hundred volts to several kV negative, interlocked to the RF (ORNL-1196).
• Drive through it. Above the extinction voltage the electrons can no longer oscillate in step and this discharge dies out, so a drive that can be brought through the susceptible band quickly starts clean — the tickler-oscillator trick, or any drive source whose excitation does not come from the load (MDDC-1045, p. 11). Doing it deliberately means doing it protected: current-limited supplies and reflected-power protection, as above.
• Kick it past the top. Rochester rejected both biasing and a tickler as too awkward on a variable-frequency machine and built a third thing: a spark gap that dumps a capacitor into a single loop coupled to the dee, shock-exciting it straight through the band. The insight that makes it cheap is that the kick only has to clear the top of the loading band, not deliver operating power — everything above a few hundred volts is the oscillator’s own job. One loop, one capacitor, one spark gap, one HV supply, and it is additive to whatever hardware is already fitted (dg-1276).

Before reaching for any of them, find out whether you have the problem. The same Rochester report gives a diagnostic that costs nothing and takes minutes. With plate power off, ring the dee — impulse it, or drive and release — and watch the pickup envelope on a scope. The decay is not a clean exponential: it falls smoothly until the voltage reaches roughly a third of its maximum, drops steeply through the loading band, then decays slowly again below it. That kink is the multipactor band, made visible on your machine rather than inferred from someone else’s (dg-1280).

Two things follow. It is experimental confirmation that the loading occupies a bounded window rather than simply everything below an extinction voltage — it refines MDDC-1045’s ~500 V figure with an observable top edge. And it answers the question that actually matters to a builder: whether nominal operating voltage sits inside the band. A dee running at a few hundred volts may be sitting in it; one running at several kilovolts is clear of it, and a stalled start is then something else’s fault. The test bounds the band but does not distinguish which multipactor mechanism produces it.

Arcs are the other ceiling. A fresh or freshly vented chamber will spark below its eventual capability, and conditioning is the only path: assemble clean — “dust should be controlled and all grease removed (even fingerprints)” — round and polish the high-field contours, then run the system up and let the sparking subside over hours of operation; no amount of polish eliminates the conditioning period (Livingston & Blewett, Particle Accelerators, p. 189). Conditioning is a deliberate, supervised procedure on an interlocked machine — current-limited, hands off, and with the shielding in place, since a sparking vacuum gap at tens of kV is also an X-ray source. Deliberate spark conditioning measurably raises a surface's breakdown field in the early-processing regime (Werner 2004, pp. 91–92). For calibrated numbers: the 184-inch team's test rig held 50 kV RF at 13 MHz across a 0.080-inch polished copper gap at 5×10⁻⁶ mm — and the production unit, with a 0.060-inch gap and discharge-roughened surfaces, held only about 30 kV (UCRL-64, pp. 10, 21). Rate the machine at what it holds quietly, not what it touches momentarily: when the 86-inch pushed beam current until “operation at this level was very unsteady due to sparking,” the sustained level became the rating (ORNL-1196, p. 24). The frequency-dependent design guideline for working surface gradient is the Kilpatrick limit — an empirical criterion, and a conservative one: conditioned surfaces routinely exceed it.

Feedline and matching practice at 5–20 MHz

The practical numbers of the 50 Ω world transfer directly to a driven cyclotron RF system. What the amateur-radio matching literature and the small-cyclotron build record agree on:

• Judge the match by reflected power — at the resonator. For a loop or tap coupled directly into the tank, a reflected-power null at resonance indicates critical coupling, the condition of maximum voltage transfer, at which the loaded Q drops to exactly half the unloaded value (Koeth 2005, pp. 6–7); a weakly coupled S21 sweep on a vector network analyzer confirms it from the −3 dB width. A null seen through a lossy tuner proves only that the tuner is matched, not that the dee is getting the power — the Houghton Q ≈ 16 chain matched perfectly and paid for it in voltage. As a working threshold, the amateur matching literature treats SWR 1.5:1 (4% reflected) as good enough for solid-state amplifiers (Blodgett 2012, pp. 6–8); the binding number is the amplifier's own rating.
• Instrument the line. A ~30 dB directional coupler brings up to about 200 W down into the range of a small log detector with negligible main-line loss (higher power needs more coupling or added attenuation), and coupler directivity floors the SWR reading — 28 dB directivity reads a perfect load as 1.08 (Blodgett 2012, pp. 18–20).
• Expect the tune to walk. RF heating expands the dee, chamber, and coil enough to shift resonance during a run; fixes in the record are water- or oil-cooling the matching coil (Heuer & Baumgartner 2009, p. 23), holding dee cooling water temperature stable (ANL-5907, p. 6), and closing the loop with a phase-comparison autotuner (Koeth et al. 2010, p. 21).
• Respect the joints and the cable. High-current RF joints want clamped, silver-plated, cooled surfaces (ORNL-1196, p. 53), and HV coax runs should be as short as the layout allows — the cable's stored energy does the damage in an arc (Ponter 2010).

Component values — dee capacitance from geometry, the resonating inductance, the series and shunt elements of the L-network, and the resulting bandwidth — are the matching calculator's job. For component-level checks on capacitance, matching, feedlines, joints, and measurement, use the design guide's RF + dee filter (365 sourced rules).

Measuring dee voltage: calibrate a pickup

Dee voltage is the number every other design decision leans on, and it has to be measured. The measurement chain that works, from the Rutgers write-up (Koeth 2005) — the single best document on this problem:

1. Direct HV probe, low power only. A high-voltage probe on the dee stem reads truth at watts — but it loads the tank (the P6015 added 3.0 pF and shifted the resonance; retune or correct) and it fails high: in the Rutgers setup the probe departed from √P scaling above ~200 W forward, behaving like a resistive breakdown (Koeth 2005, pp. 3–4). The threshold is the probe's and the machine's, and the failure mode — a reading that quietly stops rising — is generic.
2. Calibrated capacitive pickup for operations. A small plate facing the dee, calibrated against the direct probe at low power, then extrapolated on the √P law: the Rutgers pickup calibrated at 3710× with R² = 0.994 and stayed linear to at least 1300 W (Koeth 2005, pp. 3–4). The scale factor is frequency-dependent — Houghton measured 11,300× at 3.55 MHz and had to recalibrate at every frequency change (Haas 2009, pp. 65–66).
3. Let the beam check the electronics. An ion crossing the gap once at the crest carries ½qV_p-p, so the radius of its first half-revolution satisfies q²B²r²/2m = ½qV_p-p — a probe-independent “beam-inferred dee voltage” to plot alongside the pickup data (Koeth et al. 2010, p. 33).

The classic machines used peak-reading voltmeters on the oscillator electrodes for the same reason — plate swing, grid swing, and cathode excitation were each separately metered on the 184-inch (UCRL-64, p. 16). Whatever the instrument, the √P law is the sanity check: doubling dee voltage costs four times the power, and a reading that scales any other way is a warning — the probe is saturating, the tune is walking, or a discharge is loading the tank.

Go deeper

• Dee capacitance & matching calculator — the L-network numbers this page deliberately does not repeat.
• Design guide: RF + dee rules — every sourced rule behind this page, with formulas and verbatim quotes.
• UCRL-64 and MDDC-1045 — the hosted primary sources: the 184-inch RF system design report and the 37-inch FM conversion paper.
• RF amplifier construction, deliberately out of scope here: the QST 1.25 kW LDMOS amplifier and its development notebook, plus the MRF1K50H datasheet.
• How a Cyclotron Works — why the dee voltage and frequency matter to the beam in the first place.

Sources

• K. R. MacKenzie, F. H. Schmidt, J. R. Woodyard & L. F. Wouters, Design of the Radio-Frequency System for the 184-Inch Cyclotron, UCRL-64 (1948) — hosted. Architecture choice, insulator and blade-gap tests, discharge cures.
• K. R. MacKenzie & V. B. Waithman, R.F. System for Frequency Modulated Cyclotron, MDDC-1045 (1946) — hosted. Grounded-grid single-dee system, coupling loops, phase correction, low-voltage discharges.
• L. F. Wouters, General Recommendations for Design of Small Cyclotrons, UCRL-476 (1949) — hosted. Grounded-grid Hartley, dee capacitance as tank C.
• Three further hosted reports on driving the dees, added August 2026 and not yet mined into the design guide — RF power engineering: Osterlund & Smythe, A Cyclotron Power-Amplifier RF System Using a 4CW50,000C/8350 Tetrode, COO-535-543 (1963), a tube-level amplifier rebuild; Goodman, A Square-Wave Cyclotron Oscillator, ORNL-2403 (1958), arguing the accelerating waveform should not be a sine at all; and Smith, A Three-Phase Radiofrequency System for Cloverleaf Cyclotrons, UCRL-3153 (1955).
• T. W. Koeth, Theoretical Calculations and Measurements of the DEE Voltage in the Rutgers 12 Inch Cyclotron (2005) — library entry. The dee-voltage formula, √P law, coupling equivalence, probe and pickup practice.
• T. W. Koeth et al., The Rutgers 12-Inch Cyclotron for Students, Small Cyclotron Conference (2010) — library entry. Operating record, autotuner, beam-inferred dee voltage.
• R. S. Livingston & A. L. Boch, The Oak Ridge 86-Inch Cyclotron, ORNL-1196 (1952) — hosted. Biased dees, spark-limited ratings, RF joints.
• R. S. Livingston & F. T. Howard (eds.), The Oak Ridge Relativistic Isochronous Cyclotron, ORNL-2648 (1958) — hosted. High-Q necessity, quarter-wave dee systems, multipactor at design time.
• M. S. Livingston & J. P. Blewett, Particle Accelerators, McGraw-Hill (1962), pp. 185–193 — oscillator circuits, parasitics, blue-glow discharge, conditioning.
• M. Yuly et al. (Houghton College), conference talks and theses (2009–2013) — Haas 2009, Yuly 2010, Yuly 2013. Autotuner-chain Q, pickup calibration, low-power operating points.
• M. Blodgett, 3.5 to 30 MHz Automatic Antenna Impedance Matching System (2012) — library entry. L-network selection, coupler and SWR practice.
• W. J. Ramler & G. W. Parker, The Argonne 60-Inch Cyclotron, ANL-5907 (1959) — hosted. Cooling-water stability for RF tune.
• C. Chun, The Cyclotron Magnet and RF Oscillator (2003) — library entry. Measured small-tank Q, 13.56 MHz ISM choice, image-dee architecture.
• P. V. Heuer & H. T. Baumgartner, Design of a 2 MeV Cyclotron (2009) — library entry. Cooling the matching coil against thermal drift.
• T. Ponter, Beam Energy Measurements… on the Rutgers 12-Inch Cyclotron (2010) — library entry. HV coax stored energy and arc damage.
• G. R. Werner, Probing and Modeling Voltage Breakdown in Vacuum, Cornell dissertation (2004) — library entry. Spark-conditioning statistics.
• A. J. Buckler, QST LDMOS amplifier article and notebook (2015) — library entries. The out-of-scope amplifier, built and documented.