Spec Markets v3

1
SPECTRUM MARKETS
Randall Berry, Michael Honig
Department of EECS Northwestern University
May 2011
DySPAN Conference, Aachen, Germany
Spectrum Management
2
Policy Economics
Communications Engineering
Spectrum Markets Tutorial, DySPAN Conference
May 2011
Spectrum Management
3
Policy Economics
Communications Engineering Rakesh Vohra

Kellogg MEDS
R. Berry, MLH
EECS
May 2011
High Profile Issue

4
Chicago Tribune April 15, 2011
May 2011
Why This Tutorial?

5
Timely and relevant

New policies are needed for spectrum allocation. Markets are natural policy candidates.
Markets for spectrum pose unique challenges/questions.

Definition of property rights, interference externalities Efficiency, incentives, wireless system design
Interplay between economics and engineering issues

Spectrum Markets Tutorial, DySPAN Conference May 2011
This Tutorial
6
Is NOT about
Large-scale spectrum auctions Related policy issues Analytical methodology
IS about
Fundamental technical and micro-economic issues and approaches to defining spectrum rights, markets Impact on wireless systems and services
Objectives
7
Put spectrum markets into policy context (US-centric) Explain technical challenges and tradeoffs with defining spectrum property rights
Illustrate with basic models
Describe different types of market structures Contrast spectrum markets with other spectrum sharing models (commons/white space)
Outline
Background and Motivation (MH) Spectrum Market Design (RB) Market Organization (MH) Concluding Remarks (MH)
May 2011
Background and Motivation

History Spectrum sharing models Motivation for spectrum markets
May 2011
Limited Supply of Spectrum

10
good for cellular

(300 MHz to 3 GHz)
Increasing Demand
11
Spark-gap transmitter (Tesla, 1893)
May 2011
Spectrum Crunch
12
Petabytes per month
May 2011
Regulation Prior to 1927: Open to All

13
Earliest uses of wireless for ship-to-ship, ship-to-shore communications.
Broadcast radio begins in 1921. Licenses issued by the Department of Commerce.

May 2011
Two Landmark Cases

14
Hoover vs Intercity Radio, 1923 United States vs Zenith Radio, 1926 Department of Commerce has no authority to regulate licenses. Broadcasting boom: 200 new stations appeared in < 6 months. Interference created chaotic radio environment.
Herbert Hoover, US Sec. of Commerce
May 2011
Spectrum Property Rights: A False Start

15
Tribune vs Oak Leaves Broadcasting, 1926

Property right allowed based on homesteading Interfering stations could be fined.
Congress subsequently passed legislation prohibiting spectrum property rights Licenses issued for 90 days.
Two Views
16
Broadcast spectrum is a public good; regulation needed to avoid a tragedy of the commons, and police content. Efforts to privatize spectrum derailed due to rent-seeking by incumbents, special interests, and desire to control content [Hazlett, `90]
May 2011
Regulation since 1927: Command and Control

17
Federal Radio Commission (FRC) established in 1927. Federal Communications Commission (FCC) established in 1934. Maintains authority to
Grant / renew / deny licenses for spectrum use. Assign applications to particular frequencies. Police content and use
Regulation since 1927: Command and Control

18
Federal Radio Commission (FRC) established in 1927. Federal Communications Commission (FCC) established in 1934. Maintains authority to
Grant / renew / deny licenses for spectrum use. Assign applications to particular frequencies. Police content and use
Spectrum Markets Wise old man approach Tutorial, DySPAN Conference to spectrum allocation
The Spectrum Paradox

19
Spectrum is a scarce resource Spectrum is underutilized
May 2011
Spectrum is a Scarce Resource

20
Nearly $20B netted for 700 MHz auctions in 2008.

beachfront property
Spectrum is Underutilized
21
Spectrum measurements in New York City and Chicago conducted by Shared Spectrum Co. Spectrum Markets Tutorial, DySPAN Conference May 2011
Problems with Command and Control

22
An economists critique
Requires excessive information overhead Difficult to estimate value (utility) of a frequency assignment Encourages rent-seeking and facilitates entry barriers
May 2011
An Economists Proposal
23
R. Coase, The federal communications commission, J. Law and Economics, pp. 140, 1959. Introduce spectrum property rights, sell to highest bidders, do not restrict use. Coases Theorem: In the absence of transaction costs, spectrum owners will trade rights so that the outcome allocates spectrum to best use.
Ronald Coase, 1991 Nobel Laureate in Economics
May 2011
24
Role of government should be to minimize transaction costs.

25
Spectrum auctions finally introduced in the 1990s. Restrictions on use remain.

Problems with Command and Control

26
An economists critique:
Requires excessive information overhead Difficult to estimate value (utility) of a frequency assignment Encourages rent-seeking and facilitates entry barriers
An engineers critique:
Demand for different applications varies over time and geographic locations. Static assignments cannot exploit statistical multiplexing. New technologies can facilitate more efficient spectrum sharing.
Engineering Approach to Spectrum Crunch

27
Add intelligence to mobile devices

Frequency agility Wideband sensing Interference avoidance Adaptive quality of service (context aware)
Enables spectrum scavenging

Cognitive Radio
Mitola and Maguire (1999)
28

May 2011
Spectrum Sharing Models

29
Exclusive use Commons Hierarchical
May 2011
Exclusive Use
30
Spectrum owned by government

Licensed to particular application, service provider Rigid use rules
Spectrum is private property

Applications, technical constraints decided by markets
May 2011
Exclusive Use
31
Spectrum owned by government

Licensed to particular application, service provider Rigid use rules
Spectrum is private property Liberal licenses

Applications, technical constraints decided by markets
Spectrum publicly owned, but licenses can be transferred, liberal use rules Secondary markets (2003)
Spectrum Commons
32
Unlicensed Requires etiquette rules for sharing State-regulated

Spectrum owned by government Etiquette rules part of industry standard (802.11)
Privately owned
Owner sets rules, polices band Revenue from selling approved equipment
Hierarchical
33
Primary and secondary users Secondary users must not disrupt primary users Relies on cognitive radio State-regulated
Spectrum owned by government Use rules for secondary users part of standard (802.22)
Private contracts with spectrum scavengers

Interference levels/payments set by mutual agreement
Hierarchical: Technologies
34
Primary and secondary users Secondary users must not disrupt primary users Relies on cognitive radio Underlay: low-power, spread spectrum for secondary users Overlay: exploit white spaces left by primary users
Hybrid Models
35
Spectrum designated for exclusive use could be operated as a commons and/or with secondary users. Underlay/overlay can be used to facilitate further sharing. Spectrum scavenging can increase utilization.
May 2011
Current Allocations
36
Mix of:
restricted use bands (e.g., broadcast TV) liberalized licenses (cellular) state-regulated commons (WiFi)
Active trading of liberalized licenses among commercial service providers

About 10 billion MHz-pops annually since 2003 [Mayo & Wallsten `10]
US Policy trends have favored assignments of unlicensed spectrum over liberalized licenses
955 MHz unlicensed vs 422 MHz licensed in the US (2008)
Why Unlicensed Spectrum?

37
Pushed by DARPA
May 2011
Why Unlicensed Spectrum?

38
Pushed by DARPA
Military needs distributed, dynamic methods for spectrum sharing across military units
Also by Google, Apple

Facilitates 3rd-party software applications
Success of WiFi
Interference not a major issue for local coverage, light loads
Bold Predictions
39
IEEE Spectrum Magazine, March 2004
Engineering Issues
40
Difficult to guarantee Quality of Service

Limits applications
Problems with secondary user model

Sensing problematic, constraints compromises utility
WiFi does not scale; inappropriate for wide-area data in urban settings
May 2011
Economic Issues
41
No means for reallocating spectrum to applications with higher utility

e.g., wide-area data with coordinated interference management
No direct means to move incumbent applications to another band/wireline service

e.g., wireless mics, broadcast TV
Congestion effects may adversely affect competition among licensed service providers (more later)
The Case for Liberal Licenses

42
Provides incentives for service providers to invest in infrastructure for wide-area coverage, interference management [Hazlett, 2010]
Over $20B annual network capital expenditures Virtually no infrastructure investments for unlicensed bands (U-PCS, 3.5 GHz WiMax band)
Previous issues substantial opportunity costs for unlicensed spectrum Liberal licenses allow private commons, scavenging
So far, not economically attractive
Are Secondary Markets Active?

43
Service providers are not issuing short-term leases. Companies with spectrum (e.g., Boeing) are not reselling.
May 2011
Are Secondary Markets Active?

44
Service providers are not issuing short-term leases. Companies with spectrum (e.g., Boeing) are not reselling. But There are active markets for: transferring large blocks of spectrum among service providers, wholesale use of spectrum and infrastructure (e.g., Kindle)
Current State of Affairs

45
Large parts of the useful spectrum remain underutilized. Restricted supply of spectrum with liberalized licenses.
Cellular spectrum is extremely expensive. Service providers encouraged to build out national footprint. Fosters the development of expensive (spectrally efficient) systems.
Unlicensed spectrum is increasing.

46

May 2011
Do We Need Spectrum Markets?

47
A more fundamental question: Is spectrum scarce or abundant? Spectrum is abundant use Commons Model
May 2011

48
A more fundamental question: Is spectrum scarce or abundant? Spectrum is abundant use Commons Model Spectrum is scarce: Commons model tragedy of the commons
Spectrum Markets Tutorial, DySPAN Conference May 2011 NU, April 2009

49
A more fundamental question: Is spectrum scarce or abundant?

For short-range communications (< 50 meters), spectrum is abundant (>3 GHz) commons is appropriate What about for longer-range communications?
Ultimately a technical question
Rate Calculation
50
Extensive spectrum sharing

Roughly 1 GHz between 150 MHz and 3 GHz
Cellular Infrastructure System Assumptions

No intra-cell interference (time-division multiplexing) Limited inter-cell interference. All users are active all the time.
Rate Calculation: Assumptions

51
User at cell boundary (worst-case) Standard large-scale propagation model Uniform power over frequency Shannon rate with 6 dB margin Frequency reuse optimized over each 1 MHz band
Achievable Rate per User

52
Cell radius = 200 m
Rate per user (Mbps)
300 m 400 m 500 m
Worst-case rate is about 2 Mbps with cell radius of 200 m

Dense urban area
user density (Kusers/km2)

Achievable Rate: Different Radii

53
user at r/4
rate per cell (Mbps)
user at r/2
user at cell boundary (r)
cell radius (meters)

Is Spectrum Scarce or Abundant?

54
2 Mbps per user seems like a lot, but recall the assumptions:
1 GHz of shared bandwidth, no fading Infrastructure of access points (200 m radius) Optimized frequency reuse Spectrally efficient modulation
Also, less expensive spectrum encourages lower-cost, spectrally inefficient systems.

Spectrum Supply Curve

55
Quantity of spectrum (Hz)
supply
demand
Equilibrium price
Spectrum price ($/Hz)
As the spectrum price goes to zero:

The supply decreases due to the decrease in spectral efficiency. The demand increases due to introduction of new services.
Commons vs Market
56
Commons
Spectrum market supply
demand
p*
market transaction costs < cost of interference Set up spectrum market

Commons vs Market
57
Commons
demand
p*
cost of interference < market transactions costs Use commons model

Rate Calculations: Conclusions

58
With extensive sharing and an extensive infrastructure the commons model may provide for an adequate range of near-term services, but interference is likely to become a problem in the long-term. Interference at lower frequencies is difficult to manage with a commons model.
Suggests using commons at high enough frequencies, markets at lower frequencies
59
Market Design
Asset Design Market Mechanisms Examples
May 2011
Focus
60
Designing a dynamic market for spectrum.

Short-term allocations done in real-time Small spatial-scale
Consider one entity responsible for leasing/selling spectrum to multiple agents.
May 2011
Distinguishing Feature
61
Markets exists for many items. How is spectrum different? Main difference is interference.
May 2011
Externalities
62
Interference creates externalities.

One users consumption effects another users utility. Well-known in economics that externalities complicate market design.
May 2011
Asset Design
63
Basic question: What is being sold? Think of at two different levels:

Policy Specific Market
May 2011
Policy Issues:
64
What property rights do spectrum owners have?

Legal & economic views
What institutions are available for enforcing these?
Extensive literature on this: DeVany et al (1969), Faulhaber and Farber (2003), Hatfield and Weiser (2008), Hazlett (2008).
Property Rights
65
To facilitate markets want property rights to be:

Clearly defined, easily enforced Transparent: easy to determine utility from ownership Also desirable for property rights to be flexible. Allow adaptation to new technologies, varying demands.
May 2011
Interference
66
A key issue for spectrum is defining property rights regarding interference. Some possibilities:
Limits on received power Limits on transmitted power Cognitive approaches Flexible limits negotiated via bargaining.
or
May 2011
Asset Design for Dynamic Markets

67
Ideally, seems property rights should be broadly designed with a long time-horizon. In a dynamic market, it may be desirable to more narrowly define the asset being sold.
Simplify market design This could evolve at slower time-scales.
May 2011
Asset Specifications
68
Spatial/temporal scales Frequency scale Power allocation Amount of sharing Interference management Device or Technology
May 2011
Examples from Literature

69
CDMA-based approaches
Al Daoud, Alanyali, Starobinski (2007) Huang, Berry, Honig (2006)
Non-interfering spatial regions

Zhou, Gandhi, Suri, Zheng (2008)
Shared access
Kash, Murty, Parkes (2011)
Interfering spatial regions

Zhou, Berry, Honig, Vohra (2009)
70
Market Design
May 2011
Does Market Design Matter?

71
Coases theorem states that given no transaction costs and well-defined property rights, owners will bargain and reach an Pareto efficient outcome. Do we need to worry about designing a market?
May 2011
Caveats
72
Non-zero transaction costs Multilateral externalities Private information
May 2011
Private Information
73
$2
$4
If Lucy knows Charlies value can make an offer to sell at $4 - .
May 2011
Private Information
74
$2
$4
If Lucy knows Charlies value can make an offer to sell at $4 - . Efficient outcome.
Private Information
75
$2
$4
Suppose Lucy only knows that Charlies value is uniformly distributed on [0,10]. Then she would expect to get $5 from any transaction no trade.
Pathological Example?
76
No. Myerson-Satterthwaite theorem shows that with private information, under very general conditions there is no way for two parties to trade that is efficient and individually rational. Suggests market design matters.
May 2011
Markets
77
Market design has a long history in economics.

Intellectual foundations are mechanism design/game theory.
May 2011
Mechanisms
78
$4
Information
Mechanism
$2 Allocation/payments
May 2011
Mechanism Design Problem

79
Need to design
1. 2.
Rules for soliciting information Allocation/payment rule
Objectives:
Social welfare Revenue
May 2011
Example: 2nd Price Auction

80
Mechanism:
Users submit bids. Mechanism allocates good to highest bidder Users pay 2nd highest bid.
Users can be viewed as playing a non-cooperative game.

Use equilibrium concepts from game theory to study performance.
Optimal Bids?
81
$4 $4
2nd Price Auction

$2 $2
Weakly dominant strategy.
May 2011
Outcome
82
$4 Pays $2
2nd Price Auction

$2 Pays noting
May 2011
A Little Terminology
83
2nd price auction is a direct revelation mechanism.

Ask agents to bid their valuation
A direct revelation mechanism is incentive compatible if truth-telling is weakly dominant.
May 2011
Revelation Principle
84
Loosely, an equilibrium obtained under any mechanism can be obtained by an incentive compatible, direct revelation mechanisms. In terms of characterizing possible outcomes, wlog we can consider only direct revelation, incentive compatible mechanisms.
May 2011
Multiple Goods
85
For a single indivisible good, 2nd price auction gives efficient outcome. Unless we are allocating all spectrum to one user, we need to deal with multiple goods.
May 2011
Vickrey-Clarke-Groves (VCG)
86
VCG mechanisms generalize 2nd price auction to arbitrary goods.

Incentive compatible, direct revelation mechanism with the efficient outcome.
May 2011
VCG Mechanism
87
Let A = set of alternative allocations. Each agent i submits valuation vi(a) for each a A. Mechanism chooses alternative to maximize vi(a). Charge user i the marginal cost they impose on other players: maxb j i (vj(b)- vj(a))
Can modify payments by adding terms that only depend on other players valuations.
Issues with VCG

88
Complexity: VCG requires solving N+1 optimization problems for allocating goods to N agents. Overhead: Required bids may have a high communication costs. Requires agents to know values for all alternatives. May be susceptible to collusion.
May 2011
Other Market Mechanisms

89
Incentive Compatible Approximation Mechanisms

Algorithmic mechanism design
Non-incentive compatible mechanisms

e.g. proportional sharing Characterize equilibrium efficiency loss.
Dynamic auctions Posted prices

Uniform vs. discriminatory prices
90
Market Design
May 2011
Basic Model
91
Consider the allocation of C spectrum assets to A agents.

Each asset is right to transmit in given spatial region over a given frequency band for fixed time period. Model Interference among assets via an interference graph.
2 4
May 2011
Static Interference Free Allocation

92
Pick a fixed set of non-interfering assets.

Only allocate these.
If agents valuations of different assets are additive can allocate each using second price auctions.
May 2011
Static Interference Free Allocation

93
Overhead: linear in number of assets. Complexity: O(CA log(A)) Issues?
May 2011
Dynamic Interference Free Allocation

94
Let agents bid on every asset. Allocate an interference free set of assets with the highest bids.
$7, $5, $3
2 4
$3, $5, $2
$1, $3, $1
$2, $1, $2 $1, $1, $1


95
$7, $5, $3
2 4
$3, $5, $2
$1, $2, $1
$2, $1, $2 $1, $1, $1


96
Maximum weight independent set.
$7, $5, $3
2 4
$3, $5, $2
$1, $2, $1
$2, $1, $2 $1, $1, $1

VCG payments
97
Consider Agent 1?
$7, $5, $3
2 4
$3, $5, $2
$1, $2, $1
$2, $1, $2 $1, $1, $1

VCG Payments
98
Consider Agent 1 Remove Agents bids
$5, $3
2 4
$5, $2
$2, $1
$1, $2 $1, $1
VCG Payments
99
Consider Agent 1
Remove Agent 1s bids Re-calculate allocation Payment = (6-2) + (2-0) = $6
$5, $3
2 4
$5, $2
$2, $1
$1, $2 $1, $1

100
Overhead: linear in number of assets. Complexity: NP-hard!

Need to find multiple maximum weight independent sets.
May 2011
Discussion
101
Added more flexibility to the market but at the cost of increased complexity. Is this significant?
Depends on market size.
What about approximations?
May 2011
Approximations
102
Consider a greedy approximation:

Order assets by bids and assign from highest to lowest if possible.
$7, $5, $3
2 4
$3, $5, $2
$1, $2, $1
$2, $1, $2 $1, $1, $1

Greedy Approximation
103
No longer Truthful!
Truthful bids Agent 1 gets Assets 2 Suppose Agent 2 increases bid on 1 to $4
Agent 2 gets Assets 1 and 3 and pays $3 Pay-off = $4- $3.
1 2 3
True values $1, $2 New bids $1, $4
$3, $1 $3, $1
$1, $2 $1, $2
May 2011
Truthful Approximation
104
Issue with previous algorithm is that VCG payments are not suitable for approximate allocations. For some cases can get truthful approximations by changing payments charged to each agent.
E.g. VERITAS (Zhou,Gandhi,Sur,Zheng 08)
May 2011
Allocations with Interference

105
Different application can tolerate different amounts of interference. More flexible approach: allocate all assets.
agents can purchase neighboring assets to mitigate interference.
2 4
May 2011
Complementarities
106
Agents can control interference effects by owning neighboring assets.

E.g. Joint scheduling, power control
This makes neighboring assets complements.

Can also have complements in time/frequency.
2 4
May 2011
Model
107
Agent i A receives value rij if it is allocated j C but incurs an interference cost cjji if not allocated j.
If cjji is large, want interference free assignment.
May 2011
Optimal Allocation
108
Agents need to report values and costs. Allocation given by solving
xij = 1 if agent i is assigned asset j

Complexity?
109
For a general interference graph and choices of costs, this ILP is also NP-hard.
Reduction from independent set.
But would not expect arbitrary costs and arbitrary graphs.
Costs
110
Costs:
Remains NP-hard even if interference costs are constrained to be no greater than an arbitrarily small fraction of the revenue. Reduction from graph partitioning.
Constraining the topology?
Alternative Formulation
111
Let denote the min revenue agent i will receive if allocated asset j. Can then view cjji as additional revenue from owning j and j. Re-write objective as:
Linear Relaxation
112
Prop. If G is a line, the LP is totally unimodular.

Also can solve efficiently if G is a ring.
Some Simple Approximations

113
A 1/((G) +2)-approximation:
Solve LP. Divide zjji variables into (G) +1 sets {Wi} so that adjacent edges are in different sets. Let W0 be set of all xij variables. Find the set of variables that contribute the most to the objective of the LP. Round these to best integer solution & set other variables to any consistent values.
A greedy approximation:
give each asset j to agent with largest rij. gives (1+ 2) approx. where = max ratio of interference costs to rijs.
Truthfulness?
114
Again approximations with VCG prices are not incentive compatible. Truthful Approximations?
May 2011
Secondary User Model

115
Asset A
Boundary Z
Asset B
Boundary region Z supports only secondary users that must not interfere with a different primary users in A or B. Agent i gets revenue rij from asset j {A, B} Agent i gets revenue ABi from cell boundary Z Agent i gets additional revenue ZAi from owning both Z and neighboring asset A.

116
In full network, create one boundary region for each pair of assets. Secondary users occupy one or more boundaries. Allow users to be both primary or secondary. Again optimal allocation is solution to an integer program.
May 2011

117
In general problem is still computationally hard. But if agents can only be primary or secondary then easy to solve in two rounds. Likewise, if secondary revenue across agents is consistent then can also solve easily.
May 2011
Another Alternative Model

118
So far did not specify how interference reduced between assets. Suppose assets correspond to cells. Interference controlled by adapting the boundary of cells. But this effects all neighbors.
Flexible Cell Boundary Model

119
Cell A
coverage radius RiA [0, L/2]
Agents can adjust radius of assigned cell. Revenue is proportional to area: riA = 4wiA RiA2
Interference Region
120
Cell A
coverage radius RiA [0, L/2]
Agent can adjust coverage radius by changing the power. Value is proportional to area: riA = 4wiA RiA2 . Area reduced by interference footprint of neighbor.
Two Adjacent Cells

121
Cell A
Cell B
width zAB
Agent is revenue in cell A: riA = wiA (4RiA2 2RiA zAB) Where zAB = [RiA + RiB (L )]+
Efficient Allocation in a Lattice

122
x,R,z
max
XX
iA jC
2 4wij Rij
i,kA (j,j 0 )E
ik zjj 0 0
ik 2wij Rij zjj 0
Subject to:
ik Rij + Rkj 0 zjj 0 L ,
xij (L/2 ) Rij xij L/2 X xij 1, xij {0, 1}

i
xij = 1 if agent i is assigned asset j A is the set of agents, C is the set of assets, E = {(j,j): j and j are neighboring assets}
Flexible Boundary
123
Efficient allocation is a Mixed integer quadratic program. Assignment of each cell can be done greedily. But given assignment determining radii is difficult. For small guard zone solution is well approximated by mixed integer linear program.
May 2011
Sectorized Boundary Model

124
Suppose each boundary of cell can be adjusted separately.

Sectorized Model
125
Again can formulate assignment as mixed integer quadratic program. Greedy assignment still optimal. Now, given assignment boundary problem decouples.
One problem for each pair. Can solve each problem easily. VCG is computationally feasible.
May 2011
Summary
126
Illustrated several different market designs. Different definitions of spectrum assets can tradeoff flexibility/complexity. Some extensions:
Multiple frequency bands Temporal dimension Multiple sellers/two-sided markets Shared versus exclusive use.
May 2011
127
Market Organization
Market structures Competitive Behavior
May 2011
Market Structures
128
Bit pipe model: secondary markets Two-tier spectrum markets Local transactions
May 2011
Bit Pipe Model

129
Wholesale contract with cellular provider
Kindle
Bit Pipe Model

130
Wholesale contract with cellular provider Mobile Virtual Network Operators (MVNOs)
Resells mobile services (e.g., Virgin Mobile)
May 2011
Bit Pipe Model

131
Wholesale contract with cellular provider Mobile Virtual Network Operators (MVNOs)
Resells mobile services (e.g., Virgin Mobile)
Emerging model for wholesale cellular provider

Open broadband network No retail services Wide coverage
May 2011
Bit Pipe Model: Properties

132
Wide-area coverage, high mobility Interference management Quality of Service guarantees Facilitates new wide-area wireless services Well-matched to lower frequency assignments
May 2011
Higher Frequencies
133
Wide-area coverage becomes difficult Interference management becomes easier Possibility for distributed, dynamic spectrum assignments
May 2011
Spectrum Asset (Property Right)

134
Right to transmit up to specified power at specific time/frequency/location. Power limit should depend on frequency, antenna heights, proximity to other access points, time of day. Spectrum property rights can ultimately be defined by the market itself (Coasean bargaining).
May 2011
Owning vs Leasing
135
Owned spectrum asset has unlimited time duration; traded as property (e.g., land).
Leased spectrum asset has limited time duration; available through local spot market
May 2011
Owning vs Leasing
136
Owned spectrum asset has unlimited time duration; traded as property (e.g., land).
Leased spectrum asset has limited time duration; available through local spot market
Owners can deploy services or rent / lease spectrum assets. Service providers need not be spectrum owners!
Two-Tier Spectrum Market

137
band 1 2 4 5 8 9 6 10 3 7 location
cell 1 cell 2 cell 3 3.5 GHz 3.6 GHz 3.7 GHz
Owner A Owner A Owner A
Owner B Owner B Owner C
Owner A Owner A Owner C
Spectrum Broker
Service requests
Service providers
(Acme Wireless)
Owners A, B, C,
May 2011
Lower-Tier Spot Market

138
Spectrum Broker
Service providers
(Acme Wireless)
Owners A, B, C,
Managed by spectrum broker

Sets prices, attempts to clear market Auction mechanism: collects bids; determines allocation Can be automated (spectrum server)
Lower-Tier Spot Market: Properties

139
Spectrum Broker
Service providers
(Acme Wireless)
Owners A, B, C,
Immediate access, rapid (automated) transactions Low transaction costs Facilitates local services No need to build out large footprint
Interference Management
140
Spectrum Broker
Service providers
(Acme Wireless)
Owners A, B, C,
Spectrum assets predefined by owners

Service providers must comply with interference constraints
Spectrum assets partitioned, assigned by broker to satisfy service requests

Spectrum Contracts
141
Spectrum Broker
Service providers
(Acme Wireless)
Owners A, B, C,
Contracts can be arranged across:

Frequency (spread spectrum, underlay) Locations (mesh networking) Time (time-of-day, futures, scavenging)
Variable QoS guarantees (statistical)

Integration
142
Spectrum Broker
Service providers
(Acme Wireless)
Owners A, B, C,
Broker may integrate allocation of

Access points (property leases) Equipment Spectrum
Lowers entry barriers for new service providers

Commons versus Market

143
Commons
demand
p*
Commons/market boundary depends on associated costs.


144
Commons
demand
p*
market transaction costs < cost of interference Set up spectrum market


145
Commons
demand
p*
cost of interference < market transactions costs Use commons model


146
Commons
demand
p*
Can we shift the boundary to the right with distributed interference management schemes?
Local Transactions
147
Routers use the same channel, cause little interference
May 2011
Local Transactions
148
Would cause excessive interference.

Deterence Price
149
$ Pay new user to not setup access point in exchange for sharing capacity.
May 2011
Usage Price
150
$ Set up community of access points, charge fee for sharing capacity (Fonera).
May 2011
Pricing and Efficiency

151
Setup access point or share?
Deployment game: each user decides whether or not to setup an access point given a fixed deterrence price from neighbors. Deterrence pricing can substantially increase efficiency, mitigate interference [Bae et al, DySPAN `09] .
152
Competition with a Commons

Analysis of white space policy
May 2011
TV White Space
153
FCC recently announced rules for use as unlicensed commons

Devices must check data base to see if spectrum is available before using.
May 2011
TV White Space
154
FCC recently announced rules for use as unlicensed commons

Devices must check data base to see if spectrum is available before using.
Advocates: lowers entry barriers for new services Detractors: tragedy of the commons
Observations
155
Lower frequencies than WiFi

better coverage more interference
longer propagation
Incumbents will compete with services in TV white space.
May 2011
Observations
156
Lower frequencies than WiFi

better coverage more interference
longer propagation
Incumbents will compete with services in TV white space. How will additional white space affect service providers and consumers?
May 2011
Scenario
157
SP 1
SP 2
SP 3
frequency
Incumbent service providers (SPs) have exclusive licensed bands.
May 2011
Scenario
158
commons
SP 1
SP 2
SP 3
frequency
Incumbent service providers (SPs) have exclusive licensed bands. All incumbents and new entrants have access to commons (unlicensed band). How does this additional spectrum affect total welfare? Analyze using framework for competition in congested markets [Acemoglu, Ozdaglar `07]
Model: Summary
159
commons
SP 1
SP 2
SP 3
frequency
Each SP competes for pool of customers by announcing prices for licensed and unlicensed services. Customers choose SP based on Total price = Announced price + Congestion cost
Results: Summary
160
commons
SP 1
SP 2
SP 3
frequency
The equilibrium price of commons spectrum is zero. (Total price = congestion cost) Adding unlicensed spectrum can decrease total welfare (consumer + SP revenue).
Happens over a substantial range of unlicensed bandwidth.
Model: Revenue
161
commons
SP 1
SP 2
SP 3
frequency
Each SP i chooses prices to maximize revenue:
i = piqi+piwqiw
pi : price for licensed band piw : price for unlicensed band qi , qiw : quantity of customers served
Model: Congestion
162
commons
`w (
Customers in band i experience a congestion (latency) cost l (qi), which is increasing convex. Customers in commons experience a congestion cost lw(qw), where qw includes all unlicensed users. Total (delivered) price in band i: piw + l (qi)
SP 1 SP 2
SP 3
frequency
w qi ) i
`(q1 ) `(q2 ) `(q3 )
Model: Competition
163
commons
`w (
Customer chooses band with the lowest price pi +l (qi)(licensed) piw +lw(qw)(unlicensed) Equilibrium: total price is the same across all bands Demand curve: quantity q (number of customers served) as a function of total price
SP 1 SP 2
SP 3
frequency
w qi ) i
`(q1 ) `(q2 ) `(q3 )
Total Welfare: Monopoly SP

164
Price
Demand curve
Total price*
*maximizes revenue
q*
Quantity q (customers served)

May 2011

165
Price
Demand curve
`(q)
Total price*
*maximizes revenue
`(q )
q* Quantity q (customers served)
May 2011

166
Price
Demand curve consumer surplus
`(q)
P*
Total price*
SP revenue
*maximizes revenue
q*
Quantity q (customers served)

May 2011
Total Welfare: Two SPs

167
Price
Demand curve
consumer surplus Total price* P1*
SP 2s revenue P2* *maximizes revenue
`(q)
q1* q2* Quantity q (customers served)
May 2011
Price of Unlicensed Spectrum

168
In equilibrium announced price is zero.*

Otherwise SP can increase revenue by lowering.
Total price of unlicensed spectrum is congestion cost.
*Similar observation in [Maille, Tuffin, Vigne `10]

Total Welfare: SPs Plus Commons

169
Price
Demand curve SP 2s revenue
consumer surplus Total price* P1*
P2* *maximizes revenue
`w (q)
q1* q2* qw* Quantity q (customers served)
May 2011
Total Welfare vs Commons Bandwidth

170
Total Welfare
Multiple identical incumbents Box-shape demand
commons bandwidth
Single Incumbent
171
Box inverse demand Linear latency
Demand curve
`(q)
`w (q)
Total price*
q1*
qw*
May 2011
Single Incumbent
172
Theorem: As the capacity C of the commons spectrum increases, there exists constants C1, C2 such that 1. For C<C1, total welfare is constant 2. For C1<C<C2 total welfare decreases 3. For C2<C total welfare increases
Total Welfare
May 2011
Why does total welfare decrease?

173
Adding the commons takes away customers from an SP, lowers congestion cost for remaining customers. The SP increases its revenue by raising its announced price:
Extracts additional surplus from its remaining customers Increases congestion in the commons (zero welfare)
For a single incumbent consumer welfare increases with bandwidth.
May 2011
Heterogenous Customers
174
Customers may have different trade offs between announced price and delay. Commons should help differentiate delay-tolerant from delay-sensitive customers. Does the previous scenario still occur?
May 2011
Heterogenous Customers
175
Customers may have different trade offs between announced price and delay. Commons should help differentiate delay-tolerant from delay-sensitive customers. Does the previous scenario still occur?
Yes, SP may raise price to offload low-end customers to white space (zero welfare), extract more surplus from high-end customers Here customer surplus can also decrease.
Example: Total Welfare

176
Total welfare
Single incumbent Box demands
Capacity
Observations
177
Decrease in total welfare is analogous to Braesss paradox in transportation networks. To avoid, must restrict entry to commons, or price entry. Possible enhancements of model:
More general demand/latency functions Investment costs Deployment geometry
May 2011
178
Concluding Remarks
Implications for wireless system design
May 2011
Spectrum Management: Two Views

179
Spectrum is abundant
It is just poorly managed Previous computation: 1 to 2 Mbps available in urban areas, but did not account for shrinking cell sizes, offloading traffic to WiFi, Femto-cells Unlicensed commons should meet future needs
Latency in commons band will be small enough so that previous inefficiencies do not arise.
May 2011
Spectrum Management: Two Views

180
Spectrum is abundant
It is just poorly managed Unlicensed commons should meet future needs
Latency in commons band will be small enough so that previous inefficiencies do not arise.
Spectrum will remain scarce

Applications will be generated to use new spectrum Shift to cheaper, spectrally inefficient technologies Unlicensed model is unsuitable for low frequencies
Managing Spectrum Scarcity

181
Bit Pipe (< 1 GHz)

Frequency
Commons (> 3 GHz)

Distributed interference management (random access) Inexpensive
Coordinated interference management Expensive infrastructure

May 2011

182
Bit Pipe (< 1 GHz)

Coordinated interference management Expensive infrastructure
Dynamic Spectrum Markets

Frequency
Commons (> 3 GHz)

Distributed interference management (random access) Inexpensive
May 2011
Local interference management Spectrum servers Rapid transactions


183
Bit Pipe (< 1 GHz)

Macro-cells Wide-area coverage Expensive service plans
Dynamic Spectrum Markets

Frequency
Commons (> 3 GHz)

Local coverage Femto-cells/WiFi Inexpensive
Micro-cells Limited coverage Local services
May 2011
Spectral Efficiency Objective

184
Wireless systems engineering has put a premium on spectral efficiency (bits/sec/Hz). Remarkable progress:
Practical coding techniques that achieve close to the Shannon bound Creation and exploitation of degrees of freedom: frequency (OFDM), multiple antennas (MIMO), cooperative relays Opportunistic resource allocation Advances in signal processing capabilities
Transition to Spectrum Abundance

185
Tradeoff between spectrum efficiency and power efficiency

Shift emphasis towards low-power, inexpensive wideband signaling techniques
Efficiency becomes limited by transaction costs

Distributed interference management (pricing, auctions, local exchange) Transparent (standardized?) mechanism for spot markets Wireless devices: frequency agile, compatible with spot market mechanism, used by multiple service providers
Many Remaining Challenges

186
Policy Economics
Engineering
Incentives, efficiency Market design
Interference management
May 2011
Many Remaining Challenges

187
Policy Economics Transition to spectrum markets??
Engineering
Incentives, efficiency Market design
Interference management
May 2011

Spec Markets v3

Uploaded by

Copyright:

Available Formats

You might also like

Spec Markets v3

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Spec Markets v3

Uploaded by

Copyright:

Available Formats

1

DySPAN Conference, Aachen, Germany

Spectrum Markets Tutorial, DySPAN Conference

Communications Engineering Rakesh Vohra

Spectrum Markets Tutorial, DySPAN Conference

High Profile Issue

Chicago Tribune April 15, 2011

Spectrum Markets Tutorial, DySPAN Conference

Why This Tutorial?

Timely and relevant

Markets for spectrum pose unique challenges/questions.

Interplay between economics and engineering issues

Spectrum Markets Tutorial, DySPAN Conference

Background and Motivation

Spectrum Markets Tutorial, DySPAN Conference

Limited Supply of Spectrum

good for cellular

Spark-gap transmitter (Tesla, 1893)

Spectrum Markets Tutorial, DySPAN Conference

Petabytes per month

Spectrum Markets Tutorial, DySPAN Conference

Regulation Prior to 1927: Open to All

Earliest uses of wireless for ship-to-ship, ship-to-shore communications.

Broadcast radio begins in 1921. Licenses issued by the Department of Commerce.

Spectrum Markets Tutorial, DySPAN Conference

Two Landmark Cases

Herbert Hoover, US Sec. of Commerce

Spectrum Markets Tutorial, DySPAN Conference

Spectrum Property Rights: A False Start

Tribune vs Oak Leaves Broadcasting, 1926

Spectrum Markets Tutorial, DySPAN Conference

Regulation since 1927: Command and Control

Regulation since 1927: Command and Control

The Spectrum Paradox

Spectrum is a scarce resource Spectrum is underutilized

Spectrum Markets Tutorial, DySPAN Conference

Spectrum is a Scarce Resource

Nearly $20B netted for 700 MHz auctions in 2008.

Problems with Command and Control

Spectrum Markets Tutorial, DySPAN Conference

Spectrum Markets Tutorial, DySPAN Conference

Role of government should be to minimize transaction costs.

Spectrum auctions finally introduced in the 1990s. Restrictions on use remain.

Problems with Command and Control

Engineering Approach to Spectrum Crunch

Add intelligence to mobile devices

Enables spectrum scavenging

Background and Motivation

Spectrum Markets Tutorial, DySPAN Conference

Spectrum Sharing Models

Exclusive use Commons Hierarchical

Spectrum Markets Tutorial, DySPAN Conference

Spectrum owned by government

Spectrum is private property

Spectrum Markets Tutorial, DySPAN Conference

Spectrum owned by government

Spectrum is private property Liberal licenses

Unlicensed Requires etiquette rules for sharing State-regulated