>> Anyone with more knowledge care to chime in?

It isn't just a matter of altitude. The geostationary satellite is a simple bent pipe hanging in space. It returns signals to a fixed ground station that it remains in contact with 24/7. The low-orbit satellite isn't hanging off in space. It is moving, constantly connecting and disconnecting with multiple ground stations every couple minutes. The route between customer and internet is therefore unstable, akin to a cellphone on a speeding car having to negotiation with multiple cell towers. But in this case the cell towers are also each moving on different highways too.

Sometimes there is no up-and-down connection. Sometimes the satellite over the customer cannot 'see' a downlink station. Then it has to bounce your connection along a chain of multiple satellites. Theoretically this should all work, but you don't need to be a physicist to understand the complexity of chaining together so many wireless connections. Ping times will suffer.

AND... there are a bunch of issues regarding the relative speeds of these satellites. They move fast enough that Doppler shifts between them become significant. Satellites have to modulate signals depending on where they are going. That means processing times, buffering ... latency.

It's not quite as bad as the cell phone case, since the exact visibility of users to satellites and ground stations to satellites is completely predictable, since the ground stations don't move, and the satellites move in fairly stable orbits.

In the cell phone case, you really have no idea when someone is about to go behind a building or something and have to switch cells.

All of the relative dopplers are also known ahead of time, so the transmitter should be able to change frequencies on a preplanned schedule without needing much time for resynchronization.

Just nitpicking, but uppercase S is symbol for Siemens, the unit of electrical conductance: https://en.wikipedia.org/wiki/Siemens_(unit)

The symbol for second is lowercase s. The SI is funny like that.

Other than that, I fully agree with your post.

Not really just nitpicking. I actually was thinking Siemens when I first saw the post. After reading it I did quickly realize it was milliseconds.

Man, product placement has really gone overboard this time.

From the FCC quote in the article: "Propagation delay in a satellite network does not alone account for latency in other parts of the network such as processing, routing, and transporting traffic to its destination."

In other words, client and server are not both sitting right under the same satellite. The trip to and from LEO is just part of the full route; you also have server to ground station and (in future) uplink satellite to downlink satellite (AFAIK Starlink currently doesn't do that).

I mean- sure? But that's also true of terrestrial internet. It's 22 hops between my laptop and www.google.com, and it pings at 5-10ms. Certainly Starlink isn't do any fancy routing that my internet isn't already doing?

Switches and fiber are well known proven technology. Satellite laser interlinks are R&D dreams at the moment.

There are satellite laser interlinks in orbit right now.

I thought the current starlink generation wasn’t using laser links yet

I believe both your comment and the parent comment are true. Laser comm has been done at fairly high speeds (both sat to sat and sat to ground station), but it's relatively cutting edge and Starlink isn't currently doing it

Sure, operated manually by a bunch of people with PhDs.

Is there any public documentation on how much of the round-trip time the processing and routing takes up? It has had the shit optimized out of it every else, so I can't imagine it would get significantly worse for a satellite, especially one of SpaceX's which attempt to use commodity hardware as much as possible.

Satellites have to operate in a very different thermal and radiation environment than most networking hardware.

Cooling is very difficult when you're surrounded by a vacuum. Also, without the planet acting as a heat sink, the difference between daytime and nighttime temperatures is staggering.

I don't actually know to what extent radiation hardening is a concern for these satellites. They're well below the Van Allen Belt, so the background radiation is nothing like deep space. Still, it's something you need to consider for systems that are expected to have years of continuous uptime.

(The Russians use commodity PC hardware on the ISS for systems not involved in navigation or life support. But that's a situation where there are people around to hit the power button.)

Not to mention that replacing hardware is expensive, and repair is basically impossible, so there are incentives to prioritize reliability over performance.

> Not to mention that replacing hardware is expensive, and repair is basically impossible, so there are incentives to prioritize reliability over performance.

SpaceX is actually changing this equation quite a bit in case of Starlink - mass-produced satellites + much cheaper and more frequent launches = replacing hardware gets much cheaper, which means they can prioritize performance over reliability more than it's customary in the industry.

> Not to mention that replacing hardware is expensive, and repair is basically impossible, so there are incentives to prioritize reliability over performance.

I'm well aware. But SpaceX is heading in the direction of quantity over quality, and banking on the sheer volume of satellites being launched cheaply outweighing any one failure.

I don't see why the delays would scale linearly with travel time. I would naively expect them to be a constant. So geosynchronous satellites have 240 ms of travel time and 310 ms of delay, in which case we might expect LEO satellites to have 13 ms of travel time and 310 ms of delay.

Maybe it's not that bad. Maybe you have to use more robust coding with geosynchronous satellites, and that drives up latency. But i can't see a reason why it would scale linearly with distance.

It's not a linear change with distance, it's based on different technologies. Just as your wifi router doesn't add 100ms satellites don't inherently need to have that kind of latency.

However, one change that is based on distance is ground stations. With starlink they want a ground station under every satellite to minimize the number of hops required. Satellite internet however has minimal ground stations because they gain little from adding more. Similarly, less physical distance let's them more efficiently manage bandwidth reducing queuing delays.

PS: Also, it's a round trip so you subtract latency twice from the total delay. Further, it’s not just altitude GEO is above the equator at a specific point which can be significantly east or west of you. Giving Ohio worse latency than Florida etc.

Round trip latency includes two trips to the sattelite and back. So geostationary sattelites have ~480 ms of travel time and ~70ms of delay

Do you have any justification for assuming the delays scale linearly?

I'd expect them to be mostly decoupled from the uplink/downlink time (not for any particularly sophisticated reason, just because the parts of the link that aren't uplink/downlink can be optimized and the speed of light can't). So that's 70 ms.

Starlink isn't supposed to be geostationary, to my knowledge.

According to the Wikipedia page, they requested the height be ~550km as of April 17th of this year.[1]

On 17 April 2020, in documentation to the FCC, SpaceX said lower altitude will put the satellites closer to Starlink consumers and allow the network "to provide low-latency broadband to unserved and underserved Americans that is on par with service previously only available in urban areas". The change will also improve service for U.S. government users in polar regions and allow for more rapid deployment of the network, SpaceX said. The lower orbits will help ensure the satellites re-enter the atmosphere in a shorter time in case of failure, and will enable them to broadcast signals at reduced power levels, because they are closer to Earth, which SpaceX said will allow the fleet to be compliant with limits to reduce radio interference with other satellite and terrestrial wireless networks.

The specific info on height is below that paragraph in the Wikipedia article. Whether that means it's achievable in practice or not I don't know, but geostationary orbit is 35,786 km, so that puts it at 1/65th the distance.

1: https://en.wikipedia.org/wiki/Starlink

That is not the round trip time, it's the one-way delay. Also, processing time can be significant in satellite networks. Finally, given the current architecture of SpaceX without crosslinks, the total RTT will be ~30 ms + processing time + whatever is the RTT of the backbone network they deliver the traffic to.

I think the processing time might be heavily driven by the waveform and how they do the networking. As complexity increases and they move towards a software defined networking schema, the latency goes up. If they're not using ready-made/proven ASICs, the latency goes up. I don't believe any of this is public from them.

I think so too. In addition, they seem to have regenerative payloads on-board (otherwise I don't get how they get ~20Gbps/sat with a single Ka-band antenna), so they need to do full decoding + encoding on the satellite. Not sure how fast that can be done, but it will add some latency ms probably.

This is the first I've heard someone speculate about that. I didn't think they were doing 20G with a single antenna, but that they were adding all spectrum together that it's capable of processing (even if it doesn't get access to it all). On-board processing would also require a significant amount of power, and it's riskier.

Its an existential threat to some incumbent last mile providers.

I think it's less an existential threat than forcing them into specific geographic areas where they can still compete favorably. I doubt your Starlink speeds will be anything special in a heavily populated area where people are sharing the spectrum, and that heavy population means that fiber is feasible. Fiber means Starlink is competing against Gigabit plus speeds, delivered individually and not shared (to the degree that the ISP provides enough capacity where they aggregate the connections).

Rural areas and coverage across most/all the United States, even in areas where only satellite phones worked previously, are major draws IMO though.

Agree with this, there’s plenty of areas (Eg non central belt of Scotland from my own experience) outside of cities, where the current provider is the only choice even though the service is slow and unreliable. Starlink could really appeal and gain a solid user base, very quickly.

Starlink is not going to be for end users most likely. The terminal, by all accounts so far, will be far more expensive than a geo terminal due to the motor and the phased array. I guess if you're willing to pay for it, it will certainly solve that problem, but there aren't enough of those customers to be profitable.

There's no need for a motor; that's what the phased array antennas are for.

802.11ac packages already implement MIMO-based beamforming, and I think there are already speciality 802.11ac packages using phased array for backhaul radio links. Phased array beamforming is a matter of size and processing power, but nothing different from cutting edge WiFi and 5G products, and not much different from what's already deployed in many laptops and cellphones. The entire software and hardware stack, including at the consumer level, is already capable of real-time, electronic beamforming. The technology for the Starlink terminals is well established and it could be done cheaply, it's just a matter of whether Starlink has the manufacturing pipeline primed.

Just like with rockets and electric cars, all the pieces exist and are relatively mature. It's just a matter of assembling them cost effectively, and that happens to be Elon Musk's special gift.

I know that, but there IS a motor. Elon has said this himself. what you are describing is far more expensive than a phased array with the scan needed for no motor. It is not a solved, inexpensive problem. 5G and wifi use significantly lower frequencies, which is much easier and cheaper to implement the phased array. Starlink is Ku and Ka band. There are no examples of a consumer version of these that is cheap enough. The only thing that came close is the oneweb, but that's only Ku, and it only exists in a Wyler tweet.

Satellite will never beat local radio (ex: 5G, WiFi) and local radio will never beat fiber, for physical reasons.

The reason comes from the Shannon theorem: C=B*log2(1+S/N). C is capacity (what we want), B is bandwidth, S/N is the signal/noise ratio.

With modern technologies, we are very close to that limit. Both 5G and satellites are using among the highest radio frequencies the atmosphere can carry, the signal power is limited by electrical power requirements and emission power regulations and there is not much we can do about ambient noise. Fiber uses light which, in theory, can carry many orders of magnitude more information but the problematic is different.

But the thing is that this capacity is per channel, and the narrower the channel, the more we can pack in a given area, and that's where satellites are at a huge disadvantage. Satellites, even Starlink satellites, are really expensive, and they are 100s of km away at best. It means they can only project wide beams, even with they fancy beam forming antennas. Ground based antennas can cover a much smaller area and are much cheaper, as a result they have more channels and therefore more bandwidth. Satellites may prove effective in sparsely populated areas, but not in dense cities.

Fiber, and even copper wires are the extreme case. The channel is a single wire, you can pack hundreds in a cable you can hold in your hand. Compared to wireless transmission, bandwidth is essentially unlimited.

Radio, whether local or via LEO satellites, absolutely can beat fiber. Routinely. That is much of the value proposition of SpaceX's product.

Microwave links from Chicago to New York claimed hundreds of millions in investment exactly on the prospect of selling reduced transit time at a premium. Similar gains on the New York - London channel, via Starlink, would be much more valuable.

The reason fiber is slow is that the signal does not propagate in a vacuum. It loses 30+% of the speed of signals propagated through air or vacuum.

Where fiber wins is on bulk bandwidth. A radio link has limited capacity, but it can get the first few packets through soonest. Starlink might start out with some pretty draconian tradffic caps.

I have heard of hollow-core fiber, where the beam travels in what amounts to a tube in the middle of the fiber, but I don't know of any deployed.

Fiber actually has limitations based on the length of the cable. Photons that travel perfectly straight arrive faster than photons that enter the cable at a steep angle and bounce off the walls a lot. Because of this, as you increase transmission rate, you start to get interference between signals, and instead of a clear On-Off, you start to get various degrees of brightness until the signal becomes too degraded.

Fixed wireless is that too.

One thing that hasn't been mentioned yet is queuing delay. All networks contain queues to smooth over bursts of packets being sent at once, and the time a packet spends occupying a queue can be significant.

Yes, this is huge, and likely one of the least appreciated aspects. Scheduling and queueing delay (not in the switches and routers) can contribute significantly to that 100ms.

No one forgot about those factors, he was only adding up the numbers to see where the lower physical limit is.

That shouldn't be significant unless: 1. There is a significant serialization/deserialization delay, which I don't think will be the based on my naive understanding of the tech 2. Buffering in time periods even relevant to discussion should really only occur under specific circumstances (like during overutilization)

I don't think that's the round trip time, that's the time it takes to send data over the satellite (ground -> satellite -> ground). A full round trip for data would be two of those since you have to factor in the data being sent back.

No, it's more like 500ms RTT for GEO