Before You Switch Anything On

Hazards arrive in build order. The vacuum system runs for months before anything else works, and its hazards are mechanical and chemical: a flat lid carrying tonnes of atmospheric force once the air outside has nothing pushing back from inside, glass that can implode, and the oil mist from a rotary-vane pump’s exhaust, which accumulates with every pump-hour (safety page). None of those needs a radiation meter. Every powered vacuum controller is in the assessment too: an ion-gauge controller carries hazardous electrode voltages before any RF runs, and “gauge off” can mean standby rather than isolation. All of them need thought before the first pump-down. The magnet brings stored energy, crush and projectile forces, and hot coils (safety page; I've heard that switching off the magnet coils can produce a high voltage. How does that happen, and what protection do the coils and the power supply need?). The high-voltage supplies and the RF bring lethal shock, stored charge, RF burns, and the first radiation: X-rays from electrons accelerated across a vacuum gap, assessed from the highest voltage the gap actually sees, RF peaks included; the site’s 15 kV figure is the trigger for assessment and controls, not a voltage below which there is no X-ray hazard (Can the machine make radiation, or hurt me some other way, before I've even detected a beam?). Field emission is the electric field pulling electrons out of the metal; the gap voltage accelerates them, and electrons stopped in metal give off X-rays. Hydrogen brings a flammable gas into an enclosed room.

Protection means different things for equipment and for people, and the hazard list keeps them in separate columns, with a third column for when each is verified. Equipment protection is interlocks on cooling flow and temperature, surge paths across coils, current limits on supplies, and the acceptance tests that prove them before first use. Personnel protection is isolation proven dead before hands go in, covers and guards, access control, and instruments that can see the hazard. Interlocks sit in both columns, and the professional rules for them are short. An interlock is a switch or circuit that permits an operation only while the condition it monitors holds. A complete circuit permits operation, so a broken wire lands in the safe state (Martin (ed.), p. 152); that covers an open wire, and the complete safety function also covers a shorted wire, welded contacts, a failed output device and power returning after a trip, and closing an interlock never by itself restarts anything. False trips are the price of real ones (Martin (ed.), p. 153). An opened interlock is remade at the point of the break, never from the console (Martin (ed.), p. 149). And interlocks are essential but insufficient without people who know why they exist (Martin (ed.), p. 17).

Radiation protection is the element most often treated as a purchase; it is a plan. The instrument has to suit the photon energies the machine makes, tens of keV at these voltages, where thick-walled probes under-read; and it has to be used where the RF and the magnet are not corrupting it, because a meter that reads zero under RF has measured nothing (safety page; Properzio (1970), p. 46). Such X-rays are stopped by thin metal, so a probe’s own wall absorbs part of what it is meant to count. The program is the state’s radiation-control program; classification is its ruling on whether and how the machine is regulated, and a survey is the measurement of radiation levels around the running machine. The site’s recommended workflow, which is conservative rather than a legal requirement everywhere, is to put the classification question to the program in writing before the machine can first make X-rays, and to treat the survey under maximum operating conditions, run as an authorized and controlled test once the review is complete and not as a reason to energize everything first, as the document the program will ask for first (legal page); programs differ, and the legal section’s state pages carry what each one actually requires.

What needs eyes other than yours. The classification and the survey: a state program answers classification questions as routine business, and an institution routes a cyclotron through its radiation safety officer (legal page). An independent builder does not thereby become a health physicist. The professional literature lets an operations staff act as its own health-physics staff only where responsibilities are written down (Martin (ed.), p. 19); the home equivalent is a named qualified reviewer for the radiation plan, not self-certification. The mains side: the branch circuits feeding a kilowatt-class magnet supply and RF amplifier, and any home-built high-voltage supply, belong under the local electrical code and an electrician’s eye. Rigging: a magnet of hundreds of kilograms, with pole attraction in tonnes, is a lifting plan. None of these reviews waits for a beam. Radiation protection in particular costs a fraction when planned from the machine’s conception rather than retrofitted (Martin (ed.), p. 17). The RF and high-voltage circuits, the magnet’s energy-disposal path, the protective bonding and any personnel interlock also need review by someone qualified in that system before their first energized test; competence with mains wiring does not establish it.

Once any electrode in vacuum can hold tens of kilovolts, or is planned to, the radiation part of that plan is: a survey meter in hand, a thin-window probe to find radiation and a calibrated instrument to quantify it, both checked at 10 to 30 keV where thick-walled tubes under-read (safety page); RF kept away from the meter or the meter shielded (Properzio (1970), p. 46); a survey of the whole perimeter at first power-up, at every voltage increase and after every vent; and the jurisdiction’s registration rule known before the machine can make X-rays, which is before first beam (safety page, legal section). A detector’s own wall absorbs low-energy X-rays before they reach its sensing volume, which is how a thick-walled tube under-reads; keV is the energy of each photon and sets how far it penetrates. The survey only counts inside that assessment: a zero from an instrument that cannot see 20 keV photons, or one being upset by the RF, demonstrates nothing. None of this waits on a beam, and a null reading on the electrometer proves nothing about any of it.

The other hazards that need no beam, in the order they tend to be met:

• High voltage. Filament bias of hundreds of volts; arc supplies that strike at 1 to 3 kV (Clark (1981), p. 3); dee bias and deflector supplies at tens of kilovolts. To strike is to start the discharge; the deflector is the electrode that pulls the beam out of the machine.
• Stored energy. Capacitors, and cable. A coaxial cable is itself a capacitor, inner conductor and shield across an insulator, and it holds its charge after the supply is switched off. Six feet of HV coax at 30 kV stored about 0.2 J on the Rutgers deflector run; an arc from its shield reached a magnet coil and permanently damaged the magnet power supply (Koeth, p. 5).
• The magnet. Opening a coil circuit at operating current makes an inductive spike that arcs through insulation or a person, and the field turns tools into projectiles (safety page). A coil resists any change in its current; break the circuit and the voltage across the break rises until the stored energy finds a path.
• The vessel and the gas. Viewport implosion, pump-exhaust oil mist, hydrogen. A window on an evacuated chamber carries the push of the atmosphere on its outer face and breaks inward if it fails; an oil-sealed pump exhausts a fine oil mist; hydrogen in air burns.

To understand it. The X-ray mechanism is the one newcomers most reliably miss. At tens of kilovolts, microscopic points on electrode surfaces emit electrons; the electrons cross the gap at full voltage, strike the dee, the wall or a viewport, and make bremsstrahlung. Bremsstrahlung is the X-radiation a fast electron gives off as it is stopped in matter. An electron crossing the gap arrives with the gap voltage as its energy in electron-volts, and the X-rays it makes carry up to that energy. Two kinds of number exist for where this starts, and they should not be confused. Voltages at which X-rays were detected in particular experiments, with their instruments and backgrounds: fusor builders surveying their own chambers found X-rays from 18 kV on the electrodes (Kovalchick (2012)), and Michigan State calibrated its dee voltage from the X-ray spectrum made by nothing but the gap’s own electrons, which rose above room background only above about 30 kV in their setup (Miller & Kashy (Michigan State University Cyclotron Laboratory), p. 1). A conservative design assumption: this site’s safety page treats any vacuum gap above about 15 kV as an operating X-ray source. Neither kind is a physical threshold below which nothing is emitted. RF voltage on the dee counts, and two dees driven push-pull double the gap voltage. Push-pull means the two dees swing to opposite polarities at the same moment, so the voltage between them is the sum. Dark current is the electron current the gap draws from its own surfaces with no beam present, and conditioning is its decline as the gap runs. The output also moves: dark current falls as electrodes condition and jumps after every air exposure (safety page), so last month’s clean survey is not evidence about today. A 1970 Bureau of Radiological Health study of classroom discharge tubes states the general rule: X-rays appear wherever there is an electron source, a target, enough voltage and a gas pressure in the right range (Properzio (1970), p. 38), and that includes gassy rectifiers inside a high-voltage supply (ANSI N43.2-1971 (1972), p. 19).

A first machine running a few hundred volts on the dee is far below the voltages at which those experiments detected anything, and X-rays from such a gap are too low in energy to leave the chamber wall. That statement covers the dee gap only. Every other electrode and supply in the machine gets the same question, and only a survey answers it. The COLUMBUS builders judged their 2 to 3 kV dee “on the safe side” for students beside the machine; that is their judgment for their apparatus, and the 5 kV class some regulations use for incidental emitters is a legal line, not a physical one (Prechtl & Wolf (2020), p. 41). The usual next step after first beam is more dee voltage. That is when the hazard arrives, and the survey meter takes longer to buy than the amplifier.

The switch is one interruption among several. A contactor opening at current does it. So does a fuse blowing, a connector pulled, a wire breaking at a crimp, a mains failure, or a fault in the supply’s output stage. A supply fault is three different faults: an output that opens, which needs the coil’s own clamp; an output that shorts, which puts the coil’s energy through the supply; and a pass device that fails on, which keeps driving the coil and which no clamp stops, so drive removal is detected and done independently of the electronics that regulate the current. Mains failure is the one builders forget: the supply stops sourcing current and the coil’s energy looks for a path through the supply’s output, which may not survive it. Equipment protection therefore answers four questions, and the answers differ by machine.

• Where does the energy go? Into something that absorbs kilojoules at a bounded voltage: a freewheeling diode with a series resistor across the coil, a varistor or the thyrite resistor of the older practice, a dump resistor switched in by a contactor, or a discharge circuit built into the supply (Wouters (1949), p. 5). The path sits on the coil side of everything that can interrupt the supply wiring, at the coil itself if possible, stays available when mains or control power fails, and is rated for the current, voltage and energy it must take; a resistor-and-diode dump can put reverse voltage on a bench supply still connected, which that supply must be known to tolerate. The choice depends on the supply, the stored energy, the voltage the insulation permits, and how fast the current must fall.
• How does a normal shutdown happen? By reducing the current at a controlled rate, with the supply and the discharge arrangement designed to handle the magnetic energy released; a supply that can source current is not necessarily one that can absorb it, so where the energy goes during a ramp is a design question too.
• What happens on a fault? It depends on the fault. Loss of cooling water with a cool coil can tolerate a slow ramp. The Rutgers 12-inch’s protection controller ramps the magnet down and latches it off on over-temperature or loss of flow (Koeth (2011), p. 4); that is one group’s choice for that fault class. Latched means it stays off until someone resets it. A coil fault, an arc or smoke calls for prompt removal of drive power while an independent energy path handles the coil. That is why the dump path exists apart from the supply’s logic. The hazard decides the response.
• How is the coil proven before first use? By insulation tests suited to the coil, done by someone qualified to do them. An insulation test puts a voltage between winding and grounded core and measures the current leaking across the insulation; it is one required check, not proof against every interruption stress, which also falls between turns and layers and across terminals and leads. A laboratory’s acceptance test holds a kilovolt DC to ground with microamperes of permitted leakage (IUAC (2024), p. 27). The permitted dump voltage is set from the whole insulation design, and the test voltages and limits are chosen by a qualified person for this coil; the cited 1 kV is an example, not a prescription. The spike finds the weakest point. A home-wound coil is unproven until something has looked for it.

Personnel protection is separate and simpler. Nobody touches coil terminals, connectors or the supply output while current flows or is decaying. With a resistive dump path the current decays over a time set by L/R, and after one such time constant a third of it is still flowing; a nonlinear iron magnet, a diode clamp or an active discharge circuit need not follow that simple curve, so the clock is no guide at all. L is the coil’s inductance and R the total resistance of the loop the current circulates in, winding and dump path together; their ratio is the time constant of the decay. A meter showing zero current is one check, not the verification. Before anyone touches the circuit, it is also isolated from its supply, checked for residual voltage and stored energy, and secured against re-energization, which is the ordinary hazardous-energy discipline. Coils past the guarding thresholds are covered, with the covers screwed down, and the core is grounded so a failed insulation faults to earth rather than to a hand (Tanabe (2005), p. 12-13). Grounding the core gives a first winding fault a path and no more: an isolated output has no second return, so the fault may draw little current and trip nothing, and a floating magnet circuit needs its own earth-fault detection that removes drive and leaves the discharge path in place. Insulating hoses avoid a metal path, but the water inside them still conducts, so coolant isolation is engineered and checked (water-path length and resistance, water conductivity) and accessible metal plumbing is bonded (Tanabe (2005), p. 3). One documented accident runs the other way. On the Rutgers 12-inch, an arc from a deflector cable’s shield (the deflector is a high-voltage electrode inside the chamber) terminated on a magnet coil, and the magnet power supply was permanently damaged (Koeth, p. 5). Protection considers energy arriving at the coil as well as leaving it.

To understand it. The mechanism is Faraday’s law. The voltage across an inductance is V = L × dI/dt, where dI/dt is the rate at which the current changes. At steady current dI/dt is zero and nothing happens; the supply’s few tens of volts only overcome the coil’s resistance. Interrupt the current and dI/dt becomes enormous, so V does too. The energy is set by the magnetic field: about B²/2μ₀ times the gap volume (B the field strength, μ₀ the magnetic constant, the gap the space between the pole faces), plus what the coil and yoke hold (Tanabe (2005), p. 12-16, Zickler (2010), p. 18). Take one census design, the Cyclotron Kids’ 14-inch magnet: 1.6 T across a 54 mm gap, 720 turns at 110 A (Heuer & Baumgartner (2009), p. 32). Its gap alone stores about 5 kJ. For a linear inductor the stored energy is ½LI², so twice that energy divided by the square of the current, L = 2U/I², gives an energy-equivalent inductance near 0.9 H. Iron near saturation is not a linear inductor, so that is a scale, not the coil’s measured inductance. On that scale, stopping 110 A in 10 ms would call for about 10 kV, and in 1 ms, 100 kV. Those numbers illustrate the consequence of a fast interruption; in a real circuit an arc or a clamp sets the voltage that appears. Tanabe’s magnet-engineering lectures put their guarding line at 5 J of stored energy (Tanabe (2005), p. 12-13). A small cyclotron magnet is a thousand times past it.

At megahertz current hugs a conductor’s surface, and a thin wire opposes it far more than a steady current. Every current flows in a closed loop; the dee is the electrode carrying the RF accelerating voltage, and the current driven onto it needs the return half of its loop. The return has to be designed, because a joint that reads zero ohms on a multimeter can be a poor conductor at 10 MHz. The measurement reference and its shield are arranged so only the current you want reaches the instrument; the electrometer’s current enters at input high and leaves at input low, and signal-low is that low terminal. Leakage is the small current through imperfect insulation; stray capacitance, present between any two nearby conductors, passes alternating current. A biased circuit is deliberately held at a chosen potential relative to some reference: a bias supply with its positive terminal on the chamber biases the filament while being referenced to ground, and the filament heater supply must then have an isolated output, because the filament it heats sits at that negative bias. The heater’s output terminals connect to the filament, so they sit at the bias too; an output tied to earth would hold the filament at earth instead. An isolated magnet-supply output, by contrast, floats without being biased anywhere in particular.

The safety bond covers the accessible conductive enclosures and structures that need protective bonding and every instrument chassis; deliberately biased electrodes inside the vacuum are not among them. The chamber is also the RF return and the reference for the DC measurement. Those are two further jobs given to it on purpose, with contacts built for RF current (Livingston & Boch (1952), p. 53) and one defined point for signal-low. They are not consequences of the earth bond. The filament heater output floats on the bias potential, its leads bypassed to the chamber with small capacitors so RF sees ground and DC does not (Yuly (2013), p. 3). The collector plate’s only connection is its measurement path: one lead, through the RF filter and the overvoltage protection that an arc inside the RF-powered chamber makes necessary, each with its own short return to the chamber, and then to the electrometer. The choke and the bias battery are not that protection. It has to be designed for the fault voltage and energy the chamber can deliver, reviewed by someone qualified, and its leakage has to fit the picoampere measurement. Even then the meter reads net collector current, leakage and any RF-induced offset included, so background, leakage and RF susceptibility are checked with the finished interface before a reading is called beam current. Do not count signal shields, gauge cables, RF cable braids or plumbing as the chamber’s protective bond; they can hide a failed bond while carrying fault current they were never meant for. The bond is a designed conductor, reaching the lid and every removable part, and it is checked. The magnet supply’s output is typically isolated from earth while the magnet’s iron core is grounded (Tanabe (2005), p. 12-13).

The table is an example for one configuration and one instrument class, not a universal wiring diagram. The configuration: a single dee with a grounded dummy dee, a filament source on its own isolated heater supply, and an internal collector plate. The instrument: a feedback picoammeter of the Keithley 617 class in its amps mode, guard off, with the cable shield on input low as the manual specifies for current measurements (Keithley 617 manual, §2.7). While the powered meter is working within its current range and bandwidth, it holds input high close to input low; the remaining difference is the voltage burden. That does not limit the input voltage during RF pickup, an overload, loss of power or a discharge. Change any of those and the table changes.

Node	Sits at	Why, and the source
Chamber, lid, magnet iron, vacuum plumbing	Protective earth; also the designed RF return and the DC measurement reference	Every exposed conductor bonded; the core grounded (Tanabe (2005), p. 12-13); RF joints broad, clean and clamped (Livingston & Boch (1952), p. 53)
Dee	RF hot; at DC ground through the tank coil unless insulated for bias	The coil ties the dee to ground at DC; a biasable dee needs that path broken by design (dee-coupling page)
Dummy dee, liners	Chamber potential, each deliberately bonded (a liner on insulating supports or an oxidised joint can float and pick up RF)	The grounded electrode of the gap
Filament heater	Output floats on the bias potential; supply chassis earthed; each lead bypassed to chamber with a small capacitor	Houghton’s hairpin at about −90 V, 2 A, 1 nF per lead against RF pickup (Yuly (2013), p. 3); Rutgers ran −200 to −320 V (Koeth (2006), p. 5, Koeth, p. 5)
Filament bias supply	Negative to the filament, positive to the chamber; the emission meter in that return	The meter that matters for the source (My filament is glowing. How do I know it's actually making ions? And what are the filament current, the bias voltage and the emission current each telling me?)
Ion-source body or chimney	Chamber potential, deliberately	An ungrounded source floats toward dee potential and becomes a different machine (Cohen (1952), p. 9)
Collector plate	Insulated from the chamber; one lead, with an RF choke, to the electrometer’s input high	The choke reduces MHz pickup without blocking DC (Koeth (2006), p. 1); high resistance to ground, chokes and bypasses on the meter (Wouters (1949))
Collector bias (this example)	A small battery in series with the input-high lead, between plate and meter; input low at chamber potential	A feedback picoammeter has almost no voltage burden, so the plate sits at the battery voltage and the meter reads the plate current, and input low never floats (Keithley 617 manual, §2.7.5). Houghton reported a +9 V bias lowering the reading by suppressing secondaries (beam-measurement page)
Electrometer	Chassis earthed through its own cord; input low tied to the chamber reference at one point; shield to input low	Safety precaution 1 and §2.7; input low to chassis is limited to 500 V peak, and the analog output must not be earthed while the input floats
RF amplifier and generator	Chassis earthed; coax shield bonded to the tank and chamber at the RF return; the oscillator in a grounded screen box	Keeps RF out of every other circuit (Wouters (1949), p. 9)
Magnet supply and coil	Output isolated from earth; coil insulation to ground proven; core grounded; non-conducting water hoses	dg-1616, dg-073, dg-202
Gauge controllers and tubes	Controllers earthed; tube bodies at chamber potential; a hot-filament ion gauge switched off during beam runs	The Rutgers run-up shuts off the ion gauge before the filament goes on (Koeth (2006), p. 5)

Two things to check on the real instruments before any of this is built, and one on the source side: the heater supply’s output-to-chassis isolation and the filament bypass capacitors are rated for the full bias plus transients, and the bias supply is current-limited with a fault response, because a heater lead or a shorted bypass capacitor faulting to earth collapses the bias without tripping anything. The electrometer first. The 617’s input high, input low and chassis are three separate conductors. Input low may float from chassis only within the manual’s common-mode limit. That limit is the largest voltage permitted between input low and chassis, the table’s 500 V peak. It is not an input-high-to-low overload rating, not an arc rating and not permission to apply RF; the manual’s mode, frequency and accessory restrictions apply. And the manual warns that earthing its analog output, preamp output or COM while the input floats can damage it (§2.7.2). An example that floats input low to apply bias must be checked against that caution before a scope goes on the analog output. A mains-powered oscilloscope’s probe ground is earthed through the scope’s own cord, so a probe earths whatever its ground clip touches. Second, any conductor carrying DC bias through the magnetic field inside the vacuum can sustain a Penning-type discharge (Osterlund & Smythe (1963), p. 5). When a box is not in the table, the method is the same. Find its chassis-to-output relationship in its manual. Decide which of the four ideas each terminal belongs to, and connect it accordingly; the one-chosen-point rule applies to the measurement reference, not to protective bonding or RF returns, which have their own requirements. And before the table becomes a schematic anyone builds, it gets an independent electrical review that includes what protects the electrometer’s input when an unintended discharge in an RF-powered vacuum apparatus puts voltage on the collector; a correct nominal current path is not a demonstration that the arrangement is safe.

To understand it. Three puzzles the four-concept picture resolves. How can something be grounded for RF but biased at DC? With a capacitor from the lead to the chamber. It passes the megahertz and holds the direct voltage, which is what the filament-lead capacitors do (Yuly (2013), p. 3). Why does the multimeter say two parts are connected when the RF connection is bad? At megahertz, current flows in the skin of the metal. A long, thin or oxidized path has inductance and loss the meter cannot see. The Rutgers resonator’s return path measured sixteen times the coil’s computed resistance for that reason (Koeth (2005)). How do I bias the collector and still read picoamperes without reading the bias supply’s own current? Put the bias where only the plate current flows through it. In this example that is a battery in series with the input-high lead, so the meter measures what passes through plate and battery together and its input low stays at the reference. A battery has little leakage of its own; its holder, wiring and insulation can, and they sit in the measurement path. If the battery, its holder or the lead opens, the plate loses its path to the meter: the meter reads zero while the plate charges from whatever reaches it, so the plate’s voltage must not depend on that continuity for its limit, and the bias state is checked before a reading is trusted.

For a first beam: one chosen point where signal low meets earth. Shields on signal low for the measurement cables. RF cables and enclosures bonded as RF practice requires. A choke (an RF-blocking inductor) in the collector lead, and a logged RF-on, feed-gas-off background before every session, with the pressure and the source’s heater and bias states written down; residual gas and electron or ion currents can remain, so it diagnoses background and does not prove absence of beam. That removes most of the mystery offsets a builder meets. The ones it does not remove are usually not ground loops, and none of them is ever a reason to defeat an earth.

If it does not work. Recognizing one, and the three offsets that imitate it. The analog output is a voltage on the instrument that follows the reading; line frequency is the mains frequency. The manual’s test is to watch the instrument’s analog output on an oscilloscope: line-frequency signal there means interference is present (Keithley 617 manual). Before connecting the scope, check the output’s isolation and common connections, so that the scope’s own earth does not violate the input-float restriction or add a second signal-reference earth; never lift the oscilloscope’s protective earth. A ground loop tends to change when a second instrument is unplugged or re-routed, when a cable is moved, or when a nearby load switches. That narrows the diagnosis without closing it, because one action can disturb several mechanisms at once. RF pickup from the dee is rectified somewhere and arrives as a DC offset. Rectified means some junction passes the RF better one way than the other, as a diode does, turning part of it into direct current. It follows the RF power and vanishes with the drive off. The beam-measurement page’s RF-on no-gas background helps reveal RF-related background, though a reading taken that way can include other RF-dependent currents too; the cure is filtering in the collector lead and at the meter (Koeth (2006), p. 1, Wouters (1949)). Electrostatic pickup shows as readings that move when a hand or body moves near the experiment (Keithley manual). Leakage shows as a steady offset that drifts with humidity and changes when an insulator is cleaned. A dirty connector or a poor insulator is an unintended resistive path, and the current through it adds to, subtracts from or bypasses the current being measured; at the picoampere level that parasitic leakage current is comparable to the signal (Keithley manual). None of those is by itself a ground loop. Each is fixed at its own filtering, shielding, reference or insulation defect, which may include correcting a signal or RF ground; removing protective earth is never part of it.

One machine’s catalogue shows the mix a builder actually meets. The Mullins log’s interference entries, collected in its hosted instrumentation note (§8, Table 6): RF into the Hall meter, cured by reading the magnet’s field with the RF off; RF pickup raising the electrometer background when an attenuator (a device that reduces the RF drive by a fixed ratio) came out of the chain, cured by restoring it; the magnet supply’s programming input (the control terminal that sets its output current) picking up RF, and “ground noise” from a bench instrument, above a certain drive level, cured by a 9 V battery and a potentiometer as the programming source; and a baseline that stepped with gas flow or a clip-lead ground. Most of those are RF pickup and supply coupling rather than ground loops; whether the “ground noise” entry was a loop, the log does not say (hosted instrumentation note §8; /library/amateur/#mullins).

Avoiding one depends on which current you are controlling and at what frequency, and a cyclotron needs two treatments at once. For the low-frequency, high-impedance DC measurement, the electrometer manual’s practice applies: shields on signal low, signal low meeting earth at one chosen point, short cables tied down against flexing, since a flexing cable generates small currents of its own (Keithley 617 manual, §2.7.5, §2.14). For the RF, the opposite habit applies, because at megahertz a shield bonded at one end only carries a voltage along its length and radiates, while one bonded at both ends carries the return current and keeps the field inside. Shields and enclosures are bonded at low impedance at both ends and continuously, the oscillator sits in its own grounded screen box, and every heater, bias, control and measurement lead crossing the RF boundary is filtered, while protective bonds and the intended RF feed and return keep their continuity and are not (Wouters (1949), p. 9, Livingston & Boch (1952), p. 53, safety page). The two sets of cables are kept apart. The chamber serves both only because it is designed to: as the engineered RF return, and as the one point where the measurement’s signal low meets earth. A laminated or glued magnet core follows its own engineered bonding scheme, grounded at one point so the iron itself does not form a loop (Tanabe (2005), p. 23); that one point is not permission to leave frames, clamps or covers unbonded.

To understand it. The electrometer’s own manual is the worked example of advice that needs interpreting. Its older ground-loop illustration shows only one instrument connected directly to power-line ground; its power-connection instructions require a properly earthed receptacle to prevent shock (Keithley 617 manual). The manual does not itself say how those two fit together, and the one thing a reader must not do is resolve them by disconnecting a protective earth. The useful principle, which is the modern engineering reading rather than the manual’s literal text, is to arrange the signal references so that no unwanted current circulates while every safety bond stays in place. Some instruments isolate input low from their chassis and some tie the two together; the relationship varies by instrument and is found in its manual or by measurement, never assumed. Where low is isolated, the builder chooses the one place where signal low meets earth, and the other instruments reference that point through the signal cables rather than through their own chassis. Where an instrument’s low is tied to its chassis internally, the remedies are an isolated input, a differential input, or a battery-powered front end. A differential input is not necessarily isolated (check its common-mode range and rating), and a battery-powered front end stays isolated only while every charging, readout and control connection preserves it. The experiment the manual recommends, on the lowest range, is an experiment on signal-reference wiring, never on earth conductors.