Hacker News Reader

Comment by randyrand | Hacker News Reader

Keep in mind, the entire Starlink array will only be able to handle like ~1% of global consumer internet traffic. They're only a handful of gbps downlink each. ~10gbps downlink.

View on HN

Starlink apparently won't be a fixed concept. 5 year sat lifecycle means its likely we'll see rapid year-over-year turnover and hence macro-level iteration on the specs and design.

RF and the Atmosphere being what they are and the non-variable parts of this equation mean there are significant real physical and abstract legal limits for any project attempting to put "bandwidth in orbit."

What are some ballpark figures on the theoretical / practical limits?

20 Gbps per satellite, according to the FCC filings.

Which isn't that great.

So theoretically the more remote a location the faster the starlink will be because it is shared by fewer people. Which is the exact inverse for wired infrastructure.

Except that the more remote your location the further you likely are from the nearest ground station, which means more hops through the satellite network before you get back down to the ground, higher latency, and reduced total throughput of the system.

The system Starlink is trying to beat is BGAN and regional Ku/Ka providers. If you're expecting them to beat the pants off of your terrestrial wired service you're probably going to be disappointed, unless you're on the edge of the range of some old neglected AT&T DSL line or still trapped on dialup. Even then it's almost certainly going to be much more expensive but at least it will have a sizeable speed advantage.

You want to downlink near the server, not near the client. Packets will move 40% faster through vacuum than through fiber.

The more traffic routes through the satellites vs. the ground the more you'll be competing for that limited sat-to-sat bandwidth and the more queuing delays you will suffer. You want to get the bits off of the birds and onto the ground ASAP if you want to maintain consistent latency figures.

This was a big problem with Iridium. With only one ground station available your latency figures were all over the place thanks to the complicated routing that every packet had to go through.

Each RF hop is going to have more latency than fiber for RF physical layer processing. Forward error correction intentionally uses large frames to spread out burst errors.

Sounds speculative. Given link bandwidths and likely error rates, it seems better to allow the higher layers to retransmit than to optimize for correcting burst errors. Especially given Starlink's goal to control latency.

I haven't seen a link budget. Do you have a link for one about the "likely error rates"?

My guess is Starlink is going to need FEC for Doppler effects and multipath in urban environments. Maybe they will be able to overcome multipath with modulation. But Doppler shift will be significant with LEO.

The latest 802.11n standards have been increasing physical layer frame sizes, not decreasing them. I don't know of any modern wireless standard that doesn't use FEC. Coding gain is significant in most link budgets I have seen.

The last satellite RF downlink I worked with was for JPSS-1 in LEO nearly 6 years ago. There was a (255, 223) Reed-Solomon code with an interleave depth of 4 frames. So you effectively had to receive 1024 bytes, decode, and you got 4x 223 bytes worth of information. While latency wasn't a major concern, this scheme seemed pretty standard to me. If you don't believe me, the spec is public information:

https://www.star.nesdis.noaa.gov/jpss/documents/CDFCB/GSFC_4...

I'm sure SpaceX is trying really hard to control latency - so it doesn't get out of hand! One set of RF processing here, another over there, each with their own filters and decoding algorithms and then you have an extra 100 ms of delay in the system.

Not commenting on error correction, but your speculation about multipath reveals that you fundamentally fail to understand how starlink is meant to work.

Starlink allows high bandwidth density on the ground, and at the satellites because every antenna in the system is an extremely capable steerable phased array. Every transceiver can fully distinguish between signals of equal power and frequency that differ whose sources differ at least 10 degrees.

Reflections from buildings are irrelevant when the antenna must already be capable of picking the correct transmission out of the 50 or so equally powerful beams on the same frequency that only differ in angle.

Also, since every terminal already needs to be constantly aware of the position of every satellite, the correct answer to doppler is not fancy error correction, it's modulating the transmission to eliminate it, or just expecting it at the receiver.

Elon talks about 20ms or 10ms, not 100. He wouldn’t be wrong by a factor of 10. You’re really just speculating based on some non IP based overarchitected mil spec monstrosity you’ve seen before

https://mobile.twitter.com/elonmusk/status/11329039145865297...

Besides we’re talking about the latency of the satellite to satellite links, not the downlink (since that’s the difference between using a downlink close to the client vs close to the server).

Yeah, I'm speculating and so are you. Your citation is an Elon tweet.

I'm not talking satellite to satellite latency, I'm talking total latency end-to-end. I'm also not saying the system will be 100 ms, just that it takes significant engineering effort so latency doesn't get out of control. IMHO, it's a fundamentally much harder problem than existing transatlantic links, despite the greater signal speed in a vacuum. I'm skeptical of what the latency will be. You disagree and that's fine.

Maybe we'll see an actual measurement soon enough. The competition can do 60 ms round-trip to NYC to London.

I think you’re suggesting an /extra/ 100ms in addition to speed of light time, which is to speculate that they’ll do it in a way that’s needlessly crappy.

I’m speculating they’re smart guys working hard to keep latency low, and that routing will be a more challenging latency issue, while serialization should be minor.

But we’ll see and I agree we’re both speculating

Have they given a range on latency goals? Ie seconds, tenths of seconds, milliseconds, etc?

I’m seeing figures of 1 million$ per satellite. Over 10 years that’s roughly 100k/month in mortgage or around 1k customers at 100$/mo.

That's interesting, can you expand on that? Thanks.

Tell that to the HFT people.

Is this some fixed point-to-point line of sight terrestrial microwave link? I'd believe that could do it. Do HFT people ever communicate with spacecraft though?

No, the point of the HFT links was to make the most direct shot possible so they could beat the packets traveling via fiber. Going up to space would defeat the purpose.

Except if the terrestrial fiber backbone is cheaper and less congested you want to downlink near the client.

> the more remote your location the further you likely are from the nearest ground station

I thought downlinks were going to be pizza box-sized antennae. Why would it be far away from you? That could live on your roof.

Netflix claims an UHD stream is 25 Mbps. This is probably the most demanding use case for your average consumer, but it may not be a rare use case in the near future. Each satellite can support 800 UHD streams. That doesn't seem sufficient if each satellite covers a large geographic area.

This suggests to me a low data cap and/or rate, perhaps comparable to DSL.

12,000 satellites is the current FCC approved launch figures, I believe. 20Gbps per satellite but I don't know if that's full duplex or how much is divided between terminal/gateway, so lets divide that in half. For simplicity let's say they only sell consumer-grade 100Mbps connections, which are oversubscribed 100x as is typical. Divide the possible number of customers by 3 to account for uneven geographic distribution and other factors.

12000 * 20Gbps / 2 * 100 / 100Mbps / 3 = 40,000,000 customers

A service like that could sell for a lot in rural North America, but maybe not to 40m customers. Let's assume most of those customers are from lower income countries or from places with more competitive ISP environments. Maybe $20/mo average revenue per customer. That's $800m a month. That already sounds compelling to me, before considering the potential for more or higher bandwidth satellites, or for higher margin commercial or government contracts.

You're forgetting that this is a satellite constellation- I'm not sure what the orbital parameters are offhand, but I'd divide that number by at least 4 (and maybe up to 20), based on number of satellites over the service area(s) at any given moment.

Last I saw, they were reconfiguring deployment plans to go for higher density constellation in the lower latitudes before worldwide coverage, so I'm not sure if this is even knowable at this point.

I think your simple calculation assumes relatively even geographical subscriber distribution, which absolutely is not the case. Just rough estimating, assuming every other figure is approximately correct, you’ll probably have to cut that by 50-75% based on people’s natural tendency to live near other people.

A few HFT customers will probably finance the whole thing, tbh.

Another calculation is revenue per satellite per customer served. Would 100k per month per satellite finance these things long term?

This seems like the important metric.

($800m a month * 12month * 5month sat lifespan)/12000 sats = of 40m revenue per satellite

150kg sat to launch at $2,719kg/leo costs $0.4m

This first round of sats have 20Gbps but the next gen will be much faster.

But if you remove all streaming/porn/torrent traffic it would be able to handle a lot more of the global traffic.

If you remove all of that what's the point of even having Internet anymore?

Or just charge based on volume. I'm sure USAF is willing to pay a lot for true global Internet access on their planes

Actually, during the early 2010's I had heard from a reputable source that half to two-thirds of the internet traffic was "unwanted by the end user" i.e. spam, involuntary software updates, and advertising. Another source from the same time period mentioned that about a quarter of all internet traffic originated from Youtube.

That’s a bit tautological wouldn’t you say?

I'm sure people who intend to do business will accept such restrictions as long as they can access email and Salesforce dashboards. Although it might break the spirit of net neutrality (which, to be fair, isn't really a thing anymore anyways).

What's the point of such internet? I can already buy low-cap 4G cheaply.

Except where you can't. Lots places have no coverage.

Reply on news.ycombinator.com