电子倍增器高压电源在环境监测仪与粒子探测中的低噪声

Environmental monitors that count particles or ionizing events usually run unattended for long periods, and the supply's noise is what limits how small an event can be detected. The distinction that matters is between noise that appears as a continuous baseline and noise that appears as discrete pulses: a baseline that moves by a few millivolts can be averaged out over time, while a discrete pulse that mimics a real event cannot be told apart from one, and it counts as a false detection.

Electron multipliers in environmental instruments operate at lower gain than the plates in imaging detectors, typically a few hundred to a few thousand, because the input rates are low and the readout's dynamic range is limited by the counting electronics rather than the multiplier. Lower gain means the output pulses are smaller, which means the supply's noise floor is a larger fraction of the pulse height, which means the discriminator's threshold must be set higher, which loses small events. This is the chain of consequences that makes supply noise a system-level parameter rather than a component specification.
The supply's discrete noise comes from three sources. Switching transitions inside the supply radiate and conduct into the output; corona within the supply's insulation produces small pulses at irregular intervals; and the multiplier itself, being a high-field device, produces microdischarges that are indistinguishable from real events. The third is not the supply's fault, but a supply with a large output capacitance will feed more energy into each microdischarge and make the pulses larger, so the supply's design does influence the third source's severity.
Switching noise is best addressed by topology rather than filtration. A supply that switches at a few hundred kilohertz with a resonant transition produces a spectrum with narrow peaks at the switching frequency and its harmonics, and those peaks can be filtered with a notch if the frequency is stable. A hard-switched supply produces a broad spectrum that cannot be notched. For a counting instrument, the resonant topology with a stable frequency and a well-defined filter is the more practical choice, even though its efficiency is slightly lower.
Corona is the source that is hardest to eliminate and easiest to find. A supply in a clean, dry environment will produce corona at a rate that depends on the field's concentration points, and those points are usually at the edges of the divider's resistors, at the termination of a cable, or at a solder joint's sharp edge. Increasing the radius of those edges, or potting them, reduces the rate. In a counting instrument, a corona rate of a few per hour is acceptable if each pulse is below the discriminator's threshold; the same rate with pulses above the threshold produces false counts that cannot be separated from the data.
The measurement that identifies a supply's contribution is a run with the multiplier disconnected and a dummy load connected. Any pulse that appears in that run comes from the supply or the cable, not from the detector. This test takes an hour and should be part of the instrument's acceptance, because it establishes the floor against which the detector's own noise is measured. Instruments delivered without this test often spend months attributing supply noise to the detector.
Long-term unattended operation adds a requirement that laboratory instruments rarely have: the supply must not drift in a way that changes the discriminator's effective threshold. The threshold is set in absolute terms, and the pulses' height scales with the voltage, so a supply drift of a tenth of a percent shifts the threshold's position in the pulse height distribution by one percent of the gain, which for an exponent of eight is eight tenths of a percent. Over a year, that is enough to change the count rate by several percent, which in an environmental monitoring context looks like a change in the environment.
The fix is either a very stable divider or a periodic automatic calibration. Automatic calibration is practical if the instrument has a reference source, such as a built-in pulser that injects a known charge into the front end. The pulser's amplitude is derived from the supply's voltage through a divider, so a pulser-based calibration measures the whole chain and corrects the supply's drift along with everything else. Instruments that have this feature hold their calibration for years; those without it need a technician with a reference source at every annual service.
Temperature is the dominant driver of drift in unattended installations, which are often in unheated enclosures. A supply that is specified for a laboratory's ambient will have a divider that moves several hundred parts per million over a forty-degree swing, and that is more than the annual drift budget allows. Housing the divider in a small insulated enclosure with a heater, or specifying a divider with a low temperature coefficient, are the two options, and the second is usually cheaper if the specification is written into the purchase order rather than discovered in the field.
The cable's contribution is worth measuring separately. A long cable to a multiplier in a remote enclosure carries the supply's ripple and also picks up ambient fields. A braided shield with a semiconductive inner layer reduces both, and the cable's termination should be potted into the multiplier's socket rather than left as a connector. Field experience is that most supply noise cases in remote installations turn out to be cable terminations that have absorbed moisture.
For particle detectors in outdoor enclosures, the supply's input is also a path for disturbances. A nearby lightning strike induces a surge on the mains, and the supply's input filter must absorb it without passing it to the output. Metal oxide varistors and a common-mode choke are standard, but the varistor's clamp voltage must be below the input capacitor's rating with margin, and the choke must not saturate during the surge. These details are usually specified together as a surge category, and the category should be chosen for the installation's exposure rather than for the laboratory's.
Counting statistics are what make the whole chain defensible. If the instrument records both the count rate and the supply's voltage, a change in the count rate that correlates with a change in the voltage is a supply problem; a change that does not correlate is an environment or detector problem. The correlation analysis is trivial to compute and is the most useful diagnostic in the instrument's log.
The narrow conclusions: specify the supply's discrete noise, not its ripple; test the supply with a dummy load to establish the noise floor; provide a pulser-based calibration so the drift is corrected rather than accumulated; and treat the cable and the divider's temperature coefficient as part of the noise and drift budget. Those four items determine whether an unattended monitor produces data that can be trusted over a season.
Field service on these instruments usually starts with the question of whether the baseline is moving or the counts are changing. A two-hour window with the source's shutter closed answers it: a baseline that drifts during the window is electronic, and a count rate that is steady during the window while the day's data is not is environmental or detector-related. That test costs nothing and saves a site visit.
A detail about the multiplier's socket: the socket's insulation is often the first thing to absorb moisture in a humid enclosure, and a moisture-laden socket produces a small leakage current that raises the multiplier's effective cathode potential. The result is a gain that appears to rise at night when the enclosure cools and condenses. Potting the socket, or fitting a heater to the multiplier's housing, removes the effect, and the heater's few watts are cheaper than a site visit in the rain.
Spare parts strategy for a network of monitors is driven by the same logic as the mobile system: the module that stops the instrument is the one to stock. In practice that is the high-voltage module itself, because the counting electronics and the detector are usually reliable. Keeping one module per ten instruments, plus a documented swap procedure that takes under an hour, is the difference between a two-day outage and a two-week one.
On the question of how much noise is acceptable, the useful reference is the detector's own pulse height. If the supply's discrete pulses are ten times smaller than a single event's pulse at the discriminator's threshold, they will be discarded by the threshold and the count rate is clean. If they are comparable, no filtering will separate them, and the supply's insulation has to be reworked. Setting the acceptance criterion in those terms, rather than in ripple volts, makes the specification testable at the customer's site with the customer's own equipment.
For an unattended monitor, the count rate itself is the final arbiter: if the instrument's daily count rate is stable across a week of stable weather, the supply's noise is below the detector's and the problem, if any, lies elsewhere. If the rate wanders with the supply's temperature, the cause is found in a week of logging rather than in a year of speculation.