Tuesday, March 29, 2016

Show Me the Gbps: When 1 + 1 > 2

Some airlines offer passengers connectivity for free, others make the passengers pay, and in the end most will offer a mix of free and pay services.  Will the airplane radio be able to meet the needs of the passenger?  What about the radio network?  How will that evolve over the next ten years?  This feasibility analysis takes a look at the US market to reveal the technical requirements for a satellite network to meet the demand from a "typical" large US airline and from four such large US airlines.

Both Ku-band satellites and Ka-band satellites can serve any foreseeable airline passenger connectivity market.  Because of concentrated demand, a single large US airline under the heaviest demand cannot be served from a single orbital slot by 2026.  It takes a family of orbital slots with satellites offering overlapping coverage to aggregate the spectrum needed at the busiest airports.  A family of satellites offers a robust and scalable solution to grow as demand grows, applying the newest technology incrementally along the way.


Passengers expect airlines to offer a connection to the Internet that gives them an experience at least equal to the mobile Internet.  Ideally it aspires to equal the good experience in a hotel or airport, but frankly can only dream of matching the home or office. This analysis starts from what is considered a good Internet experience of 150 kbps and associates that with typical take rates for successful products (free and pay).  The average data rate is modeled to increase 20% per year, to 929 kbps in 2026.

A typical airplane fleet from a large US airline is modeled at 500 airplanes in-service in 2016.  8% of the fleet may congregate at one hub, or about 40 airplanes in 2016.  Two fleet "growth rates" are modeled (low/high), with the low fleet growing at 3%, the high fleet starting out a bit busier and growing at 5% per year.

A four-airline model is built by simply multiplying the one airline fleet numbers by 4.5, except that only 6% of the composite fleets congregate at a given hub.

The analysis is purposefully skewed to the most passengers in the busiest national route structure using the data network to their heart's desire.  Other airlines will have much smaller fleets and operate over very different route structures. These techniques can be applied to any market.  The point is to see what it takes to solve the biggest problem (is it feasible?), and then everything else will be easier. 

The analysis is in combination of: 
  • Four take rates (10%, 20%, 30%, 50%)
  • Two airline fleet growth rates (3%, 5%)
  • Hub Concentration (8%, 6%)
  • Data rate growth rate (20%)
  • Streaming Rate Max (5x Average Rate)
  • Peak Rate from one airplane (3x Average Rate) 
The free scenarios create the greatest challenges, especially at highest (50%) take rate.

A basic building block for the models is the average data rate (Mbps) from one airplane across the four take rates and ten year span.  In this context, the units of airplanes are abstracted to about 160 passengers.  More or less passengers would scale the associated data rate per airplane by take rate.

In 2016, one airplane is well served with an average of 2 - 5 Mbps (passenger pays) to an average of 6 - 12 Mbps in a free model.

In 2026, an average of 30 Mbps (passenger pays) or 74 Mbps (free) is consumed by one airplane.

The summary data assumes some number of average airplanes.  The average airplane model can be used in any scenario to judge overall and regional satellite capacity requirements.

Three basic parameters are revealed for each scenario: 
  1. Continental Data Rate (how many Gbps for the fleet)
  2. Peak Data from One Airplane
  3. Peak MHz for One Region

Continental Data Rate 

Near-term continental data rate for four airlines of 26 Gbps is hardly a challenge for any HTS satellite being deployed today.  Growth to 266 Gbps can be met by one to a handful of HTS satellites.

Peak Data Rate from One Airplane

The average data rate to one airplane drives network capacity calculations.  An airplane terminal should accommodate a heavier than average usage, here modeled as three-times the average.
Maximum data rate to one airplane drives antenna instantaneous bandwidth, maximum symbol rate in the modem, and adequate network capacity.  Near-term peak rates of 36 Mbps can be met easily by any terminal, whereas rates over about 45 Mbps requires the most modern and capable equipment.

Peak MHz in One Region

The greatest challenge is serving the peak demand to a small, localized region.  The concentration of airplanes can be significant in the vicinity of a busy airport, especially considering gate-gate operation.  The past application of fixed 36 MHz wide-band transponders will be replaced with flexible bandwidth and power application of enormous resources using beam hopping.  

Both Ku-band and Ka-band transponders have spectral limits.  It may be possible to apply between 1,500 to 3,000 MHz from one satellite into one beam, but today it is more likely between 250 - 1000 MHz.  By 2026 no single satellite can serve even one large US airline under the heaviest concentrated demand.  The demand from four airlines in 2026 to serve the heaviest demand will take many satellites with overlapping coverage. 

For example, ten satellites each contributing 20 Gbps to the coverage and 1,500 MHz to the busy terminals offers 200 Gbps overall and 15,000 MHz in any region.  Alternatively, two satellites each contributing 100 Gbps to the coverage and 3,000 MHz the busy terminals offers 200 Gbps but only 6,000 MHz in any region.  Projections for the four-airline, heaviest take rate in 2026 are paced most by spectral capacity in the busy regions rather than data rate capacity in the overall coverage area.

Only a family of satellites with overlapping coverage can aggregate the spectrum needed at the very busiest airports.  

A family of satellites offers a robust and scalable solution to grow as demand grows, applying the newest technology incrementally along the way.   

The use of open-standards offers more options when assembling overlapping coverage or fail-operational capability, by leveraging a bigger marketplace across more than one satellite operator.

With multi-engine airplanes, the unrelated loss any single engine is minimized the more engines there are to start with.  Single-engine operations leave the least options.  The more engines you have, the more likely one will fail.  A balance is found for every market.  

The loss of passenger Internet access coupled with non-safety airline applications drives assurances for reliability where the unexpected in-orbit loss of a single satellite may cause the network to dim just a little, rather than going altogether dark.  Flexible satellites with beam forming and beam hopping give rise to a generic, open payload that can be built and stocked for launch-on-demand, or even by shuttling across flexible in-orbit spares. 

Both Ku-band satellites and Ka-band satellites can serve any foreseeable airline passenger connectivity market.

Users per Airplane

As a parametric aid, two candidate business models include free or pax-pay.  If the passenger pays for the service, the take rate is expected to be (optimistically) in the range of 10% - 15%.  In the service is free, take rate may be expected to be as high as 30% on average.

Taking a 150-160 seat average airplane, the lower bound of a successful pax-pay scenario is about 15 passengers or 10%.  An upper bound for the same service is a dreamt-of 30 passengers, or 20% take rate.  These are shown in red.  A passenger pay service should normally operate between the two red lines.

45 passengers (30%) will likely take a free product, with an upper bound of about 80 passengers (50%).  A service free to the passenger should normally operate between the two blue lines.

Average Data Rate per User

An average Internet access session is modeled at 150 kbps.  For the purpose of this analysis, 120-150 kbps is available on the forward channel and 30 kbps on the return channel (4:1).  Average implies some user behavior will be lighter and some will be heavier.

Estimating future behavior is never precise.  For this analysis, the average data rate for an Internet session grows at 20% per year, starting at 150 kbps.  This matches up with other models, notably the Cisco mobile forecast.

In 2026, an average Internet session is expected to be 929 kbps.

Data Rate from One Airplane

Knowing how many users on one airplane and the data rate per user allows a simple calculation of the average data rate from one airplane.  

A pay service starts between 2 and 5 Mbps in 2016 and ends needing between 14 and 28 Mbps in 2026.

A free service starts between 7 and 12 Mbps in 2016 and ends needing between 42 and 74 Mbps in 2026.

While the average data rate is nominal, the real world creates "corner" conditions.  To manage demand and fair-use, a streaming rate limit (and for download or any continuous heavy demand) is applied equal to five times the average data rate.  Given natural variations, the resultant peak data rate is modeled at three times the average data rate.

A pay service starts between 7 and 14 Mbps in 2016 and ends needing between 42 and 84 Mbps in 2026.

A free service starts between 20 and 36 Mbps in 2016 and ends needing between 125 and 223 Mbps in 2026.

Data rates above about 45 Mbps require modems with high symbol rates, antennas with high instantaneous bandwidth, and satellites that can power large swaths of spectrum.  Data rates over 100 Mbps is available with equipment available in 2016.  Equipment and networks are already planned and being deployed for receiving 250 MHz carriers that will support more than 223 Mbps.  Airplanes offering free services should already be provisioned for high data rates, whereas an airplane with passenger pay service only has until about 2022 before mandating high data rate terminals.

Capacity for Peak Usage over Coverage Area

A model of a "typical" airline is based on a rough average of four large US airline fleets totaling about 2,250 in-flight in 2016.  Two fleet models are created to represent typical fleet growth (Low) and by increasing share of the market (High).  The Low fleet starts with 500 airplanes in-flight (peak) and the High fleet starts at 550 airplanes.

Applying the four take rates to the two fleets gives rise to the boundary conditions for each business model.  The lower take rate combines with the low growth airline fleet; the higher take rate associates with the high growth fleet.  The feasibility analysis offers a range for each business model, where the passenger pays lies between the red borders and the free model between the blue borders.

The long-term capacity for the most demanding free model peaks at 67 Gbps; achievable from a single HTS satellite. 

A progressive capacity build-up can start in 2016 with a demand of up to about 3 Gbps for passenger-pays and about 7 Gbps for a free model. 

A four-airline model is applied to determine if there is any performance factor that emerges as a significant concern when scaling up to serve the majority of the US airline fleet.  In this context, the four-airline US model is scaled up by a factor of 4.5 times the one airline US model described above.  Where one airline (Low) starts with 500 airplanes, the four-airline (Low) model starts with 2,250 airplanes.

The long-term capacity for the four-airline, most demanding free model peaks at 223 Gbps; achievable from a single HTS satellite or a family of several HTS satellites.  

A progressive capacity build-up can start in 2016 with a demand of up to about 10 Gbps for passenger-pays to about 26 Gbps for a free model, all within reach of even one HTS satellite.

Capacity for Peak Usage in One Beam

Passenger connectivity services are expected to begin from the moment of boarding until the moment leaving the airplane and rejoining the airport terrestrial public networks.  This gate-gate service model intensifies the concentration of airplanes on the airport with those arriving and departing the terminal area.

Spot beam sizes are shrinking with larger (unfurlable) satellite apertures and also when comparing Ka band to Ku band.  Without knowing for sure, beam sizes are expected in the range of a few hundred miles to maybe as little as one hundred miles across.  An estimate of about 8% of the fleet from one airline is expected to concentrate close enough to be served in the same satellite beam.   The Low airline 2016 estimate is 40 airplanes in one beam.

Spectral efficiency converts data rate (bps) into required spectrum (Hz).  For the purpose of feasibility, spectral efficiency is set to 1.5 bps per Hz, or that 1.5 Mbps would require 1 MHz.  With this in mind, the spectral requirements for a single beam (region) can be estimated. 

Traditional wide beam transponders typically span 36 MHz.  Even the most conservative 2016 demands are well beyond one wide beam transponder.  Assuming four-color reuse provides for between 250 and 1,000 MHz in one beam, from one satellite slot.  Beam hopping may provide higher capacity, perhaps 1,5000 - 3,000 MHz in one "region" from one satellite slot.  For feasibility, it will be assumed that one satellite can deliver between 1,000 and 1,500 MHz to one region.

It Takes a Village

2016 free models may require more than one satellite slot, whereas all models may require more than one satellite slot by 2026. 

For example, in the 50%-H model in 2026, an estimate of about 3,500 MHz would probably take between three and four satellite slots. 

The four-airline model concentration is modeled at 6% of the four-airline fleet.  The Low model assumes 135 airplanes in one small region (one beam) in 2016.

The potential for a large number of airplanes operating simultaneously is slightly moderated than when looking at one airline in isolation, but the combination still represents a formidable challenge.  

The demands in 2016 are achievable from one to two satellite slots.  It may take more slots in the 2016 time-frame until HTS satellites with sufficient bandwidth are in-service.

The demands in 2026 are considerably higher.  The passenger pay model may need 5,325 Mhz, which could require 4 - 6 satellite slots.  The passenger free model may need as much as 14,200 MHz, which would could require 10 - 15 satellite slots.  

Concluding Remarks

The data rate to any given airplane is a unique function of that collection of passenger usage patterns.  The highest combination of take rate and usage can be supported by existing airplane satellite radios today assuming sufficient transponder power and spectrum.  However, even the lowest long-term projection demands performance from modems with high symbol rates and antennas with large instantaneous bandwidth.

Demand estimates out to 2026 project capacity well within the range of a single HTS satellite across all of the coverage area.   Most concerning is the need for multiple, overlapping, satellite coverage to aggregate sufficient capacity for the busiest regions.   While one US airline might be served in 2026 from less than five satellite slots, four US airlines may require ten or more satellite slots.  At least two satellite operators each have access to more than ten slots with US coverage.

Serving the US airline passenger market is feasible with: 

  • Multiple overlapping slots
    • One or more satellite operators working together
      • Open standards (antenna, IP network)
      • Multiple modems (teleport network compatibility)
  • HTS satellites (Ku or Ka)
    • Large, unfurling apertures (5m+)
    • Generic beam-forming arrays (hundreds of overlapping beams)
    • Beam hopping with 1+ GHz in a given beam
  • Mobile antennas and modems supporting high symbol rates (100-250 MSps)

Other terminal areas around the globe may present concentration of airplanes or concentration of larger airplanes that surpasses service from a single satellite slot.  The methods described are applicable elsewhere and likely far less demanding.

Keep in mind this is a relatively simple model to judge feasibility.  Actual forecasting using take rates adjusted for each flight and for airplane size and and load factor would add a necessary degree of fidelity.  Actual growth rates are subject to non-linear factors not modeled here.   The good news is everything looks technically possible and consistent with the going trends.

The next real question is how much will it cost and will that match the product visions - especially the free model.  With this analysis as a baseline, cost will be the subject of an upcoming blog.

Stay tuned,

Peter Lemme
peter @ satcom.guru

Copyright 2016
All Rights Reserved

Attachment 1 - Tabular Data

Passenger Pays

Free to Passenger 

Attachment 2 - Estimating Airplanes in Coverage Area

A feasibility analysis focused around serving peak demand depends on estimating the peak number of users in a coverage area.  As a first-order, the marketplace is simplified into a generic airplane of about 160 passengers with an average take rate assumed at peak time (daytime) across most dense number of airplanes.  Gate-Gate operations brings airplanes on the ground into the mix.

Rather than getting bogged down in airline schedules, I chose to use flightradar24.com to take a snapshot of four large airlines traffic on a typical weekday.  The number of airplanes doesn't change dramatically over much of the day - the peak just moves a bit from the east to the west.  Peak demand is really not just one time and one place.  It is most of the time and moves around.

The following are snapshots for AAL, DAL, SWA, and UAL.  The total inflight was about 2250 airplanes.  I took 500 airplanes as the "generic" airline, and 2250 for the "four airline" totals (4.5x).

For the feasibility analysis, I established a low and high fleet profile.  Starting for the low airline model at 500 airplanes and growing at a 3% annual rate.  I looked at models from Boeing and from the FAA and found 3% is the highest projected amount for all US airlines together.  Some airlines grow market share along with gross passenger traffic trends.  With that in mind, I set the high airline growth rate to 5% and started it with 550 airplanes (busier fleet).

 American Airlines - 704 airplanes in coverage area


Delta Air Lines - 524 airplanes in coverage area

 Southwest Airlines - 390 airplanes in coverage area

 United Airlines - 525 airplanes in coverage area

Peak Airplanes in a Region

Each airline has a few airports that are busiest, usually at a hub with connecting traffic.  The size of a spot beam is shrinking.   A Ku HTS spot beam might cover several states, whereas a Ka HTS spot beam might not fully cover one state.  With unfurling antennas to support beam hopping, the expectation is for ever-shrinking spot beams.  The spot beam can shrink down to maybe one hundred miles across.  

The Dallas region is the most dense traffic for any one region, from one airline (AAL).  This snapshot is across about 400 miles.  

Most terminal regions peaked at around 40 airplanes from one airline.  

Without a firm grasp on beam size I settled on 8% of one airline as a peak concentration (or 40 airplanes in the peak beam).  Given the likelihood that at most two of the four airlines would share one hub, I assumed 8% from two the airlines, and 4% from the other two, or 6% of the four airline fleet as a peak concentration. (Thanks to Tim Farrar for that suggestion).

Dallas appears to be a especially challenging, to which I have left as an exception that could be nearly twice the projected MHz if for American Airlines.

American Airlines traffic near Dallas - 75 airplanes


  1. Nice work. I agree that diversity of access will be far more scaleable than trying to do this with a mega-sat. Have you looked at future Air-to-Ground as a solution to grow capacity with demand in the most dense regions? Seems like a good high-bandwidth complement to satellites, which are better at coverage.

    1. From a technical standpoint, ATG is another option to add capacity, but it will suffer as well. For example, the 14-14.5 GHz allocation offers 500 MHz x 2 polarities or about 1,000 MHz into a given cell site. I am projecting needing over 14,000 MHz into one small region, which would drive at least 14 "cells", to which I am skeptical. Practically, carrying a second comm radio system is not advantageous if you can do it all with one radio system. However, the future will bring new options, both ATG and satcom, so my conclusions are temporal and worthy of revisiting as we go down the road.