USB works off 5V DC. Unfortunately, as cables get longer, so the voltage drop along them increases. To get round this, either cables must be limited in length, or made with conductors of much lower resistance – thicker copper (making the cables heavier and less flexible), or use low resistance metals such as silver or gold.

There is an alternative, however.

Many systems use what is called a “Kelvin connection”. This is a very simple trick which allows the resistance of an intervening conductor to be “nulled out” in a system. The two main conductors are connected between source and load; across the load are also connected two arbitrarily thin conductors which return to the source where they are connected across a very high resistance.

Since only a very tiny current flows though the return conductors, the voltage across this resistor is for all practical purposes the voltage applied to the load.

This allows the source to be dynamically adjusted to maintain a constant voltage at the load, irrespective of the resistance of the power cable (as long as R(k) << R(p))

Now, the data cores of a USB cable could be used as a Kelvin return without affecting their digital functionality; all that is required is that they transmit a low, constant DC voltage on which the data is superposed.

So, for USB 4.0, we propose that the Kelvin functionality is incorporated into both host devices and cables; the system is totally transparent to all client devices.

Using a higher drive voltage at the host port – 12V would be the logical choice – very much longer USB cables could be used with no degradation in performance and, most importantly, no necessity for user understanding of or involvement with the technology.
-- 8th of 7, Apr 28 2019

One of the arguments for Firewire being better than USB was that the differential signalling had no DC bias, which meant the power supply was more stable for things like audio interfaces. It's a pretty interesting idea to try and use the DC bias as a feedback system.

With long cables, ground differences do become significant. It may not be much of a problem if you're providing all of the power, but some devices will need more power than you can provide, and need plugging into another source, which means problems.

The main reason we cannot have long USB cables has nothing to do with power delivery, but with the design of the protocol. Each packet must be acknowledged within a fixed amount of time, and at the longest allowed configuration of hubs and cables, the tolerance is only a few nanoseconds, so even adding a few more feet of cable would cause packets to drop.

It is hard to come up with a better idea than Power-over-Ethernet. There are magnetics in every cat5 socket which isolate the signal so you don't need to worry about ground differences even over really long distances, and because it's isolated, you can stick a massive DC bias on it without doing any harm to equipment that isn't compatible with it. And of course no packet acknowledgement and full duplex means you can have very long cables.
-- mitxela, Apr 28 2019


[+], though I think in the power supply world this is more commonly called "remote sense" than "Kelvin connection". Also, IIRC, gold is less electrically conductive than copper, silver being the only non-superconductor better than copper. But the multiplexing of the sense voltage on the data lines is clever. I think I had thought in the past of using the data lines as voltage sense lines, but didn't realize it could be done without losing the use of them for data. (USB uses mostly differential signaling, but it does use one common-mode signal. Still, low-pass filtering or active removal of that signal can easily keep that from interfering with voltage feedback.)

// One of the arguments for Firewire being better than USB was that the differential signalling had no DC bias, which meant the power supply was more stable for things like audio interfaces. //

But with this scheme, the data line common mode voltage is coming from the peripheral only. That would seem to make the grounding better, because it takes some current out of the ground return line.
-- notexactly, Apr 28 2019


Time for rise of computer semaphore, just watch them sticks wiggle.
-- not_morrison_rm, Apr 28 2019


The alternative, that I thought of last night but then forgot before I could write it, is to have the peripheral just tell the host/power adapter what voltage it's receiving, using normal USB data communication. But: that would require a voltage communication protocol to be defined (in such a way that it can coexist with the device's main functions, probably resulting in the peripheral becoming a USB composite device if it isn't already one), and it would require power adapters to add a USB host chip and a DAC, which would make them much more expensive than adding a simple analog feedback path to the voltage regulator that's already there.
-- notexactly, Apr 28 2019


Anything over 50V is going to make the health & safety fascists all twitchy.
-- 8th of 7, Apr 28 2019


I love seeing that massive spark between the cable and the socket when I unplug my 100m USB4.0 cable to my backup RAID disks in the shed at the bottom of the garden...
-- hippo, Apr 29 2019



random, halfbakery