mPOWER

High throughput satellites (HTS), phased arrays, software defined antennas and LEO constellations are not quite popular subjects like Mars missions or Big F*** Rockets. Rockets are far more popular than satellites. But while images of lunar landings, Pluto and Saturn rings capture popular imagination, true value of space lies in obscure satellites that provide location, observation and communication services.

In the communication domain we are witnessing a "perfect storm" which will reduce cost of space communication for two orders of magnitude in two decades. It started with launch of the first high throughput satellite (IPSTAR) in 2004, Intelsat EpicNG series, three generations of Viasat satellites, O3b, OneWeb, and its latest newcomer, SES mPOWER constellation.

When I calculated expected cost of per megabit, I was amazed how SES got everything right with this constellation, which makes it a likely big winner in the future. In this post I will try to articulate reasons how mPOWER really beats all competition by a wide margin.

A bit of history

SES has been traditionally slow adopter of HTS. Actually their first HTS satellite (SES-12) is yet to be launched next year. Which seems a bit slow compared to Viasat and Intelsat, as the major HTS competitors. But it was an early investor (and now sole owner) of 03b MEO constellation, that can be viewed as low-scale HTS solution, with its six beams per satellite and 700km beam spot size. From historic perspective, it seems obvious that SES would try to maximize the throughput and value of its MEO network.

Assumptions

SES did not provide many details about their next generation constellation, so I will provide assumptions here.

First, initial SES order was 7 satellites, which is an odd number for a constellation. Based on their MEO location (some 8000km above equator), it is logical to assume that they aim for six operational satellites (spaced 60 degrees apart) and one in-orbit spare.

SES has (in their FCC application) requested that their constellation should allow up to 44 satellites in equatorial orbit (similar to 1st generation 03b constellation, named O3bN) and up to sixteen satellites in two near-polar orbits (O3bI). In their application, they have foreseen using a total of 2.5GHz bandwidth for TX and 2.5GHz for RX in the same bands as existing O3b constellation.

Based on published information, I assume that 10Tbps capacity will be spread among six satellites (using one as in-space spare), and that 30000 beams will be provided by six operational spacecraft (5000 per satellite). 

Throughput

On the merit of throughput, it seems that mPOWER satellites will be on the same level as the Viasat-3 GEO satellites, providing well over 1 terabit per second each. At the same time, SES plans to spend less than half per satellite than Viasat (around 200 vs 400 million USD).  All this throughput will be managed through unparalleled frequency reuse, over 30000 narrow beams (5000 per satellite). Original O3b had already quite narrow 700km beams, but each satellite could carries just ten downlink beams.

Since mPOWER is a successor to 03b constellation, it is expected that they will retain the same channel width per beam (2x216MHz), and possibly the same encoding scheme using roughly 1.9bps/Hz, while retaining compatibility of ground segment installations, and giving upper limit  It could also provide upper limit of 10Gbps of capacity for a single spot on the Earth. Individual gateway capacity would also be limited to 10Gbps, unless some additional technology like laser communication terminals would be used.

Latency

MEO orbit seems to be in the sweet spot for the TCP/IP round trip latency issue, as real life measurements from FCC report 10-50ms delay for ground based services and whopping 600ms for GEO satellites. 03b has reported latencies of 140ms, of which around 60ms can be attributed round trip travel between ground stations and satellite. It might seem much worse than 8-20ms delay from LEO satellites, but here MEO provides additional advantage that it may effectively reduced number of hops for IP packets to reach their destination. If we measure well-connected sites located on other continents, round trip time is 150-200ms using fiber links. For the local sites (less than 3000km away) fiber connectivity can go as low as 30ms. So MEO introduces noticeable delay into the network, but brings it into acceptable range for regular browsing experience. It also reduces any voice related delays for long distance calls.

Ground segment infrastructure

Now this is a segment where MEO really beats LEO constellations. It greatly reduces needed number of gateways compared to LEO and also provides greater flexibility where gateways will be built. LEO constellations either have to build large number of teleports close to their customers or design their satellites to perform as smart network switches/routers using inter-satellite links. Important reason why GEO sats remain relevant for 15 or more years in space is, that by design they are built to be "stupid" analog repeaters. Operators try to move as much of the intelligence and processing requirements to ground segment, where equipment can be replaced and improved as soon as it is available. That approach, for example, enabled bits/Hz measure to four times without making a single change on the satellite.

So MEO constellation does not need to provide inter-satellite connectivity, which greatly reduces system complexity. SES, using their decades long experience in operating communication satellites, is obviously well aware of this issue. It does not mean that such capability will not be provided, but it is not necessary or crucial for MEO or GEO constellations.

Another (possible) advantage of mPOWER is that it will use the same frequency band as 03b. Which likely means that existing 03b customers will not need to change anything on the hardware side to use the new system. That will help in customer retention, as the existing customer base will (inevitably) be moved from older 03b satellites to new constellation.

Light me up?

Beside technical risk, mPOWER constellation as described has one Achilles heel: large number of required gateways. If gateways would use the same frequency band as remote sites, their individual throughput would require at least a thousand 10Gbps gateways dispersed over wide geographical area in fully utilized system. That represents huge financial investment and logistical hurdle. 

This problem is not unique for mPOWER, but for any ultra-HTS system. Viasat built 45 gateways for their Viasat-2, and plans to build hundreds of geographically dispersed gateways for Viasat-3. Could such a model be expanded for world-wide network with ten times more capacity? Also note that full full constellation would have 60 satellites, or 100Tbps capacity. With ten thousand stations? 

So it would be ideal to use some other transport mechanism for communication between gateways and satellites. Since no higher frequency bands (such as V-band) were not requested by SES in the FCC application, we can assume that there mPOWER will use laser based communication. Such approach would enable much lower number of gateways (less than 100 worldwide), and would be scale-able to full constellation capacity of 100Mbps. Such links have been demonstrated up to 1.6Gbps capacity from GEO and tested on EDRS satellites. And interesting enough, ESA and DLR sponsored development of OPTEL-µ, 2Gbps laser terminal for small satellites intended for commercialization. And who supports development and integration of ground terminal? SES Techcom. This piece of equipment is scheduled to launch in early 2018 and be tested on ISS. 

Now, 2Gbps does not seem like much for a satellite handling 1.6Tbps. That is a bandwidth if you use single color laser (essentially one frequency), so there is definitely a space for much larger throughput possible. Although given terminal is marketed as lightweight (around 8kg and 45W), scaling it to Tbps throughput until 2021 seems kind of far fetched. Furthermore, gateways would be build gradually around the world, as demand picks up in different areas. But long term, build-up costs of huge number of gateways needed for HTS shall have similar issues for all providers, and it would not be surprising to see some kind of collocation agreements between major HTS systems in the future in order to reduce costs. 

Kessler Syndrome

MEO constellations would operate in space that is the least polluted by debris. LEO constellations would be located at the worst possible place where they would be affected by space debris from early space age. On the other hand, MEO is far bigger and was not popular orbit. Also number of objects in MEO would always be at least an order of magnitude smaller than LEO.

The same issue applies to GEO, where practically all satellites reside in the same arc and could be affected by the same issue. There are over 1550 objects in GEO and GSO, with total mass of at least 2000t. And we launch 50t of additional satellites to GEO each year. Some of these objects are there for over 50 years, and there events of sudden loss of function.

Therefore MEO constellations should be far more resilient to space debris collisions than both LEO and GEO satellites. Therefore they should have lower insurance requirements.

Number of units

This is another advantage of MEO. GEO is ideal for needing just a three satellites to cover all the Earth surface (with the exception of polar areas), but we can see that mPower is designed to cover similar area with just six satellites. Unlike LEO, where it is necessary to build whole new factories and manufacturing processes (take original Iridium, or OneWeb as examples), MEO satellites can use the same manufacturing and design process used for GEO satellites. mPower itself  will be based on the existing Boeing 702 satellite bus, which decreases operational risks and reduces investment into new manufacturing facilities.

Secret sauce

So if MEO satellites are that good, and have so many advantages compared to both LEO and GEO satellites, why there were not used before? The answer lies in the mechanics of steerable antennas. Simplz put, mechanically steered antennas are expensive to make and maintain. But the advent of phased arrays, software defined radios (and antennas) finally enable flat panel, software controlled beam shaping both on the space and ground segments. There is no way SES could pack 5000 small mechanical antennas on a single satellite. But with electronic/software defined steering, using 30x30 cm phased array antenna, you could pack 10 beams on one square meter. And high power GEOsat with 15kW electrical power has roughly 50m2 of solar panels. Who most often are oriented away from the Earth. There you have it - space for 500 beams just like that. But, I noticed that mPower would operate from 17.7GHz. That is less than 2cm wavelength and thus can be adjusted to work with 10x10cm phased array antenna. Which gives us the needed 5000 beams per satellite.

Another advantage of MEO compared to LEO is that beams are highly steerable. MEO constellation satellites would spent most of the time over the oceans (70%) where their capacity would not be used. But each MEO satellite can reach over 10% of Earth Surface directly (and possibly close to 20% depending on possible beam steering angles supported). Which means that each satellite would be
able to reach much larger set of customers at any given point in time.  So their beam utilization could easily reach over 50% just by connecting continental customers, ignoring maritime, air traffic and island connectivity.

The same secret sauce (phased array antennas) would work to reduce wear and tear on the ground segment, which is a real reason why GEO prevailed to LEO/MEO applications to this day.  But there is an additional benefit - such antenna can also easily switch between MEO and GEO satellite, providing backup capability for ground (or sea or air) infrastructure that is out of reach equatorial MEO satellites. Software controlled antennas could possibly support wider frequency ranges without changes of the ground equipment, which provide greater resilience.

And what about SpaceX?

SpaceX constellation has been considered as a possible game-changer, but to make it work, SpaceX has to design huge new rocket (BFR), provide unparalleled mass production of the satellites (thousands per year), provide advanced cross satellite (laser?) communication links and all that at price point comparable to SES? Their cost estimate is 10 billion USD, ten times more than SES plan. And it would be realized several years AFTER OneWeb and mPower and Viasat-3 are launched and in operation. To make them viable, they would need to target bulk cost of 1USD per Mbps per month. And with shorter life span of 7 years. So their network would need to be sized for order of magnitude higher throughput (100Tbps) than SES mPower. How much is that? Well, total internet traffic per month was estimated at 122EB (exabytes). That is around a billion terabits per month. Or 376Tbps on average. So SpaceX constellation would provide capacity for a quarter of a whole Internet traffic today. Good luck with that!

All things put together

Of all announced HTS LEO/MEO projects, mPOWER has the lowest risk (by using a proven satellite bus architecture) and the lowest price/cost structure of 6USD per Mbps per month. SES plans to invest one third of the money and has projected cost per Mbps at one third of OneWeb and Viasat-3 projects. Plus initial existing 03b customer base should be quite easy to migrate to new service as some of these satellites will reach end of projected life by 2023. The frequency bands will be inherited from O3b and customers should not require significant upgrades. Therefore SES has a clear winning stack of cards in their hands.

Of course, we have not yet seen announcements regarding Viasat MEO, next generation OneWeb or SpaceX Starlink. But the bar has been set quite low, and competitors will need to provide something even better.