Your prototype radio worked fine on the bench. Then you enabled the buck converter, and the link margin dropped by 6 dB. The receiver didn’t fail. The transmitter didn’t fail. The power supply started switching, and the radio started listening to its own power rail.
This is self-desense: a receiver’s sensitivity degradation caused by an unwanted signal generated inside the same product. At 1k–10k unit volumes, it is one of the most common reasons a working prototype becomes a non-shippable product. The failure is rarely a single bad component. It is usually a planning gap: nobody mapped the converter’s switching harmonics onto the radio’s receive band, nobody measured the spur table before the enclosure closed, and nobody owned the layout decisions that determine whether the noise couples or stays put.
What the converter actually produces
A buck converter is a pulse-width-modulated (PWM) switch. The switch node swings between the input voltage and ground at the switching frequency, and the inductor current ramps up and down around the DC load current. Texas Instruments’ Switching Regulator Fundamentals describes the basic mechanism: the feedback loop adjusts output voltage by changing the switch ON time, and the inductor ripple current is typically kept below 20–30% of the rated DC current (TI SNVA559C, 2019).
That switching action creates a periodic waveform with fast edges. A periodic waveform has a fundamental at the switching frequency and harmonics at integer multiples. A 2.2 MHz buck converter, for example, produces energy at 2.2 MHz, 4.4 MHz, 6.6 MHz, 8.8 MHz, and so on. The amplitude of each harmonic depends on the rise and fall times of the switch node, the parasitic inductance and capacitance in the power loop, and the load current.
The problem is not the fundamental. The problem is where the harmonics land. A 2.4 GHz radio with a 2.2 MHz converter sees harmonics near 2.2 MHz × 1091 ≈ 2.4002 GHz. That is inside the receive band. A sub-GHz radio at 915 MHz sees harmonics near 915 MHz / 2.2 MHz ≈ 416, so the 416th harmonic. The exact harmonic number matters less than the fact that some harmonic will land in-band unless the switching frequency is chosen to avoid it.
This is harmonic planning: choosing the switching frequency, the radio channel plan, and the filter corner frequencies so that converter harmonics fall outside the receiver’s passband, or are attenuated enough that they don’t matter.
Why the datasheet ripple number is not the spur table
A converter datasheet might specify 20 mV peak-to-peak output ripple. That is a time-domain number measured with a specific probe and bandwidth limit. It does not tell you the amplitude of the 416th harmonic at the radio’s input. The spur table is a frequency-domain measurement: amplitude versus frequency at the receiver input, with the converter running and the radio in receive mode.
The difference matters because conducted emissions and radiated emissions are different coupling paths. A converter can meet its output ripple specification and still produce a conducted spur at the radio’s supply pin that is 20 dB above the receiver’s sensitivity floor. The ripple number is a proxy, not a guarantee.
TI’s application note on step-down converter PCB layout (SLYT614, 2015) is explicit about the mechanism: the switch node is the main generator of electromagnetic interference (EMI) in a switch-mode power supply, and the input capacitor is the single most important component for reliable operation. Extra parasitic inductance between the input capacitor and the IC’s PVIN and PGND terminals creates excessive voltage spikes due to V = L × dI/dt. The note recommends placing the input capacitor as close to the IC as manufacturing rules allow, using wide and short plane connections, and keeping the switch-node copper area minimal because all copper connected to the switch node is one plate of a parasitic capacitor whose other plate is every other node in the circuit.
That parasitic capacitor is a coupling path. At 1k–10k unit volumes, the layout that worked on the eval board may not be the layout that goes to production. The contract manufacturer may rotate a component, change a via pattern, or substitute a capacitor with different equivalent series resistance (ESR) and equivalent series inductance (ESL). Each of those decisions changes the parasitic network and therefore changes the spur table.
Bench measurement: separating converter spurs from receiver desense
The first measurement is not a radiated emissions scan. It is a conducted measurement at the radio’s supply pin, with the converter running and the radio in receive mode. The goal is to see the spur table at the point where the noise enters the receiver.
A practical setup:
- Power the product from a clean DC supply through a line impedance stabilization network (LISN) if you need to compare against CISPR or FCC conducted emissions limits. The LISN provides a defined impedance and isolates the product from the supply.
- Use a spectrum analyzer with a resolution bandwidth (RBW) narrow enough to resolve individual harmonics. For a 2.2 MHz switching frequency, a 10 kHz RBW is a reasonable starting point. Set the span to cover the radio’s receive band plus at least one harmonic on either side.
- Measure at the radio’s supply pin with a high-impedance probe or a coaxial pigtail with a DC block. Do not use a standard 10:1 passive probe for this; the ground lead inductance will corrupt the measurement above a few megahertz.
- Record the amplitude of each harmonic in dBm or dBµV. This is your conducted spur table.
Then measure the receiver’s sensitivity with the converter off and with the converter on. The difference is the self-desense. If the sensitivity drops by 6 dB when the converter wakes up, and the conducted spur table shows a harmonic 6 dB above the receiver’s noise floor at the same frequency, you have a correlation. If the sensitivity drops but the conducted spur table is clean, the coupling is radiated or common-mode, and you need a near-field probe to find the path.
A near-field probe (a small loop or stub) connected to the spectrum analyzer can locate the source. Move the probe along the power loop, the switch node, the inductor, and the radio’s input matching network. The amplitude will peak near the coupling path. This is not a calibrated measurement, but it is a useful localization tool.
Input filter topologies: what actually reduces spurs
The default fix is to add a ferrite bead or an LC filter between the converter output and the radio supply. That can work, but it can also destabilize the regulator or create a new resonance that makes the problem worse at a different frequency.
A ferrite bead is a frequency-dependent resistor. At low frequencies it is nearly a short; at high frequencies it dissipates energy as heat. The bead’s impedance versus frequency curve is the relevant specification, not its DC resistance. A bead that is 1 kΩ at 100 MHz may be only 50 Ω at 2.4 GHz. Check the impedance at the harmonic frequencies you care about.
An LC filter is a second-order low-pass filter. The corner frequency is 1 / (2π√(LC)). The attenuation above the corner is 40 dB per decade, but only if the filter is properly damped. An undamped LC filter has a high-Q resonance at the corner frequency. If a converter harmonic lands near that resonance, the filter amplifies the spur instead of attenuating it.
Damping is the fix. A series resistor with the capacitor, or a parallel resistor with the inductor, reduces the Q. The trade-off is DC loss and cost. For a 1k–10k unit product, the damping component is often a 0.1–1 Ω resistor in series with the filter capacitor, or a lossy ferrite bead that provides both inductance and damping.
The input filter has the same problem. TI’s layout note recommends the input capacitor be placed first and routed immediately to the IC. If you add an input filter, place it before the input capacitor, not between the input capacitor and the IC. The input capacitor is the local energy source for the switching loop; adding impedance between it and the IC increases the loop inductance and makes the switch-node ringing worse.
Manufacturing decisions that change the spur table
At 1k–10k units, the contract manufacturer’s floor decisions are part of the RF design whether you treat them that way or not. The most common prototype-to-production divergences:
- Ground stitching. A prototype may have a continuous ground plane under the converter and the radio. Production may add stitching vias for mechanical or thermal reasons, or may remove them to save board space. Stitching vias change the return path inductance and can shift the resonant frequency of the ground plane.
- Shield can placement. A shield can over the converter or the radio changes the radiated coupling path. If the prototype had no shield and production adds one, the spur table may improve or get worse depending on where the can is grounded and what it encloses.
- Component tolerance. The inductor’s inductance varies with temperature and DC bias. A 2.2 µH inductor may be 1.8 µH at full load and 85°C. That shifts the ripple current and the harmonic amplitudes. The capacitor’s ESR and ESL vary with dielectric type and package size. A production substitution from X7R to Y5V changes the filter impedance at the harmonic frequencies.
- PCB stackup. A prototype may use a 4-layer stackup with a solid ground plane. Production may move to a 2-layer stackup to reduce cost. The return path for the switch-node current changes, and the loop area increases. TI’s layout note is clear that keeping the power loop area small is the primary layout goal; a 2-layer stackup makes that harder.
None of these decisions is inherently wrong. The failure is making them without re-measuring the spur table. A 6 dB desense can appear from a stackup change alone if the return path inductance doubles.
Regulatory limits and radio desense: related but not equivalent
CISPR and FCC conducted emissions limits define the maximum allowable disturbance voltage on the power leads. The ITU-R SM.329 recommendation defines unwanted emissions in the spurious domain for radio systems. These are regulatory floors, not radio performance guarantees.
A product can pass CISPR 32 or FCC Part 15 conducted emissions and still desense its own receiver. The regulatory limit is typically measured with a LISN and a quasi-peak detector over a specified frequency range, often 150 kHz to 30 MHz. A 2.4 GHz harmonic is far above that range. The regulatory measurement does not look at the radio’s receive band.
Conversely, a product can fail conducted emissions and still have acceptable radio sensitivity if the offending harmonic is outside the receive band. The two measurements answer different questions. The regulatory measurement asks whether the product pollutes the power line. The spur table asks whether the product pollutes its own receiver.
For a 1k–10k unit product, both matter. The regulatory test is a legal requirement. The spur table is a functional requirement. Budget for both, and measure both before the enclosure closes.
A practical procedure
1. Before layout, list the radio’s receive bands and the converter’s switching frequency. Calculate the harmonic numbers that land in-band. If any harmonic is within 10% of the receive band center, change the switching frequency or the radio channel plan.
2. During layout, follow the datasheet’s recommended layout for the converter. Place the input capacitor first. Keep the switch-node copper area minimal. Use a solid ground plane under the converter and the radio. Do not route the switch node under the radio.
3. After the first prototype, measure the conducted spur table at the radio’s supply pin with the converter running. Measure the receiver sensitivity with the converter off and on. Record both.
4. If the desense is unacceptable, add an input filter or an output filter. Damp the filter. Re-measure. Do not assume the filter works because the simulation says it does.
5. Before production, review the contract manufacturer’s stackup, component substitutions, and shield can placement. Re-measure the spur table on the first production units. If the desense has changed, find out why before shipping.
The 6 dB loss is not a mystery. It is a measurement you didn’t take yet.
FAQ
Can I just use a linear regulator for the radio supply? Yes, if the current draw and thermal budget allow it. A low-dropout (LDO) regulator does not switch, so it does not generate switching harmonics. The trade-off is efficiency and heat. For a 1k–10k unit product, the LDO may be the simplest fix if the radio current is low.
How do I choose the switching frequency? Start with the radio’s receive bands. Avoid switching frequencies whose harmonics land in-band. If the radio has multiple bands, you may need to filter rather than avoid. A spread-spectrum converter can reduce peak harmonic amplitude but does not eliminate the harmonics; it spreads them over a wider bandwidth.
What is a spur table? A table of unwanted signal amplitudes versus frequency at a defined measurement point. For self-desense, the measurement point is the radio’s supply pin or antenna port, and the frequencies are the converter’s switching harmonics and the radio’s receive band.
Does a shield can always help? No. A shield can changes the coupling path. If the dominant coupling is conducted through the power rail, a shield can does nothing. If the dominant coupling is radiated from the switch node, a shield can may help if it is properly grounded and does not create a new resonance.
How much desense is acceptable? That depends on the link budget. A 1 dB desense may be acceptable if the link margin is 10 dB. A 6 dB desense is usually not acceptable unless the link margin is very large. The receiver’s sensitivity specification and the system’s fade margin determine the threshold.
Where can I find the primary sources? TI’s Switching Regulator Fundamentals (SNVA559C) covers converter operation and ripple current. TI’s Five steps to a great PCB layout for a step-down converter (SLYT614) covers layout and EMI. ITU-R SM.329 covers unwanted emissions in the spurious domain. CISPR and FCC Part 15 cover conducted emissions limits. Your converter and radio datasheets cover the specific numbers for your design.
