MIL-STD-461 EMI/EMC Pre-Compliance on a Small Budget: Nine Weeks from Contract to Lab Pass
The email arrived on a Monday: the EMC laboratory had a slot in nine weeks, and the contract said MIL-STD-461G.
The team had never taken a product through that standard. They had built a rugged GNSS and LoRa tracker for a defence customer, it worked, and it had passed every commercial test they had thrown at it. What they did not have was a spectrum analyser that went below 9 kHz, a screened room, or any idea what the lab would find. This is the story of the nine weeks, told in order, with the technical detail where it appeared. It is a representative composite, genericised, drawn from several programmes, and the numbers are illustrative.
Day One: A Contract That Says Four Letters and a Table
The contract line was short: the equipment shall meet MIL-STD-461G for its platform. That line is the start of the work, not a description of it. MIL-STD-461 is not one test but a menu of requirements, each with a code, and an applicability table in the standard decides which requirements apply to which platform and installation. A contract's reference to the standard means the rows that apply to your equipment, which a good lab will help you read and your customer's EMC authority has to agree.
The codes that matter for most small electronics are these. CE101 and CE102 limit conducted emissions on power leads, CE102 covering radio-frequency currents from about 10 kHz up to 10 MHz. CS101 injects audio-frequency ripple onto the power input. CS114 injects radio-frequency currents onto cable bundles, CS115 uses a fast impulse, and CS116 uses damped sinusoids at spot frequencies. RE101 limits magnetic-field emissions at low frequencies, RE102 limits radiated electric-field emissions from 10 kHz to 18 GHz, and RS103 exposes the equipment to radiated electric fields from 2 MHz upward, at levels up to 200 volts per metre depending on the platform. Revision G removed an older power-lead transient test and tightened rules about how some susceptibility levels are calibrated above 1 GHz.
The team read the platform's rows and wrote the list: CE102, CS101, CS114, CS115, CS116, RE102 and RS103. Seven tests. The nine weeks were for those seven.
Reading the Applicability Table: A Worked Example
The applicability table is the most important page in the standard for a product team, because it turns a menu of thirty-odd requirements into the handful your equipment must meet. It is organised by platform, such as ground equipment, aircraft, surface ships and submarines, and by whether a given requirement applies in every case, or only in some, depending on the equipment and the installation.
The table below describes the seven requirements the team in this story faced. The frequency ranges are as commonly published for the 461G family, and they should be confirmed against the revision in your contract. The last column records, from experience, what usually causes a miss.
Test | Range or waveform | What it checks | Usual cause of a miss |
CE102 | 10 kHz to 10 MHz on power leads | RF current the unit puts back on its supply | Switching-regulator harmonics, weak input filter |
CS101 | 30 Hz to 150 kHz ripple on the power input | Immunity to supply ripple | Regulator with poor rejection, unstable feedback loop |
CS114 | Bulk cable injection, 10 kHz to 200 MHz | Immunity to RF currents induced on cable bundles | Unfiltered interfaces, sensitive analogue inputs |
CS115 | 30 ns impulse, 5 A peak, 30 Hz repetition | Immunity to fast transients on cables | Weak reset pull-ups, unprotected inputs |
CS116 | Damped sinusoids at spot frequencies up to 100 MHz, up to 10 A peak | Immunity to ringing transients | Resonant filters, long cable loops |
RE102 | 10 kHz to 18 GHz radiated electric field | Radiated emissions from the unit and its cables | Clock harmonics, switching loops, cable resonance, enclosure seams |
RS103 | 2 MHz upward, up to 200 volts per metre depending on platform | Immunity to radiated fields | Poor shielding, sensitive analogue front ends, unfiltered cables |
Read across the table and a pattern appears. Three of the seven tests are about energy leaving the unit, and four are about energy entering it. Emissions are measured with a receiver and can be previewed on a bench. Susceptibility needs amplifiers, injection probes and fields that a small team rarely owns. That asymmetry shapes the preparation: do the emissions work yourself, and spend the susceptibility effort on design discipline.
Emissions can be measured on your bench. Susceptibility has to be designed in, because the first time you see the injection equipment will be at the lab.
Defining Pass Before the Lab Does
Susceptibility tests do not have a simple numeric limit. The equipment is exposed to a stress and its behaviour is judged against a performance criterion, and the criterion belongs in the test plan, agreed with the customer before the first day.
A good criterion is specific. It names the functions that must continue, such as position reporting at the stated rate, and the tolerance for each. It states which anomalies are acceptable and which are not: a momentary dropout of a status LED may be fine, a reset of the processor may not be. It defines recovery: a unit that stops during the stress and resumes by itself within a stated time may pass or fail depending on the wording.
The team's trouble with the fast impulse test came from an assumption. Their engineers regarded a one-second reset as a pass, because the unit recovered. The test plan, written earlier from a template, said no resets. The wording had cost them an afternoon, and in a different lab could have cost a retest.
Write the criteria in plain language for each test, have a software engineer and a hardware engineer review them, and get the customer's authority to sign them. The cost is an afternoon. The benefit is that on lab day nobody argues about what pass means.
Week One: Draw the Product as an EMC Engineer Would
Before touching a probe, the team drew the product the way an EMC engineer sees it. Power enters on a two-wire cable. A switching regulator converts it. A GNSS receiver, a microcontroller, and a LoRa radio share a board. An antenna cable of about a metre leaves for each radio. A data cable leaves for the host. Every cable that leaves the box is an antenna, and every switching edge inside it is a source.
The drawing was more useful than any measurement. It showed that the regulator ran at 500 kHz, which put its harmonics squarely inside the CE102 band, where each multiple of the switching frequency can appear as a spike on the power lead. The 25 MHz clock for the microcontroller put harmonics at every multiple of 25 MHz through the RE102 band. And a one-metre antenna cable has a quarter-wave resonance near 75 MHz, which would turn any common-mode current on it into an efficient radiator right where radiated limits are tight.
The arithmetic fits in a few lines, and running it before any measurement tells you where to look.
C = 299792458
def harmonics(f0_hz, lo_hz, hi_hz):
n = int(lo_hz // f0_hz) + 1
out = []
while n * f0_hz <= hi_hz:
if n * f0_hz >= lo_hz:
out.append(n * f0_hz)
n += 1
return out
def quarter_wave_mhz(length_m, vf=1.0):
return C * vf / (4 * length_m) / 1e6
print(len(harmonics(500e3, 10e3, 10e6))) # 20 regulator harmonics inside 10 kHz to 10 MHz
print([f / 1e6 for f in harmonics(25e6, 30e6, 300e6)]) # 50, 75, 100 ... 300 MHz
print(round(quarter_wave_mhz(1.0), 1), round(quarter_wave_mhz(0.5), 1)) # 74.9 and 149.9 MHz
Twenty harmonics of the regulator sat between 500 kHz and 10 MHz. The clock produced eleven lines between 50 and 300 MHz. A one-metre cable resonated near 75 MHz, which coincided with a clock harmonic at exactly 75 MHz. That coincidence was not luck. Clock harmonics are spaced at regular intervals, and with a few cable lengths in play, some of them are bound to align.
What to Own and What to Rent for Pre-Compliance
A small team does not need to buy a screened room. The equipment divides into what pays back from owning and what is better rented for the few days you need it.
Item | What it does | Own or rent | Note |
Line impedance stabilisation network | Presents a defined impedance to the power lead and couples out the noise for CE102 | Own if you test often, otherwise rent | Match the network the standard specifies |
Spectrum analyser, 9 kHz to at least a few GHz | Sweeps the emissions | Own for general bench work | An EMI receiver is the lab's tool, and a spectrum analyser gives a good preview |
Near-field probe set | Locates sources on a board and along a cable | Own | Inexpensive and the most useful item |
Current probe | Measures common-mode current on a cable | Own | Shows the current that makes cables radiate |
Pre-amplifier | Raises sensitivity for radiated sweeps | Rent or own | Check the noise floor with and without |
Ground plane and cable supports | Sets the standard's geometry on the bench | Build | Print the set-up figures and follow them |
Ferrite and filter parts kit | Lets you try fixes in minutes | Own | A tray of chokes and capacitors pays for itself in a day |
Small shielded enclosure or cell | Reduces ambient pickup | Rent or build | Helps in a noisy building |
Amplifier, signal source and injection probe | Rehearses susceptibility tests | Rent | Needed only if you plan a rehearsal |
A practical rule: buy the instruments you will use on every project, and rent anything that sits in a cupboard between campaigns.
Week Two: A LISN, an Analyser and a Bench
The pre-compliance kit was modest: a line impedance stabilisation network that presents a defined impedance to the power lead, a spectrum analyser with the right resolution bandwidths, a set of near-field probes, a metal sheet as a ground plane, and a few metres of cable laid out the way the standard describes. The standard's own set-up details, the height above the ground plane, the cable length and the bonding, matter, and the team printed the relevant figures and taped them to the wall.
The first CE102 sweep showed what the drawing predicted. Spikes at every multiple of 500 kHz, with the worst around 2 to 3 MHz. A reading from the unscreened bench has noise from every radio and switching supply in the building, so the team ran the unit powered and then unpowered to separate the product from the room.
The analyser's own numbers deserve respect. A pre-compliance sweep tells you the shape and the rough size of the problem, with an uncertainty of several decibels, because the bench is not a screened room and the probes are not the lab's calibrated equipment. The team used a rule they would keep: aim for at least 6 dB of margin below the limit on the bench, because the lab will be a little different and a pass by one decibel on a bench is a failure waiting for the lab.
def margin_report(meas, limit, need_db=6.0):
rows = []
for f, m in sorted(meas.items()):
mg = limit[f] - m
rows.append((f, m, limit[f], round(mg, 1), "FAIL" if mg < 0 else ("MARGINAL" if mg < need_db else "ok")))
return rows
meas = {100: 38.0, 200: 49.5, 300: 41.0} # dBuV/m, illustrative bench readings
limit = {100: 44.0, 200: 48.0, 300: 47.0} # illustrative limit values, replace with the standard's curve
for r in margin_report(meas, limit):
print(r) # the 200 MHz point fails by 1.5 dB
The Harmonics Were Not Where We Looked
The surprise came from the near-field probe. The team had expected the 25 MHz clock to be the main radiated problem, and moved the probe over the microcontroller first. The strongest signal was not the clock. It was a loop formed by the regulator's input capacitor, its switch and its return path, radiating a broad spread of harmonics near 150 to 250 MHz, with edges so fast that the harmonics reached well above what the switching frequency suggested.
This is the common story. The loud source is rarely the one the design review worried about. A fast switching edge on a power stage generates energy to hundreds of megahertz, and the area of the current loop sets how well it radiates. The team measured the loop area on the board, found it was larger than needed because the capacitor sat a centimetre away from the switch, and drew the fix: put the capacitor against the pins and shrink the loop.
They also learnt something unwelcome about the unit's enclosure. A seam in the metal case, where two halves met without a gasket, leaked at the same frequencies. Near-field probing along the seam produced a clear peak. A small conductive gasket would fix it, but not without a redesign of the lid.
Cable Resonance: One Metre, Seventy-Five Megahertz
With the board's loop improved, the pre-compliance sweep still showed a stubborn peak near 75 MHz. The probe moved to the antenna cable, then the host cable, and the peak followed the cable. A cable attached to a box is a monopole waiting for a driver, and common-mode current from the board's ground noise was driving it.
The standard's set-up places the cables at specified lengths, so the team could not shorten them to dodge the resonance. They had to reduce the common-mode current at the source or choke it on the cable.
Three measures worked in sequence, and the order is worth recording. First, a ferrite choke on the host cable close to the connector cut the peak by about 8 dB. Second, a better ground connection between the connector shell and the case, with a short, wide strap instead of a long thin wire, brought another 5 dB. Third, filtering on the interface lines at the connector, a small common-mode choke and capacitors to the case, took the rest.
A ferrite alone is rarely a cure. It works over a limited band and can saturate. The designers learnt to treat it as a diagnostic first: if a clip-on ferrite changes the peak, you have found a common-mode path, and the real fix is to shrink the path's driving voltage or give it a lower-impedance route.
A clip-on ferrite is a stethoscope before it is a medicine. If it moves the peak, you have found where the current flows, and the cure is to starve it at the source.
The Fix List, in the Order We Tried It
By week five the team had a list of fixes ranked by cost and effect, and they tried them cheapest first.
The inexpensive ones came first and delivered most of the improvement. Moving the input capacitor against the regulator pins, a layout edit. Adding a small series resistor and a snubber across the switch node, two components. Rerouting the antenna cable ground, a strap. A pi filter on the power input, an inductor and two capacitors, sized from the measured spikes at the 500 kHz harmonics, which took the CE102 sweep from failing by 10 dB to passing by 8.
The mid-cost fixes followed. A conductive gasket and a tighter seam on the enclosure. A shielded cable for the host interface with a proper 360-degree termination at the connector, in place of a drain wire. A change of the regulator's switching frequency to 450 kHz, which moved every harmonic and, as it happened, aligned fewer of them with the worst parts of the limit curve.
The expensive fix, a full board respin, they avoided. Each change was logged with before and after sweeps, so a regression could be traced and the lab could be told what had changed.
Pre-Compliance Is Not Compliance: What the Lab Sees That You Do Not
Walking into the lab with a bench-passing design is the right goal and a poor guarantee. Several differences between the bench and a formal test deserve respect.
The lab's measurement system is calibrated, the antennas are specified and the receiver follows the standard's bandwidths and dwell times. Your analyser may use a different detector or resolution bandwidth and report slightly different numbers.
The formal tests include susceptibility tests that a bench rarely reproduces. CS114 injects current onto cables at levels that need an amplifier, a coupling probe and calibration. RS103 needs a field generator and an anechoic or screened chamber. Few small teams have either, so those tests are the unknown, and the best preparation is design practice: filters on every input, no unprotected high-impedance nodes, a watchdog that recovers cleanly, and software that does not trust sensor readings blindly.
The lab sees the equipment in its worst operating mode, often with every interface active, at maximum transmit power. Your bench test may have run in a friendlier mode. List the modes in the test plan, and test the worst on the bench.
Bench Readings Against Lab Readings: Calibrating Your Own Optimism
The team's bench passed at 200 MHz with a thin margin, and the lab measured a few decibels higher at the same frequency. That difference is normal, and understanding it turns a pre-compliance bench from a source of false comfort into a useful instrument.
Several things separate a bench from a laboratory. The lab's antenna and its distance, the cable layout against the ground plane, the receiver's bandwidth and detector, and the ambient noise floor all differ from yours. Cables laid a few centimetres differently change the radiated result by several decibels, particularly near a resonance.
You can measure the offset instead of guessing. Before the lab slot, take a stable reference source, such as a battery-powered comb generator that emits regularly spaced lines across the band, and measure it on your bench. At the lab, ask whether the same source can be measured in their set-up, which many labs will do in a few minutes. The difference at each frequency is your bench-to-lab correction, and applying it to your own sweeps shows what the lab is likely to see.
After the formal test, record the correlation: the reading on your bench and the reading at the lab for each frequency that mattered. The team found its bench read about 2 to 4 decibels lower than the lab near 200 MHz, an illustrative figure for one set-up and not a rule. That one table of differences made the next project's pre-compliance far more reliable, because it replaced hope with a measured correction.
Keep the guard band even so. A 6 dB margin on the bench is still a sensible target, because production units vary and a correction applied from one test has its own uncertainty.
A bench that passes by one decibel is a coin toss at the lab. Measure the offset between your bench and theirs, then keep a margin larger than the offset.
Lab Day One: Emissions Pass, the First Susceptibility Run Does Not
The unit passed CE102 with 7 dB of margin and RE102 with just over 3 dB at its worst point, near 200 MHz. That is thin, and the lab engineer said so, but a pass is a pass, and the engineer agreed that the seam gasket should go into the next revision.
The susceptibility tests started the next morning. CS101 and CS114 passed. CS115, the fast impulse, caused a reset of the microcontroller. The unit recovered in a second, which was outside the performance criterion the test plan had agreed, because the plan said no reset.
The cause was a pull-up on the reset pin sized for low power. The impulse coupled through the cable and a stray capacitance onto the reset line, and the weak pull-up let it dip below the threshold. The lab allowed the team an afternoon to try a quick fix, and a stronger pull-up and a small capacitor on the reset line cured it.
Notice what that episode needed. It needed a margin in the schedule: the team had left half a day open for exactly this, and the lab was willing to give it. Labs vary in how flexible they are, so ask at booking.
What Failed Was Not the Radio
The failures across the campaign had little to do with the radio sections the team worried about most. The LoRa transmitter's harmonics were within limits from the first try, thanks to the module's filtering. The failures came from the plain parts: the switching regulator's loop, the cable's ground strap, a reset line's pull-up, a seam in a lid.
This is a general pattern for small electronics. The intentional radiators are usually well behaved, because their makers care about spectral purity. The unintentional radiators, the power supply and the clock, the cables and the case, decide the result. A design review for EMC should spend its time on those.
The Second Slot and a Margin You Can Defend
The team took a second lab slot to re-test RE102 after the gasket change, and to repeat the test at a thicker margin. This time RE102 passed with 8 dB at the worst point. That margin does something beyond passing: it gives room for production variation. Boards from a different batch, a different regulator lot or a slightly different cable length will vary, and a design that passes by 1 dB in the lab may fail in production by 2.
They also ran a short post-test review, a page listing what changed in hardware between the first and second slots and why, which went into the report and into the production drawing set. A qualification pass applies to the unit that was tested. Any later change, even a replacement part, needs a judgement on whether the pass still holds, and the page made that judgement easy.
Layout Rules That Would Have Saved Two Weeks
Hindsight on this project produced a short list of board and enclosure habits, each cheaper at the layout stage than in the lab.
Keep switching loops small. Put the input capacitor against the switch and the diode or synchronous device, so that the high-frequency current stays in a tight loop. The area of that loop sets how well it radiates.
Filter at the point of entry. Every cable that enters the enclosure should meet a filter immediately at the connector, before the signal reaches the board, with the filter's ground tied to the enclosure with a short, wide connection.
Bond the connector shell to the case, with a 360-degree termination for shielded cables and a short strap for others. A long thin wire to a ground pin is an inductor and a poor one.
Give return currents an obvious path. A signal that crosses a gap in the ground plane sends its return around the gap, and the loop radiates.
Treat the enclosure as a component. Seams are slots, and a slot radiates at the frequencies where its length approaches half a wavelength. Gaskets, overlapping lips and closely spaced fasteners all help.
Choose a switching frequency with the limit curve in mind. Moving the fundamental so that fewer harmonics fall on the tight part of the curve is a free change in firmware or in a resistor value.
Every one of these costs nothing at layout time and costs a respin in the lab. The EMC review belongs before the board goes to fabrication.
What It Cost in Days, Not Rupees
Costs vary with the lab and the city, so this account counts calendar days. The preparation took nine weeks including two weeks of waiting for the lab slot. Bench pre-compliance took about three weeks of one engineer's time with a part-time helper. The fixes took two weeks, including a board edit. The first lab slot ran three days, the second one day. The report arrived ten days later.
The biggest saving came from the cheapest step: drawing the product as an EMC engineer would, and computing where the harmonics and resonances would fall. It cost an afternoon and pointed at the right places on the board in week one.
What We Would Do on the First Day Next Time
Read the applicability table with the customer's EMC authority and write the test list in the first week.
Draw the product with every cable, every switching edge and every enclosure seam, and compute the harmonics and resonances.
Choose parts and layout with EMC in mind from the start: small switching loops, a filter at every entry, a solid ground path to the enclosure at each connector.
Build a bench for conducted emissions first, since it is the cheapest to set up, and a near-field probe kit for the radiated ones.
Aim for 6 dB of bench margin, list the worst operating modes and test them.
Plan for the susceptibility tests by design: filtered inputs, debounced and well-pulled-up resets, a clean watchdog recovery.
Book the lab with a spare half-day, a second slot in the schedule and a clear performance criterion in the test plan.
The Quiet After the Test
The unit was delivered. The report ran to eighty pages and one page mattered most: the table of margins, with every test, its limit, its worst reading and the difference. Nobody reads the rest.
Looking back over the nine weeks, the product never changed in any way a customer would notice. A capacitor moved a centimetre, a gasket appeared in a seam and a resistor changed value on a reset line. Those small changes were the whole qualification.
Related service: Certification and Compliance Support
EurthTech is an engineering-first product design and embedded systems company with a development center in Hyderabad. To discuss a project, write to connect@eurthtech.com or see https://www.eurthtech.com/case-studies.
Comments