Saturday, May 12, 2018

Economic rivalry, interference and spectrum allocation

I’m intrigued by the questions:
  • What is the relationship (if any) between rivalry and licensing regime in spectrum regulations to date? 
  • And what should it be in the future? 
 I think such work could shed light on efforts to allocate and assign millimeter-wave spectrum (roughly, above 30 GHz).

My working assumption is that very rivalrous (and effectively excludable) resources are best managed as private goods, e.g. through a property rights regime. In spectrum, the paradigmatic example is exclusive, flexible use cellular licenses assigned by auction.

Rivalry/excludability yields the well-known 2x2 matrix. Another important cell for spectrum policy is public goods (non-rivalrous and non-excludable), e.g. unlicensed.  Both rivalry and excludability take on a continuum of values, so a 2x2 matrix is just a rough approximation; for example, unlicensed is excludable to the extent that one need to have a certified device to operate in it.

Two key characteristics of mmWave spectrum are (1) that the wavelength is small (millimeters!) so that highly directive antennas can be compact, and (2) that signal attenuation is much higher than at longer wavelength, necessitating such antennas.

Work I've seen and been involved in seems to suggest that interference between pencil beam mmWave systems isn’t much of a problem; that is, rivalry is low. Indeed, the light licensing of the E-band (70/80 GHz) by the FCC, Ofcom and ACMA was premised on the fact that pencil beams and large path losses in mmWave would allow multiple concurrent operators to be supported without interference concerns (E-Band Communications Corp. white paper, FCC information page).

Empirical calculations of rivalry (i.e., the degree to which operation by one party prevents simultaneous operation by other another) could suggest policy actions:
  • If mmWave is non-rival, then assigning exclusive licenses by auction seems like overkill. Further, from a technical angle, the fit between large area licenses and short-range highly directed beams since imperfect, to say the least. An unlicensed regime would seem to be indicated in this case.
  • If it’s somewhat rival, then a light licensing regime (operators have to register links, but don’t require separate licenses for most links, or traditional frequency coordination) sounds more appropriate. If rivalry is moderate, then one could require frequency coordination, as with mid-band point-to-point microwave links.
  • If rivalry is high, then assignment to the highest value user, e.g. through exclusive licenses assigned by auction, would seem most appropriate.

A research program might have the two major components:

1. Study existing allocations

Looking at current services would give one a sense of what the link has been between rivalry and licensing regime, and help define useful rivalry metrics for Part 2 below.

Some services that come immediately to mind; in order of decreasing rivalry/exclusivity
  • Cellular – exclusive use licenses
  • Land Mobile Radio (Part 90 rules in the US) and point-to-point microwave (Part 101), both of which have frequency coordination regimes
  • E-band (70/80 GHz) – light licensing
  • 2.4 GHz and 5 GHz unlicensed – common pool resources
For each of these, one would calculate rivalry metrics, and see if there’s a correlation with the chosen licensing regimes.

Possible engineering proxies for economic rivalry include
  • Spectrum load metrics, a la Kruys, Anker & Schiphorst: quick and dirty; combine RF power density, median duty cycle and occupied frequency range; Kruys & Co developed theirs for unlicensed, but should be readily applicable to wide-area cellular as well; less clear how this would work for directed-beam applications, where focusing on interference seems more productive.
  • Risk metrics that calculate the likelihood and severity of interference: more complicated than spectrum load metrics, but provide a richer assessment of rivalrousness (and perhaps more suited for integration with cost-benefit analysis); the key choice will be of the interference severity metric
Interfering signal strength distributions over geography form a bridge between these two metrics families. Spectrum load could be seen as a simple spatial average of interference power; and a spatial distribution is a risk metric where severity is measured by interfering signal power.

2. Apply lessons to current/future allocations

One would now turn to current allocation questions; I think mmWave cellular is an obvious and important case.

Building on existing research, one would examine the interference mechanisms and severity in mmWave cellular deployments of various kinds, e.g. fixed wireless access and mobile broadband. This could include measurements of channel and antenna performance, and link and system modeling of interference and throughput. Using this information, fine tune (or invent from scratch) measures of economic rivalry that one could apply to mmWave cellular.

Once one has a robust metric that hopefully shows some correlation with prior license exclusivity decisions, one can apply it to contemporary questions. (Non-correlation between a rivalry metric and licensing regimes would be a very interesting result, too!)

The window for an effective intervention in mmWave policy decisions is closing, but it’s still open. While the FCC auction process is already under way for 24, 28, 39 and 47 GHz, and 57-71 GHz has been assigned to unlicensed, the debate has just started for 95+ GHz. Ofcom had a consultation on 26 GHz and 30+ GHz late last year, but I don’t think it’s made any decisions yet. I don’t know about EU countries, but their policy makers seem to have drunk the 5G Kool Aid, which means cellular-friendly licensing is going to be the default.

No comments: