There are underlying physical limitations to spectral efficiency. The best modulation schemes in use today can hit about 60 bits per second per hz of bandwidth. Meanwhile, single fibers top out above 40 terabit. To duplicate the bandwidth of just a single fiber link with a radio downlink, you'd need over 600 ghz of available spectrum. That'd require advances pushing deep into the thz gap, but beyond that, the physics of propagation are so different between 6 ghz, 60 ghz, and 600 ghz that it's hard to conceive of a single system covering so much spectrum. And that's not even getting into issues with geometric limitations of radio beam widths.

While change is constant, this isn't an argument that any change you can imagine is physically realizable.

Fiber is here to stay for a very long time precisely because it starts out with such massive fundamental physical advantages. It's also a mistake to assume innovation only benefits one alternative. We're still learning how to make and run fiber cheaper too.

> 60 bits per second per hz of bandwidth

Between one transmitter and receiver. Starlink is designed to cheat Shannon by utilizing SDMA on a truly massive scale. Every antenna in the system, both on the satellites and on the ground stations, is a large phased-array antenna, capable of simultaneously emitting different signals on the same wavelength in different directions, and capable of discriminating between multiple different signals on the same wavelength that come from different directions.

This is how Starlink intends to push so much throughput over just ~50 GHz of bandwidth. Not by pushing more bits per hz per link, but by filling the sky with hundreds of simultaneously visible satellites, and having each of them to all use all of the available spectrum at the same time on every link between any of them and any of the ground stations.

Phased arrays aren't magic. The directivity of the beam scales logarithmically with array size / element count. The limitations are such that you should think of it as being able to send Nebraska one signal while you send Kansas a different one. You can't get much more specific than that.

Elon has been very explicit about starlink never being competitive vs telcos in urban areas.

And independent people have done the math too. The system geometry maxes out at less than a single fiber: http://www.mit.edu/~portillo/files/Comparison-LEO-IAC-2018-s...

Phased arrays are not magic, but they are better than you make them sound like. When the directivity of the beam scales logarithmically with the element count, it just means you need a whole lot of elements.

The beam sizes are in the FCC filings, and they will be able to fully re-use the frequencies for ground stations that are >20km apart. (Not with the same satellite, you want adjacent beam spots on the ground to come from different satellites, but with literally hundreds above you at any time, that's not a problem.) Availability of bandwidth does not really limit them until the satellite counts are up to really hilarious numbers.

The aggregate throughput numbers in the paper you linked are correct, for the initial FCC filing. Since then they have multiplied the size of the constellation by ~7 and the throughput per satellite by ~3 (for 2.0 satellites). (ISL's are still missing, though.)

For what it's worth, satellite is typically around 1-4 bits per second per Hz.

Context is important, though. Network performance advances are always relative. If satellite fully-congested download speed improves 10x while hardwire link performance improves 10x, satellite will still be perceived as "too slow" by some segment. Our experience thus far has been that advances in applications tend to consume the bandwidth available. Whether that trend continues remains to be seen. (To be fair, I have never saturated my bidirectional 1GBps fiber link and I don't foresee ever doing so except for very brief periods of time.)

> Encoding and compression software and hardware continue to improve.

But these would apply to both wired and wireless, no?

If encoding tech is specific to wireless then not really , just like improving cable medium , number of cable pairs etc will not improve wireless , better transmitter or receiver will not impact wired

What encoding technology is specific to wireless? (Hint: none)

You are thinking just software , Hardware tech is very different for wireless. The benefits from beam forming, MIMO or any of the advancements in 3G/4G/5G only benefit wireless. There has been similar technical improvements in wired which only benefit that medium, they keep cramming more signal on the same size cable every generation.

Even in software at L3-7 the encoding tech is probably same of course, however protocols L0-2 is very medium dependent. The loss , noise, signal strength patterns all change depending on the medium so does

Even at L4 and above UDP based stacks benefit in wireless compared TCP based transport, if your application can handle it of course. The next gen HTTP is getting designed around UDP in some part to reflect the changing way people access the internet , a lot more traffic is mobile and wireless today so designing around wireless makes sense.

Personally even as app developer who works only with L7(HTTP) and L6 (SDP) I have specifically developed non trivial optimizations, workflows and APIs for mobile clients which were not needed or used by web clients due to different connection parameters of wired/ WiFi as opposed to wireless 2G/3G .

So yes there is lot of difference and most people working at any layer cannot ignore the medium.

I'm very aware of MIMO and the other physical layer technologies. But the original topic was about how wireless could somehow progress faster in these metrics than physical mediums. Mediums like fiber have progress much more in the last decade by adding more wavelengths and different modulation technologies compared to space technologies. SpaceX is only using MIMO in the sense that they have a phased array antenna, but that's not the same as your typical wireless router at home with multiple antennas. The other issue is space is severely power limited, so they are not going to get really high modulation schemes that fiber or cable have enjoyed (256+QAM) for many years.

> Encoding and compression software

What does that have to do with bandwidth ?

Sending less data over starlink (which is not tcp/ip btw).

I’d you can make 10Mb suddenly only require 900k of actual raw bandwidth by say adding a fpga or something then you can increase throughout to the users.

Technology only improves compression ratios to a certain point, and we passed that point with lossless compression decades ago. We're approaching that point with lossy compression now.

Going beyond that point requires the laws of physics (more precisely, the laws of information theory) to change, and that's not going to happen.

Hopefully in 2020 most internet traffic is encrypted. If encrypted traffic is compressible you're doing it wrong.

It would help if it was compressed pre-encryption, by the server.

If that big compression rates where achievable why wouldn't we be already using them? They would be standard in any network.