Sunday, August 12, 2018

Satellite constellation license auctions

Broadband satellite constellations in non-geostationary orbits (NGSO) will share frequency bands. They will interfere with each other from time to time. How should such conflicts be resolved? FCC rules encourage constellations to coordinate their operations, treating the shared NGSO bands as something like an FCC-supervised commons. However, spectrum regulators have increasingly used auctions to assign radio operating rights (cf. cellular licenses), and largely left it to the market to solve coordination problems. Could spectrum auctions be used for NGSO operation?

(Update October 1, 2018: A subsequent post explores treating satellite constellation spectrum as a common pool resource.)


Absent prevention, NGSO constellations will sometimes interfere with each other. For example, two communication systems will suffer reductions in data throughput when transmit paths between their respective ground terminals and satellites overlap (i.e., when terminals are close together, and satellites are in the same part of the sky as seen from the ground) and both are using the same frequency channel. In economic jargon (see the Postscript), the right to transmit is a partly rival good.

However, link-path/frequency coincidences will only happen occasionally. Most of the time, paths are not aligned since there are many satellites that can serve the same terminal (handfuls to dozens), and terminals are some distance from each other (i.e., not co-located). Also, the chance of both constellations choosing the same channel for an overlapping link is small since there is so much bandwidth available: 2 GHz space-to-Earth and 2.5 GHz Earth-to-space in Ka-band, for example; more in V-band.

Susan Tonkin is figuring out how often this happens, as part of a risk assessment study of NGSO coexistence; the results will be presented at TPRC in September 2018. Mean throughput loss without interference, mainly caused by rain outage, varies widely depending on location and system design, but is typically less than 10%. This increases by around five percentage points up to 15% when an interfering constellation is added.

Broadly speaking, one can resolve contention through rules created by a regulator, or in the market. (That’s an over-simplification, of course, since markets need regulation to function, and regulation often delegates implementation and enforcement details to commercial operators.) This post will explore market mechanisms as an alternative to the regulator-driven status quo.

Status quo: Management by regulation

There are currently two models for the coordination of inter-NGSO satellite interference.

Under the ITU regime, the operator with lower ITU priority (determined by the date on which an application is filed) has to avoid harmful interference to a network with higher priority, unless a prior coordination agreement has been reached. (Technically, the operator with the later filing date has to initiate coordination with operators with earlier filings, cf. ITU Radio Regulations Articles 9.12 and 9.53.)

The FCC rejected this approach for NGSO operation (see §50 in FCC-17-122 for its reasoning). It chose “band splitting” instead: if coordination isn’t achieved, and interference exceeds a specified level, interfering operators must split the band between them (47 CFR § 25.261). This is a strong incentive to coordinate, since our TPRC work-in-progress suggests that the FCC’s interference criterion will often force band-splitting; if the band is split between two operators, each suffers 50% throughput loss.

However, neither of these regulatory methods takes any account of the amounts of capacity actually needed – let alone the value placed on that capacity – by different operators. It’s either “first come, first served” (ITU), or “divide by the number of interferers” (FCC), both rather crude ways to structure coexistence.

Licenses acquired at auction are a way to take into account the differing values that operators place on their licenses. Before I outline a constellation license auction approach, I’ll explore the menu of radio operating rights.

Options for defining radio operating rights

Let’s posit that some kind of license regime will help prevent and resolve interference between NGSO constellations. Successful precedents include point-to-point microwave frequency coordinators, mobile operators negotiating interference at cell boundaries, and the agreement between LightSquared and Inmarsat to partition the MSS band.

An unlicensed regime where anyone using rule-compliant equipment can operate (such as the 2.4 and 5 GHz bands where we find Wi-Fi, Bluetooth, etc.) doesn’t seem to be a good fit. Satellite operators serve vast areas, unlike the small hotspots in unlicensed. Huge capital investments are required to launch a service, so that operators will want some assurances about protection from interference. Since there is no limit on the number of participants, negotiations involving unlicensed interests can be unwieldy; examples include the arguments between Bluetooth and Wi-Fi back in the 2000s, and the ongoing dispute between Globalstar and various 2.4 GHz unlicensed stakeholders. In contrast, there are only a few, easily identifiable NGSO players because of the scale of investment required.

The NGSO satellite case is more tricky than Part 101 frequency coordination, inter-cellco arrangements or the LightSquared/Inmarsat L-band carve-up. NGSO operators will have asymmetrical interests unlike microwave link operators or cellular companies that use similar technologies under similar business models. These include very different orbit designs, different business models, and large differences in uplink transmit power. Inter-constellation interference could occur anywhere the US, rather than being localized; and finally, there will likely be more than two parties.

One possibility is exclusive licenses such as those used for terrestrial cellular or broadcast services (to be precise: exclusive, geographic area, frequency block licenses). By analogy to cellular licenses, one could divide an NGSO band into a predetermined number of blocks, and then auction nationwide (or maybe regional) licenses to operate exclusively in that block. Exclusive frequency blocks make little sense to me, though:

First, co-channel interference will be pretty rare, so it’s overkill: it solves something that isn’t much of a problem. Two NGSO constellations serving customers in the same place using the same channel don’t always (or even often) interfere with each other – unlike cellular or broadcast, where overlapping licenses would cause continuous interference.

Second, spectrum utilization, defined as the fraction of possible link-path/frequency combinations used for communications, will be low in exclusively-owned blocks: Susan Tonkin’s system modeling suggests that that most link-path/frequency combinations will be unused most of the time. This will be the case even when multiple operators share the same frequencies. Spectrum efficiency, defined as percentage utilization of time-frequency-path resources, will be very low. In principle, licensees could unwind exclusivity and improve efficiency by subsequent inter-licensee agreements to operate on a frequency-shared basis; but why create an inefficient regime only to spend resources to deconstruct it?

Third, an exclusive license regime requires upfront decisions about the number of frequency blocks that a band should be divided into. Nobody knows how many of the more than a dozen NGSO constellations that have applied to the FCC will actually be built and deployed; it probably won’t be all or even most of them (cf. NSR’s prediction that two out of the five NGSO constellations it analyzed do not have a viable business case). The number will be probably be something between two and ten. The FCC determines who may operate on technical grounds; it doesn’t take a view on who’s most serious, or likely to succeed. The application window is open to all comers who pay the processing fee. In practice it’s therefore impossible to guess before the fact how many sub-band frequency blocks there should be.

Two existing approaches that are neither exclusive licenses nor unlicensed don’t seem to fit the bill either:

Frequency coordination, such as that between point-point microwave operators under Part 101 rules, prevents local interference problems between links that don’t move or change over time – but for NGSO, links are constantly changing as satellites can move across the sky and disappear, in some cases in a matter of minutes.

Light licensing in the E-band (70/80/90 GHz) leverages the very narrow pencil beams in the high millimeter wave bands. Once service operators have obtained a non-exclusive nationwide license, they can register specific point-to-point links in a database and get first-in-time protection. Just like Part 101, though, and unlike NGSO service, the links are fixed.

A market alternative: NGSO license auctions

To recap: I submit that one needs a license regime to avoid interference between NGSO constellations; that license assignment should consider the differing values placed on spectrum bandwidth by operators; and that neither exclusive geo area licenses, existing non-exclusive licenses, nor unlicensed fit the bill. This suggests assigning licenses by auction. But what kind of license, and how should it be auctioned?

What kind of license

Given that interference is relatively unlikely, I’ll assume some kind of band sharing. Two possibilities immediately come to mind: licenses that confer interference priority rights, or that define the amount of bandwidth an operator gets if band splitting occurs.

Under the first possibility, the license confers a priority ranking. If there’s a conflict, the operator with the higher priority wins; the loser might have to stop operation for the duration of the interference event. The priorities are similar to ITU ones, but in this approach they will be assigned according to the value licensees place on operation in an auction, not by who happened to file first.

In the second possibility, the license confers a fraction between zero and one, with the total of the issued fractions summing to one. This fraction can be used in various ways. First, it could be used to resolve priority disputes, and is thus a generalization of the previous possibility. Second, it could be used to determine the fraction of the band an operator gets under band splitting. For example, let’s say there are three licensees and their assigned fractions are A: 0.6, B: 0.3 and C: 0.1 (A+B+C=1.0). Consider the case where A and C interfere. In the current FCC band splitting regime, each would get 50% of the band. In this new arrangement, A would get 86% (0.6/[0.6+0.1] = 0.6/0.7), and C would get 14% (0.1/[0.6+0.1] = 0.1/0.7). This band split reflects the relative value the licensees placed on operation.

I’m assuming here that the rights we’re talking about here are part of the authorization a company needs to operate a satellite system. In other words: if a company doesn’t win a license at auction, it can’t operate. (Satellite licensing can get pretty complicated; check out Jordan Regenie’s handy primer on commercial satellite licensing for details.) When I say “license” in this piece, I’m referring loosely to either a license or market access for space stations, depending on whether the company falls under US jurisdiction for these purposes or not; and ground station licenses.)

In the alternative, the interference priority rights I’m describing could be ancillary to the license. A company would obtain their license (or market access) under the current NGSO “processing round” mechanism (§ 25.157), but they would not be entitled to any protection against interference, and would not be allowed to interfere with another licensee; the interference rights would be obtained separately in an auction.

How to auction

For the purposes of illustration – I’m no auction expert! – here are three auction designs that would let market determine differential license values. (I’ve omitted details about how to handle squirrely eventualities.)

Ascending clock auction.

Divide the band into N small slivers, where N is much larger than the number of possible bidders. N is chosen to approximate the continuum in a tractable, discrete way. For example, N could be 500; large enough that the regulator doesn’t need to guess how many will ultimately deploy.

Start with a low opening price per sliver; one can think of it as the reserve price.

In each round, parties can bid for some number of slivers at the current price. If the round is oversubscribed, i.e. there are bids on more than N slivers, one moves to the next round. The price increases in each round until the number of slivers bid for is equal or less than N. The fractions are calculated in the final round as each successful bidder’s ask as a fraction of the total demand.

For example, let’s say the opening price is $1/sliver, with 500 slivers. Player A, B, and C bid for 300, 200 and 150 slivers respectively, for a total demand of 650 slivers. Since that’s 150 more than the 500 that’s available, the price increases to $4.

In the second round, both A and B bid for 250 slivers at $4 ($1,000 each), and C bids for 100 slivers ($400). That’s a demand for 600 slivers, still more than the 500 available, so we go to the next round, at $8/sliver.

In the third round A, B and C bid for 250, 150, and 100 slivers ($2,000, $1,200, and $800 respectively). Since the total demand is for 500 slivers, the auction closes.

The awarded fractions for A, B and C are 0.5 (=250/500), 0.3 and 0.2, and they pay the amount bid in the final round: $2,000, $1,200, and $800 respectively.

Dutch auction (descending clock auction).

Start with N slivers as in the previous example. In this case, though, we start with a very high (rather than very low) opening price per sliver, and the price decreases in steps in successive rounds.

Any bidder can bid for any number of slivers in any round; they’re assigned the slivers as they bid. The auction ends when all the slivers have been purchased. If in the final round there are more bids than available slivers, the slivers are assigned pro rata to bidders in that round.

For example, let’s say that there are again 500 slivers available, and the opening price is $10/sliver. That’s too high, and nobody bids.

In round 2 the price drops to $8, there are still slivers available. Player A bids for 50, B for 40, and C doesn’t bid. The bidders get the slivers they asked for at $8; 410 (=500-50-40) slivers are left

In round 3, at $6, the bids are A:75, B:50 and C:40 slivers, and they’re awarded this ask at $6. Now 245 (=410-75-50-40) slivers are left.

In round 4, the price drops to $4, and the bids are A:125, B:60, C:60, for a total of 245 slivers, exactly what was left over at the end of round 3. This concludes the auction.

In the end, A gets 250 (=0+50+75+125) slivers, B gets 150 and C gets 100; their fractions are thus A:0.5 (=250/500), B:0.3 and C:0.2. The amounts paid over the course of the auction were A:$1,350, B:$860, C:$480.

First-price sealed-bid auction (FPSBA).

In this case, a single-round auction, all participants simultaneously submit sealed bids. Each gets a band fraction proportional to the price paid. For example, let’s say there are three participants who bid as follows: A: $2,000, B: $1,200, C: $800. The total amount bid is $4,000. Their assigned fractions are A: 0.5 (=2,000/4,000), B: 0.3, C: 0.2.

There is of course a multitude of possible auction mechanisms, and there may be better ones than I’ve suggested. I prefer clock auctions to FPSBA, since multiple rounds allow for preferences to be revealed, and for players to drop in and out of the auction. If one assumes that the auction is for operating licenses rather than just for interference protection, then market forces rather than mere application to the FCC determines how many parties are licensed.

The ascending and descending clock auctions differ in the uniformity of slice prices: all the slices are sold for the same price in the ascending auction, but license may have paid different prices for different slices in the Dutch auction

(My thanks to Susan Tonkin and Mark Bykowsky for their help here, and thinking about the problem in general. Susan suggested Dutch auction mechanism, and Mark pointed out that the NGSO licensing scenario resembles the problem that new firms have in selling a finite number of shares in an initial public offering where the objective of the firm is to maximize the amount of capital it obtains from selling its shares.)


Once an auction has assigned differential rights, they could be used for purposes other than determining interference priority and/or unequal band splits. One of the striking characteristics of NGSO operation, compared to most other service allocations (including geostationary satellites, cellular, broadcast, and point-to-point microwave), is the huge differences between different constellations. The numbers range from three to many thousands of satellites; orbit altitudes range from 340 km to more than 43,000 km; orbit inclinations include equatorial, polar and highly elliptical. This leads to asymmetries in vulnerability to interference. Differential rights could provide a basis for operators to negotiate accommodations that would be much more difficult to agree if everyone held generic, nominally equal rights. An operator that was particularly vulnerable would have the incentive to bid more for greater interference protection priority.


I’ve argued that licenses can be useful to manage coexistence between NGSO constellations; that license assignment should consider the differing values placed on spectrum bandwidth by operators; and that neither exclusive geo area licenses, existing non-exclusive licenses, nor unlicensed fit the bill. I’ve proposed differential non-exclusive licenses where parties can all operate concurrently in same band across a country but with different levels of interference protection rights, and I’ve offered three auction mechanisms by which such licenses can be assigned.

Update, 20 September 2018

It's a commonplace that there's a statutory prohibition on auctioning spectrum for international space stations. It's instructive, however, to look at the language of the ORBIT Act (3 Pub. L. 106-180, 114 Stat. 48), in 47 U.S.C. § 765f:

“Notwithstanding any other provision of law, the Commission shall not have the authority to assign by competitive bidding orbital locations or spectrum used for the provision of international or global satellite communications services.”

As I read it, the prohibition is only on assigning orbital locations or spectrum, not something else like (say) interference protection rights among multiple NGSO licensees all authorized to use the same band.

It also doesn't help that “international or global satellite communications services” doesn't seem to be a defined term. One could argue that an ISP (internet service provider) connecting U.S. user traffic to the internet at an IXP (inter-exchange point) in the United States isn’t offering international communications.

Postscript: Club goods and common-pool resources

In the jargon of public finance, goods that are strictly rival (aka rivalrous; use by one party can preclude the use by another) and also excludable (one can prevent someone who has not paid for the good from using it) are classified as private goods. They’re typically traded in a market; think food, clothing, cars, real estate, etc. So-called public goods like environmental protection or national defense are both nonrival (use by one person in no way limits its use by another) and non-excludable, and are usually provided by the state.

There are also two other permutations. Rival goods that are not easily excludable (like oil reserves and fisheries) are sometimes called common-pool resources, and nonrival but excludable ones (like athletic clubs or movie theaters) are called impure public goods or club goods. The categories aren’t strict – the degree of rivalry and excludability of various goods is a continuum. Club goods like golf courses, for example, are nonrival if usage is low, but become rival when they become congested.

NGSO transmit rights are partly rival goods since one operator can sometimes impede another operator’s full use of its rights through radio interference. The rights are excludable, since the FCC bars anyone without a suitable authorization from operating. They can thus be analyzed using club good theory (see e.g. Sandler’s 2013 review article), which enable one to optimize the number of club members and the level of membership fees, given members’ preferences (their marginal utilities, in the jargon) and the degree of congestion.

Transposed to the NGSO satellite case, one can imagine optimizing the number of licensees and associated spectrum fees given a model of operators’ preferences, and congestion calculated in terms of interference risk. This would be a significant departure from the status quo in the US, where the FCC does not limit the number of NGSO operators in a band, and fees are fixed.

The disadvantage of using club good theory is that it’s built on the usual heroic assumptions of neoclassical economics, and determines club rules (e.g. prices and membership numbers) up-front, based on more or less wild guesses about economic utilities.

Another way to treat partly rival goods that aren’t fully excludable is as common pool resources, aka CPRs). A few spectrum scholars, notably Martin Weiss, have tried to treat unlicensed bands as common pool resources; as far as I know, no one has analyzed NGSO bands in this way yet. This approach, an alternative to auction-assigned differential rights, would shift the regulatory regime from an FCC-supervised commons to a stakeholder-managed CPR. It’s worth trying, since the analogy to other CPRs is plausible. For example, the right to use an NGSO band where any operator with a license or market access can operate is only partly exclusive, and resembles an oil-field where many properties lie over a reservoir, giving any owner the right to pump. A first step would be to analyze NGSO coexistence in terms of Elinor Ostrom’s eight design principles which are prerequisites for a stable CPR arrangement, and seeing whether her strategies for adaptive governance are applicable (see Wikipedia); stay tuned for more.

No comments: