Dark Clouds on the Spectrum Horizon

Technology issues can be hard and complicated. In the absence of genuine understanding, many people become True Believers, embracing positions they don’t really understand for easy answers and standing in particular political subgroups.

One such person is former Google CEO Eric Schmidt, an expert on Internet advertising who fancies himself a spectrum guru. Schmidt has no educational or professional engineering experience with radio-based data networks, but he expresses strong opinions about how to build them in the Wall Street Journal.

Schmidt’s bias emerged when President Obama put him in charge of the 2012 PCAST spectrum report. Rejecting the engineering consensus on spectrum auctions, licensing, and the efficient operation of radio networks, Schmidt directed the group to recommend a plan for dynamically sharing spectrum in multiple complex and unpredictable ways in the hope of achieving high efficiency.

The Appeal of Fuzzy Notions

The PCAST plan made for nice slide presentations, but it was fundamentally nothing more than a way to inefficiently share spectrum assigned to an incumbent (DoD) with no interest in improving its legacy systems. This is not the general case for spectrum use.

Democrats have a fairly deep bench of lawyers and economists, but they’re extremely weak on spectrum technologists. This is because of the outsized role played by self-styled public interest advocates in Democratic policy circles; effectively, they do their best to push private sector technologists off the stage.

This creates a vacuum for people who present themselves as experts when they really aren’t. Hence, dynamic spectrum sharing systems grew legs because semi-pro tech wonks liked the way it sounds. Dynamic sharing is good, right? It is, but there are many ways to do it, some more powerful than others.

How Radio Networks Share

Every radio network is shared by multiple users. The conventional cellular network built on licenses auctioned for exclusive management by a network operator implements a sharing protocol.

AT&T and Verizon don’t buy spectrum licenses to talk to themselves, they get them so they can support the concurrent use of their infrastructure by millions of users. Cable networks operate in the same way, with hundreds of users sharing a cable that carries radio frequency signals around a neighborhood.

3GPP cellular networks implement sharing with techniques known as scheduling, beam-forming, frequency division multiplexing (FDM), statistical multiplexing, spatial division multiple access (SDMA), code division multiple access (CDMA), orthogonal frequency division multiple access (OFDMA), multi-user multiple input/multiple output (MU-MIMO) and more.

These techniques work best when the network is managed by a single conductor to keep all the players on key and on tempo.

Not All Systems of Sharing are Equal

The Internet’s core protocol, Internet Protocol (IP) enables multiple users to share communication links as well. The IP system of multiple use – simple statistical multiplexing – is the basis of Internet economics.

Sharing is everywhere in networks, so we have to view people claiming to bring it to any engineering problem for the first time with suspicion. What we need is powerful, efficient sharing, not haphazard, chaotic, and unpredictable sharing.

Single conductor systems are best because such an arrangement – and only such an arrangement – permits resource sharing decisions to be made in ridiculously small units of time, thousandths of a second or less.

When we can allocate resources on tiny time scale, we can pack a link to capacity, and when we can’t our systems become more costly and less reliable because they become inefficient. PCAST-style sharing is an extra layer of overhead on top of an extremely efficient system that makes everything worse under heavy load.

The conductor approach also unlocks the non-temporal sharing modes listed above (SDMA, CDMA, OFDMA, and MU-MIMO et al.)

The PCAST Model Limits Application Choice

Citizens Band Radio Service (CBRS) is an attempt at implementing the PCAST dynamic spectrum access concept. It adds a spectrum access server to each location where a legacy network exists that is potentially sharable with CBRS-aware spectrum managers but does not require the legacy network to be CBRS-aware.

This is a compromise that fits with the aging DoD procurement model, where equipment has to achieve twenty years or so of utilization, but isn’t at all friendly to the agile, dynamic, Moore’s Law approach to radio system employed by the private sector and regular people.

With CBRS, each spectrum band has primary manager that can do all the things networks normally do, such as scheduling packets for transmission in advance to prevent collisions and packet loss. Secondary and tertiary user fight over table scraps.

While single conductor networks support the whole range of applications from vehicle safety to video conferencing to the web, CBRS is only reliable for web browsing. If your business is advertising, this is fine. If it’s not, it’s not.

MVNOs are More Efficient than CBRS

Cellular networks commonly host multiple service providers by creating virtual networks. This is a construct that distributes billing and some aspects of network security without perturbing the conductor’s capacity to achieve harmony.

The FirstNet system for public safety is an MVNO, a master network with multiple virtual networks. It is able to shift and allocate communication capacity it at least two dimensions, location and urgency. MVNOs do generally require some technology uniformity, which CBRS doesn’t.

With no apparent regard for the distinction between dynamic sharing and MVNO, Schmidt and his DoD sponsors now propose the military operates an MVNO for 5G networks, ostensibly to enable the US to compete with China for control of 5G standards evolution.

An Unsound Plan

The advantage to DoD for taking the MVNO path is to retain control over its spectrum assets rather than releasing them to the private sector and buying back whatever level of service it may require for its own operations.

The disadvantage for 5G consumers is that DoD has no competence at network operation, so it would necessarily have to issue a contract to one favored vendor to operate the network on its behalf. The preferred vendor that it appears to favor is Rivada Networks, a company with no experience in network operations but strong ties to the outgoing Trump Administration.

The alternative is to auction DoD’s mid-band spectrum to the competent firms that have expressed interests in it, the same companies that offer cellular services today. This approach is just as dynamic as the DoD approach since the commercial operators are free to make MVNO agreements with each other where and when they’re needed.

So: we’re really talking about issuing a single government contract that leaves the Pentagon in control of a critical communication resource (one that it’s unable to manage) vs. an auction for licenses that preserves market competition and enables technology upgrades.

What Does the Rivada Plan Have to do with China?

Not much. To the extent that the US is behind any other country with 5G deployment, the bottleneck is mid-band spectrum. In most nations, the regulator simply auctions it off because there’s no significant incumbent other than satellite TV.

But in the US, the incumbent is the recalcitrant military, a slow-moving behemoth that has acquired a patchwork quilt of applications, many of which are past their sell-by dates. If motivated, the military can upgrade its apps to use commercial carriers or to shift them to other frequency bands.

It can also redesign them to coexist with the commercial networks it’s likely to encounter in all theaters outside the US. The military does do most of its warfighting outside the US, so it has to be prepared to fight in places where it does not control spectrum policy after all.

What to Do Next

Dr. Schmidt needs a course on spectrum management to learn the differences between DSA, 3GPP standards, and MVNOs. The Pentagon needs to understand that taking control of American spectrum policy will not do it a bit of good in actual combat.

And we all need quick access to mid-band spectrum for 5G. Schmidt and like-minded sharing advocates have long claimed that it takes ten years or more to repurpose spectrum by auction.

The C-Band auction proves that is not the case, is it will repurpose satellite spectrum to 5G use on a one-to-four year timeline; see this Fierce Wireless story:

Specifically, the satellite companies must first clear 120 megahertz of spectrum in 46 Partial Economic Areas (PEAs) by December 5, 2021. In the second phase, they’re due to clear the lower 120 megahertz in the remaining PEAs, plus an additional 180 megahertz nationwide, by December 5, 2023.

If the companies fulfill these commitments, they’ll be eligible for up to $9.7 billion in accelerated relocation payments plus “reasonable” relocation costs, paid for by the new flexible use licensees. If the operators had chosen not to accept accelerated relocation payments, the deadline for clearing the lower 300 megahertz of the band would have been December 5, 2025.

Let’s stop playing around and do what needs to be done.