Juggling Pipes: orchestrating scarce radio resources to serve multifarious applications

I concluded in Cisco’s Fascinating Flaky Forecast that the impending supply/demand mismatch in wireless data services presents opportunities for “innovations that improve effective throughput and the user experience”. This post explains one example: a software layer that that matches up various applications on a device to the most appropriate connectivity option available, mixing and matching apps to pipes to make the cheapest, fastest, or most energy efficient connection. (In academic terms, it’s a version of Joe Mitola’s Cognitive Radio vision.)

Peter Haynes recently prompted me to ask some experts what they thought the most exciting wireless technology developments were likely to be in the next decade. Mostly the answer was More of The Same; a lot of work still has to be done to realize Mitola’s vision. The most striking response was from Milind Buddhikot at Bell Labs, who suggested that the wireless network as we know it today will disappear into a datacenter by 2020, which I take to mean that network elements will be virtualized.

I don’t know about the data center, but from a device perspective it reminded me of something that’s been clear for some time: as a device’s connectivity options keep growing, from a single wired network jack to include one or more cellular data connections, Wi-Fi, Bluetooth, UWB, ZigBee etc., as the diversity of applications and their needs keeps growing, from an email client to many apps with different needs including asynchronous downloads, voice and video streams, and data uploads, and as choosing among becomes more complicated, such as trade-offs between connectivity price, speed, quality of the connection, and energy usage, there is a growing need for a layer that sits between all these components and orchestrates all these connections. Can you say “multi-sided market”?

The operating system is the obvious place to do such trade-offs. It sits between applications and peripherals, and already provides apps with abstractions of network connectivity. As far as I know, no OS provider has stepped up with a road map “smart connectivity.” It’s decidedly not just “smart radio” as we’ve heard about with “white spaces”; the white space radio is just one of the many resources that need to be coordinated.

For example, one Wi-Fi card should be virtualized as multiple pipes, one for every app that wants to use it. Conversely, a Wi-Fi card and a 3G modem could be bonded into a single pipe should an application need additional burst capacity. And the OS should be able to swap out the physical connection associated with a logical pipe without the app having to know about it, e.g. when one walks out of a Wi-Fi hotspot and needs to switch to wide-area connectivity; the mobile phone companies are already doing this with Wi-Fi, though I don’t know how well it’s working.

That said, the natural winner in this area isn’t clear. Microsoft should be the front-runner given its installed base on laptops, its deep relationships with silicon vendors, and its experience virtualizing hardware for the benefit of applications – but it doesn’t seem interested in this kind of innovation.

Google has an existential need to make connectivity to its servers as good as it could possibly be, and the success of Android in smartphones gives it a platform for shipping client code, and credibility in writing an OS. However, it is still early in developing expertise in managing an ecosystem of hardware vendors and app developers.

The network operators don’t much end-user software expertise, but they won’t allow themselves to be commoditized without a fight, as they would be if a user’s software could choose moment-to-moment between AT&T and Verizon’s connectivity offers. The telcos have experience building and deploying connectivity management layers through orgs like 3GPP. Something like this could be built on IMS, but it’s currently a network rather than device architecture. And the network operators are unlikely to deploy software that allows the user to roam to another provider’s data pipes.

The chipset and handset vendors are in a weaker position since they compete amongst themselves so much for access to telcos. Qualcomm seems to get it, as evidenced by their Gobi vision, which is several years old now: “With Gobi, the notebook computer becomes the unifying agent between the different high speed wireless networking technologies deployed around the world and that means freedom from having to locate hotspots, more choice in carrier networks, and, ultimately, freedom to Gobi where you want without fear of losing connectivity – your lifeline to your world.” As far as I can tell, though, it doesn’t go much beyond hardware and an API for supporting multiple 3G/4G service providers on one laptop. Handset vendors like

Vendors like Samsung or HTC could make a go of it, but since network operators are very unlikely to pick a single hardware vendor, they will only be able to get an ecosystem up to scale if they collaborate in developing a standard. It’s more likely that they will line up behind the software giants when Google and/or Microsoft come forward with their solutions.

It is also possible that Cisco (or more likely, a start-up it acquires) will drive this functionality from the network layer, competing with or complementing app/pipe multiplexing software on individual devices. As Preston Marshall has outlined for cognitive radio,* future networks will adapt to user needs and organize themselves to respond to traffic flow and quality of service needs, using policy engines and cross-layer adaptation to manage multiple network structures. There is a perpetual tussle for control between the edge of the network and the center; smart communications modules will be just another installment.

* See Table 4 in Preston F Marshall, “Extending the Reach of Cognitive Radio,” Proceedings of the IEEE, vol. 97 no. 4 p. 612, April 2009

Better Radio Rights

Demand for wireless services is growing relentlessly, but the ambiguous definition of rights and unpredictable enforcement has led to prolonged inter-service interference disputes that impede innovation and investment.

Silicon Flatirons organized a conference on this topic in DC a couple of weeks ago. The goal was to explore how radio operating rights could best be defined, assigned and enforced in order to obtain the maximum benefit from wireless operations. The event web site has links a fascinating set of position papers prepared by the panelists. There’s also a compendium that collects them all in one place (PDF).

Kaleb Sieh and I proposed (position paper, full paper on SSRN) an approach to radio operating rights based on three principles: (1) aim regulation at maximizing concurrent operation, not minimizing harmful interference; (2) delegate management of interference to operators; (3) define, assign and enforce entitlements in a way that facilitates transactions.

We argue that radio rights should be articulated using transmission permissions and reception protections, defined probabilistically (the Three Ps): transmission permissions should be based on resulting field strength over space and frequency, rather than radiated power at a transmitter; reception protections should state the maximum electromagnetic energy an operator can expect from other operations; both are specified probabilistically. This formulation of operating rights does not require a definition of harmful interference, and does not require receiver standards.

Since any initial entitlement point is unlikely to be optimal, the regulator should facilitate the adjustment of rights by: limiting the number of parties to a negotiation should be limited by minimizing the number of recipients, and enabling direct bargaining by effective delegation; recording a complete and current description of every entitlement in a public registry; stipulating the remedy (injunctions or damages) that attaches to an operating right when it is issued; the regulator refraining from rulemaking during adjudication; leaving parameter values unchanged after an entitlement has been defined, although values may be adjusted though bilateral negotiation between operators, and the regulator may add new parameters at license renewal.

Who gets the apple?

I’ve been looking for a metaphor to illustrate the weaknesses I see in the FCC’s “you two just go off and coordinate” approach to solving wireless interference problems among operators.

Let's think of the responsibility to bear the cost of harmful interference as an apple.*  It’s as if the FCC says to Alice and Bob, “I've got an apple, and it belongs to one of you. I’m not going to decide which of you should have the apple; you decide among yourselves.”

Now, if Alice were the owner of the apple and valued it at 80 cents, then the answer would simply depend on how much Bob valued having the apple (and rational negotiation, of course). If having an apple was worth 90 cents to him, he’d get it for some price between 80 and 90 cents; if it was worth only 60 cents to him, Alice would keep it. Problem solved.

Trouble is, the FCC doesn’t tell them who actually owns the apple, and even if it did, it doesn’t tell them whether it’s a Granny Smith or a Gala. The odds of Alice and Bob coming to an agreement without going back to the FCC is slim.

The analogy: The FCC’s rules often don’t make clear who’s responsible, in the end, for solving a mutual interference problem (i.e. who owns the apple); and it’s impossible to know short of a rule making by the FCC what amounts to harm (i.e. what kind of apple it is).

-----------------------
* There's always interference between two nearby radio operators (near in geography or frequency).  While the blame is usually laid on the transmitter operator, it can just as reasonably be placed on the receiver operator for not buying better equipment that could reject the interference.

How I Learned to Stop Worrying and Love Interference

(With apologies to Stanley Kubrick.)

Radio policy is fixated on reducing or preventing harmful interference. Interference is seen as A Bad Thing, a sign of failure. This is a glass-half-empty view. While it is certainly a warning sign when a service that used to work suddenly fails, rules that try to prevent interference at all costs lead to over-conservative allocations that under-estimate the amount of coexistence that is possible between radio systems.

The primary goal should not be to minimize interference, but to maximize concurrent operation of multiple radio systems.

Minimizing interference and maximizing coexistence (i.e. concurrent operation) are two ends of the same rope. Imagine metering vehicles at a freeway on-ramp: if you allow just one vehicle at a time onto a section of freeway, people won’t have to worry about looking out for other drivers, but very few cars would be able to move around at one time. Conversely, allowing everybody to enter at will during rush hour will lead to gridlock. Fixating on the prevention of interference is like preventing all possible traffic problems by only allowing a few cars onto the freeway during rush hour.

Interference is nature’s way of saying that you’re not being wasteful. When there is no interference, even though there is a lot of demand, it’s time to start worrying. Rather than minimizing interference with the second-order requirement of maximizing concurrent operation, regulation should strive to maximize coexistence while providing ways for operators to allocate the burden of minimizing interference when it is harmful.

I am developing a proposal that outlines a way of doing this. Here are some of the salient points that are emerging as I draft my ISART paper:

The first principle is delegation. The political process is designed to respond carefully and deliberatively to change, and is necessarily slower than markets and technologies. Therefore, regulators should define radio operating rights in such a way that management of coexistence (or equivalently, interference) is delegated to operators. Disputes about interference are unavoidable and, in fact, a sign of productively pushing the envelope. Resolving them shouldn’t be the regulator’s function, though; parties should be given the means to resolve disputes among themselves by a clear allocation of operating rights. This works today for conflicts between operators running similar systems; most conflicts between cellular operators, say, are resolved bilaterally. It’s much harder when dissimilar operations come into conflict (see e.g. my report (PDF) on the Silicon Flatirons September 2009 summit on defining inter-channel operating rules); to solve that, we need better rights definitions.

The second principle is to think holistically in terms of transmission, reception and propagation; this is a shift away from today’s rules which simply define transmitter parameters. I think of this as the “Three P's”: probabilistic permissions and protections.

Since the radio propagation environment changes constantly, regulators and operators have to accept that operating parameters will be probabilistic; there is no certainty. The determinism of today’s rules that specify absolute transmit power is illusory; coexistence and interference only occur once the signal has propagated away from the transmitter, and most propagation mechanisms vary with time. Even though US radio regulators seem resistant to statistical approaches, some of the oldest radio rules are built on probability: the “protection contours” around television stations are defined in terms of (say) a signal level sufficiently strong to provide such a good picture at least 50% of the time, at the best 50% of receiving locations. [1]

Transmission permissions of licensee A should be defined in such a way that licensee B who wants to operate concurrently (e.g. on nearby frequencies, or close physical proximity) can determine the environment in which its receivers will have to operate. There are various ways to do this, e.g. the Australian “space-centric” approach [2] and Ofcom’s Spectrum Usage Rights [3]. These approaches implicitly or explicitly define the field strength resulting from A’s operation at all locations where receivers might be found, giving operator B the information it needs to design its system.

Receiver protections are declared explicitly during rule making, but defined indirectly in the assigned rights. When a new allocation is made, the regulator explicitly declares the field strength ceilings at receivers that it intends to result from transmissions. In aggregate, these amount to indirectly defined receiver protections. Operators of receivers are given some assurance that no future transmission permissions should exceed these limits. (Such an approach could have prevented the AWS-3 argument.) However, receivers are not directly protected, as might be the case if they are given a guaranteed “interference temperature”, nor is there a need to regulate receiver standards.

While this approach has been outlines in terms of licensed operation, it also applies to unlicensed. Individual devices are given permissions to transmit that are designed by regulator to achieve the desired aggregate permissions that would otherwise be imposed on a licensee. Comparisons of results in the field with these aggregate permissions will be used as a tripwire for changing the device rules. If it turns out that the transmission permissions are more conservative than required to achieve the needed receiver protections, they can be relaxed. Conversely, if the aggregate transmitted energy exceeds the probabilistic limits, e.g. because more devices are shipped than expected or they’re used more intensively, device permissions can be restricted going forward. This is an incentive for collective action by manufacturers to implement “politeness protocols” without regulator having to specify them.

Notes

[1] O’Connor, Robert A (1968) Understanding Television’s Grade A and Grade B Service Contours, IEEE Transactions on Broadcasting, Vol. 47, No. 3, September 2001, p. 309, http://dx.doi.org/10.1109/11.969381

[2] Whittaker, Michael (2002) Shortcut to harmonization with Australian spectrum licensing, IEEE Communications Magazine, Vol. 40, No. 1. (Jan 2002), pp. 148-155, http://dx.doi.org/10.1109/35.978062

[3] Ofcom (2007) Spectrum Usage Rights: A statement on controlling interference using Spectrum Usage Rights, 14 December 2007, http://www.ofcom.org.uk/consult/condocs/surfurtherinfo/statement/statement.pdf

Alfred Kahn, SURs, and new approaches to radio regulation

I have at last finished writing up a Silicon Flatirons meeting on rules for inter-channel radio interference (web page, PDF). Reflecting on the event, I was struck again by the contrast Phil Weiser noted last year [1] between the success of airline deregulation, and the halting progress in doing the same for “spectrum”.

The meeting showed there was broad support for taking receivers into account more explicitly when drafting rules, for example by regulating resulting signal levels rather than the customary approach of specifying rules for individual transmitters. This approach focuses on the results of transmission – which includes interference, the bone of contention in most radio regulation debates – rather than the transmission itself.

Ofcom, the UK communications regulator, took an interference-based approach to licensing by creating Spectrum Usage Rights, also known as SURs [2]. However, SURs were roundly rejected by the cellular operators. Ofcom chose not to impose SURs on the mobile industry, on the premise that the goal of SURs was to improve the certainty of the license holders for their benefit, not the regulator’s benefit.

While there are many other reasons for the cellcos to reject SURs (the problems SURs address, like uncertainty about likely uses and technologies, or disparate uses in adjacent channels, are largely absent in cellular bands), it is clear that Ofcom deferred to the interests of incumbents – potentially at the cost of consumers or new entrants. One of the conclusions of a 2007 report for the European Commission on radio interference regulatory models [3] came to mind:

“Technology and service-neutral licensing (as would be supported by interference-based licensing techniques) offers significant benefit for end-users but not necessarily for spectrum owners and network providers.”

In an essay in honor of Alfred Kahn’s 90th birthday, Phil Weiser observed that airline regulation (where Kahn, the "Father of Airline Deregulation," made his name) and spectrum regulation share some basic characteristics: both regimes emerged from an effort to protect established interests; both limited output by restricting the use of the resource in question; and in both cases, early academic criticism calling for regulatory reform went unheeded. In making the case for Kahn as a political entrepreneur, Weiser argues that he “pursued the objective of eroding the airline industry’s commitment to the legacy regulatory regime by both undermining the manner in which it protected established incumbents and bolstering the strength of those interests that would benefit from deregulation.”

The radio incumbents Weiser had in mind were the broadcasters and not the cellular companies – but it’s not too much of a stretch to attribute at least some of the resistance to new methods of radio regulation to the New Incumbents.

REFERENCES

[1] Phil Weiser (2009), “Alfred Kahn as a Case Study of a Political Entrepreneur: An Essay in Honor of His 90th Birthday.” Journal Network Economics, 2009. Abstract at SSRN. The paper was first delivered at a conference at Silicon Flatirons in Boulder on September 5, 2008.

[2] See e.g. William Webb (2009), “Licensing Spectrum: A discussion of the different approaches to setting spectrum licensing terms” (PDF); and Ofcom (2008), “Spectrum Usage Rights: A Guide Describing SURs” (PDF)

[3] Eurostrategies and LS telcom, “Study on radio interference regulatory models in the European Community” 29 November 2007 (PDF)

Protection Payments: Licensed vs. unlicensed radio rights

I have just realized the blindingly obvious: the main value of a radio license is the right not to be interfered with, rather than the right to transmit.

This claim should be testable by establishing the degree to which the protection against interference influences the prices of licenses sold at auction. I’m working with Johnny Chan at the University of Washington to generate some results in this area. It’s certainly true anecdotally; M2Z has argued that T-Mobile knew its AWS-3 license, which had an adjacent band which would generate more interference, was worth less and so paid less at auction.

Proponents of license auctions charge that unlicensed allocations mean that “people” (they’re thinking of Google and Microsoft) are getting something without paying for it. It’s true that users of unlicensed radios don’t pay for a license, and it’s also true that both kinds of licenses confer some permission to operate a radio.

But there’s a big difference: a licensee can stop others from interfering with their operation, whereas an unlicensed user not only may not interfere with licensees, but also has to accept interference from all comers.

The big difference between an exclusive-use license and an unlicensed regime is excludability rather than autonomy, to use terminology I defined in an earlier post (Protecting receivers vs. authorizing transmitters). (Regarding a property, exclusivity means an owner can control what other people do, while autonomy allows an owner to act without hindrance.)

However, since radio licenses are defined in terms of transmission rights rather than receiver protections, what’s being sold is autonomy rather than exclusivity.

In practice there is a gamut of license types, with increasingly strong excludability rights: from unlicensed, to licensed by rule, to secondary licenses, and then primary licenses. The more excludability you get, the more a license should be worth. We’re planning to do regression analysis on US auction results to see if this is the case.

If, as we’re working to show, the main benefit of a radio license is protection from interference rather than the right to transmit, then current radio policy is misconceived in focusing on transmit rights rather than receiver protection rights. While it’s true that defining transmit rights implicitly defines the receiver rights (again, see Protecting receivers vs. authorizing transmitters), not making receiver rights explicit guarantees downstream conflict, as the M2Z/T-Mobile argument over AWS-3 has shown.

Protecting receivers vs. authorizing transmitters

When governments hand out permissions to operate radios – licenses, for example – they think in terms of transmitters: within the licensed frequency range you can broadcast at such-and-such a power, and outside those frequencies you can transmit only at some other, much lower, power. [1] This distribution of broadcast power is often called a “transmission mask”.

In thinking about new ways of regulating radio, I’ve come to believe that a transmission mask is not sufficient; it helps to include receiver parameters. [2] But that’s not the point of this post; if you’re interested, read my paper at SSRN.

Today’s question is: if transmitter parameters are not sufficient, could one do without them completely? Can you go the whole hog, and only specify a receiver mask? I think you can, and I’m encouraged that Dale Hatfield tells me Robert Matheson concluded this some time ago, though I haven’t found a reference yet.

Receiver and transmitter parameters are figure and ground. Assume a steady state where all systems operate without interfering with each other. Transmissions will by definition have been chosen to prevent interference with receivers in adjacent frequency bands. The result of all the transmissions is a level of electromagnetic energy at every receiver which is low enough that no harmful interference results. (I’m ignoring receiver specifications [2] and the geographical distribution of transmitters in this discussion.) Each transmission propagates through space to a receiver, resulting in the allowed receiver mask:

{all transmission masks} -> {propagation} -> {resulting receiver mask}

One can also invert the calculation: given a receiver mask and propagation model, one can determine what the allowed transmissions should be.

A license defined in terms of receiver masks would allowed the licensee to transmit any amount of energy as long as it does not exceed the mask of anyone else. It would guarantee that nobody else is allowed to radiate energy which leads to the allowed levels being exceeded at the licensee’s receiver.

One can compare reception-based vs. transmission-based licenses by thinking about property rights. The two important attributes here are exclusivity (if I’m a licensee, I can prevent anyone from transmitting in my frequency “band”), and autonomy (within my “band”, I can do what I like, notably acting in a way that makes money). [3], [4]

A transmitter-based license focuses on autonomy by defining transmission parameters. It specifies what a licensee is allowed to do, but it doesn’t provide any guarantee of exclusivity. A receiver-only right reverses this emphasis: by specifying what would amount to harm to a receiver, it provides a way to make exclusivity real in practice. The constraints on autonomy are implicit in the receiver-rights of others: as long as a licensee doesn’t interfere with other licensees’ exclusivity, it can do what it likes.

Anything transmitter-only rights can do, receiver-only rights can also do. They are mirror images of each other. The information burdens are also mirror images. Receiver rights place a burden on all the other rights holders to figure out if their transmissions will transgress a receiver spec. While transmission rights don’t impose an information overhead upfront, the rights holder bears the burden of uncertainty: they may be blind sided at some future date by a new transmission right that reduces the value of the system they’ve deployed. The fight between M2Z and T-Mobile is a good example: T-Mobile claims that the proposed terms of a proposed new cellular license M2Z seeks (AWS-3) will cause harmful interference to their operations under a current license (AWS-1).

I like receiver-only rights because they put the focus on the ability of a wireless system to operate in the presence of noise, which one can only do by taking receiver parameters into account. However, enforcement proceedings may appear more complicated in this case, since it isn’t obvious which of many transmissions is responsible for a receiver mask being exceeded: a spectrum analyzer at a receiver can only measure the sum of all radiated power. Today the regulator has what looks like a big stick: it can objectively check whether a licensee’s equipment meets or violates its approved transmission mask. The stick doesn’t actually help solve the most difficult interference problem, the case where all transmissions meet their masks, but it gives the regulator power that it will be loath to give up.

Conclusion

There a choice in creating radio rights between protecting receivers and authorizing transmitters – or some mixture of the two. The current approach limits the discussion simply to ways of authorizing transmitters. A more nuanced analysis of the trade-offs is required, and was begun here.

Notes

[1] Since this is going to get pretty geeky, so I’m going to leave out a lot of important other stuff, including geography (radio licenses are typically restricted to a certain area) and duty cycles (how often transmitters operate).

[2] Receiver parameters. I distinguish between a receiver mask, by which I mean a distribution of RF energy (i.e., a spectrum) which represents the worst-case noise environment in which a receiver needs to be able to operate, and a receiver specification, by which I mean the ability of the receiver to detect and decode wanted signals in the presence of unwanted ones.

[3] According to Gary Libecap in Contracting for property rights (1989), p. 2 “. Private ownership . . . may involve a variety of rights, including the right to exclude nonowners from access, the right to appropriate the stream of rents from use of and investments in the resource, and the right to sell or otherwise transfer the resource to others.”

[4] “Band” here means the collection of constraints on operation. In the conventional approach, it’s usually taken to mean a frequency band and geographical region.