Need help debugging EMI interference of a circuit powered by switching supply

Question

I am testing a home-built amplifier setup, and this amplifier is powered by switching power supplies. Let me first try to summarize my setup below:

I have two 15 V 1 A off-board AC/DC converters. For what I want to do, I need low output noise, so I filtered it with some home-built power filters right after the AC/DC converters. These filters are 470 uF capacitors, followed by 2 mH common mode chokes, followed by two 2 uH inductors in series, followed by another 470 uF capacitor.

After these filters and about 1m of twisted cables, I arrive at a home-built PCB. The power lines go through smaller common-mode chokes again, then the grounds of positive and negative supplies are combined here (brown arrow). The power line then goes through another LC filter, and positive and negative LDOs respectively (LM317 and LM337). In the end, these power the amplifier I have.

I tie the input of the amplifier to 0, and the amplifier consumes 0.1-0.2 A from the supply. There are two ports (A, B) and their impedance are around 4 Ω each. I have some home-built low-noise preamp to amplifier AC-coupled small noise and it eventually goes to a USB scope and PC for taking noise spectrum. This pre-amp board uses all coax connections.

Problem

Here is what I observed: When I measure the noise from A or B with respect to ground (by connecting pin A to center of BNC connector, and ground pin to shield of BNC connector, as close as possible to the board), I got something pretty quiet (10~30 nV/root Hz). But if I measure A to B, or if I measure ground to port A or B, I got peaks at frequencies (~100 kHz and several harmonics) that look like switching power supply, and the added noise is ~70 uV RMS. To make sure that it is not coming from my amplifier design, I later soldered two 4 Ω resistors to the ground pins, and connected my measurement setup to the resistors. I also pick up similar levels of switching noise. So it seems that the key factor is some impedance between the ground of measurement setup to the ground of the board.

Why is that? I should not have any ground loop. The two AC/DC converters are isolated and connected only at the board, while my measurement setup are all on battery, so I see no DC ground loop. I have common-mode chokes and inductors on the power filter and on the board so I am trying to suppress AC coupling. Adding more chokes or increasing capacitance at the power filter doesn't seem to help. Where is it coming from and how can I beat it down further, say by 10X?

Sorry for a long post, but I want to describe the situation clearly, and I hope this question can also be helpful for others.

*As a background, I am a physics grad student and have worked on custom electronics. No formal power engineering training, but I can understand most general concepts and have some amateur experience. I will also add real photos below:

---

My amplifier board in KiCAD. Power in lower right, output on the top.

---

Power Part (click image for larger version):

(Original image size is too large for Stack Exchange. External link here.)

---

Amplifier and measurement part (click image for larger version):

(Original image size is too large for Stack Exchange. External link here.)

Responding to Tim's comment, I tried to summarize the essentials of the measurement setup and in what condition I see the noise. HOMEAMP are op-amp with transistors at the output to boost the current. PREAMP is ADA4898-1.

Picture of CM path (I probably got something wrong) per Tim's request.

Answer 1

Isolated SMPS produce lots of common-mode noise which means the two DC outputs have noise in common with respect to the AC input, so eventually with respect to the Earth.

You mention several "common-mode filters", but judging from your images I don't see any place where you actually shunt the common-mode noise to Earth using either capacitors or direct connections. Common-mode chokes are extremely ineffective without such shunting elements.

Even if you use a floating scope, I guess the large common-mode noise on your circuit (a huge antenna) will couple into the differential scope measurement.

Answer 2

The CM equivalent circuit looks more like this:

(Disclosure: I work at 7TL.)

The power supply example is for a flyback converter, which is likely the case here. The exact form is not material; only notice that, because it uses a transformer, which is two (or more) pieces of wire separated by insulation, and upon which some AC is applied -- there is a necessary and unavoidable capacitance, and noise current through it. Typically, the peak voltage on Q1 would be on the order of 400 to 700V, and while CS may be small, the waveform is square so the noise consists of sharp edges and ringing.

By reducing H and N in parallel, and OUTP and OUTN in parallel (which is justifiable because the impedance between them is small (large X caps or output filter caps), and the voltage difference between them is also small), we can draw a simplified common mode equivalent circuit.

Note that several of CYns may be absent, depending on particulars of the PSU's design and construction. All the positions shown, may be used from time to time, but not necessarily all would be used in practice.

We can further reduce this equivalent to a different Thevenin voltage and capacitance source, and place it in your system.

Note that the grounds are being used loosely here, as well; the mains impedance RS is grounded in the sense that it has an impedance to... somewhere. Presumably there are reactive elements due to facility wiring, branches, etc. The "resistor" can be considered a general impedance here. (During EMI testing, a LISN would be used, which largely presents a resistance.)

The line input filter module is "grounded" inline between mains and the PSU's ground, i.e. the ground itself is another wire.

The PC is grounded... somewhere else (so, still more wiring length), perhaps in the same receptacle, perhaps a different one. Or if still completely floating (laptop/tablet off charger), it at least has some self-capacitance to the surroundings (on the order of 100pF).

All together, there is a voltage source in a chain of impedances, with few shunting impedances -- mostly inside the PSU itself (CY1-4), one external but maybe not very relevant, and several series impedances (CMCs). Also in series is the amplifier's output impedance RO, which is in parallel with the probe J3, in the relevant configuration(s).

Of note, the series impedances are basically all in series (somewhat give or take which Y caps are in use), so they simply add. Which is why the inline filter doesn't have much effect, it doesn't have anything to filter against.

The most likely improvement would be to add the CY6 as shown, or perhaps another filter stage, with Y cap returned to the signal/cable ground path. A "Y" cap between amp ground (represented by the AMP net) and instrument ground wouldn't be very effective as that path is already a low impedance.

The ground ambiguities can be resolved by placing everything on a metal ground plane, for example installing the line filter, PSU, filter module ground, and instrument grounds, as directly as possible on the plane. Sheet metal, foil, blank copper clad PCB, etc. will do as a plane. Conductive (aluminum or copper) tape can be used to make low-impedance connections, or bolted or soldered joints where applicable. (Beware of tape with not-actually-conductive adhesive; it may have to be burnished into the surface, or an edge curled under or something, to actually make contact.)

The especially low impedance of the plane will also make it obvious where and why you can't connect your instrument backwards across ground and signal; you must have a true differential probe to do those sorts of measurements.

(A final note, I've used "Y cap" loosely here to refer to any capacitor shunting the common-mode path, not necessarily one which requires a safety rating for line-to-ground use, as the actual X1/Y1 standard specifies. In particular, any filter caps on the secondary/ground side can be anything at all, say, 50V ceramic or film, doesn't really matter.)