Look to the Stars…for Internet?
Back in 2015, SpaceX announced plans to develop a global satellite-based internet service. It won't be alone in that space; HughesNet, along with some other options in partnership with Dish Network or DirectTV, are established players in the satellite internet provider game. Frankly, satellite-based internet is unacceptable for most consumers—especially for those gerbil in nature—owing to its extremely limited bandwidth and incredibly high latency. Because of the extreme latency, anything that needs near-real-time communication is pretty much out the window. That mostly affects online gaming, but even things as simple as VOIP can fail. Some services, like the PlayStation Network, will simply refuse to operate with ping times in the four-digit range. The reason for the extra latency is simple physics.
Satellites that provide internet are largely based way out in GEO. This means that the satellite is at roughly the same place in the sky relative to a single point on the ground. That's a good thing for most use-cases. A consumer will have a dish located somewhere on their property that is pointing to a specific place in the sky, because that's where the satellite is orbiting. The trade-off for maintaining a fixed place in the sky is that you have to be pretty far away above the surface of the earth: around 36,000km, to be semi-specific.
So, why not put satellites in LEO? The upside to LEO-based satellite internet is that the latency is around an order of magnitude lower. Instead of talking 100's of ms, you're talking 10's of ms. In fact, by some calculations, it may be faster to communicate long distances via satellite than fiber due to fiber's approximate 30% speed of light slow down (by my quick math, if the linear distance is greater than four times the satellite's height, satellite would be faster). That all sounds great, but of course there's a catch: LEO is far from stationary, so you need a lot of fast-moving satellites to maintain coverage. This setup is called a "satellite constellation."
From a ground-tracking standpoint, a single dish pointed at a fixed point in the sky won't cut it for constellation communication. Instead, the receiver would have a much wider view, something like 45° of the sky. Over time, multiple satellites would pass in and out of that cone. Such a client device would have logic built-in that would let it know where the satellites are, where they're going, and handle seamless handoffs between them. Those of you who have worked with Wi-Fi zero-handoff know how tricky this was, until recently.
In addition to the latency benefits, Starlink could offer significantly higher bandwidth. Rather than having a handful of satellites serving the entire continental United States, you might have a couple hundred that are covering that same area. As a result, the bandwidth isn't spread nearly as thin, so more is available to any individual client.
SpaceX has worked toward deploying such a satellite infrastructure for a while now. The company's intent is to have thousands of satellites in LEO, between 500-700km about the earth, with a user-base of a million clients. So far, Starlink is still in the development phase, but that's about to change. Back in February 2018, SpaceX launched 2 prototypes, called Tintin-A and TinTin-B. They were proof-of-concept satellites that piggybacked on a large communication satellite launch (Paz, a Spanish imaging satellite). For the last 15 months, SpaceX has been testing various aspects of their system. While many details of the overall project are still unknown, 2019 is supposed to be the year that SpaceX launches a 1.0 version of their communication hardware.
Future Starlink satellites are expected to make their way to space on dedicated SpaceX launches, rather than as secondary payloads. Most of SpaceX's payloads have been either two small or medium size satellites or, more commonly, a single large satellite. These are mounted on top of the second stage of the rocket and are kept protected inside a fairing.
Most communication satellites, which rest in GEO, tend to be big and fill up the mass and volume restrictions of a rocket by themselves. A larger satellite dish is required to receive signals in GEO, and more onboard fuel is needed for station keeping. For most launches, the chief limitation is the rocket's mass-lifting ability, rather than volume-carrying capacity. Along those lines, a SpaceX Falcon 9 rocket can deploy a payload with three times the mass if it is going to LEO instead of GEO. The farther in orbit your payload is meant for, the more energy it takes to launch the same payload. The same math means that the further out your payload is going, the less mass you can launch with the same energy.
SpaceX designs its own rockets, and the company knows exactly what their capabilities are. Instead of large communication satellites, what SpaceX has done instead is develop small satellites. SpaceX can clearly optimize the dimensions, the center of mass, and other physical properties of the rocket to suit the needs of launching Starlink. It can also intentionally design the shape of the individual satellites to maximize the amount of space inside the rocket fairing. SpaceX's current fairings are 5m in diameter and 13m in length; for the initial Starlink launch, SpaceX found a way to Jenga sixty satellites inside the fairing.
Another difference between LEO and GEO is the rate of decay. A typical GEO satellite can stay in its orbit for over ten years before it runs out of the fuel it uses to stay still. Most LEO satellites have a lifespan closer to five years. There are negatives and positives to this limitation. The positive is that a LEO satellite will de-orbit itself at the end of its life, or, in the event of a complete failure, will deorbit after a few years with no input needed. In GEO, reserve fuel must be kept until end-of-life, at which time the satellite burns its thrusters to put the satellite in a "graveyard orbit" where it is out of the way, and can't interfere with other satellites. Of course, the big negative to LEO is that five-year lifespan.