The unit that passed your bench and failed theirs
A regulator that works on your bench supply and oscillates on the customer’s bench is not a random defect. It is a predictable interaction between the converter’s negative dynamic input impedance and an input filter that your bench did not present. The customer’s long cable run supplies inductance; the source at the far end supplies capacitance; together they form an undamped LC network. The converter, behaving as a constant-power load, presents a negative resistance across that network. Under the right conditions, the pair oscillates.
This is a system-level failure mode. It is not evidence that the regulator IC is defective, and it is not evidence that the customer’s power supply is bad. It is evidence that the input filter formed by the cable and source was never part of the stability analysis.
Why a constant-power load looks like a negative resistor
A switching converter that holds its output voltage while the load draws constant power has an input characteristic in which input current rises as input voltage falls. The slope of that voltage-current line defines the dynamic input impedance. As Texas Instruments’ Robert Kollman writes in the input-filter damping note, with Pin = V · I, we have V = Pin / I, so dV/dI = –Pin / I², or approximately dV/dI ≈ –V / I. The slope is negative (Kollman, “Power Tip #3: Damping the input filter — Part 1”).
That approximation is a simplification because the control loop shapes the input impedance versus frequency, but it is good enough to explain the mechanism. A negative resistance in series with a positive resistance of equal magnitude has no net damping; the circuit tends toward oscillation. Kollman states the consequence directly: where the negative resistance of the supply is sourced by the equivalent resistance of the input filter at resonance, “the circuit will tend to oscillate.”
TI’s application note SLVAFE0 makes the same point in converter terms: the DC/DC converter acts as a load for the input filter and provides negative variable resistance; connected to an undamped input filter, it “acts as a negative oscillator circuit” and causes system instability.
The cable is an input filter you did not design
On your bench, the regulator sees a lab supply with short leads and a large output capacitance. At the customer site, it sees a cable with distributed inductance and resistance, terminated by a source with its own capacitance and possibly a long return path. That combination is a second-order low-pass filter. At its resonant frequency, the filter’s source impedance peaks. If that peak is comparable to or larger than the converter’s closed-loop input impedance, the negative resistance wins and the loop rings.
The design rule follows from the mechanism: keep the filter’s peak output impedance well below the converter’s closed-loop input impedance, evaluated at the worst case. Kollman puts the worst case at minimum input voltage and maximum load, because that is where the input impedance is lowest. A TI E2E forum response from Timothy Hegarty gives a practical approximation for input-filter design: assume closed-loop Zin = –Vin / Iin, note that current-mode control keeps that impedance relatively flat up to the loop crossover frequency, and keep the EMI filter’s output impedance below it with preferably 6 dB of margin. SLVAFE0 offers a more conservative rule of thumb: the filter output impedance should be less than one-eighth of the converter input impedance.
Those two margins are not identical, and neither is a standard. They are engineering recommendations from the cited documents. The point is that some margin must be designed and verified, not assumed.
What the datasheet stability plots do and do not tell you
Regulator datasheets and application notes discuss loop stability in terms of the output capacitor’s ESR, the power-stage pole, and the compensation zero. TI’s AN-1148 (SNVA020B) is explicit that most LDO oscillation cases come from output capacitor ESR being too high or too low, and that large ceramic capacitors with ESR below 20 mΩ will cause most LDOs to oscillate unless the part is designed for them. That is real and worth checking.
But it is a different failure mode from the one in this article. Output-capacitor ESR instability is a property of the regulator’s own control loop. Input-filter oscillation is a property of the source impedance the regulator sees. A datasheet’s typical application circuit assumes a low-impedance source; it does not model the customer’s cable. If your bench test used the datasheet’s recommended input capacitor and a short lab lead, you have verified the regulator, not the installation.
The LMR36503-Q1 datasheet’s input-capacitor guidance is a useful counterexample because it acknowledges the field condition: it recommends an electrolytic capacitor in parallel with the ceramics when long leads or traces connect the supply to the regulator, because the electrolytic’s moderate ESR helps damp ringing caused by the long power leads. That is a manufacturer telling you the cable matters.
Reproduce the customer’s cable before you touch the regulator
The first diagnostic step is not to swap the regulator. It is to reproduce the customer’s source impedance on your bench. That means the cable length, the conductor gauge and spacing, the source capacitance, and the return path. If the oscillation appears only with the cable emulator in place, you have confirmed the mechanism.
A practical bench checklist:
- Measure the cable. Inductance and resistance per unit length, plus the source capacitance at the far end. If you cannot measure inductance directly, a rough estimate from geometry is enough to locate the resonant frequency; the exact value matters for the fix, not for the diagnosis.
- Compute the filter resonance and characteristic impedance. For an LC filter, the resonant frequency is f = 1 / (2π√(LC)) and the characteristic impedance is Zo = √(L/C). The peak source impedance is roughly Zo times the circuit Q, as Kollman describes.
- Compare peak filter impedance to converter input impedance. Use Zin ≈ –Vin / Iin at minimum input voltage and maximum load. If the filter peak is within a few dB of that value, the design is marginal.
- Add damping and re-measure. Do not declare the fix until the loop phase margin is verified at the customer’s operating point.
The hypothetical example in the outline is worth stating as an illustration, not a measurement: a 24 V industrial sensor on 10 m of two-conductor cable can form a resonance in the tens of kHz, and if the converter’s negative input impedance at that frequency is comparable to the filter’s peak impedance, oscillation is expected. The numbers depend on the actual cable and converter; the mechanism does not.
Measure the loop, not just the output ripple
Output ripple on a scope can tell you that something is oscillating, but it does not tell you whether the loop has adequate phase margin. TI’s AN-1889 (SNVA364A) describes a loop measurement method that uses an audio generator, an injection transformer, and a two-channel oscilloscope. The steps that matter here:
- Break the loop at a low-impedance node above the high-side feedback resistor and insert a small resistance, about 20 Ω, in the feedback path. This creates an injection point without materially changing the output voltage.
- Inject a small sine wave through a transformer, typically 30–100 mV across the measurement resistor, small enough not to change the large-signal behavior of the supply.
- Find the frequency where the injected and returned sine waves have equal amplitude. That is the 0 dB crossover. The phase difference between the two signals at that frequency is the phase margin.
AN-1889 states that a minimum phase margin of 45 to 50 degrees is necessary, depending on how conservative the design is, and that more is better. It also states a constraint that matters for field debugging: loop measurements can only be taken successfully on a design that is not oscillating or in hysteretic over-voltage protection. If the unit is already ringing, you must first stabilize it enough to measure, or measure the input filter impedance separately and reason from the comparison.
SLVAFE0 shows the complementary measurement: output impedance versus frequency. In its worked example, an undamped second-stage output filter produced a resonance peak near 83.4 kHz and a control-loop phase margin of around 26°, with an oscillatory transient response. That is the signature to look for: a resonance peak in the impedance curve that lines up with a phase-margin collapse in the loop plot.
Damping the filter is usually cheaper than changing the converter
Once you have confirmed that the input filter resonance is the problem, the fix is to damp it. SLVAFE0 compares two techniques. A series damping resistor Rd in the filter reduces the resonance peak but also affects attenuation after the cutoff frequency. A parallel Rd-Cd branch damps the resonance peak without impacting the filter’s attenuation level. The note gives the design equations: for a critically damped series RLC, Q = 1 and Rd = √(L/C); for the parallel branch, the damping capacitor should be at least five times the filter capacitor so that the cutoff frequency is not shifted.
In the SLVAFE0 example, the calculated series damping resistor was 0.23 Ω and the parallel damping capacitor was 75 µF for a 10 µF filter capacitor. Those values are specific to that example’s 530 nH inductor and 10 µF capacitor; they are not a recipe for your cable. The method is the transferable part.
There is also a simpler field fix that the LMR36503-Q1 datasheet endorses: add an electrolytic capacitor in parallel with the ceramic input capacitor. Its moderate ESR provides damping, and it also helps with voltage dips caused by high-impedance sources. This is not a substitute for calculating the resonance, but it is a low-cost measure that addresses the same mechanism.
What to change in the design review
The failure mode in this article is created by a design process that treats the input source as ideal. At 1k–10k unit volumes, the customer’s installation is part of the product’s operating environment, and the cable is part of the power delivery network. The design review should include:
- A stated maximum cable length and source impedance for the installation, with the filter resonance computed for that case.
- A comparison of the filter’s peak output impedance to the converter’s closed-loop input impedance at minimum input voltage and maximum load, with a defined margin.
- A damping component selected for that resonance, not copied from a datasheet’s typical application.
- A bench test that reproduces the cable, not just the lab supply.
- A loop measurement at the customer’s operating point, with phase margin recorded.
None of this requires a new regulator. It requires treating the cable as a component and the source impedance as a specification.
FAQ
Is this the same as output capacitor ESR instability?
No. Output capacitor ESR instability is a property of the regulator’s control loop and is covered in AN-1148. Input filter oscillation is a property of the source impedance the regulator sees. They can coexist, but they have different fixes.
Can I just add a big electrolytic and call it done?
It may help, and the LMR36503-Q1 datasheet recommends it for long leads. But without computing the filter resonance and comparing it to the converter input impedance, you do not know whether the damping is adequate or whether you have moved the problem to a different frequency.
Why did the unit pass our bench test?
Because your bench supply did not present the cable’s inductance and the source capacitance that form the resonant filter. The regulator was stable into the source it saw. The customer’s installation is a different source.
Do I need a network analyzer?
No. AN-1889 describes a loop measurement with an audio generator, an injection transformer, and a two-channel oscilloscope. A network analyzer makes it faster, but the method works with basic lab equipment.
What phase margin should I target?
AN-1889 states that a minimum of 45 to 50 degrees is necessary, depending on how conservative the design is, and that more is better. That is a recommendation from the cited application note, not a regulatory requirement.
