Starlink's Hidden Secrets: What 15-Second Handover Means for YOU!

everyone knows about the LEO sata network and there's a four major players like out there with different number of satellites running and some of them are already in service as well and but what I want to know see is like kind of uh LEO is more than just let's say broadband internet access uh we have let's say kind of direct satellite to cell phone and we also have like internet things uh in the space as well and this Thursday uh we have another panel like kind of about the kind of um data center in space as well which I'm part of in that. So I just mentioned like those four leading players and particular like kind

of we're focusing on Starink because that's the biggest elephant in the room right now and it claims it has like a 10 million users by the beginning of the year and by developing a special technique from the internet we can measure these 10 million users and we can successfully reach let's say six million of them. So this is one of the purpose we see like this is the largest test pad because like you know we can cover 60% even more than that uh of the 10 million user out there. So you can see like know from this figure we show like the majority of starting users are coming from US and Canada and also like kind of um in South America as well. So in this case like kind enough we know let's say kind of where the user distributed because stunning only tell us how many users in total but we also know like which country has how many users in that sense. So because it's like a portup covered we also want to measure them like around the world as well. So this is done like

by uh this winter it's winter in the northern hemisphere so it's summer like now done in the Antarctica. So we have like a many scientists working there in the research stations. Some of them like know brought the stalling dish with them and they have much better connectivity than they whatever they have before. So we can see like for these stalling dishes they actually clustered in the research stations which is quite reasonable because you stay most of time like in the research station. So this is the behavior for one particular stalling dish. So here is the wall clock the real time and the yaxis here is the latency. So basically of what we measure is the time uh between the port presence for that particular dish to the dish itself. You can see like a very interestingly the minimal latency changes over time in a periodic way. So this because like for the low earth object satellite we can see them about maybe few minutes. That's what our inhitial sort as well. So maybe after few minutes you have to switch to another satellite but once we have done some measurement we found actually kind of starting does hand over much sooner

than that and every 15 seconds starting switch to another satellite because like if you have more than let's say kind of 10,000 satellites in the space but you do have like 10 million users on the ground so that's why kind of you have to shuffle these satellites between users to make sure like everyone has a fair share of the satellite capability as well. So that's like first discovery we found. Okay, this is quite different from what our sort in terms of visibility. Next we also put like all these 14 let's say starting dishes we discovered in the antactica in the same graph. And again this is the walk clock and this is their latency. So here you can see like not only they switch satellite every 15 seconds they also switch at the same time or synchronized way at least like in the antactical later going to see actually around the water as well. So you can see these delay delay spikes is a moment when the satellite switch between those like uh for these dishes. You can see all these foring dishes they actually kind of synchronize the switch over behavior as well. And this quarter here uh is the styling behavior let's say kind of in the Arctic circle because up there we only see a small portion of the satellites in going to the like a polar orbits. So this like a black line uh the uh white line here is the trajectory of the satellite. So basically enough we can not only know this dish is in the actctic circle but we also know which satellite you're talking to as well. Uh same thing like for the dishes like near the equator here and you also see a black band here that's because for both the LEO satellite system and the traditional geo they use the same frequency bands and the geo has priority. So that's why kind of when the satellite go over the equator above the equator there they have to shut down for a while because otherwise they're going to interfere with the geo system. So that's why kind of the user always complain that why there's like a black bound in my kind of uh field of view as well. That's because that's geostatist thing there. So this is the um test bed we're

building right now. It's called like a CCC LEO stands for the coast to coast coast let's say test bed. It covers all 10 Canadian provinces and territories as well. So you can see like know they're located like know in each province and territory and then now we're also expanding them into the US and other countries around the world as well. So there are some dish for this one is the dish like of at uh in Victoria we put them into the hot building and we also have some like hosts in different parts of the world like this one in Ottawa and this one in Dambo. Basically they host our dishes and we also put a small mini PC behind it to do all the measurement automatic measurement like a taskus and from these dishes you can see they see different part of the sky because for example this part uh is obstructed by the roof uh someone just put a dish on the backyard and the way radio kind of like penetrated the roof. So you see the like a red dots here and we also see like the trees and things like that as well. So this test bed has starting let's say operating from 2023 and by now we have collected more than one terabyte of the compressed trace data and this has been downloaded around the water for 6,000 times. So that's a lot of data like accumulated and uh next I'm going to show you some of the work based on the data we collect uh by ourself and also some by other people as well. uh all our work uh paper data and the tools can be found at this particular like portal link uh with the link to the test bed itself as well. Uh for example like when we collect enough data from the starting users we can also map the styling backbone around the world. Staling is quite unique in a sense like it's not only provide access network but also provide let's say backbone network around the world as well. So this is the starting backbone in let's say around the world and this part is just I showed you about the handover behaviors. So initially we saw like kind of you could hand over every let's say few minutes when the satellites are visible but the sting in decided to hand over every 15 seconds later going to show like what's the implication of this like fast hand over and here is the stalling like kind of um satellite you're talking to right now for example like our dish is in Victoria so here is Victoria on the Vancouver Island and the dish is here and here we can see like at a certain moment of time we're talking to a particular satellite and after 15 second like we switch to another one so that our delay goes up and our signal tones ratio could drop down. So in this case like you may switch to older generation slightly. So that's why kind of the um signal tones ratio is lower in this case. So for

comparison I mentioned like there's four major players right now in addition to star link we also have some uh thing called one web uh which they have about 600 satis running there in polar orbits. So their handover behavior is quite different because for one web the beam is fixed right now. So that's why when they go over the earth surface they just scan through earth surface. So that's why you can see their beam behavior is much more regular than the stalling. But because they're a little bit higher in let's say elevation and there's a few let's say satellites out there with limited ground infrastructure the latency is slightly higher than stalling but relable like know more stable as well. So once we have the big picture about these like Leo stat networks out there, we're going to dive into let's say some of them for let's say deep understanding as well. Uh one thing we did is like

know for example like if this is your stalling dish and we want to figure out which satellite talking to right now before Staling provides such information is in GPC interface but for whatever reason later they removed it. So for us what we can do is like kind of if you know like kind of you're talking to a satellite in this direction and then you can also link to the latency you're experiencing right now. For example like this is moving in this direction and this is your dish it actually moves closer to you. So that's why you see the minimum latency here goes down a little bit as well. That's because the propagation delay goes down and after 15 seconds starting switch you to another satellite. So this one is like the green one here and after another 15 seconds it's moved to even closer satellite the red one and then they realize they made a mistake because this particular satellite actually is moving away from you. So in this case like it's not going to stay with you for a long time. So that's why the starting moves you back to the this particular satellite which is the same as the previous one here. So now we know like this particular hand

over is useless actually you waste like kind of handover opportunities and you introduce like latency changes as well. So that's how we communicated this back to like starting. So now they remove this kind of like unnecessary handover in this case. So this is work done by by my student Ali and he had now graduated like with the math degree. So once we know the location of the setter we're talking to and then also use the TIE data which is public from uh the um because we all we have to let's say kind of track all the objects running in the space. So in this case like kind of we know we're talking to satellite in this direction because we know our GPS location we also know the kind of orientation of the dish that's why kind of we know these are the satellite we're talking to and we can map the location of the satellite to the identify identity of the satellite as well. In this case like we know not only the setup right now but also two minutes or even two hours after right now.

So because the handover behavior we found like kind of this actually introduce new challenges to the network protocols and applications as well. So one things like kind of for the uh protocol we are running we assume like kind of the latency will be stable but if the latency jumps up you have like a timeout if the latency jumps down you have like a pack ordering which all cause like a TCP through going down so that's why kind of we found these handover behaviors they actually can align with the lost events in the TCP protocol as well not only in Victoria in BC but also in Victoria in Chile which is another place in the middle ocean so in well once we've identified the root the cause of the problem. The fix actually is quite simple because we just need to know like know we're going to have a regular handover right now. We're going to freeze our congestion control algorithm for a little bit of while about 100 millisecond. So in this case we can improve the throughput and we can also reduce the transfer time which is like improved the network protocol application and performance as well. So by applying the same idea we can use

this for the video streaming over let's say kind of stalling or like a leo seta network. So this is a test by our students get in our lab. We basically going to build a small Netflix in our lab with the kind of virtual machine in Seattle. We are starting PPS and we do video streaming and by using the same technique we can improve the like quality and also reduce lit as well. So these are some of the work we do but we also collaborate with other people using like the infrastructure and also collect more data as well. So one example is

like we put those stalling dishes on vehicles because so far you've seen like these dishes which is stationary. So on Vancouver Island we have some indigenous kids they have to spend like a two and a half hour every day in the morning and a two and a half hour day in the evening going to another part of the island for school. So in this case like the bus going through some like kind of uh winding let's say kind of highway which is like no coverage. So the parents and teachers get worried about the weather and other things as well. So what do we do is like we put a starting a dish on the school bus. We also put a camera there because we want to like a capture the obstruction on the dish by using a video camera to compare the like kind of estimation about the obsession versus what's the visual let's say kind of gu is and we also collabor with people like from university of um Minnesota to do let's say kind of this uh mobile styling dish on a vehicles in the city and also collaborate with people with the northern east university to let's say kind of have a crossount uh test among the kind of uh dishes along the way because we with different terrains and different user population and a different like kind of uh performance as well. Not only on the ground, we got into the

air space as well. So this is like know one of the things we do like kind of uh we basically kind of build a small test bed node which is about the size of this much with the battery with the mini PC and all the communication necessary there. What do you need to just like plug in the power plug and then it does the automatic measurement by itself. So basically for example like in this case we see like kind of the um aircraft used to have let's say kind of a very bad internet because they are using the geo satellite systems. Now some of them have moved to LEO uh stalling or one web you can see like the throughput jumps up a lot but more importantly the latency let's say goes down dramatically because low latency there. So there's some experimentation we do like kind of how to do let's say video streaming or video conferencing on aircraft although of course sometimes it's not allowed uh by the current regulation as well and I was like kind of booked for a flight between let's say Seattle to Dha like kind of at the beginning of the March to 6D hour let's say round trip to test the kind of sting performance on kata areas but unfortunately the trip get canceled due to the uh things happening in the Middle East but that's why kind of we also develop a new techniques to measure the kind of um styling performance on aircraft from the internet. So in this case we don't have to take the trip but we can measure the performance from outside in as well. So this is something like kind of we're doing right now and if time allows later I can show a quick demo as well. So far all the things I have shown you to is by the regular consumer level let's say user dish. In addition to the user dish

stalling also has something called a communic gateway. Uh they about to change the name as well. So in this case like kind of if you have a small ground station look like uh infrastructure for a small community most likely will be like island nation in the Pacific. So that's why kind of they use different frequency bands that's why kind of they can reach like 10 Gbit per second let's say by symmetric link as well. So uh we can compare the performance with user dish versus like communic gateway because we have like a more bandwidth and also less competition because you can like have a very few ground stations or communic gateways versus like a thousands of let's say user dishes just like in this room if we double the Wi-Fi devices it's the performance is only just half but even lower because the computation as well. So in this case like you can see the uh community gateway now has much more better uh performance. Uh these kind of gateway are not only like know for those island nation that also moving on some of the biggest cruise ships like the uh icon the utopia and the star of the seas and the star is the first one to confirm it has the kind of um communicateway in addition to the like a regular let's say uh high performance dish as well. uh of course like not only for cruise ships but also for many like kind of shipping vessels close the different oceans there. So basically kind of you can think about these uh dots here are the vessels with the stalling dishes they cross the ocean as well. uh in this case like they traverse different pops uh of the stalling ground infrastructure because I mentioned like for stalling it's not only like of satellite in space but also like kind of ground infrastructures and when you are like kind of in uh let's say Boston your stalling dish will be associated with the kind of uh pop in let's say New York City but your ground station could be like either nearby or could be very far away but nevertheless let's say kind of your um pop will be like stay in like uh New York City. Uh we're not only just like doing the

things like know on the ground and the into the air. We're also trying to like reach the space with a cub set to be launched by the end of next year. So this is like a new contract from DRDC the defense research development Canada which is like the DAPA in the US to develop a test satellite to be launched and this test satellite is going to collect data from the space from the air from the ground. Meanwhile, it's going to use the laser communication like links uh just as the this morning mentioned like it's pretty hard to point out to like laser links in the space because they moving things around with the commercial let's say kind of Leo con translation there and then we kind of down link the data to the ground to our like kind of center for aerospace research at the University of Victoria but we also have like a radio links from the test to do let's say simple TTC uh things as well. So this is my last slide uh and also my contact information as well. So one thing I want to bring attention to is like kind of uh we have all the information here at this location uh with the data set with the kind of a test bed as well and you're welcome to

use data set but meanwhile kind of uh if you can uh you're encouraged to maybe host a virtual machine or maybe bring something like this to your next trip like on pen or maybe like kind of on cruise ship as well and next year I probably can talk about like more about our test satellite to be launched by next year and it'll be in space for at least like a six months is uh to do all experimentation as well. So that's what I have and this Friday I will give guest lecture in like 1689 which I can give more details about all the things we have been doing.