[Starlink] Comprehensive Measurement Study on Starlink Performance Published

[David Lang]

On Tue, 27 Feb 2024, Ulrich Speidel wrote:

> All good points ... see below.
>
> On 27/02/2024 2:13 pm, David Lang wrote:
>> On Tue, 27 Feb 2024, Ulrich Speidel wrote:
>> 
>>> On 27/02/2024 12:19 pm, David Lang wrote:
>>>> On Tue, 27 Feb 2024, Ulrich Speidel via Starlink wrote:
>> 
>>>> There are large areas with poor or non-existant cell coverage.
>>>> 
>>>> Outside the US, scaling of Starlink can happen just by providing coverage 
>>>> to locations that don't yet have coverage with no additional satellites.
>>> But what's the population of these areas? Generally quite sparse. (Or 
>>> politically disinclined to accept Starlink service)
>> 
>> per square mile? low. But there are a LOT of square miles, and those areas 
>> are ones where it's very expensive per-user to run fiber (even to cell 
>> towers or wireless ISP towers)
> The point though is that these sparsely populated areas aren't where the 
> scalability issue arises. Capacity needs to be where the demand for it is.

I only partially agree with you here. Yes, capacity that isn't needed doesn't 
matter, but I think that the capacity where there aren't other options matters 
more than in the more densly populated areas where there are other options (and 
I say this as a starlink user living in the Los Angeles area, a fairly densely 
populated area) I use Starlink as a backup now, but I do periodically test it 
and verify that it is acceptable for work + other uses.

>>> If you put in fibre today, you know that by upgrading the endpoints over 
>>> time, you can get orders of magnitude of extra bandwidth if needed.
>> 
>> If you can get fibre, you should get fibre (with starlink as a possible 
>> backup). SpaceX has said many times that Starlink is never going to be 
>> competitive to fibre
>> 
>> if you can get fibre, you aren't under-connected.
> Tick. But 2 billion plus can't, or at least not yet. The question is how many 
> of them might Starlink & Co be able to assist in due course?

what information do you have about the distribution of these 2B under-connected?

(and as someone who just a couple years ago was on a 8m down/1m up connection, 
what is the definition of 'under-connected'?, see the revoking of the grant to 
SpaceX for providing such service after they changed the definition of minimum 
acceptable connectivity and then proactively declared that Starlink will not 
meet it several years out)

>>> If you can reduce distance between satellite and ground station by a 
>>> factor of 2, all else being equal, theoretically you'd also reduce 
>>> footprint to a quarter, but that's assuming you don't need to worry about 
>>> antenna sidelobes. But say we can, and then that gives us a factor of 4 in 
>>> terms of capacity as long as our user density is the same. It also buys us 
>>> an extra 6 dB in received signal power and hence an extra 2 bits per 
>>> symbol. That's another factor of 4 at best if you go from 1 to 3 
>>> bits/symbol. Larger antennas: Doubling antenna size gives you 3 dB in gain 
>>> or an extra bit per symbol. So that turns into a game of diminishing 
>>> margins pretty quickly, too.
>> 
>> add in the ability for multiple satellites to serve a single cell and you 
>> can get a noticable multiple as well
> Not in terms of capacity - but in terms of a better distribution of it. 
> Already happening - most places can now "see" dozens of Starlink satellites 
> above the horizon.

but as I understand the reverse-engineering of the starlink system, a given cell 
is currently only serviced by one satellite at a time.

>>>  But now you want to serve cellphones on the ground which have smaller 
>>> antennas by a factor of I'd say about 16:1 aperture-wise. So you need to 
>>> make your antennas in space 16 times larger just to maintain what you had 
>>> with Dishy.
>> 
>> the cell service is not intended to compete with the Dishy, just be an 
>> emergancy contact capability
>
> Here's how one of the local partner organisation here spins it. Much more 
> than just an emergency contact capability:
>
> https://one.nz/why-choose-us/spacex/
>
> (Judge for yourself whether this instils the impression that you're going to 
> get 5G level service off this. You really need to read the small print!)

yeah, that does seem to imply more than it can offer. Elon has been pretty vocal 
that each cell is something like 70 miles in diameter, and the available 
bandwidth needs to be shared across all users. Text messages should always work, 
voice will probably work, and as the system gets built out, data will happen, 
but will be slow due to the sharing.

>>> That's a far cry from what is needed to get from two million or so 
>>> customers to supply two billion unconnected or under-connected. For that, 
>>> we need a factor of 1000.
>> 
>> without knowing what the user density of those under-connected are, it's 
>> going to be really hard to get concrete arguments.
>> 
>> Spacex is intending to launch ~10x as many satellites as they have now, and 
>> the full 'v2' satellites are supposed to be 10x the bandwidth of the V1s 
>> (don't know how they compare to the v2 minis), that's a factor of 100x 
>> there. 
>
> It's not that easy. Adding satellites in the first instance is just adding 
> transmitters, and unless you have spectrum to accommodate these, then even if 
> the satellites' on-board bandwidth is higher, it doesn't translate into as 
> much extra capacity. Spectrum comes in terms of extra Hertz, and in terms of 
> spatial beam separation. The former is limited in that they don't make any 
> more of it, and the latter is a matter of antenna size and getting antenna 
> side lobes sufficiently far down. And we know that SpaceX are running close 
> to spectral capacity in some areas.

I am assuming that the fact that they have planned for 10x satellites (~45k 
satellites up from the ~5k they have now) means that they have a plan to be able 
to use that many efficiently. I have no inside knowlege, so I am speculating 
about this.

> Assuming that the spatial distribution of the under-connected is somewhat 
> similar to Starlink's current customer base in terms of densities, we need 
> that factor of 1000.

I think there are a lot of early adopters for who Starlink is a luxury, not a 
lifeline. I think the under-connected are going to be in more sparse areas than 
the early adopters. I have friends and family in rural areas, and awareness of 
Starlink is only slowly penetrating there.

I'm seeing increased use for mobile applications here in the US, including in 
built-up areas.

>> Is another 10x in the under-served areas being in less dense areas really 
>> that hard to believe?
>> 
>> And Starlink will hopefully not be the only service, so it shouldn't have 
>> to serve everyone.
> So which factor in terms of capacity growth should we expect of Starlink & Co 
> over today?

not enough information. I agree it's something to watch, I'm just more 
optomistic about it than you are.

>>> And then you need to provision some to compete with extra capacity you 
>>> wanted, and then some to cope with general growth in demand per client. 
>>> And then you have to transmit that same viral cat video over and over 
>>> again through the same pipe, too.
>> 
>> True, although if you can setup a community gateway of some sort to share 
>> one satellite connection, you gain efficiency (less housekeeping overhead 
>> or unused upload timeslots), and have a place that you can implement 
>> caches.
>
> Indeed. Or if you provide a feed to a local ISP. But Starlink still focuses 
> on direct to site, as does every other LEO provider FAIK.

SpaceX is diversifying thier offerings, including boats, planes, and very 
high-performance community gateways.

I'd love to see more tech folks supporting this sort of thing.

I would especially like to see us put together disaster kits that can take one 
uplink and spread it around. We've seen SpaceX being willing to donate dishy 
kits, but being able to spread the hotspot island out from direct wifi range of 
the dishy to be able to cover a larger area would be worth quite a bit (and 
don't forget the need for power for the system)

David Lang