AC-DC and DC-DC Common-Mode Noise Discussion
for DIY-SMU and other projects
DIY-SMU project

The Problem: Common-mode noise

DC-DC converters are specified for output noise, but almost never for common-mode noise.  Common-mode noise applies to all AC-DC converters and to all isolated DC-DC converters. You can think of it as the noise on the 'common' or output return. Or as noise that is 'common' to all outputs. Output noise is technically "normal-mode" noise, the noise on an output with respect to the common. It does not apply to non-isolated DC-DC supplies such as Boost, Buck, and Buck-boost. Also if an isolated supply is not used in an isolated application (input and output grounded together) then common-mode noise is much less of an issue.

A difficulty in specifying common-mode noise is that it cannot just be measured directly as a voltage. The noise contains frequencies as low as the AC line and up to many harmonics of the switching frequency. Take any isolated AC-DC or DC-DC switching power supply, and measure the output ground with a 10Meg 'scope probe whose ground is connected to the Chassis or input Ground. You will observe significant junk. Pulses, noise, ringing, and noise at the switching frequency, riding on line-frequency noise. It tends to be more of a current than a voltage. A current measurement would be more accurate than a voltage, A sensitive, high-bandwidth current probe could measure it, but I don't have one of these. So I use a simple resistor to convert the current to a voltage.  Common-mode noise may be best thought of as a current, not a voltage.   The load has a large effect on the value.

All AC-DC power supplies use a capacitor from Input (chassis) ground to the output common to reduce the high frequency noise. Many DC-DC switchers also have capacitors from input to output ground.

Here is a typical 3Watt DC-DC's common-mode noise, measured across a 50 Ohm resistor. It is about 1 volt peak-to-peak. As a current that would be I = V/R or 1V/50ohms, or 20mA p-p. Ideally the current wold be in the uA range.

noise

This common-mode noise can show up as audible or ultrasonic noise (higher frequency than audible) on audio or other critical analog circuits. It can causes an unavoidable AC ground loop at high frequencies. It can cause measurement errors in sensitive electronics. This noise is one reason that the highest-performance Audio and instrumentation often will not use a switching power supply. Instead they often use line-frequency transformers. In fact they often use single or multiple internal shields between the primary and secondary windings.

It is very appealing to use these low cost AC-DC and DC-DC switching supplies in many applications. Just be careful about the common-mode noise, particularly if you need the switching power supply output to float. If your system can be grounded (non-isolated), grounded supplies generally reduce most of the common-mode noise to acceptable levels.

Switching Power Supply Common-mode Noise Issue

With any power supply, either lab or otherwise, common-mode noise is an issue. When you float a power supply, there is always some AC current flow from the power supply ground to the chassis (AC) ground. With an AC transformer based supply, this is usually a small amount of 60Hz current due to the inter-winding capacitance between the primary and secondary windings of the power transformer. With a linear supply, the frequency (60Hz and some harmonics) and the capacitance (20-200pF) and the 240VAC input causes about I = V / Xc = 240/(1/2*pi*60Hz * 200pF) or tens of microamps. No big problem, and the typical .01uF safety capacitor to ground shunts out most of this current.  In the case of a high-class power supply or a precision instrument, the AC transformers are usually double or even triple shielded with metal foil between the windings. This shielding reduces the common-mode noise current significantly.

However with a switcher and its high frequency transformer, the frequency is not 60Hz, but the harmonics of the the fast rise-time switching waveforms: rise times of 300V pulses can be about 100ns causing pulses with harmonics of 10MHz or more. The transformer windings are usually smaller, so the inter-winding capacitance is a bit less. Just to meet radiated and conducted EMI, the transformers are often shielded. You will sometimes see copper foil on switching transformers.

Still, I see some pretty ugly looking common-mode switching noise on many switchers.  Manufacturers do not specify common-mode noise, so how do you deal with it? How do you quantify it? Search for this problem on line and you will find no specific data or techniques. In fact, to meet EMI, Switchers are often tested with a short and heavy wire from their DC common to the chassis ground. This effectively shunts any common mode noise to ground. But if your application requires a floating supply, you are on your own dealing with this issue. Measuring the open-circuit voltage is interesting. You will typically see a few volts of high frequency crud plus some AC. Why not ~100V, since the switcher is switching hundreds of volts? The answer is that AC-DC switchers do have a capacitor from AC to DC ground, typically between .01 -.05uF. This is often a Y safety-rated cap in case the ground of the system is accidentally opened up.

Measuring Common-mode Noise 

To measure the common-mode noise of a power supply, I use a simple current measurement. A 10 or 50 ohm resistor has bandwidth out to the GHz range. Wire a 10 or 50 ohm, 1/4W resistor between the chassis ground pin and the DC common, usually V-. Measure the voltage across this resistor with a 20MHz or 100MHz scope, and you have a good indication of the high frequency common-mode currents flowing through the supply. I did this on several switchers and as expected, most had about 1V p-p of crud across 10 ohms or 100mA of switching currents.  But to my surprise, I found some switchers are quiet, measuring less than 20mV across 10 ohms or just 2mA! What do they know that the other guys don't?

To investigate this, I first measured the capacitance from GND to V-. All supplies measured about .02uF, meaning that the manufacturer typically uses a .022uf capacitor there. So I opened up the bad and two good ones to see what the difference was. The bad ones use a safety-rated, thru-hole, ceramic disc cap from V- to GND. Seems reasonable. But the good ones use an array of 3x2 surface mount capacitors and much shorter and thicker PC traces.  And they mount the capacitors directly between the V- and GND pins of the supply. This approach minimizes the circuit inductance and therefore the high frequency noise. Nice.

Who is good and who is bad? All the CUI (V-Infinity) supplies I measured (n=3) were bad. All of the TDK-Lambda supplies (n=2) were good. I will be using TDK-Lambda switchers from now on when I am concerned about noise.

This is the TDK-Lambda LS50-24, showing the common-mode measuring circuit. The red and black clip leads are for a 100 ohm (.24A) load resistor. The resistor for measuring CM noise is 47 ohms.

setup

Here is the common mode noise waveform for the CUI VGS50-24 50W AC-DC switcher. Lots of ringing (600mV p-p) and 1.5V spikes.

CUI CM

This is the same Common-mode noise measurment on the TDK LS50-24. The voltage range is 4x lower, for 125mV p-p noise. Also no nasty spikes.

LS50-24 CM

Here is the back side of the CUI VGS50-24, showing the grounding layout. The small white circle marks the location of a ceramic 10nF disc capacitor between the heavy DC and AC ground traces.
 cui

Here is the TDK grounding layout. Notice the 2x3 array of SMT capacitors between the heavy DC and AC ground traces. The capacitance measures 22nF, so the 6 caps are about 120nF each. This provides very good filtering of high-frequency common-mode noise.
TDK

DIY-SMU Rev2 Amp and Oscillation

When I built DIY-SMU, after fixing a few instability (oscillation) issues, and one persistent noise. The outputs were clean if they are allowed to float. But when I grounded the -OUT, generally to the scope ground, the outputs showed about 50mVp-p sine-ish wave  at 300-350KHz. I worked on this on-and-off for weeks, but it remained. After eliminating all the loop stability issues by removing or changing parts, it remained. One problem is that the overall control loop is complicated. It has a half-dozen opamps, plus the high voltage amplifier. It has both voltage and current loops. I tried improving my Spice simulation, but no instability. I built a second unit with minimum circuitry: no ranges or modes, voltage loop only. My debug approach was to reduce the circuit complexity to the minimum that exhibits the problem. The stripped-down version worked in FV mode, but the oscillation remained. After much debugging,  I noticed that the oscillation frequency varies slightly with the +12V input voltage. This pointed to the DC-DC power supply. I measured the inputs to the DC-DC and saw 3V p-p of 300KHz! Finally the problem! Turns out the 3W Meanwell DC-DC does not perform well with the common-mode (CM) choke. I removed and bypassed the choke with a fixed ferrite bead, and the output quieted considerably. Normally a common mode choke is a good thing, reducing EMI on an output. However in an SMU like this, that inductance plus the DC-DC capacitance is in the ground return path. I decided the overall design was better off without it. I replaced it with a 600ohm ferrite in the + input only, and a short circuit in the ground. That's when I measured the common-mode noise of several DC-DC converters.

DC-DC converter Common-mode noise Measurements

After I found that the major output noise source on the DIY-SMU project was caused by common-mode (CM) noise of the DC-DC converter, I tested a handful of different manufacturer's parts. No manufacturer specifies the common-mode noise of their parts. CM only affects applications that require isolation. If your application connects the output ground to the input common (with a short wire), then this is less of an issue. And the value of any 'Y Capacitor', a capacitor that connects the input to output commons, also affects the noise.  The effects of CM noise depend on your system. It can cause measurement errors, audible noise, radiated or conducted EMI, or other problems.

Here is a test summary spreadsheet, plus a few scope shots.
summary

Meanwell DPBW03, the worst at 62mV
Meanwell DPBW03

Meanwell DPAN02 comes in at 30mV
DPAN02

CUI PQMC3: Clean at 10mV p-p
CUI PQMC3

And the cleanest, Recom RS3-1215D, with only 4.8mV
Recom RS3-1215

These were measured on a stripped-down DIY-SMU Main Board: only the DC-DC and its associated input and output filter components, plus resistor loads are installed on the test board:
    DC-DCs: 12V to +/-15V
    2W or 3W models
    Original input filter Common-Mode choke replaced with a 600 ohm 1210 size ferrite bead
    Output loads are 220 ohm 1/2W resistors, +/- 68mA, 2.05W total
    SIP 8 pin package
    20 MHz scope bandwidth, ~20mV range, 10x probe

I also measured a handful of the smaller 1-2W 7 pin SIP devices. These are 5V to +/-15V which I use on my 18b DAC project.

I use 2 different loads between the input and output common pins.
    1) Relatively large 10nF Cap (Y-Cap) to filter common-mode noise. This is what I intend to use in the product.
    2) 50 ohm resistor load to observe the high-frequency common-mode current waveform.
   
With the 50 ohm resistor, the waveform exhibits the common mode current waveforms. These are narrow bipolar pulses ~100-500mV that correspond to the switching transistor output transitions in the DC-DC. With the 10nF cap, these fast bipolar pulses (impulses) waveforms are integrated: V = 1/C * Int(I) which results in the observed square waves. The amplitude of the square waves is the integral of the pulse currents. It shows the energy (I * T) of the pulses. 

I only measured 2 devices for the 50 ohm test, the worst (Meanwell),  and the best (Recom). Waveforms are attached. The Meanwell has a 590pF Y cap and exhibits a large area pulse.

Measurements are:
    Input-to-output common capacitance, pin 1 to pin 7, measured at 1KHz
    Switching frequency
    Common mode voltage and waveform, 20MHz BW
    Output DC voltages
    CM waveforms on some
   
Other stuff in the spreadsheet:
    Qty 100 Price

Conclusions:

Noise seems to be a function of price. High cost ~= low noise and vice-versa. Since I originally picked the lowest cost device, I got the one with the highest noise (Duh!). I plan to change from the MeanWell to the Recom part.

The combination of the Meanwell DC-DC and a common-mode choke is a bad choice. The MeanWell without the choke is an improvement. The Recom with the choke is even better. Next I will try the Recom without the choke.

The Cpri-sec capacitance indicates the value of the built-in Y-caps. Units with higher value caps tend to have lower high frequency spike noise. Spike noise can be harder to filter out with an external Y-Cap.

TDK-Lambda CC6-1212SF

cc6
I tested a TDK-Lambda CC6-1212SF dc-dc module. It is a 6W, 12V to 12V, single output supply. The CC1 through CC10 products feature a wide 2:1 input range, and adjustable and regulated output. They are flyback types with a switching frequency of 640KHz. The common mode noise is pretty high. I measured 200mV p-p of spikey common mode noise across a 50 ohm resistor. The capacitance of the input to output is about 100pF, probably from transformer inter-winding capacitance. With a 10nF Y capacitor, the CM noise is a smoother but still large 100mV. Even with a 50 ohm resistor.

Building a quiet switcher

There are several common approaches to making switchers quieter.
At Analogic, I worked on very precision 20 bit 2MHz digitizer and waveform generator instruments for a big ATE company. It required multiple supply voltages, all at very low noise. It provided +/- 17V, and every critical amplifier had its own +/- 15V power supply regulators and filtering, built with dozens of tiny 78L15 and 79L15 regulators. This is the only time I have seen this approach used. Anyway, the power-supply consisted of an off-board box that provided a clean, isolated 100HKz sine-wave AC via a single 3-wire shielded-twisted cable. If I remember correctly, it was center-tapped. On the instrument was a small, shielded module that provided transformers, rectification and filtering using many, large ceramic capacitors. This approach worked well, but if you looked real close, there was always a tiny 100KHz noise spur on the instrument outputs or ADC input caused by the power supply.

I think that in addition to lower common-mode noise, a resonant switcher also tend to have lower diode switching noise, since the output side rectifier diodes are not being asked to turn on with nice slow sine-waves instead of 10-100nS rise-time input waveforms.

The Keithley 236 SMU uses an AC transformer to provide its +/- 120V supplies as well as other internal supplies. I think the main AC transformer is triple-shielded to allow this instrument to measure and source currents down to the 1 pA levels. The 237 High-voltage version adds a +/- 1200V power supply. It uses a Resonant design which drives the primary with diode rectifiers and heavy L-C filters.  Resonant converters apply a high-frequency sine wave to the transformer, not the usual square wave. The 237 supply starts with a PWM full bridge, then filters it with and L-C filter to convert to a sine wave. For a great discussion of this power supply check out  Marco Reps. He built his own 237 power supply board in order to upgrade his 236 to a high-voltage 237. Marco is my hero.

LT1533 Low Noise DC-DC

Jim WIlliams at Linear Tech (R.I.P.) recognized the problem of isolated switching supply noise. In his excellent: Application Note 70, "A monolithic switching regulator with 100μV output noise: Silence is the perfectest herald of joy..."  He uses a LT1533  switching regulator which offers reduced rise and fall times. This is a decent approach, and can be built with an off-the-shelf transformer. This IC requires feedback to operate though, and most of the app-note circuits do not have isolated feedback.This excellent treatise shows some old tektronix techniques to reduce power supply noise. It doesn't really address the common mode issue, but his techniques to make low noise supplies also will help reduce common mode currents.

The lT1533 IC uses a push-pull transformer. Unlike flyback and h-bridge types, push-pull transformers tend to reduce common mode noise simply by having the two halfs of the primary out-of phase. One side's common mode noise cancels out the other side....

Push pull converters can be open-loop (unregulated)  or closed loop (regulated). For closed loop design, a ''forward converter" design is used. The transformer drives a D-L-C filter where the L-C filter converts the PWM duty-cycle to an output voltage, similar to a Buck converter.

For push-pull open-loop (unregulated) design, the transformer is driven with a 50% duty cycle and the output filter is D-C with optional D-D-L-C.

Art of Electronics: The X-Chapters

Another excellent source for info on low noise power supplies is Art of Electronics: The X chapters, 9x.14 discuss common-mode noise,  one of the few places where this subject is discussed.  They dove into the problem big-time, using:


Maybe I have to build the AoE X-files solution. To get minimum CM noise, they applied all of these techniques. They got the common mode noise down to 1mV p-p across 50 ohms. 20uA is quite good for high frequency common mode current.

They address it using a differential sine-wave driving two Class-B amplifiers. Pretty brute-force and high component cost IMHO. And inefficient due to the linear amplifiers. A more efficient way to get a sine wave is to use a resonant driver.... How about a class D audio amplifier? These are pretty cheap and efficient.Getting clean-ish sine waves at 100KHz or so, not so good.

Ancient semi-resonant HiV power supply design

Here is an ancient Circa 1981 high voltage power supply I designed in my youth for a medical monitor, the Corometrics 505/506. It provided high voltages +/- 2KV (I think) for a 5" CRT, and +150V for the deflection amplifiers. What is cool about it? The entire medical monitor drew ~6W including several isolated medical front-ends and this CRT. It was portable and battery powered from a 12V gel-cell. This DC-DC ran open loop, constant frequency, constant +12V in. The transformer was a big 3019 potcore. See the schematic below in its yellowed blue-line and hand-drawn glory. The tricks were:

This worked because of the relatively high inductance of the +150V secondary winding. The 10pF capacitor was selected to make the switching waveform ideal. Resonance was about 25% faster than 29.7MHz or about 36MHz. Don't laugh at my 4000-series logic. Two cascaded counters were used to make sure the Q0 (U902-13) output that sets the pulse width was a few nS early, to avoid decoding glitches. Transformer T6 is a 3019 potcore, carefully wound to handle the high voltage. All the logic and power supplies in the system were derived from a single clock crystal to eliminate any random frequency beating effects. Had some pretty sensitive analog circuitry.

Not a bad design for a kid 5 years out of WPI. I did have a predecessor's design as a starting point, which is always a good idea. I'm not sure if such an approach has any practical use on modern low or mid-power DC-DC converters.
 

hiv PS

As project lead on this design in 1981, I'm still pretty proud of what we built back then. Here is our Corometrics 506 Neonatal Monitor. Couldn't have done it without a great team: Jeff S, Dave L and Wayne C.

Coro 506

Building a low noise AC-DC power supply

I recently (2022) built a music server using a Raspberry Pi and Volumio Software. I wanted a quieter power supply than the usual Raspberrry Pi switching wall-wart. After watching Marco Reps excellent video on the subject of common mode currents in isolated supplies, I realized that there is a trick that can reduce AC transformer common mode current considerably. Here in the US, line voltage is 120VAC. Since many AC transformers have dual primaries, one of the 120VAC primary windings can used for power, and the other winding is wired out-of-phase to the other. The second primary's voltage cancels out the main primary from a capacitive coupling point of view.  Split-core transformers (with primaries isolated from secondaries)  already have pretty low common mode currents, and this trick reduces the current by another 10x. It's a bit of a hack because it only works on 100-120V AC systems. To worh at 240VAC, it would be necessary to have two 240V windings on a transformer. It could be done, but I know of no off-the-shelf transformers like that. Check it out on my Raspberry Pi Power page.

This trick demonstrates that push-pull power supplies have lower common mode currents than unbalances types. 

Further work to do

I would like to experiment with the LT1533 and some off-the-shelf and custom transformers. Unfortunately this part is 12V maximum supply (limited by switch output voltage) and is $15 at Digikey.

I would like to continue investigating simple and low power resonant power supplies to make sine-wave drive.  Or even higher power supplies.

I would like to investigate DIY electrostatic shields on custom dc-dc transformers.

I would like to build a quiet, higher power, push-pull driver. Preferably that is less dependent on a specific IC. Maybe:

For a low power quiet DC-DC:

Since layout is critical, maybe design / build a few custom boards that can be used as modules


Here is a first pass low noise DC_DC LT1533 board. The proto area is an attempt to accommodate different transformer footprints. RM cores will fit since their pins are on a 0.1" grid. Same with many Wurth transformers. Others will need some hand wiring. This project is mostly an exercise in transformer experimentation. I left out optical feedback (closed-loop) and LDO regulators, too many choices, and some won't need them. You probably want the LDOs on your application board.  It's a starting point... Time to order boards and parts.

The BOM cost including a $3 PCB is about $33. Assumes $10 for a transformer. Wurth ones are cheaper.

1533 board

1533 sch

Transformers

I've been looking at various off-the-shelf transformers for low power DC-DCs. MIdcom/Wurth, and Eaton look promising. These are specified in various TI and Linear App notes to be used. Many of these transformers are designed for Gate-drive applications or communications. So the turns ratios are generally pretty limited. Many are not push-pull on both sides. Also these are not shielded, and most use E cores or toroids. I rarely find one that meets my needs. If you want something done right, do it yourself. I'm an old potcore guy. potcores are great, easy to wind and assemble,  contain the B-field well, and are pretty flexible. I have a 40 year old assortment of them, old 3B7 and 3B9 ferrites. RM cores are kind-of the modern upgrade to potcores. Similar to potcores, they offer more flexible bobbins, simple clips for assembly, lower cost and more efficient board space. In AN70, Jim Williams does a nice comparison of core types.

However... here is the Wurth Push-Pull transformer listing, pretty comprehensive. Some decent and interesting turns ratios here. 5V input to +/- 5, 12, 15 and 17V are interesting.  Some are E-cores, some are toroids. Digikey stocks about half of them, prices are pretty low.

However I found a fundamental problem. The LT1533 that I am using operate up to 250Khz clock rate. But the Wurth transformers, with their tiny toroid cores and low inductance, need 200Khz minimum. But, but, it's a trap! The transformer specifies the voltage drive frequency, and the LT specifies the clock frequency. The voltage output of the LT1533 is 1/2 of it's clock. So it can only go down to 250KHz / 2 = 125Khz. Darn. I tried overclocking the LT1533 to 300KHz, a compromise frequency for both it and the transformer, and got OK  results. Looking at LT3999 and

In searching for ways to reduce transformer leakage inductance, I came across this excellent book on transformer design: Transformer and Inductor Design Handbook (Electrical and Computer Engineering) 4th Edition by Colonel Wm. T. McLyman, The author was a transformer guru at JPL and many big space programs. There is a section on low noise switchers, using a semi-resonant design used in various JPL programs including Hubbell telescope and Mars rovers. I'm working on a rotary transformer project, and sure enough, there is a chapter on rotary transformers! Great resource. The paper version is expensive, the Kindle version is reasonable. With some rooting around on-line and signing up for some spam,  I found a free PDF. I read all 700 pages and it is very good. Technical, but it steps you through the math. Lots of fairly simple math for any transformer design.

LT1533: No 12V operation

I received the LT1533 boards, built them up, and found a few small design errors. Mostly missing grounds which were fixed with rework wires. I forgot to add the 25nH inductance (wire) on pin 16. To get 50% fixed duty-cycle, ground the DUTY pin, and the FB pin must be below 1.2V, about 1.0V. But if FB is below 0.4V, the clock oscillator slows down. This is a slow-start feature that runs it at lower power until the part is fully powered up. It operates with a lovely and clean trapezoid waveform. Input and output currents are clean. The  rise time of the voltage and current are controlled by resistors.

The part works with 5V in, and up to about 7-8V. But I couldn't get it to work reliably at 12V in. The part is rated at 12V in, and the switches at 30V. Some of my applications need 12V. At 10 or more volts, it was going into thermal overload. If you look closely at the LT1533 data sheet and at AN70, all the application circuits are 5V input. At 12V, even with a high inductance transformer, it goes into current limit. With a beefy, high inductance (540uH per side transformer, the SOIC-16 package overheats in a few seconds. With even higher 1.2mH per side, it runs for about about a minute at +12V before overheating. This is with or without the output winding connected and loaded.

Using a 1 ohm resistor in series with V+ to the transformer center tap, I was able to measure the primary current, and found the problem. The current pulses high during the switching transitions, which were about 15% of the clock period. This was too long for 12V operation. Rvsl (voltage slew rate, pin 13) controls the output slew rate. At 12V, the transition time is 2.4x more than at 5V. Reducing the Rvsl  resistor from 15K to 6.8K made a big difference.  Still I'm surprised at the high current draw during the transitions. But it now works at 12V! It still gets hot, over 70C with no load. But it can put out as 5W or so without overheating or current limiting.

In troubleshooting the problem, I got the LtSpice LT1533 simulation fixture working. Of course it doesn't go into thermal shutdown like the real circuit does. I modeled the output transistor power (Vcola *Icola) and it seems fine with a 1.2mH transformer. No explanation yet why it gets so hot in the real world. Apparently the Spice model doesn't model the transistor slew current very well.

I am thinking of a big power supply / transformer test board, following in the footsteps of the AofE X-Chapters work.
I'll use my SG3525 proto board for now, and plan to build up a LT1683 board for the higher power applications. It should do voltages up to 24 and 48V and at much higher power. 


   Dave's Home Page
Last Updated: 10/2/2023