20 MHz is the standard bandwidth cap used across the DC/DC converter industry when measuring output ripple and noise, applied via the oscilloscope's built-in analog low-pass filter or an external RC/LC network so that switching spikes above the fundamental are attenuated and cross-vendor numbers stay comparable [S2][S4].
Ripple on a switching converter's DC output typically sits between 10 mV and a few hundred mV peak-to-peak, generated by charging and discharging of the output filter capacitance at the converter's fundamental switching frequency or a multiple of it, with additional high-frequency noise spikes produced by parasitic inductance during the commutation of tens to hundreds of amps [S2]. A 20 MHz cap removes the noise spikes while leaving the ripple fundamental largely intact, which is exactly what converter datasheets and qualification reports quote [S2][S5].
Why 20 MHz Became the Industry Reference Bandwidth
There is no industry-wide mandatory standard for output ripple and noise measurement, and test methods vary from vendor to vendor, which is why Flex Power Modules' design note 022 (revision R2C) describes 20 MHz filtering as "the most commonly used practice in the power conversion industry" rather than a regulated requirement [S2].
The practical origin of 20 MHz is straightforward: most converter datasheets have historically been written assuming the scope is bandwidth-limited, and the cap is wide enough to pass the fundamental ripple of common topologies (100 kHz to 1 MHz fundamental, with harmonics reaching low MHz) while rejecting the GHz-class edge ringing that a 100 MHz or 1 GHz scope would otherwise pick up [S2][S3][S5]. Analog Devices' AN-1144 states explicitly that "changing the oscilloscope bandwidth setting to 20 MHz removes the high frequency component of the signal leaving the output ripple itself" [S5]. The cutoff is also still embedded in most modern oscilloscopes: even a 4 GHz instrument will typically offer 20 MHz and 200 MHz analog band-limit selections, because the filter is useful for any low-frequency measurement on a noisy rail, not just PSU testing [S4].
Scope-Internal Filter vs External RC/LC Network
Two implementations dominate. The first is the scope's internal analog 20 MHz low-pass filter, engaged from the channel setup menu, which reduces measurement bandwidth and therefore broadband noise proportionally to the square root of the bandwidth ratio for a spectrally flat noise source, so a 4x bandwidth reduction cuts integrated noise roughly in half [S4].
The second is an external filter built on the bench, of which there are two flavours: a 50 ohm-terminated BNC-T low-pass filter, typically a pi or L-section with roughly 1 ohm DC source impedance to mimic the standard 20 MHz test fixture, and a higher-impedance LC network when the filter is placed before the scope's 1 Mohm input [S1][S2]. Flex's note describes the same 20 MHz low-pass RC filter as the canonical "20 MHz filter" topology used for datasheet verification across the industry [S2]. The LC variant, suggested in the EEVblog 594 thread, can be built from the same C1/C2 values converted to an LC topology, and the trade-off is insertion loss versus stopband sharpness: an LC network gives a steeper roll-off and a cleaner stopband than a single RC, but a simple RC is enough to make one supply's ripple number comparable to another's [S1][S2].
Who Should Use 20 MHz Filtering, and Who Should Not
The 20 MHz cap is the right call for anyone characterising a converter for a datasheet, comparing two supplies, or verifying a rail against a published ripple and noise spec, because it produces the same numbers the datasheet author used [S2][S3][S5]. It is also the right starting point for general low-voltage rail integrity work, where broadband noise above 20 MHz is usually radiated pickup or scope-channel noise rather than a real signal on the rail [S4][S7].
It is the wrong call for EMI/RFI compliance, radio-frequency power supply design, and any rail that drives a high-speed ADC, PLL, or RF stage above 20 MHz. In those cases, the high-frequency content is the signal, not the noise, and a 20 MHz cap will hide real failures. As one experienced engineer on the EEVblog thread put it, "pretending that anything above 20 MHz doesn't matter does no favours to anyone" for supplies that feed RF stages, and modern SMPSUs with sharp edges can carry significant energy well into the hundreds of MHz [S1]. The rule of thumb from the same thread: if the supply is in a screened enclosure, the rail is the dominant noise source, and the 20 MHz cap may discard real failures; if it is in free air, ambient pickup above 20 MHz is usually larger than anything the supply is actually producing, and a 1 nF decoupling cap on the load will typically knock the 20+ MHz content down by a factor of 10 to 1000 [S1].
Measurement Setup Decisions That Move the Reading
Probe tip ground length dominates high-frequency noise pickup above the filter, and a long ground lead acts as a 100 MHz-class loop antenna that no 20 MHz filter can remove cleanly, so the recommended practice is a short ground spring directly on the probe tip rather than the alligator clip lead [S4][S5][S7].
Termination impedance is the second largest variable: a 50 ohm scope input or feed-through termination at the probe tip is the standard for datasheet figures, but most inline 50 ohm terminators are rated for small-signal use, so testing a high-voltage or high-current rail through a built-in scope terminator can overheat the resistor in minutes [S1][S2]. A third variable is AC versus DC coupling: AC coupling strips the DC rail so the scope's dynamic range can be set for the mV-level ripple, which is the standard configuration shown in the EDN Asia 20 MHz filter example [S4]. Adding extra external capacitance on the converter output will reduce the ripple fundamental slightly but does very little to the high-frequency noise, because the dominant parasitic inductance is already inside the converter, not on the test board [S2]. The practical sequence recommended across the sources is therefore: set the scope to 20 MHz bandwidth limit, use AC coupling, terminate in 50 ohms at the probe tip with a short ground spring, and keep the measurement loop under roughly 1 cm² to avoid picking up radiated noise that no filter can subtract [S4][S5][S7].
Verification and Cross-Reference Points
For an engineer choosing an oscilloscope, the key spec to confirm is that 20 MHz is implemented as an analog hardware filter ahead of the ADC, not as a digital FIR applied after acquisition, because analog filters do not add aliasing and have no group-delay artefacts that would distort a switching waveform [S4]. Most scopes with 100 MHz or higher native bandwidth include the 20 MHz option as a standard, and a 4 GHz class scope typically extends the list to 20 MHz, 200 MHz, and full bandwidth, with an additional user-selectable digital noise/ERES filter layered on top [S4].
For an engineer building an external 20 MHz filter, Flex's design note 022 and the EEVblog thread agree on the topology envelope: an RC low-pass with 50 ohm source impedance, or an LC conversion of the same C1/C2 values when a steeper roll-off is needed, and the cutoff should be set so that the converter's fundamental switching frequency and its first few harmonics pass unattenuated, with the corner at 20 MHz, so the filter passes the signal the datasheet is meant to capture [S1][S2]. The bigger picture on whether 20 MHz is even the right number for your rail is covered in detail in the power integrity measurement solution walkthrough, which contrasts bandwidth-limited ripple measurement against full-bandwidth PDN probing for high-speed digital rails [S6]. For readers working on adjacent power supply characterisation, the switching power supply reference covers the topology side of where the ripple fundamental actually comes from, and the DC power supply entry covers the linear-rail case where 20 MHz filtering is usually unnecessary because the dominant ripple frequency is already at 100/120 Hz. Bench engineers trying to decide whether the noise above 20 MHz is the supply or the room will also find useful background in this practical vibration meter tracking filter note, since the underlying question, isolating a known fundamental from broadband pickup, is identical in both domains.
Two trackable signals to confirm the standard remains in force through 2026: first, the 20 MHz analog filter continues to appear as a default option on every new mid-range and high-end oscilloscope platform from the major T&M vendors, including the 4 GHz-class units documented in EDN Asia's 2023 scope-filter write-up [S4]; second, no major converter vendor has publicly migrated their datasheet ripple and noise specification away from the 20 MHz bandwidth cap, so any supply comparison done in the next procurement cycle should still be qualified against the same 20 MHz reference, not against full-bandwidth scope captures [S2][S5].