Thursday, March 14, 2013

Using an auction to decide the number of 3.5 GHz spectrum access administrators

The FCC faces a choice of whether to authorize one database administrator or many to run the spectrum access system (SAS) that will manage small cell operation in the 3.5 GHz band. This resembles the choice between an exclusive-use licensing or unlicensed regime. The FCC could use an auction to let the market decide by using a simplified version of the 2008 Bykowsky, Olson and Sharkey proposal.


The FCC’s notice of proposed rule-making (NPRM) on the 3550-3650 MHz band  (Proceeding 12-354) proposes that small cell operation be managed by a spectrum access system (SAS) that uses a dynamic database, along the lines proposed by the PCAST report (pdf), and inspired by the database systems authorized for the TV white spaces.

Under the NPRM, “the SAS would authorize existing licensees as GAA users in the larger, combined band, and would authorize higher power levels in less congested areas, provided there is no risk of harmful interference to Incumbent Access or Priority Access operations.”

Once such a system is in place, it could do many more cool things, at least in principle. For example, Vanu Bose suggested at the FCC Small Cell Workshop on March 13, 2013 that “Sensing data from mobiles and infrastructure in the local area can be used to adjust power levels appropriately for each area” (pdf, slide 4). Used in this way, the spectrum access system could not only coordinate small cell devices to limit their interference into primary users, but could also manage interference among these devices by limiting band access when too many want to operate. Doing this requires that the SAS database orchestrates collecting data and sending power level adjustments to end user devices at very short intervals – something that cellular systems do as a matter of course.


However, a cellular system has one master database coordinating all the devices. I find it hard to imagine that (say) ten independent SAS databases (the number in play in the TV white spaces) could coordinate their instructions to devices at the required 100 millisecond refresh rate. It’s even harder to see how one could make real-time spectrum access decisions for the small cell clients of ten independent SAS databases; each database administrator would naturally want to give priority to its own clients.

The FCC therefore faces a choice: should it allow only one SAS database administrator, or many?  This question was raised by workshop moderator (link TK once FCC uploads the video), and the answer was, “Let the market decide.” The panelists probably mean that the FCC should authorize many administrators, and to let them fight it out in the marketplace; perhaps only one will be left standing. However, if there's great value that can only be realistically achieved by having just one database administrator, the FCC forecloses that possibility as soon as it allows more than one. As we’ve seen so many times, incumbents don’t give up their positions, even it barely makes commercial sense.

The FCC’s preference is to avoid such choices by allowing both alternatives, as it has done when faced wit the analogous choice between exclusive-use licenses and unlicensed. Unfortunately that option is not available in this rulemaking unless it arbitrarily splits the band between a one-administrator and many-administrator sub-bands.

The FCC therefore has to make the one/many decision, and it has to make it upfront. But how?


The market can decide upfront by using an auction. A simplified version of the approach proposed by Bykowsky, Olson and Sharkey in OSP Working Paper 43 (2008) (pdf) could be used. They described how to allocate spectrum between licensed and unlicensed operations (called L-type and U-type) in four spectrum blocks by creating a “market” for licensing rules. L-type and U-type bidders would place bids conditional on their preferred license regime; U-type bids would be aggregated and compared to individual L-type bids. If the aggregate(s) of U-type bids were among the top four amounts bid, some number of the blocks would be assigned to unlicensed.

In the one/many administrator choice, there’s just one “block” at issue: the 3550-3650 band as a whole. (The FCC could offer the one/many choice in more than one sub-bands of the overall 3550-3650 band if it chose, e.g. 3550-3600 and 3600-3650, but I’ll stick with one band for simplicity.) Bidders would place One-type and Many-type bids. If the highest bid was One-type, that bidder would become the administrator. If the sum of Many-type bids was the highest, there would be many administrators.

One of the objections to the Bykowsky, Olson and Sharkey proposal is that U-type bids necessarily underestimate the value of unlicensed allocation because a player could refrain from bidding, and thus paying, for an unlicensed allocation but still benefit because an unlicensed allocation by definition doesn’t exclude anyone from the band. U-type allocations would therefore be under-provided relative to their actual social value. That problem could be avoided in the one-many choice auction if only parties who made Many-type bids were allowed to apply to become database administrators.

A second-order issue, then, is whether to exclude bidders who make only trifling bids from eligibility. There could be a threshold bid value that would trigger administrator eligibility, e.g. giving eligibility only to the N highest Many-type bids, or to the bidders who make up X% of the total bid – but picking N or X will be difficult. On the basis of the experience in the TV white space, N=10 seems reasonable; and the Pareto Principle suggests X=80% wouldn’t be crazy.


16 March 2013
An economist pointed out to me that a single database service provider may pose competitive problems, i.e., may charge a supra-competitive price for its database services. He noted that in such an instance, the FCC will need to identify the minimum number of firms it believes must exist in order to substantially reduce competitive concerns. For example, he suggested, the FCC may determine that the choices in the auctions are: (1) three database service providers versus (2) many database service providers.

He went on to provide the following excellent mini-tutorial on auctions, monopoly rents and the government’s slice: When parties are bidding in such an auction, their bids reflect, in part, the profits they obtain from participating in the market. The fewer the database service providers, the higher the prices the service providers will be able to charge. Interestingly, in such an auction, competition for the monopoly causes the U.S. Government to obtain a portion of the supra-competitive profits that the winner of the auction would earn. For example, in a second price auction where everyone is bidding to be the monopolist, the winner might be willing to pay 10 for the right to be the monopolist. Assume that the second highest bid is 8. If a competitive market structure would have generated a high bid of 6, then "the competition to be the monopolist" enabled the government to extract 2 units from the monopolist. The monopolist's payoff is the difference between 10 and 8 (i.e., 2), plus any additional gain the monopolist obtains from not bidding its true value in the auction. The Government's ability to extract a portion of the gains from the monopolist doesn't mean that everything is fine. Indeed, the supra-competitive prices for the database service still results in a welfare loss for society. However, in this instance, the "surplus" goes not to the shareholders of the monopolist, but to the US Treasury.

16 March 2013
An engineer told me that he thought that only slow updates to coarsely adjust base station power levels are needed. The update rate could be slow; he noted that the 100 millisecond update in cellular systems is to handle high speed handover for mobiles.

(I think that even if one doesn’t need a single SAS administrator to do real-time transmit power coordination, it may still be needed to do admission control. If there’s an aggregate interference problem from a sea of secondaries into a noise-limited primaries like radar or satellite earth stations, just setting EIRP ceilings on the secondaries isn’t sufficient. Once you reach a critical number of operating devices in an interference zone, you have to stop admitting new ones. This isn’t necessarily a sub-second coordination problem, but rather a commercial coordination problem: every administrator will want to privilege its own clients. You need one arbiter in the end.)

No comments: