common-modecommon-mode-chokeswitch-mode-power-supply
I am designing a hi end SMPS and I have 1 CMC on the input (ac line) and a second on the DC output.
My question is where to place this second CMC.
Usually its best to place filters right at the source of noise. I assume that source of CMN is the secondary output of the transformer . Of course I cannot place it after the transformer , so I am thinking to place it after diode + pi filter and Cout.
Is the above better or I should place it at the cables (DC output) ?
The trifilar transformer you have drawn will couple all three of its windings together. In the ideal transformer case, that means the voltage on each set of windings will be identical.
It will therefore effectively parallel all the signals together. The fact that you have drawn load impedances of Z on the end of each winding means that the transformer will not actually short your two noise sources together, which without the Zs could have made them malfunction. The average noise source voltage, ie half the sum, will be added to your signal.
Nevertheless, effectively connecting your two noise sources in parallel like this is a rather 'dirty' way to do it.
You can design a three input combiner, that isolates each signal from the other. Alternatively cascade a pair of two-input combiners. The design equations for these things are readily available, or you can buy the finished part, see MiniCircuits for instance.
If the sources are low impedance, then it would be better to cascade a pair of simple transformers, adding one noise source in each. The transformers will put the noise voltage in series with the signal voltage, so simply adding them. As the two noise series are then in series, they won't load each other.
The transformer will induce eddy currents into any metal around it; planes will be proportionally more vulnerable; I'd have a square of copper under the Xformer to intercept the Hfields, and tie that square at ONE place to the surrounding "GND planes."
How low is "ultra low noise opamps"?
Here is what Signal Chain Explorer predicts (we have NOT included a transformer flux-leakage candidate in the Hfield (HFI) interference table).
Input from sensor is 1 millivolt; each opamp is 3,000MHz UGBW so the various interferers get amplified and passed along; GainStage 1 and 2 use 100 ohm and 11 ohm resistors, to get 20dB gain in each. High values of resistors cause massive phaseshift, and the opamp gain stages become oscillators out at 500MHz.
The "Gargoyles" are enabled (top right button), but only HFI interferers are active; I disabled EFI, PSI, GPI. All but PSI require the "interconnect" button active; the default interconnect between stages is 14mm long PCB trace, 1mm wide and 1.5mm above the GND plane. The Magnetic field loop is 14mm high and 1.5mm above plane. The various Hfield aggressors are modeled as infinitely long straight wires. The 2 HFI aggressors are switchreg at 2MHz and ARM clock line at 100MHz.
Note the SNR, with Gargoyles active, is -48dB. With Gargoyles off, SNR is +29 dB defined by the 12 millivolt RMS output thermal noise.
Here are details of the HFI table of available (its editable) aggressors, plus output trash magnitudes and which stages produce the trash.
==========================================
What to do, to improve the circuit? ---Do not place a MCU clock trace 1mm away from 300MHz bandwidth Signal Chain. {this is the dominant aggressor: 90 volts RMS output}
---Do not place a switchregulator 10mm away from 300MHz BW signal Chain.
---The random noise peaks near 300MHz (opamps are 3,000MHz UGBW; Av = 10X); Reduce the bandwidth to 3MHz (add a final R+C passive low-pass-filter: 1Kohm and 47pF) and the noise power is reduced by 100:1 and the noise voltage is reduced by sqrt(100) or 10:1, and you have 20dB better SNR and 3+ bits more ENOB.
---What happened when ALL FOUR of the Gargoyles (interferers) are enabled? The 2nd strongest is electric-fields [also MCU clock at 1mm distance] causing 0.944 volts RMS trash floor. The 3rd is Ground Plane currents [from 0.1 amps of SwitchReg return-currents sharing 5 squares (0.002 ohms) GND plane resistance]. The 4rth is PSI --- power supply trash and finite OpAmp PSRR [ the active Power Supply trash sources are: 60Hz, 120Hz, and 100MHz ringing of SwitchReg, each at 10mV level].
Does SNR degrade when all four Gargoyles are enabled? No. Unless we move the MCU clock far away from our Signal Chain. And move the SwitchReg away.
With no MCU and no SwitchReg interference, what remains? the 120Hz power supply ripple into gain stage#1, which produces 110 uV RMS on Signal Chain output.
======================================
How to design the GND plane, once MCU clock and SwitchRegs are TAKEN FAR AWAY? Use slits to guide aggressor currents away from the Signal Chain. This requires you to understand what aggressors remain, how those currents need to flow, and sketch out a finite-element-model of the GND plane and how the bad currents need to flow; add slits to isolate your gain-of-1,000X circuit from GND voltates: V = Ignd * Rgnd, at 500 microOhms per square of copper foil.
Best Answer
Many devices have 2 external cables: a power supply and a signal. They make 2 good antennas for EMI. The purpose of a CM choke is to prevent wiring (external and internal) to act as antennas, thus suppressing RF emissions. Therefore the logical place for CM chokes is where the wires leave your device or PCB. The entire common mode current must pass (actually: not pass) through the choke, so make sure that there is no parasitic bypass path, like a capacitance or 2 magnetically coupled loops. This risk is greater if you put the choke earlier in your circuit.
Our EMC experts used self-made current loops with BNC connectors, connected to an oscilloscope or spectrum analyzer, to locate the sources of EM emissions. Then you know which loops to make smaller, or which common mode currents to suppress. A decoupling capacitor is for making RF current loops smaller. The problem with CM current in external cables - antennas - is that it flows into an very large loop, so suppressing it with chokes is the only option.
The easy way out is to put chokes on the actual cables, but then the user must not substitute cables. If you see a choke near or inside the mains plug then that is for suppressing conducted emission, to satisfy legal rules about not polluting the mains. Normally radiated emission and conducted emission go hand in hand, think antenna and ground. If 'you' only stop the CM current at the end of a cable then I think 'you' are just making a dipole antenna and fooling yourself.