Power and Load Provocation — Margins, Brownout, and Load Steps

Section Overview

The power-and-load intermittent fails when its source weakens or its demand peaks, and this section drags it to that edge deliberately — the campaign's third provocation (why-intermittents-are-the-hardest-faults). The mechanism is eroded margin. A tired electrolytic, a regulator near dropout, or a resistive connector consumes the reserve the designer left, so the device works in generous conditions and fails at the edges — which is why the history says battery, heavy task, or cheap adapter. The source is squeezed by margining. Voltage margining sweeps the input down through the rated range toward brownout and finds the exact voltage at which the fault appears — a number a known-good unit turns into a verdict (load-and-regulation-testing). The demand is spiked by stepping. A load step — a dummy load switched onto a rail, or the device's own heaviest function run on command — recreates the sag-and-recover transient under an armed tripwire instead of the owner's timing (capturing-transients-and-single-shot-events). The bench's generosity is the trap. A stiff supply, short fat leads, and a fresh battery hide exactly this class, so source substitution — the owner's real adapter, tired battery, and long cable — is often the whole summons. Discipline bounds the squeeze. The sweep stays inside absolute ratings, the current limit protects the patient, and the bench's own artifacts — a limiting supply's sag, a lead's drop — are kept out of the evidence (measuring-rail-voltage-ripple-and-noise). And the margin proves the repair. The failing voltage or load level is the fault's measure, and the fix is proven when the same sweep and steps show the margin restored. Squeeze the source, spike the demand, respect the ratings, subtract the bench, and measure the margin — and the edge-of-tolerance fault performs on schedule.

Why This Matters

Power-and-load intermittents are the great impersonators — they present as software crashes, random reboots, and no-fault-founds, while the real story is a margin quietly consumed by aging parts (why-intermittents-are-the-hardest-faults). This matters because the bench hides this class better than any other: the stiff bench supply and fresh cabling are a transplant of generous conditions, so the device that fails nightly at home passes every bench test until the owner's own source is put back in the loop. This matters because the margin is a number: "fails below 4.6 volts in" or "resets at the second load step" is quantified evidence that survives arguments, compares across units, and measures the repair — the rarest luxury in intermittent work (load-and-regulation-testing). It matters because demand peaks are schedulable: the crash that needed the owner's video call happens on command once a dummy load or the heaviest function stands in for it, with the tripwire armed on the rail (capturing-transients-and-single-shot-events). It matters because the provocation can lie: a bench supply hitting its current limit sags exactly like a failing regulator, and long thin test leads drop voltage the home wiring never did — artifacts that misdiagnose the healthy (measuring-rail-voltage-ripple-and-noise). And it matters because margin restored is proof in a number: a repair that moves the failing voltage from 4.6 back below the rated minimum is verified arithmetic, not hope. Learn to squeeze margins on purpose, and the class that impersonates everything else becomes the easiest to measure.

Required Prerequisites

Why Intermittents Are the Hardest Faults — Section 9.1 built the failure window, the reproduction recipe, and the instrumented observable; this section executes them for the power-and-load class.
Load and Regulation Testing — Section 5.3 taught dummy loads and how healthy regulation behaves under demand; this section turns those tools from characterising rails into summoning the faults that hide at their edges.

Recommended Consumables

Power resistors or an electronic load — to step demand onto rails in known increments (load-and-regulation-testing)
A notebook or worksheet — to record margins, failing voltages, load levels, and the shrinking recipe
The owner's adapter, battery, and cabling — to substitute the real source the bench supply replaced
Clip leads rated for the currents involved — to switch loads without adding accidental resistance
A printed copy of the device's input ratings — to bound every sweep before it starts

Recommended Practice Hardware

An adjustable bench supply with a settable current limit — to sweep inputs and protect the patient (load-and-regulation-testing)
A device with a known power-window fault — or a healthy one to baseline — to practise the sweep-and-step cycle (why-intermittents-are-the-hardest-faults)
A scope with single mode and a tripwire on the rail — to witness every sag, step, and failure point (capturing-transients-and-single-shot-events)
A selection of adapters and cables of varying quality — to feel how much source substitution changes the story
A partly discharged battery or battery simulator — to recreate the weak-battery window on demand
A notebook of margin baselines — to build known-good failing points, per device family

Real-World Applications

Margin work turns the vaguest complaints into the crispest numbers on the bench. A technician with a tablet that "crashes randomly, but only unplugged" sweeps the input down on a battery simulator and finds it resets at a voltage a healthy unit sails through — the margin names the fault class in one sweep (load-and-regulation-testing). A repairer facing a router that reboots when the refrigerator starts recreates the mains dip as an input step on the bench supply and catches the reset with a tripwire on the 3.3-volt rail — the compressor never needed to attend (capturing-transients-and-single-shot-events). Someone whose console fails only in the loading screen steps a dummy load onto the suspect rail and watches the sag bottom out below the reset threshold — the demand peak was the summons all along. A technician about to condemn a healthy board notices the bench supply's current-limit light flicker during the sag and moves the limit up — the "failing regulator" was the bench protecting itself (measuring-rail-voltage-ripple-and-noise). And a bench proving a recap job re-runs the margin sweep and shows the failing voltage moved from inside the rated range to well below it — the repair measured, not presumed (why-intermittents-are-the-hardest-faults). The failures this prevents: a battery-window fault chased as software, a mains-dip reboot blamed on firmware, a bench artifact condemning a healthy supply, and a recap shipped with no number showing it changed anything.

Common Challenges

The bench replaces the suspect without noticing. Plugging into the bench supply removes the owner's adapter, battery, and wiring from the circuit — often removing the fault with them — the difficulty is remembering that the source is part of the device under test (why-intermittents-are-the-hardest-faults).
Provocation artifacts imitate the fault. A current-limited supply sags, long leads drop, and thin clips add resistance — the difficulty is proving the squeeze came from the fault and not the bench, which is what lead placement and limit checks exist for (measuring-rail-voltage-ripple-and-noise).
The window may need a partner. Margins shift with temperature — regulators lose headroom warm, and high-ESR capacitors are at their worst cold — the difficulty is recognising power-plus-thermal combination windows in either direction instead of acquitting on one room-temperature sweep.

Safety Notes

Risk Level: Medium. Deliberate supply manipulation and load switching on live equipment adds electrical stress and heat to the standing live-work discipline.

Squeezing margins is deliberate electrical stress — bound every sweep by the ratings and let the current limit protect the patient

Margining means driving a device toward the edge of its ratings on purpose, so the sweep's bounds are written down first — never above the rated maximum input, and downward sweeps stop at the point of failure, not beyond it — because operating outside absolute ratings creates faults instead of finding them (load-and-regulation-testing). The bench supply's current limit is set before power is applied — generous enough not to impersonate a sag, tight enough to protect a faulted board — and any limit event during a run is logged as a possible artifact (measuring-rail-voltage-ripple-and-noise). Dummy loads get hot: size resistors for the power they will dissipate, mount them clear of the board and anything flammable, and switch them with leads rated for the current. Mains-side margining — variacs and input dips on line-powered equipment — is mains work: a variac is an autotransformer and isolates nothing, so the isolation transformer is separate and mandatory, on the circuit under test and never on the scope, with the board treated as live and no improvised mains switching (capturing-transients-and-single-shot-events). Repeated brownout resets can also corrupt a device's stored firmware or settings if they strike during writes — avoid margining through write-heavy operations, and back up configurations where corruption would be costly. The takeaway: the provocation is controlled starvation, never a feast — squeeze downward inside the ratings, cap the current, and keep the hot loads away from the patient.

Professional Tips Before Starting

Write the ratings before the sweep. The input range and absolute maximum come from the label or datasheet, into the notebook, before the knob moves — bounds decided in advance don't get improvised past (load-and-regulation-testing).
Baseline a known-good first. A margin means little alone — the same sweep on a healthy unit turns your number into a verdict.
Put the meter at the device, not the supply. Leads drop voltage under load — the margin that matters is at the input connector, so sense there (measuring-rail-voltage-ripple-and-noise).

The Power Summons — Squeeze, Step, Substitute, and Measure

Recap and Frame

Section 9.1 named the class that answers to demand; this section supplies its two levers — squeeze the source and spike the load — and its special prize: a fault that reduces to a number (why-intermittents-are-the-hardest-faults). The mechanism is a spent reserve. Designers leave margin between worst-case conditions and failure, and aging spends it — capacitors dry out, regulators drift toward dropout, contacts grow resistance — until ordinary edges of use fall outside the shrunken reserve. The tools are already on the bench. The adjustable supply and dummy loads from power-rail analysis characterised healthy regulation; here they run the same moves with hostile intent, hunting the level at which behaviour breaks (load-and-regulation-testing). The observable stays armed. Failures at the edge are sags and resets — fast and unannounced — so the tripwire on the rail and the envelope measured beforehand are what turn the failing moment into a record (capturing-transients-and-single-shot-events). The bench must subtract itself. This class is the one the bench hides best and imitates most, so the discipline of limits, lead placement, and source substitution runs through every move (measuring-rail-voltage-ripple-and-noise). And the margin is the measure. Where other provocations end at a place — a joint, a zone — this one ends at a number: the voltage or load level where failure begins, compared against known-good, restored by repair. Hold the frame — spent reserve, hostile characterisation, armed observables, a subtracted bench, and a numeric verdict — and the class that impersonates everything becomes the most measurable fault on the bench.

The Power Window — Source, Demand, or Both

The history sorts this class into two windows, and the sorting chooses the first lever (why-intermittents-are-the-hardest-faults). Source windows read as circumstance. Fails on battery but never plugged in, fails on the travel charger but not the original, fails when the lights dim or the compressor starts — the device is innocent of demand and guilty of a weakening source, so the summons squeezes the input. Demand windows read as activity. Fails during the game, the print head's return, the loading screen, the transmit burst — the source is steady and the demand peaks, so the summons steps the load. Mixed windows read as both. Fails on battery during the game but survives either alone — a source already sagging meets a demand spike, and the summons must combine the levers to recreate it. The impersonations are worth knowing. Source and demand faults present as software crashes and random reboots because the failing rail resets the processor mid-task, so a "software" complaint with a circumstantial pattern belongs in this class until the rail is cleared (measuring-rail-voltage-ripple-and-noise). Thermal partnership is common. Margins erode further warm — regulator dropout rises with temperature and semiconductor leakage grows — so histories that say "on battery, after a while" are power-plus-thermal windows that need the soak before the sweep; and the partnership runs the other way too, because high-ESR electrolytics are at their worst cold, so a fails-at-cold-start history pairs the chill with the sweep instead. Circumstance to the source, activity to the demand, both to the mixed window, impersonations recognised, and the thermal partner remembered — and the power window is read. Sort the history by what was weak and what was hungry, and the first lever chooses itself.

Margining the Source — The Sweep and the Substitution

The source-side summons has two moves: sweep the input, and put the owner's real source back in the loop (load-and-regulation-testing). The sweep walks the input downward. From nominal, the bench supply steps down in small increments through the rated input range and on toward brownout, with a pause at each step for the device to show itself, and the observable armed the whole way — the voltage at which the fault first appears is the number the session exists to find (capturing-transients-and-single-shot-events). The known-good calibrates the number. A healthy unit swept the same way fails somewhere too — below the rated minimum, where it is allowed to — and the gap between the patient's failing point and the healthy one is the eroded margin in volts. Dips recreate the compressor. Beyond the steady sweep, brief input dips — a step down and back, timed like a motor start or a lights-dim event, within the bench supply's slew limits — recreate the transient version of the window for the faults that ride disturbances rather than levels. Substitution recreates the home. Source substitution puts the owner's actual adapter, the tired battery, and the long thin cable back into the circuit, because the weakness may live in the source itself — and a fault that appears with the owner's adapter and vanishes with the bench supply has just been localised without opening the case. The sensing point is the device. The voltage that matters is at the input connector under load, not at the supply's display — leads drop, and the margin is measured where the device feels it (measuring-rail-voltage-ripple-and-noise). Step down with pauses, calibrate against known-good, add dips for the transient windows, substitute the real source, and sense at the device — and the source-side summons is complete. Walk the input to the edge on purpose, and the fault's address becomes a voltage.

Stepping the Load — Demand on Command

The demand-side summons makes the hungriest moment schedulable (load-and-regulation-testing). The device's own function is the first load. The heaviest thing the device does — the game, the print cycle, the transmit burst — is run deliberately and repeatedly, because it is the exact demand the history names and it exercises the true load path. Dummy loads make demand adjustable. Power resistors or an electronic load switched onto a rail add demand in known increments, so the sag that once needed the loading screen's timing happens at a chosen level, on a chosen schedule (why-intermittents-are-the-hardest-faults). The step is the event, not the level. Marginal supplies often survive steady load but stumble on the transition, so the switching itself — load in, load out, repeated — is the provocation, and the scope's capture of the sag's depth and recovery is the evidence (capturing-transients-and-single-shot-events). The tripwire stakes below the healthy envelope. The rail's normal worst-case dip is measured first, the trigger goes just below it, and every capture after that is a genuine excursion — the staking discipline from rail measurement, reused (measuring-rail-voltage-ripple-and-noise). Steps combine with the other levers. A load step at a lowered input, or after a warm soak, recreates the mixed and combination windows that single levers miss — and the recipe records which combination finally summoned. The real function first, dummy loads for adjustability, the transition as the event, the tripwire staked under the envelope, and levers combined for the stubborn windows — and demand answers on command. Schedule the hungriest moment, and the fault that hid behind the owner's timing loses its calendar.

The Discipline — Ratings, Limits, and the Bench's Own Artifacts

This provocation manipulates the same quantities it measures, so its discipline is about not poisoning the evidence — or the patient (load-and-regulation-testing). Ratings bound every sweep. The input range and absolute maximum are written down before the knob moves; downward sweeps stop at the failing point rather than grinding past it, and upward excursions never exceed the rated input, because stress beyond ratings manufactures faults. The current limit is set deliberately. Too tight and the bench supply sags on its own schedule, impersonating the fault; too loose and a faulted board cooks — the limit is chosen for the device's real draw plus its transients, remembering that a switching front-end draws more input current as the input voltage falls, so the limit must cover the sweep's bottom end, where a trip is expected physics rather than fault current — and any limit event during a run is logged as a possible artifact. Leads and clips are part of the circuit. Long thin leads drop voltage under load and thin clips add resistance, so the bench can accidentally perform its own source-side provocation — heavy leads, short runs, and sensing at the device keep the squeeze intentional (measuring-rail-voltage-ripple-and-noise). Artifacts are ruled out by isolation. A response that might be the bench is retested with the variable removed — the limit raised, the leads shortened, the supply swapped — and only responses that survive belong to the device (why-intermittents-are-the-hardest-faults). Hot loads stay clear of the patient. Dummy resistors dissipate real power: sized for it, mounted clear of the board and anything flammable, and switched with rated leads. Ratings written and obeyed, the limit chosen and logged, leads made honest, artifacts isolated away, and heat kept clear — and the squeeze stays diagnostic. Starve with intent and never past the ratings, and the evidence stays clean.

The Measure — Margin as Diagnosis and Proof

This class pays the campaign back with the cleanest currency in diagnostics: a number (why-intermittents-are-the-hardest-faults). The failing point is the fault's measure. "Resets below 4.6 volts input" or "fails at the second load increment" is a quantity — repeatable, comparable, and honest in a way a description never is. Known-good turns it into a verdict. The same sweep on a healthy unit — or the datasheet's rated minimum — shows where failure is allowed to live, and the patient's gap from that reference is the eroded margin the repair must restore (load-and-regulation-testing). The margin points at suspects. A shrunken source-side margin points at the reserve holders and the path — dried electrolytics, a regulator near dropout, resistive connectors and jacks — and the scope's view of the sag's shape under a load step narrows which (measuring-rail-voltage-ripple-and-noise). The repair is measured, not declared. After the recap, the reflow, or the jack replacement, the same sweep and the same steps run again, and the failing point must move — back below the rated minimum, back beyond the demand the history names (capturing-transients-and-single-shot-events). The margins file with the family. Failing points for healthy and faulty units of a model are bench gold: the next device of the family gets compared against a known number instead of a fresh campaign. A failing point as the measure, known-good as the reference, the gap naming suspects, the repair moving the number, and the numbers filed — and the power window closes on arithmetic. Measure the margin before and after, and the repair proves itself in volts.

Common Mistakes

Testing only on the stiff bench supply. The bench's generosity removes the owner's weak source — often removing the fault — substitute the real adapter, battery, and cabling before acquitting (why-intermittents-are-the-hardest-faults).
Sweeping without written bounds. Improvised excursions past ratings manufacture faults — write the input range and absolute maximum down first, and stop downward sweeps at the failing point (load-and-regulation-testing).
Blaming the device for the bench's sag. A tight current limit and thin leads impersonate a failing supply — log limit events, sense at the device, and retest artifacts with the variable removed (measuring-rail-voltage-ripple-and-noise).
Watching levels and missing transitions. Marginal supplies survive steady load and stumble on the step — make the switching itself the provocation, with the sag captured (capturing-transients-and-single-shot-events).
Repairing without re-measuring. A recap with no before-and-after margin is a hope, not a repair — re-run the sweep and steps, and show the failing point moved.

Troubleshooting Guidance

The power campaign runs window, squeeze, step, discipline, measure. If the history says battery, cheap adapter, or dimming lights: the window is source-side — sweep the input down with pauses and the observable armed (capturing-transients-and-single-shot-events). If it says the game, the print cycle, or the heavy task: the window is demand-side — run the real function, then step dummy loads in increments (load-and-regulation-testing). If neither lever alone summons: combine — a step at lowered input, or either lever after a warm soak, because mixed and power-plus-thermal windows are common (why-intermittents-are-the-hardest-faults). If the fault appears with the owner's source but not the bench's: the source itself is a suspect — test the adapter and cabling as components before opening the device. If a response coincides with the supply's limit light or heavy leads: treat it as a possible artifact — raise the limit, shorten the leads, and retest before believing it (measuring-rail-voltage-ripple-and-noise). If the fault appears: record the failing voltage or load level as the measure, and capture the sag's depth and shape for the suspect list. If a known-good unit is available: run the same sweep on it — the gap between failing points is the eroded margin in numbers. If the repair is in: re-run the identical sweep and steps and show the failing point moved below the rated minimum and past the history's demand. The throughline: sort the window, squeeze or step to the edge inside the ratings, subtract the bench, and let the margin be both the diagnosis and the proof.

Verification & Testing Methods

Confirm the power campaign was measurement, not roulette:

[ ] I sorted the power window from the history — source-side circumstance, demand-side activity, or both — and chose the first lever accordingly, remembering the thermal partner for "after a while" histories.
[ ] I ran voltage margining with written bounds: stepped down through the rated range with pauses and an armed observable, stopped at the failing point, added dips for transient windows, and recorded the failing voltage sensed at the device's input.
[ ] I made demand schedulable — the device's heaviest real function first, then a load step in known increments with the transition treated as the event — with the tripwire staked just below the rail's measured healthy envelope.
[ ] I kept the provocation honest: ratings never exceeded, the current limit chosen and its events logged, leads heavy and short with sensing at the device, artifacts retested with the bench variable removed, and hot dummy loads mounted clear of the patient.
[ ] I used source substitution to put the owner's adapter, battery, and cabling back in the loop before any acquittal — and I proved the repair by re-running the same sweep and steps, showing the failing point moved back where it belongs, with before-and-after margins filed.

Then try the practice exercises below — hands-on margin work; scenarios differ from the quiz.

Practice Exercises

Baseline a healthy margin (5 minutes, hands-on). On a known-good low-voltage device, write down its rated input range, then sweep the bench supply downward in small steps with pauses, and record the voltage where it quits — sensing at the device's connector — to build the reference a patient would be compared against (load-and-regulation-testing).
Catch a load step's sag (5 minutes, hands-on). Measure a rail's healthy worst-case dip, stake a tripwire just below it, then switch a sized dummy load in and out and read the captured sag's depth and recovery — the transition, not the level, as the event (capturing-transients-and-single-shot-events).
Feel the bench's generosity (5 minutes, hands-on). Power the same device three ways — stiff bench supply with short leads, the same supply through long thin leads, and a real wall adapter — and measure the input voltage under load at the device each time, watching the "same" source become three different sources (measuring-rail-voltage-ripple-and-noise).
Sort windows and plan the proof (3 minutes, paper). For three histories — fails on battery in games, reboots when the pump starts, dies on the travel charger only — name each window (source, demand, or mixed), the first lever you would pull, and the before-and-after margin measurement that would prove the eventual repair (why-intermittents-are-the-hardest-faults).

These core steps — sorting the power window, the bounded sweep, the scheduled load step, the artifact discipline, and margin as proof — are tested in the Chapter Quiz at the end of this chapter, where a score of 80% is required to continue.

Key Takeaways

The power-and-load intermittent is eroded margin — dried electrolytics, regulators near dropout, resistive connectors — so it fails when the source weakens or the demand peaks, presents as software crashes and random reboots, and hides best of all classes on a generous bench (why-intermittents-are-the-hardest-faults).
Voltage margining walks the input down through the rated range with pauses and an armed observable, stops at the failing point, and yields the class's special prize: a failing voltage, calibrated against a known-good unit into an eroded margin in volts (load-and-regulation-testing).
The load step makes demand schedulable — the device's heaviest real function, then dummy loads in known increments — with the transition treated as the event and the tripwire staked just below the rail's measured healthy envelope (capturing-transients-and-single-shot-events).
The bench must subtract itself: source substitution puts the owner's real adapter, battery, and cabling back in the loop, sensing happens at the device's connector, current-limit events are logged as possible artifacts, and suspect responses are retested with the bench variable removed (measuring-rail-voltage-ripple-and-noise).
The margin is both diagnosis and proof: the failing point names the suspects, the gap from known-good sizes the erosion, and the repair is verified when the identical sweep and steps show the failing point moved back below the rated minimum and past the history's demand.

Skills Learned

You can now read a fault history for its power window — source-side weakness, demand-side peaks, or both.
You can now margin the source: sweep the input through its rated range toward brownout and substitute the owner's real source.
You can now step the load — dummy loads and the device's own heaviest function — with an armed observable on the rail.
You can now keep the provocation inside absolute ratings and keep the bench's own artifacts out of the evidence.
You can now quantify the fault as a margin, and prove the repair by showing the margin restored under the same sweep and steps.

Glossary Additions

voltage margining — deliberately sweeping a device's input voltage through and below its rated range, in small steps with pauses and an armed observable, to find the exact voltage at which a fault appears. The failing point is the technique's product: a number, sensed at the device's input connector under load rather than at the supply's display, that a known-good unit or the datasheet's rated minimum turns into a verdict — the gap between where the patient fails and where failure is allowed to live is the eroded margin in volts. Sweeps are bounded by written ratings — never above the rated maximum, and downward only to the failing point — and brief input dips recreate the transient windows (motor starts, dimming lights) that steady levels miss. After repair, the same sweep must show the failing point moved back below the rated minimum, making the margin both the diagnosis and the proof.
load step — a sudden, deliberate change in demand on a supply or rail, used as a provocation: a sized dummy load switched in and out, or the device's own heaviest function started on command, so the sag-and-recover transient that once needed the owner's timing happens on a schedule. The transition is the event — marginal supplies often survive steady load but stumble on the step — so the switching is repeated with a scope tripwire staked just below the rail's measured healthy envelope, and the captured sag's depth and recovery shape feed the suspect list. Load steps combine with the other levers for stubborn windows: a step at lowered input recreates mixed source-and-demand faults, and a step after a warm soak recreates the power-plus-thermal pairings that single levers miss.
source substitution — putting the owner's actual power source — the adapter, the tired battery, the long thin cable, the power strip — back into the circuit in place of the bench's stiff supply and short leads, because the bench's generosity is precisely what hides source-side faults. Substitution works in both directions as a localiser: a fault that appears with the owner's source and vanishes with the bench's has been traced to the source side without opening the case, while a fault that persists on the bench supply has been cleared of the source and pinned on the device. The substitution includes the wiring — leads and clips are part of the circuit, and the voltage that matters is measured at the device's input connector under load, where the device actually feels it.

Suggested Next Sections

Must read next:

The Long Watch — Monitoring and Logging Over Time — Section 9.5 arms the campaign for the windows no lever can compress: turning instruments into unattended sentries — logged meters, scope tripwires, and event counters — that gather evidence over hours and days while the bench does other work.

Recommended:

Capturing Transients and Single-Shot Events — the tripwire and envelope-staking discipline every sweep and step in this section leans on.
Measuring Rail Voltage, Ripple, and Noise — the healthy-envelope measurement that stakes the tripwire, and the sag-reading skills the load step's captures depend on.