Designing a Clean Power Supply for an RF Front-End

Every spur you cannot explain is a power-supply question you have not asked yet. Supply noise does not stay on the supply: it phase-modulates your oscillators, amplitude-modulates your amplifiers and shows up in the spectrum exactly where you cannot afford it.

How supply noise becomes RF noise

Three coupling paths do most of the damage:

VCO pushing. A VCO with 10 MHz/V pushing turns 100 µV of ripple into 1 kHz-deviation FM sidebands — visible directly in phase noise at the ripple frequency offset.
Amplifier AM. Gain-versus-supply slope converts ripple to AM sidebands on every carrier through the chain.
ADC reference contamination. Reference or supply ripple on a high-resolution ADC appears as spurs folded across the Nyquist zones.

A −80 dBc spur budget with a 10 MHz/V oscillator implies tens of microvolts of allowable ripple. That number — not "low noise" as a vibe — is your specification.

Switchers are fine; naked switchers are not

Banning switching regulators costs you efficiency and thermal headroom that you will miss at the power amplifier. The workable architecture is almost always switcher → filter → low-noise LDO:

Run the switcher at a frequency your system can tolerate — above the loop bandwidth of your PLLs, and planned, so its harmonics land between your channels rather than on them.
Give it an LC output filter with a damped ferrite, and mind layout: the hot loop area matters more than the part numbers.
Follow with an LDO that has real PSRR at the switching frequency. Check the PSRR curve, not the headline: many "60 dB PSRR" parts are 60 dB at 1 kHz and 15 dB at 2 MHz, which is where your switcher lives.
Leave 300–500 mV of LDO headroom — PSRR collapses in dropout.

Measuring it honestly

You cannot see 10 µV RMS with a standard 10:1 probe on a scope. Use a low-noise preamp or a spectrum analyser with a DC block, measure at the load pins, and check the supply under a representative RF load — class-AB stages modulate their own supply and make bench-idle measurements flattering.

This architecture — quiet pre-regulator, damped filter, wide-band LDO, telemetry — is exactly what our SL-PWR-RF5V packages: 5 V at 3 A with 10 µV RMS measured noise, because we got tired of debugging spurs that were really power supplies.

How supply noise becomes RF noise

Three coupling paths do most of the damage:

VCO pushing. A VCO with 10 MHz/V pushing turns 100 µV of ripple into 1 kHz-deviation FM sidebands — visible directly in phase noise at the ripple frequency offset.

Amplifier AM. Gain-versus-supply slope converts ripple to AM sidebands on every carrier through the chain.

ADC reference contamination. Reference or supply ripple on a high-resolution ADC appears as spurs folded across the Nyquist zones.

A −80 dBc spur budget with a 10 MHz/V oscillator implies tens of microvolts of allowable ripple. That number — not "low noise" as a vibe — is your specification.

Switchers are fine; naked switchers are not

Banning switching regulators costs you efficiency and thermal headroom that you will miss at the power amplifier. The workable architecture is almost always switcher → filter → low-noise LDO:

Run the switcher at a frequency your system can tolerate — above the loop bandwidth of your PLLs, and planned, so its harmonics land between your channels rather than on them.

Give it an LC output filter with a damped ferrite, and mind layout: the hot loop area matters more than the part numbers.

Follow with an LDO that has real PSRR at the switching frequency. Check the PSRR curve, not the headline: many "60 dB PSRR" parts are 60 dB at 1 kHz and 15 dB at 2 MHz, which is where your switcher lives.

Leave 300–500 mV of LDO headroom — PSRR collapses in dropout.

Measuring it honestly