Selecting a High-Voltage Pulse Generator

1Start With the Load

A pulse generator is only as good as the match between its output stage and what it drives. So the first question is not "how many volts" but "what does the load look like." Loads fall into two families that place very different demands on the pulser.

A resistive load draws current in proportion to voltage, steady for the duration of the flat top, set directly by Ohm's law. A capacitive load draws current only while the voltage is changing, and that current is set by how fast you change it: I = C times dV/dt. The faster the edge and the larger the capacitance, the more peak current the output stage must deliver, even though the steady-state current at the flat top is near zero.

This is why capacitance, not just voltage, sets the current demand for fast pulsing. A Pockels cell, a deflection plate, an extraction grid, or a length of coaxial cable all present capacitance. Drive a 100 pF load to 5 kV in 25 ns and the edge demands roughly 20 A of peak current, all of it just to charge and discharge the capacitor.

For a deeper treatment of charging a reactive load and keeping a fast edge clean into capacitance, see the companion note on pulsing capacitive loads. Get the load right first, and every specification that follows becomes a check rather than a guess.

2Voltage Range and Bipolar Offset

The voltage specification of a pulser is really two numbers, not one. A modern high-voltage pulser uses a half-bridge output stage: an upper switch ties the output to a positive high rail (VHigh) and a lower switch ties it to a low rail (VLow). The output node sits between them, and the two switches operate in complement, so the output swings between VLow and VHigh rather than switching a single rail on and off.

That architecture is what gives a pulser its flexibility. Set VLow to ground and VHigh positive and you get a conventional unipolar positive pulse. Set VHigh positive and VLow negative and the output swings through zero into both polarities, a true bipolar swing. Many applications need it: bias networks driven above and below a reference, electro-optic cells that respond to field direction, and grids that require an active reset to a negative level rather than a passive decay.

When you read a model's voltage line, note whether it is quoted as a single magnitude, a plus-or-minus range, or independent VHigh and VLow settings. A unit listed at +/-2,000 V can place its high and low levels anywhere in that window, which is what lets one instrument serve unipolar and bipolar jobs alike. The PVX-4000 and PVM-1001 both expose independent level settings of this kind.

3Rise and Fall Time vs Your Timing Budget

Rise time and fall time set how quickly the output transitions between levels, and they matter for two reasons. First, they define the sharpness of the event: a Pockels cell that gates a laser pulse needs an edge faster than the optical window you want to open. Second, the edge speed drives the peak current into a capacitive load, as Section 1 showed, so a faster edge is not free.

Read the edge specification against your timing budget. If the experiment needs a 100 ns gate with 10 ns of margin at each end, an edge well under 10 ns leaves room; an edge of 60 ns does not. Where the application demands the fastest possible transition, the 765-HV provides the fastest edge in the DEI line, and the PVM-1001 reaches single-digit-nanosecond rise times at lower voltage. A slower edge is sometimes preferable, though: it lowers peak current, reduces ringing into a reactive load, and eases the average-power burden at high repetition rates.

4Pulse Width and Duty Cycle

Pulse width is the time the output holds at the active level; duty cycle is the fraction of each period spent there. Together they connect the shape you want to the power the pulser must handle. A 1 microsecond pulse at 1 kHz is a 0.1% duty cycle and gentle on the output stage; the same width at 100 kHz is a 10% duty cycle and a very different thermal proposition.

Check three things. First, that the pulser can produce the minimum width you need without the edges merging, which sets a floor near the sum of the rise and fall times. Second, that it supports the maximum width or full duty cycle you require, since some fast pulsers are optimized for short pulses and droop badly when held long. Third, that width and repetition rate together stay inside the duty-cycle and average-power limits below. Pulse width, duty cycle, and average power are one connected budget, not three independent numbers.

5Droop and Flat-Top Fidelity

An ideal flat top holds a constant level for the full width of the pulse. A real one sags. Droop is that downward slope, and it comes from the finite energy stored in the output stage discharging into the load over the pulse duration. For a short pulse into a light load it is negligible. For a long pulse into a heavy or leaky load it can reach several percent, enough to matter when the flat-top level is the quantity your experiment depends on.

The same edges that carry the pulse can also carry overshoot, a brief excursion past the target level at the top of the rising edge, followed by ringing as the output settles. It is the interaction of the fast edge with the inductance and capacitance of the output and load. A clean flat-top specification accounts for all of it: the settling after overshoot, the droop across the hold, and the return to baseline. When flat-top accuracy drives your result, weigh droop and overshoot as heavily as the headline edge speed.

6Jitter

Jitter is the shot-to-shot variation in when the output edge arrives relative to the trigger. It is small, measured in picoseconds to nanoseconds, but it sets the floor on timing precision in any synchronized system. If a pulser gates a camera, fires a laser, or interleaves with other instruments on a shared timebase, jitter blurs the alignment between events.

For a single isolated pulse it rarely matters. For a system where many channels must land inside a tight window, or where you average many shots and need them to stack cleanly, low jitter is essential. Read the jitter figure against the timing window your experiment can tolerate, the same way you read edge speed against the gate width. A pulser with a fast edge but loose jitter can still miss a narrow coincidence window.

7Average-Power Limits

Every pulse deposits energy in the output stage, and the stage can only shed heat so fast. Average power, not peak voltage, is what usually limits sustained repetitive operation. It scales with energy per pulse times repetition rate, so a unit that drives a load comfortably at 1 kHz may overheat at 100 kHz even though every pulse is identical.

This is why a model's voltage rating and its maximum repetition rate are quoted together, and why the highest-voltage parts often carry the lowest rep-rate ceilings. Energy per pulse rises with the square of voltage and with load capacitance, so a 10 kV unit reaches its thermal limit at a far lower rate than a 1,500 V unit into the same load. When you size a pulser for repetitive work, multiply through (energy per pulse times rate) and confirm the product stays inside the datasheet average-power envelope. The capacitive-load note covers this energy budget in more detail.

8Built-In Monitors

High-voltage, fast-edge signals are awkward to probe directly. A standard scope probe loads the node, and a high-voltage divider adds its own capacitance to the very thing you are measuring. That is why many DEI pulsers provide built-in monitor outputs: a voltage monitor (V monitor) that reports a scaled replica of the output voltage, and a current monitor (I monitor) that reports the current the output stage is delivering.

These are more than a convenience. The V monitor lets you watch the actual flat-top level, droop, and overshoot without perturbing the output. The I monitor reveals the peak charging current into a capacitive load, the quickest way to confirm that your edge speed and load capacitance are within the output stage's reach. When you commission a setup or chase a marginal pulse, the monitors turn an opaque high-voltage node into something you can read on a low-voltage scope channel. Check that the models on your shortlist provide both, and note their scale factors.

9Single-Shot vs Repetitive Operation

The last axis is how often the pulse fires. Single-shot operation produces one pulse on command, with as much recovery time as needed before the next. Repetitive operation runs a continuous train at a fixed rate. The distinction matters because the limiting specification changes between the two modes.

In single-shot or low-rate work, peak performance dominates: you can use the full voltage and the fastest edge because there is time to cool between events. The PVX-2506, for instance, drives pulsed current-voltage tests where each measurement is a short burst with ample settling time. In high-rate work, average power dominates, as Section 7 described, and usable voltage or edge speed may have to back off to stay inside the thermal envelope. Some units add a burst mode that allows a short run above the continuous ceiling, useful when the experiment needs a brief fast train rather than indefinite operation. Decide which mode your application lives in before you compare numbers, because the same pulser can look very different in each.

10A Selection Starting Point

The table below maps representative DEI models against the three numbers that most often narrow a search first: peak voltage, maximum repetition rate, and edge speed. Read it as a starting point, then return to the sections above to weigh droop, jitter, monitors, and the load match for the specific job. A row that clears your voltage and rate is only a candidate until its flat-top fidelity and average-power envelope clear your application too.

11Talk to an Engineer

The right pulser is the one whose weakest relevant specification still clears your application. That depends on details a table cannot capture: the exact load capacitance, the cable run, the timing window, the duty cycle, and how much droop or jitter your result can tolerate. Working those through with someone who knows the line is faster than reading datasheets alone.

For a quick question, an engineer can usually tell you in minutes whether a model fits or whether a different one in the line is the better match.

For help choosing, contact a BNC applications engineer at info@berkeleynucleonics.com or 800-234-7858.