Dynamic Spectrum Access in 5G
Narayan B. Mandayam WINLAB, Rutgers University
[email protected] winlab.rutgers.edu/~narayan
1
WINLAB
What is 5G ?
Wide range of spectrum choices 100s of MHz to 100 GHz, Flexible BW, Licensed, Unlicensed
Wide range of device choices Low power, Mid-to-high power Low complexity, High complexity
Wide range of networking choices Mesh, Capillary, Phantom, HetNets
Wide range of application choices IoT, M2M, D2D V2V Wide range of QoS requirements Ultra low latency Very high data rate, Best effort
Wide range of networking paradigms ICN, MF, NOM, User-centric
5G: Anything you want it to be!
5G: Academic’s dream! 2
WINLAB
5G DSA: What’s out there ?
Three distinct approaches to DSA have been proposed Agile/cognitive radio – autonomous sensing at radio devices to avoid interference
Spectrum Access System (SAS) – centralized Database to provide visibility of potentially interfering networks and/or global assignment Distributed inter-network collaboration – peering protocols to support decentralized spectrum assignment algorithms 2. SPECTRUM SERVER 3. DECENTRALIZED NETWORK COLLABORATION (Collocated Networks)
1. AGILE RADIO Spectrum Server RF sensing
Net B
Internet RF sensing
Net A
Query/ Assignment
AP/ BS A
Distributed Algorithm
Net C
AP/ BS B
WINLAB
5G DSA: Agile radio
Cognitive radio networks require a large of amount of network (and channel) state information to enable efficient Discovery, Self-organization Resource Management Cooperation Techniques
Cost of Cooperation?
Scalability? PHY A
PHY C
PHY B
Control (e.g. CSCC)
Multi-mode radio PHY Ad-Hoc Discovery & Routing Capability
Functionality can be quite challenging! 4
WINLAB
5G DSA: Spectrum Access System (SAS) SPECTRUM SERVER
Internet
Query/ Assignment
Primarily in 3.5 GHz spectrum Small Cells for Cellular Coexistence with Navy Radar
Design Principles and Architecture Registration with Spectrum Server/Database Tiering and Prioritization of users Protect Incumbents Wide range of technical issues related to access Licensed Shared Access Generalized Authorized Access Control and Network State Information Radio and Network parameters exposed Coordination across databases Monitoring and Enforcement
5
WINLAB
5G DSA: Network Cooperation SAVANT: Spectrum Access Via Inter-Network Cooperation Focus on decentralized architecture for sharing spectrum info
Parallels with BGP exchange of route information between peers Architecture enables regional visibility for setting radio parameters Further, networks may collaborate to carry out logically centralized optimization for max throughput subject to policy/technology constraints
Local Adaptation to Observed Spectrum Use
Cooperative Regional Optimization of Radio Parameters
Net B
Net A
Distributed Algorithm
Radio MAP Information Exhange
Net C
*Supported by NSF EARS grant CNS 1247764 WINLAB/Princeton Project
WINLAB
SAVANT: Inter-Network Protocol Architecture involves two protocol interface levels between independent wireless domains: • Lower layer for sharing aggregate radio map using technology neutral parameters • Higher layer for negotiating spectrum use policy, radio resource management (RRM) algorithms, and controller delegation
WINLAB
Elephant in Room: WiFi Smart Phone growth is the U.S. from 2013 to 2015 is ~300% Smartphone data consumption in 2015 ~10 GB/user/month ~85% over WiFi and ~15% over Cellular
WiFi AP density in cities ~100-200 per sq km % of Enterprise/SP APs
25 San Francisco New York Chicago Boston
20 15 10 5 0
01/2009
01/2010
01/2011
01/2012
01/2013
Date
Licensed Assisted Access (LAA) and other cooperative methods including aggregation/integration with WiFi
WINLAB
8
5G DSA: Technical Challenges
Noncontiguous Spectrum Transmission
Control Plane Design
TX power is no longer “King”!
Scalability, Performance
Distributed/Hybrid Algorithms for Spectrum Coordination
Stability, Convergence of Algorithms
9
WINLAB
Case for Noncontiguous Transmission - I C
?
• Three available channels
1 23 • Node A transmits to node C via node B. • Node B relays node A’s data and transmits its own data to node C. • Node X, an external and uncontrollable interferer, transmits in channel 2.
?
B
X
A
If we use max-min rate objective and allocate channels, node B requires two channels; node A requires one channel Scheduling options for Node A and Node B? 10
2
Case for Noncontiguous Transmission - II #1: Contiguous OFDM
#2: Multiple RF front ends
C
C
1 3
1 3
A • Transmission in link BC suffers interference in channel 2
B
B
2 X
3
Nulled Subcarrier
C
12
B
#3: Non-Contiguous OFDM (NC-OFDMA)
2
2 X
2
X
2 A
A
NC-OFDM accesses multiple • Spectrum fragmentation limited by number of radio fragmented spectrum chunks with single radio front end front ends 11
11
NC-OFDM Operation Non-Contiguous OFDM Nulled Subcarrier
X[2] = 0 X[1] X[1] X[3]
Serial to Parallel
IFFT
x[1] x[2] x[3]
AP
1 3 Parallel to Serial
D/A
Modulation
B
X[3]
• Node B places zero in channel 2 and avoids interference • Node A, far from the interferer node X, uses channel 2. •
Both nodes use better channels.
2 X
2 A
NC-OFDM accesses multiple fragmented spectrum chunks with single radio front end
• Node B spans three channels, instead of two. • Sampling rate increases. 12
Resource Allocation in Noncontiguous Transmission Benefits: Avoids interference, incumbent users Uses better channels Each front end can use multiple fragmented spectrum chunks
Challenges: Increases sampling rate Increases ADC & DAC power Increases amplifier power Increases peak-to-average-power-ratio (PAPR)
Multiple RF Front Ends vs Single RF Front End ? Centralized, Distributed and Hybrid algorithms for carrier and forwarder selection, power control ? 13
Spectrum Allocation under Interference and Spectrum Span Constraints Controller
Available channels
How to allocate noncontiguous channels subject to ADC/DAC power constraints?
Radio nodes Interference nodes
Maxmin Rate Allocation (Integer Linear Program)
A
n1
n2
n3
n4
B
n5
n6
n7
n8
L1
n1-n2
L2
n3-n4
L3
n5-n7
L4
n6-n8
C
Control Plane Design: Noncontiguous Transmission CDMA is back!
Short PN-seq
Long PN-seq
Control Channel Data
16
WINLAB
Experimental Results from ORBIT testbed Result 1: Spectrum assignment while minimizing span of assigned subcarriers (reduces ADC/DAC power consumption)
USRP ORBIT testbed
Reassigned subcarriers with minimal loss (< 10%)of throughput
Result 2: Reliable timing and frequency recovery from underlay control channel in the presence of primary transmissions correct timing instance peak indicating timing instance detection peak detection threshold
Network Setup: • Multiple p2p secondary links operating in the presence of a primary transmission • 1 MHz BW, 64-subcarrier NC-OFDM with CDMA-based underlay (spreading sequence length 40-160) • Underlay to noise ratio ~ 0 dB, primary transmission to noise ratio ~ 10 dB
Result 3: Control channel BER as a function of primary signal strength with underlay to noise ratio set to 0 dB; Control channel rate = 30 kbps Primary Signal SNR
BER
3 dB
< 1e-3
6 dB
6.3*1e-3
7.7 dB
2.6*1e-2
9.2 dB
9.2*1e-2
Network Coordination: LTE/WiFi Conventional LTE
Conventional Wi-Fi
Spectrum
Exclusive licensed
Shared unlicensed
Operation technique
OFDMA: channel hopping over time to exploit good channel condition
CSMA/CA: Channel sensing before transmission to avoid packet collision
Controller entity
A single licensed carrier
No common controller
Advantage
Packet efficient
Cost effective, fair sharing
WINLAB
18
Formulating LTE/WiFi Cooperation as an Optimization problem Objective: Downlink power control optimization using Geometric Programming Maximize sum-throughput across Wi-Fi and LTE
maximize
ab 1 S 1 S l j w i w i i
iW
l
jL
subject to w (1 log 2 w Si ) ri ,min , i W ,
Minimum SINR requirement for data rate transmission
l (1 log 2 l S j ) rj ,min , j L,
Pk Gik Pj Gij N 0 C , i W ,
CCA threshold requirement at Wi-Fi
jL
kM ib
0 Pi Pmax , Controlling variables : Pi ,
i W , L
Range of Tx power
i W , L
Tx power
where Si : SINRatlinki
b 1 1 | M |, M
ai 1 1 | M ia | , M ia : i
b i
b i
Set of Wi-Fi APs in the CSMA range of AP
:
Set of Wi-Fi APs in the interference range of AP
i, i W i, i W
WINLAB
19
LTE/WiFi Scenario • UE – Associated AP: either Wi-Fi or LTE link, interfering AP is of
other technology • Varying parameters: • dA = distance(UE, Associated AP)
• dI = distance(UE, interfering AP) • Assuming UE at (0,0): if interfering AP on the (1) –X axis, dI = -| dI|,
(2) +X axis, dI = +| dI| • Reason: inter-AP distance matters due to WiFi clear channel assessment +| -| dI| dI | dA
+x-axis
- x-axis Interfering APj
(0,0) UEi
Interfering APj
Associated APi
WINLAB
20
Example LTE/WiFi Coordination Results – Performance of LTE 100
100
100 100 60
50
0
-50
-100
20
40
60
80
AP-UE dist [m] No coordination
100
Interfering AP-UE dist [m] Interfering AP-UE dist [m]
Interfering AP-UE dist [m]
Interfering AP-UE dist [m]
60 5050 40
0
20 -50 10
-100
505050
5050
40
4040
30
30
20
40
60
80
AP-UE dist [m] Power control optimization
100
6060
0 0
3030
20 -50-50
2020
10
1010
-100 -10020 20 4040 6060 8080 100 100
AP-UE dist [m] AP-UE dist [m]
Time division channel access optimization
Sagari, Baystag, Saha, Seskar, Trappe & Raychaudhuri, “Coordinated Dynamic Spectrum Management of LTE-U and WiFi Networks” IEEE Dyspan 2015 (to apear)
WINLAB
21
Example LTE/WiFi Coordination Results: Performance of WiFi 100 100
100
100
0
-50
-100
20
40
60
80
AP-UE dist [m] No coordination
100
Interfering AP-UE dist [m] Interfering AP-UE dist [m]
Interfering AP-UE dist [m]
Interfering AP-UE dist [m]
50
6060
60
60 5050 40
0
5050
40
4040
30
30 20 -50 10
-100
505050
20
40
60
80
AP-UE dist [m] Power control optimization
100
0 0
3030
20 -50-50
2020
10
1010
-100 -10020 20 4040 6060 8080 100 100
AP-UE dist [m] AP-UE dist [m]
Time division channel access optimization
Sagari, Baystag, Saha, Seskar, Trappe & Raychaudhuri, “Coordinated Dynamic Spectrum Management of LTE-U and WiFi Networks” IEEE Dyspan 2015 (to apear)
WINLAB
22
End-User Behavior and Radio Resource Management Differentiated Pricing Higher speed
Increasing significance of end-user decisions Can we influence enduser behavior and improve RRM?
Lower guarantee
How does uncertainty in the service affect enduser decisions and the network?
Figure from www.fcc.gov Measuring Broadband America 2
WINLAB
Prospect Theory: An Alternative to Expected Utility Theory for Modeling Decision Making Probability Weighting Effect Framing Effect
“Overweigh” low probabilities “Underweigh” moderate and high probabilities
24
Losses usually “loom larger” than gains
WINLAB
Prospect Pricing for Radio Resource Management
User preferences, biases can be “mitigated” by pricing
Can be used to improve RRM
Under EUT, loss is 0
Deviation from EUT results in loss, pricing reduces loss
Psychophysics Experiments
Measured Probability Weighting Function for video QoS
Yang, Park, Mandayam, Seskar, Glass and Sinha “Prospect Pricing in Cognitive Radio Networks” IEEE Trans. on Cognitive 25 Communication Networks, To Appear
WINLAB
Rural Broadband: LTE-U based Backhaul in TVWS with Local WiFi Access
LTE-U BS 1 Coverage Area
BS 1 WiFi Coverage Area
BS 4 WiFi Coverage Area
BS 2 WiFi Coverage Area
BS 5 WiFi Coverage Area
BS 3 WiFi Coverage Area
BS 6 WiFi Coverage Area
BS 7 WiFi Coverage Area LTE-U BS 6 Coverage Area
LTE-U BS 4 Coverage Area
BS 8 WiFi Coverage Area
Backhaul Tower with WS Radio and WiFi AP for local distribution
Tower with Fiber Access
LTE-U Link
WINLAB
LTE in TVWS: FCC Guidelines LTE Attributes
FCC TVWS Rules for 6 MHz Channel
LTE eNodeB DL Transmitter Power
2W EIRP for LTE FDD 3 MHz
LTE eNodeB UL Transmit Power
2W EIRP for LTE FDD 3 MHz
LTE eNodeB Transmitter Height
30 meters HAAT
LTE eNodeB Antenna Gain
0 dBi
LTE in TVWS: Simulation throughputs with multiple channels LTE FDD Throughput with multiple TVWS channels vs InterDL TP @ 1 TVWS Tower Distance
Chan DL TP @ 2 TVWS Chan DL TP @ 3 TVWS Chan DL TP @ 4 TVWS Chan DL TP @ 5 TVWS Chan DL TP @ 6 TVWS Chan DL TP @ 7 TVWS Chan 18 Mbps Load
Throughput (Mbps)
150
100
50
35 Mbps Load
0 0
5
10 Distance (km)
15
20
Estimated Rural Demand Mean Estimate of Rural Demand
Generic Scenario : E.g. Wichita, KS
• Area: 423 square km2 • Population: 385,577 (2012 Census) [1] • Available white space for fixed devices [2] Location (MHz)
57
79 85
491 527 533 671
Maxmin Rate Backhaul
3 Fiber BS can cover 144 sq km
Data Rate (Mbps)
Throughput vs Demand for Various Cell Size 100 90 80 70 60 50 40 30 20 10 0
91.31
91.31 85.02
60.87
60.87 54.78 46.86
36.52
36.52
36.52
26.36 15.98
15.98
15.98
11.72
lnter-tower distance = 2 Km
Traffic Demand
A = {5}
lnter-tower distance = 3 Km
A = {1,9}
lnter-tower distance = 4 Km
A = {1,5,9}
A={1,3,7,9}
References • R. Kumbhkar, T. Kuber, G. Sridharan, N. B. Mandayam, I. Seskar, “Opportunistic Spectrum Allocation for Max-Min Rate in ” DySPAN 2015, October 2015 • S. Sagari, Baystag, D. Saha, I. Seskar, W. Trappe, D. Raychaudhuri, “Coordinated Dynamic Spectrum Management of LTE-U and WiFi Networks” DySPAN 2015, October 2015 • S. Pattar, N. B. Mandayam, I. Seskar, J. Chen, Z. Li, “Rate Optimal Backhaul and Distribution using LTE in TVWS” SCTE Cable-Tec Expo’15, October 2015 • R. Kumbhkar, M. N. Islam, N. B. Mandayam, I. Seskar, “Rate Optimal design of a Wireless Backhaul Network using TV White Space,” COMSNETS 2015, January 2015 Y. Yang, L. Park, N. B. Mandayam, I. Seskar, A. Glass and N. Sinha “Prospect Pricing in Cognitive Radio Networks” IEEE Trans. on Cognitive Communication Networks, To Appear
31
Acknowledgments • U.S. National Science Foundation
• Office of Naval Research • WINLAB Collaborators: Ratnesh Kumbhkar, Gokul Sridharan, Neel Krishnan, Ivan Seskar, Dipankar Raychaudhuri, Arnold Glass
• Qualcomm: Nazmul Islam • NRL: Sastry Kompella
32