Hacker News Reader

Comment by 15155 | Hacker News Reader

Fascinating that they chose to use modulated board-to-board Ethernet instead of just running RGMII from MAC to MAC.

Ethernet seems far easier to prototype with. There's almost no off the shelf stuff for talking to RGMII whereas Ethernet you can just plug into your laptop for testing. If it's two different teams building things it seems like it would be a lot easier to just agree on Ethernet as the interface and then delay integration testing or release earlier.

A $3 breakout PCB with an RGMII PHY and MagJack on it would solve this problem without resorting to analog communication.

Assembly isn't free, either an engineer or the PCB fabricator has to put that together. Also the design isn't free and it's certainly not necessarily going to match the behavior of the device on the other side.

But your laptop's Ethernet adapter comes free with your laptop (both in terms of money and waiting to get it since it's already on your desk) and possibly even more importantly you know the laptop manufacturer and users have QAed it for you so it's absolutely going to behave the way you expect which is important when the device you're designing isn't behaving.

> Assembly isn't free, either an engineer or the PCB fabricator has to put that together

> your laptop's Ethernet adapter

The device as-designed likely wouldn't work with your laptop's ethernet adapter - hence why the author of TFA placed an isolation transformer and jack ...on a breakout board.

Heh I didn't notice it didn't have the isolation transformer. That is odd.

RGMII is not some obscure competitor to Ethernet, but rather, Ethernet was designed to be a modular two-part design with "MAC" and "PHY" chips connected via "MII" interface. RGMII is simply the latest version of it.

Many Ethernet-supported SoCs still use various MII style interfaces because it makes more sense to outsource the physical layer to some external chip especially if not everyone is going to use Ethernet.

It's perhaps like the difference between using Thunderbolt vs raw PCIe. You technically shouldn't need Thunderbolt if you're just permanently connecting two things inside a same machine.

Is it smarter to do it proper and make it silicon efficient than just shipping the darn thing ASAP? idk. We'll see.

RGMII requires way more work to run board-to-board (heaps of signals, quite precise length matching, impedance control, etc. on the boards, better board-to-board connectors etc.) and at the end of all that will likely be less robust than just running Ethernet. I'd much rather use SGMII just because it's far fewer signals to match (even if it runs way faster) instead of RGMII.

The chips they're using might already have Ethernet PHYs built in anyway which might also be part of the reason they're using Ethernet.

Eh, it's just a half dozen signals in each direction running at 125Mhz DDR. That's in spitting distance of being able to be bit banged by something like an rpi pico.

RGMII isn't really designed to go board-to-board, fairly high data rates, and ideally all of the signals should be delay matched. That gets a bit trickier when there are two boards involved. Also I would expect EMI/EMC issues.

I know people do that sort of thing for evaluation kits, but it doesn't seem like a good idea for production.

A lot of this is pretty POC-y. Agree digital to analog to analog to digital is kinda inefficient, and in the abstract MAC to PHY (which is probably what you mean when you say MAC to MAC) with RGMII is probably better. My off the cuff guess is that it is likely the written-up interface is easier to access or requires less diving into internals. Not sure where the RGMII lines are, and depending on the design of the Starlink mini itself (I am ignorant of this) the lines might have been buried deeper and less accessible, who knows.

I'm not sure I understand the entire point of the exercise. There's already an RJ45 jack on the Mini, so no need to hack the unit to get access to an Ethernet PHY. And the WiFi router can be turned off via the setup page.

Did they remove support for the Ethernet jack on the Minis available in Ukraine? It looks like it's still present on the WiFi board, next to the power jack.

You might imagine some use cases where mass is a critical concern.

They may want to make absolutly sure no wifi signal emit from the device. Turning it off in the setup page is definitely not enough.

The wifi chip may emit signal during boot. The device may get accidentally reset in the field. SpaceX may push an update that messes with the settings.

I mean, more power to them, certainly, but WiFi emissions seem like the least of your concerns when you're operating an antenna for satellite comms. There will be no shortage of side lobes at Ku band for anyone who cares to listen.

Cutting down on mass would make sense, though.

It will still draw power with wifi turned off, though much less. The most effective way of reducing the P in swap is to remove the unit entirely

it is not just the wifi. its the router.

What does that mean?

That connector is way cheaper than something that could carry an RGMII signal without major reflections. It's probably cheaper in the end to have the extra silicon than a better connector, and I don't think you need the magnetics if you don't need isolation.

It could just be that using Ethernet allows for more flexibility in what is connected to that board-to-board connector. In this instance, it's a wifi router but a customer may instead want a bare port to connect to their existing infrastructure (just the modem). It also means you don't have to worry about which MII standard both boards use, they can be upgraded independent of each other. The physical interface for Ethernet remains the same whereas MII has different pin counts.

Sure, then you get to write an MDIO emulator so that it actually detects link and since it's a proprietary system you can't exactly comment out the SMI code.. fun!

Reply on news.ycombinator.com