Matched impedance differential pair routing what's the most important to match?

[Northy]

I'm routing some matched impedance differential pairs (USB 3.1 SS lines) to a USB Type C connector (so there's two 4 pairs).

If I've got length mismatch within the P & N pair (Intra Pair Skew) then doesn't adding a meander to one of the traces in the pair affect the impedance matching?
What is the most important. matched length or matched impedance?

Should I be looking to match over the entire length? So Connector to IC, or the separate sections? So Connector to TVS, then TVS to IC?


Parts used:
Microchip LAN7800
TI HD3SS3212 Mux
CUI CUI_UJ31-CH-G1-SMT-TR Type C hybrid connector

Thanks,

G

[T3sl4co1l]

The coupling between pairs is typically quite poor; normal mode is dominant.  Doesn't much matter how much coupled length there is, the important thing is not introducing CM-DM conversion by travelling through noisy regions with mismatched delays.

Tim

[Benta]

If the meander is done correctly (eg, chamfered corners), impedance matching shouldn't suffer.

[Northy]

The coupling between pairs is typically quite poor; normal mode is dominant.  Doesn't much matter how much coupled length there is, the important thing is not introducing CM-DM conversion by travelling through noisy regions with mismatched delays.

Tim

Sorry Tim, I'm not sure I understand 
I might have to do some reading up.

G

[Northy]

If the meander is done correctly (eg, chamfered corners), impedance matching shouldn't suffer.

Hi Benta, what would "correctly" be?

Thanks,

G

[Benta]

If the meander is done correctly (eg, chamfered corners), impedance matching shouldn't suffer.

Hi Benta, what would "correctly" be?

Thanks,

G

https://www.ti.com/lit/an/scaa082a/scaa082a.pdf
page 14

[T3sl4co1l]

Meh, corners are highly overblown.  The discontinuity is a tiny percentage of any aspect of the waveform (edge rate or pulse width), and only amounts to a modest fraction of the impedance within that short length.  There are certainly RF circuits where that (and more) matters, but that's well into the GHz.

Tim

[T3sl4co1l]

Sorry Tim, I'm not sure I understand 
I might have to do some reading up.

More specifically, the differential impedance is very close to twice the normal mode impedance for a single trace in isolation.

Consider two coax cables, 50 ohms each.  If you apply a positive signal to one and negative to the other, the signal will experience a total of 50 + 50 = 100 ohms differential.  The common mode is 50 || 50 = 25 ohms.  There's no differential mode coupling.

Suppose you slitted the shield on both cables, lined them up and soldered them together.  Now you get a figure-eight sort of cross section, nonzero coupling between signals, and the impedance drops slightly.  Widen the slit and smoosh the cables together more, and the wires inside get closer, while the shield gets closer to an ellipse, then a circle.  When the center conductors are very close together, they couple strongly to each other, like twisted pair.

The amount of differential coupling in microstrip is typically small, say 5-30%.  (The USB spec specifies what this should be.)

I forget offhand what USB3+ signalling is, and how sensitive to distortion it is.  If it's binary (two levels) like its predecessors, then there's a wide margin for distortion, say 30% of the signal.  Not that you want to burn up all this margin at one end, but the cable is the elephant in the room, followed by the connectors.  PCB traces aren't very important.

Tim

[jonpaul]

Hello: USB 3  has many implimentations: Which one?

USB 3.2 Gen 1×1 – SuperSpeed, 5 gigabits per second (Gbit/s) data signaling rate over 1 lane using 8b/10b encoding (effective 500 MB/s) , the same as USB 3.1 Gen 1 and USB 3.0.
USB 3.2 Gen 2×1 – SuperSpeed+,[56] 10 gigabits per second (Gbit/s) data rate over 1 lane using 128b/132b encoding (effective 1,212 MB/s), the same as USB 3.1 Gen 2.
USB 3.2 Gen 1×2 – SuperSpeed+, new 10 gigabits per second (Gbit/s) data rate over 2 lanes using 8b/10b encoding (effective 1 GB/s).
USB 3.2 Gen 2×2 – SuperSpeed+, new 20 gigabits per second (Gbit/s) data rate over 2 lanes using 128b/132b encoding (effective 2,424 MB/s).

The USB standards are easily found:

https://www.usb.org/documents

  Good overview at the wiki:

https://en.wikipedia.org/wiki/USB_3.0

Like all transmission of digital data, it is binary data but analog design of transmission, with defined levels, risetimes, hysterysis, etc, see for instance the old RS232 and RS485 differential TX and RX.

After you determine the speeds and which version, suggest to check out the excellent gigaherz layout and guidelines of Dr Howard Johnson: eg High Speed Digital Design: A Handbook of Black Magic

Good Luck,

Jon

[Northy]

This is the spec of the LAN7800:

USB 3.1 Gen 1 Device Controller
- Supports SS (5 Gbps), HS (480 Mbps), and
FS (12 Mbps) modes

So up to 5 Gbps.

However, it's a USB to Ethernet bridge, and the Ethernet side is Gigabit (1000base-T), so that would be the maximum wouldn't it?

Thanks,

G

[Northy]

So is the general consensus: "do the best you can, but don't sweat too much about it"?

G