Noise on switch mode power supplies (SMPSs) sometimes gets a bum rap.

I was evaluating the voltage noise on a simple low-cost switch-mode power supply (SMPS) and almost fell for the widespread poor reputation these supplies have for noise.

 

Output Noise in Switching Regulators

By their nature, there will be some switching noise on the output of a nSMPS. After all, they are designed to switch the current from a higher DC source using a pulse-width-modulated (or pulse-frequency-modulated) signal, and then filter this using a 2-pole LC filter.

The switching action of the MOSFET creates alternating periods in which first current flows into the inductor and then the inductor discharges. This results in large dI/dt’s and large voltage spikes. We expect this sort of noise. It’s a question of how effective the LC filter is at preventing these large voltage spikes from transmitting into the rest of the circuit.

The typical output voltage of an SMPS will show ripple at the switching frequency. An important metric is how much ripple there is when the regulator has no load and then when it is loaded with the typical load resistance in the application.

 

Measuring Noise in Switch-Mode Power Supplies

I recently had a low noise application where I wanted to try to use a very low-cost 3.3 V SMPS; only 50 mA of load current was required. I had an evaluation board which I wired up to power from a 5 V wall wart supply and measured the output with a simple 10× probe. My measurement configuration is shown in Figure 1.

 

Figure 1. Measuring the output voltage rail with a 10× probe.

 

The DC level was just fine at 3.3 V. With the 12-bit resolution and large offset capability on my Teledyne LeCroy HDO 8108 scope, I was able to offset this voltage so that I could zoom in on the ripple noise and also look for slow DC drift. Figure 2 shows the measured voltage noise on a 10 mV/div scale.

 

Figure 2. Measured noise on the SMPS output with 10× probe on a scale of 10 mV/div.

 

The switcher’s 20 μsec period—corresponding to a switching frequency of 50 kHz—is clearly evident. The triangle pulses are expected from the charging and discharging cycles of the inductor current. But, on top of this expected signature, there are two types of high-frequency noise.  There is 10 mV peak to peak noise in the flat regions, and sharp, spiky noise that sometimes ramps up to 60 mV peak to peak.

The high-frequency noise and the sharp spikes of noise were troubling. This wasn’t being filtered out by the 2-pole LC filter. If I used this supply, how was I going to ensure that my board would maintain adequate functionality despite all this noise?

However, it turns out that this noise was not actually voltage noise on the power supply output. It was all RF pick up in my probe.

 

Distinguishing Voltage Noise from RF Pick-Up

The large dI/dt’s passing through the inductor in the LC filter result in large magnetic fields that are generated in the vicinity of the SMPS. Any loop with a low-inductance path will have magnetically induced currents that generate voltages which we measure with the scope.

The 10× probe that I connected to the leads of the SMPS makes a loop antenna that picks up these spikes. Your first thought might be, but doesn’t the 10× probe have a 9 MΩ resistor in the tip? Isn’t this a large impedance that would prevent any AC currents from being induced in the loop?

There is a 9 MΩ resistor in the tip, but there is also a 10 pF shunt capacitor, part of the equalizer circuit through which the high-frequency currents flow. At 100 MHz, the 10 pF capacitor has an impedance of only 160 Ω, surprisingly low.

To test the idea that some of this noise was really RF pick up in the probe and not the actual noise on the power rail, I soldered a small SMA connector to the output of the board to reduce the loop antenna area and the sensitivity to radiated fields. In addition, I added another 10× probe in the vicinity of the one measuring the SMPS output voltage, but with this second probe the tip was shorted to the ground lead. This setup allowed me to simultaneously measure the output rail with a 10× probe, the output rail via an SMA connector, and the local RF noise (which is picked up by the probe with the tip shorted to the ground lead). This is shown in Figure 3.

 

Figure 3. Using two 10× probes and a coaxial 1× connection to measure the voltage noise on the SMPS output.

 

Figure 4 shows the noise measured using these three methods.

 

Figure 4. Measured voltage on the SMPS output. All channels are on the same 10 mV/div scale.

 

Probe Attenuation Affects SNR

There are two important observations. First, the general noise level on the 1× coax is much lower than on the 10× probes. This is really due to the fact that the 10× probe is not a 10× probe, it is a 0.1× probe. It attenuates the signal by a factor of 10, reducing its amplitude by 20 dB. When we are measuring small signal levels, such as tens of millivolts, the measured voltage is sensitive to the scope’s amplifier noise.

Most scopes are smart enough to recognize that there is a 10× probe attached to the channel. They automatically adjust the displayed voltage scale to compensate for the factor-of-ten attenuation and display the tip voltage. Thus, when the scope displays the signal on a 10 mV/div scale, it is actually using a 1 mV/div scale at the amplifier. What we are seeing as almost 10 mV peak to peak of noise at the tip is really about 1 mV peak to peak noise at the scope amplifier.

The coax cable using the SMA connection is effectively a 1× probe. This trace is also displayed on a 10 mV/div scale. In this case the 1 mV peak to peak amplifier noise is more or less contained within the line width of the trace.  

This suggests an important best measurement practice: when we are looking at low-amplitude signals, such as power rail noise, any 10× attenuating probe reduces our SNR by 20 dB. When every dB counts, don’t use an attenuating probe.

 

Coaxial Connection vs. Scope Probe

The second observation is that the large, sharp spikes are not present in the coax connection but are present in the two 10× probe measurements. Since one of the probes is not even touching the rail output, this is a strong indication that the sharp spike noise is due to RF pick up and is not voltage noise on the SMPS output.

This suggests the second important best measurement practice: when measuring low-amplitude signals, use a measurement setup that is as close to a coax connection as possible to reduce the probe’s loop area and its effectiveness as an antenna.

If we implement these two best measurement practices, we have 30 mV peak to peak ripple noise, out of a 3.3 V rail. This is 1% ripple, pretty good for a low-cost SMPS. Furthermore, the high-frequency noise is greatly reduced, and the short-duration transients—which in reality are present as RF pick-up noise but not as rail voltage noise—are no longer displayed as part of the switcher’s output signal.

 

Noise in the Frequency Domain

As long as I use a ground plane in close proximity to my power and signal paths, which is an important best design practice, the devices powered by this SMPS and the signals on my board will see just the harmonics of the 50 kHz ripple generated by the SMPS.

Using the direct coaxial, low-noise connection, I measured the spectrum of the noise on the power rail from the SMPS. An example is shown in Figure 5.

 

Figure 5. Spectrum of the noise on the power rail. Top is the time-varying spectrogram, over 10 seconds, showing very steady amplitudes. On this scale, 0 dBmV is 1 mV amplitude noise.

 

The peaks in the spectrum are the 50 kHz harmonics of the switching frequency. The amplitude of the first harmonic is about 10 dBmV, which is 3 mV. This is much less than the 30 mV peak to peak voltage measured in the time domain. This is because the ripple noise has such a low duty cycle. There is not much of a sine wave in the short-duration triangle pulses at the first harmonic. The large number of higher harmonics is an indication of the odd shape of the waveform in the time domain and its high frequency content.

All switching noise is below 10 μV amplitude above about 3 MHz. For my application, this is an acceptable noise level, and actually it is very low for such a low-cost SMPS.

 

Conclusion

This article discussed important considerations regarding the voltage noise that is actually generated by a switch-mode power supply, and it presented two best measurement practices that will help you to perform accurate scope measurements of a switching regulator’s output rail.

 

Comments

4 Comments


  • recklessrog 2018-11-13

    I have found that a great deal of probe pickup comes from the removable clip end that contains a spring and a long usually unshielded flat metal strip. results are better by using the probe with the connector clip removed and just contacting with the needle point tip.

    • bogatin 2018-11-14

      The connection between the signal and return on ANY probe is what influences the rf pick up. Using a spring clip on the return reduces the loop area an d the rf pickup, but using a coax does it even more.

      • supamas 2018-11-20

        Yeah, I imagine you won’t get a lot of critical information out of that set up shown in the pictures!

  • adx 2018-11-16

    The next step that came to mind after seeing your ground leads twisted together was some “channel math” or differential mode (if possible), but I scrolled down and only saw the superior technique of 1x connection with some nice SMA hardware (must get me some for hacking now it’s so cheap). While this did not leave me disappointed, channel math is a slightly dubious technique that could help dig down to the difference between what is being picked up by the probes, and what is just scope noise.

    I find it helps lower the ‘noise’ a fair bit, at least to help get a better look at what the output is really doing. If you can get a good trigger (external / 3rd channel) then it also takes out scope noise using averaging (enough to see why the scope grounds need to be twisted together).