Should the USB shield be connected to chassis at both ends? What if the chassis has no other path to earth?

Question

I am in the process of redesigning the electrical components of a system built around a control PCB inside a solid metal enclosure. The system is usually powered with a Class II desktop AC-DC power brick outputting 24 V 65 W on a barrel connector. (One model is the exception: it uses a Class I at 48 V with a buck converter inside the chassis.) These systems also interface with a PC via USB 2.0 (high speed / 480 Mbps).

Internally, chassis ground is connected to the power brick DC-, to all internal power supply negative terminals (two switching DC-DC converters producing 5 V and 3.3 V), to one end of the 40 W resistive heater load, to USB analog (pin 4) ground, and to USB shield. USB +5V is used only to detect the cable presence (about 130 μA constant draw). However, this setup means that the only possible connection to earth ground is via the USB shield, which doesn't seem ideal. Nor is it guaranteed -- though I expect USB shield to be connected to earth ground in a desktop computer with an ATX power supply, it's probably not connected through a laptop.

Here is a diagram of the system as it's currently being built:

M for motors, H for a heater, and LC for the analog load cell + amplifier system. Power brick is 2 wires, presumably isolated + rectified SMPS to 24 V. It's the SDI65-24-UD-P6, datasheet here.

Note that the external power supply is isolated; the only path to earth ground may or may not be connected through the USB shield.

Some considerations that may also be worth noting:

1. It's a mixed-signal system with analog load cells connected through shielded cables, which are also mounted to the metal chassis. Should I be worried about ground loops? Load cells are about 30 cm - 50 cm away from the control board and the signal from the load cell realistically cannot exceed 10 Hz.
2. There are lots of inductive loads and multi-amp devices in this system and voltage spikes have caused problems in previous revisions.
3. Some connections lead to external components and user-accessible ports. I'm not overly familiar with transient voltage suppression on signal or power pins, and so far I just have lots of TVS diodes, one for each conductor in each connector. They're closely coupled to a solid ground plane (and thence to chassis) but if chassis is floating, what happens?
4. These machines have given me some wicked static shocks, particularly in winter.
5. It's going to have to pass EMC testing, which I've never been through before.

Though it's an LVDC system and doesn't require a UL/CE certification, it's going to have to pass EMC testing and in any case be relatively noise immune for the sake of the analog components.

I've heard so much conflicting advice about this topic. Should I:

• leave the system floating, or give it a solid connection to earth ground? If so, how is this usually done in metal-enclosure systems with two-conductor barrel jack power? Do Class I AC-DC desktop bricks normally connect DC- to mains earth ground? If instead DC- is left floating, what happens to ESD when the system's connected to a laptop sitting, for example, on a metal table?
• Do, or don't connect the USB cable shield at the device end? Without a shield connection, wouldn't that that just send all transients down the USB analog ground? That seems a lot worse. Or should the shield connect through a resistor? Capacitor? If so, how do I determine the right values?

Answer 1

Not sure I answer to all points because it is a rather large question.

Ground loop and impact on signal integrity

There is a ground loop as soon as you connect the USB cable between your device and PC - regardless of Class I or II AC-DC converter. The ground loop is completed by the AC-to-DC-capacitance in the PSU. 50 Hz current will flow through your ground. If you have any signals referred to your ground, this ground current will impact their integrity. If the signals are extremely small, it can become an issue. Even if you leave the USB shield unconnected, you still don't avoid this because the other USB signals are Earth-referenced if you connect them to a PC.

The right step to take here is to not allow precision signals to be single-ended. Period. If you have transducers or whatever that generate a "single-ended" output, then force them to be differential by using dedicated signal return traces for each of these signals.

ESD

Although the AC/DC converter can have isolation, that doesn't mean that ESD current doesn't flow through them. There are capacitors of at least ~1 nF to Earth which the ESD charge will use for initial dispersion. But the charge cannot actually leave through this capacitor, so in the end it will flow down the USB cable, because the USB is Earth referenced as we have established above.

If the shield (low impedance) is connected, it will do so rather fast, and retain minimum voltage on your ground node. This is good, because it means that the USB signal lines don't have large common-mode voltages, which can upset communication or damage the PHYs. With the shield unconnected, you only have the signal lines, which is not a good scenario. The PC likely has ESD protection on these lines, so it may be still tolerable after all. However, the impedance on these lines is higher, so it will take longer for the strike voltage to disappear, which will leave your box at a few 100s V of common-mode voltage for that time. This is usually not an issue, but just saying.

EMI

There are two good ways to tackle macroscopic EMI in general: Tightly and solidly connect everything with low common-mode impedance to an EMI-ground (e.g. PE) or make all connections to said EMI-ground high impedance and make sure that the device is radiation shielded. That way, you create a local EMI-ground (e.g. chassis) that coherently floats on the external EMI-tides.

As you have many cables and varying external connections, the first scenario is somewhat hard to achieve. The latter case is also not trivial because you have cables to a PC for example. You have two low impedance connections to the Earth ground at high frequency: the Y-caps in the PSU and the USB shield. As discussed before I advocate leaving the shield connected.

In order to make these two interfaces high impedance for EMI, I suggest putting cable beads on both of them. USB cables often have them integrated. For your DC power, you can also find such cables (with barrel plugs) or clamp one on yourself.

Answer 2

It's a mixed-signal system with analog load cells connected through shielded cables, which are also mounted to the metal chassis. Should I be worried about ground loops? Load cells are about 30 cm - 50 cm away from the control board and the signal from the load cell realistically cannot exceed 10 Hz

If the load cells are grounded to anything other than chassis ground, yes, you will get current from a ground loop. Ideally the load cell would be only connected to the PCB ground and whatever it is relaying signals to, that way any noise from electric fields will be common to all signals, and any magnetic loop noise will not create currents on the grounding cables, which will lead to voltage error on ground.

There are lots of inductive loads and multi-amp devices in this system and voltage spikes have caused problems in previous revisions.

This is probably due to poor layout IMO from what I have seen in the past. The design probably has the high currents from motors or switching loads running through areas with sensitive analog electronics. One sure way to avoid this is to isolate the analog electronics from the digital section. The other is careful planning of layout, which requires visualizing return currents on the PCB, and ensuring that no analog electronics will be affected by the voltage generated from the resistance of the ground and the high currents.

The other problem is insufficient grounding with not enough copper or a segmented design that increases resistance of the ground.

Some connections lead to external components and user-accessible ports. I'm not overly familiar with transient voltage suppression on signal or power pins, and so far I just have lots of TVS diodes, one for each conductor in each connector. They're closely coupled to a solid ground plane (and thence to chassis) but if chassis is floating, what happens?

One apparent problem with this design is there isn't a good current return path for ESD back to earth ground because it's 'isolated'. The only return path in the diagram is through the USB or perhaps by the chassis. This means that any ESD on the chassis will be shunted to the PC, or it will jump to ground by some means along the chassis, which may or may not be touching something that is grounded.

Engineers like things that are determinate, and a determined path to earth ground would be best. Typically products have their chassis connected to earth ground via mains ground. Not sure how this would change the IEC safety or classification, most products like this have a built in power supply with it's chassis connected to AC mains ground for fault safety.

Another thing is it is best to connect the TVS diodes to chassis ground and not PCB ground, this way any ESD on the cable can be shunted to chassis ground and not PCB ground (where ESD voltages could create problems or noise).

These machines have given me some wicked static shocks, particularly in winter.

If it's happening in winter it's more likely that you are shocking the machine and not the reverse, due to low humidity.

It's going to have to pass EMC testing, which I've never been through before.

I'd recommend getting a consultant and/or doing some pre lab testing