低纹波高压电源在电容测试仪与精密仪器中的纹波管理
Ripple is the one specification that a high-voltage supply cannot fake. It can be measured at the output, it appears in every reading that the instrument takes, and it sets the smallest quantity that the instrument can resolve.
A capacitor test set applies a known voltage to a device under test and measures the charge or the leakage. The leakage of a good film capacitor at rated voltage is in the nanoampere range, and a supply whose output carries a ripple of a few millivolts will drive a displacement current through the capacitor that is far larger than the leakage.
The displacement current is the product of capacitance and the rate of change of voltage. A capacitor of one microfarad and a ripple of one volt at ten kilohertz carries sixty microamperes of ripple current, which is a thousand times the leakage that the instrument is trying to measure.
Ripple management therefore begins with the definition. Ripple can be specified as a peak-to-peak value, as an RMS value, or as a spectral density at a particular frequency, and the three are not interchangeable. A supply quoted at one percent peak-to-peak may have an RMS value of a tenth of a percent or a third of a percent depending on the waveform.
For a measurement instrument the useful specification is the spectral density in the band where the instrument's integration time is shortest, because that band is where the ripple is least averaged.
The second element is the architecture. A switching stage followed by a linear post-regulator gives low ripple with reasonable efficiency, because the switching stage does the bulk conversion and the linear stage removes the residue. The linear stage dissipates the difference between its input and its output, so the input must be only a few volts above the output.
A post-regulator with a hundred volts of drop at a milliampere dissipates a tenth of a watt, which is trivial, and the same regulator at ten milliamperes dissipates a watt, which needs a heatsink. The design therefore depends on the current that the instrument actually draws.
Passive filtering alone cannot reach the levels required. A capacitor multiplier of three stages can reduce a ripple by sixty decibels, and a switching frequency of a hundred kilohertz multiplied by three stages of RC gives an attenuation that is adequate for a general-purpose instrument and inadequate for a precision one.
The third element is the layout. A ripple that measures well at the output terminals can still appear across the load if the cable between them carries the switching current. A four-wire connection with the sense leads taken at the load is the standard remedy, and it must be applied to the ripple measurement as well as to the regulation.
The fourth element is the reference. A supply whose output is low in ripple but whose reference drifts will show a slow change that a fast measurement averages away and a slow measurement does not. Instruments that integrate over seconds need a reference with a temperature coefficient below a few parts per million.
Oven-controlled references are the practical answer for the precision end. A small oven at forty degrees holds the reference within a fraction of a degree, and the drift falls to a part per million, at the cost of a warm-up time and a few watts of standby power.
The fifth element is the ground. A supply that is grounded at the instrument's chassis will inject its switching current into the chassis, and the chassis carries that current past the sensitive front end. A star ground with the supply's return brought to a single point, and the front end's return brought to the same point, keeps the current out of the measurement path.
The sixth element is shielding. The supply's transformer radiates a magnetic field that couples into the instrument's input leads, and the coupling is worst when the leads are long and parallel to the transformer's axis. A mu-metal shield around the transformer, or a change in orientation of the leads, reduces the coupling by an order of magnitude.
A capacitor test case illustrates the chain. An instrument measured leakage with a resolution of a hundred picoamperes and found that its readings depended on the range setting. The cause was a ripple that varied with the range because the output capacitor changed, and the displacement current scaled with the capacitance of the device under test.
The cure was a post-regulator with a fixed output capacitor and a range-independent ripple, plus a measurement window synchronized to the switching frequency so that the ripple averaged to zero within each window.
A second case involved a drift that appeared after an hour. The reference had a temperature coefficient of ten parts per million per degree, and the instrument's internal temperature rose fifteen degrees during warm-up. An oven-controlled reference removed the drift.
A third case involved noise that appeared only when a particular cable was used. The cable's shield was connected at one end only, and the other end floated at a potential set by the switching field. Connecting the shield at both ends, with the chassis as the reference, removed the noise and did not create a ground loop because the chassis was already the star point.
Measurement of ripple should be done with a probe that is adequate to the task. A probe with a bandwidth of twenty megahertz will show the switching residue and miss the fast edges, and a probe with a bandwidth of two hundred megahertz will show both. The instrument's own bandwidth should be quoted with the measurement.
The probe's ground lead is the usual source of error. A long ground lead forms an inductor with the probe's tip capacitance, and the resulting resonance shows a ripple that is an artefact. A short ground spring, or a coaxial connection, removes the artefact.
Temperature testing of ripple is worth the time. A supply's ripple at twenty degrees can double at sixty degrees because the output capacitor's ESR rises and the control loop's gain falls, and an instrument specified at room temperature will fail in a warm cabinet.
Ageing is the slow version of the same effect. Electrolytic capacitors lose capacitance and gain ESR over years, and the ripple rises by a few decibels annually. A supply that meets its specification at delivery may not meet it at the end of a five-year calibration cycle.
Spares for a precision supply are small: one post-regulator board, one reference, one output capacitor set. The reference is the item that must be handled with care, because a replacement requires a recalibration of the whole instrument.
Cost comparisons must include the calibration. A supply with a post-regulator and an oven reference costs perhaps twice a basic unit, and the calibration interval of the instrument doubles, which is a saving that most buyers underestimate.
Limits should be honest. No supply can deliver a ripple below the noise floor of the instrument's own front end, and a specification that promises more than the front end can resolve is a specification written for a brochure.
The design intent is a sentence. Make the ripple small enough that the instrument's resolution is set by its front end rather than by its supply, and make that true at temperature, at the end of a calibration cycle, and across every range.
Integration time is the instrument's own filter, and it should be matched to the ripple. An instrument that integrates for ten milliseconds attenuates a hundred kilohertz ripple by a factor of a thousand, and one that samples at a point attenuates nothing. The choice of measurement method is therefore part of the ripple specification.
A synchronous sampling scheme is the cleanest solution where the instrument controls the timing. Sampling at the ripple's zero crossing or at the midpoint of a full period removes the first harmonic entirely, and the residue is the second and higher harmonics, which are smaller by an order of magnitude.
Implementing that requires a stable switching frequency, which is another argument for a fixed-frequency design rather than a variable one. A supply that frequency-folds to reduce loss will defeat synchronous sampling, and the instrument's firmware will not know why its readings wander.
Leakage measurement of a large capacitor takes seconds, and a supply that is quiet for a millisecond is not sufficient. What matters is the integral of the ripple current over the whole measurement, so a slow drift in the average output counts as much as a fast ripple.
Average stability is therefore specified separately, usually as a long-term drift of a few parts in ten thousand per hour after warm-up, and it is measured with an instrument that is slower than the supply's own loop.
Thermal EMF is the last source of error that looks like ripple and is not. A junction of two dissimilar metals in the measurement path generates a few microvolts per degree, and if the supply's heat warms one junction more than the other, the instrument reads a voltage that has nothing to do with the device under test.
Keeping the measurement path isothermal, and keeping the supply's heat away from it, is a mechanical design task that no electrical specification can replace.
A practical acceptance test for a precision supply includes four measurements: ripple peak-to-peak at the load, ripple RMS in the instrument's band, long-term drift over an hour, and thermal EMF at the measurement terminals with the output shorted.
The four numbers, taken together, tell a buyer more than any single figure in a datasheet, and they cost an afternoon to obtain.
Documentation should state the measurement conditions. A ripple figure without a bandwidth, a load and a temperature is a number without meaning, and comparing two such numbers is a source of endless argument between suppliers and users.
The habit that settles arguments is a shared test method. A one-page procedure, with the probe, the bandwidth, the load and the temperature specified, allows two parties to obtain the same number and to discuss it rather than to dispute it.

