![Page 1: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/1.jpg)
Georgios B. Giannakis
Acknowledgments: Prof. G. Mateos, MURI Grant No. AFOSR FA9550-10-1-0567 NSF Grants ECCS-1202135, AST-1247885, ECCS-1002180
1
Cartography for Cognitive Networks
A Coruña, Spain June 22, 2014
![Page 2: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/2.jpg)
Network traffic growth
2
Communication networks today
Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012-2017
“Smart” devices multiply traffic
Large-scale interconnection of “smart” devices Commercial, consumer-oriented, heterogeneous
Projected IP traffic in Exabytes/month IP traffic is growing explosively
![Page 3: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/3.jpg)
3
Service diversification
Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012-2017
Residential services
Mobile services
![Page 4: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/4.jpg)
4
Dynamic network cartography
Network cartography: succinct depiction of the network state
Offers situational awareness of the network landscape
Network state Impact
Information dissemination
Routing Congestion control
Spectrum allocation
Risk analysis Security assurance
Network health monitoring
Interference
Link utilization
Path delays
Anomalous flows Traffic volume
Topology Coverage
QoS
Hierarchy/Reputation Vulnerability
Tool for statistical modeling, monitoring and management
Accurate network diagnosis and statistical analysis tools
Secure and stable network operation
Seamless end-user experience in dynamic environments
G. Mateos, K. Rajawat, and G. B. Giannakis ”Dynamic network cartography,” IEEE Signal Processing Magazine, May 2013.
![Page 5: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/5.jpg)
5
Tutorial outlook
RF cartography for cognition at the PHY
Map ambient RF power in space-time-frequency
Identify “crowded” regions to be avoided
0100
200300
400500
0
50
100
0
1
2
3
4
5
6
Time
Flows
Ano
mal
y vo
lum
e
Dynamic anomalography for IP networks
Reveal where and when traffic anomalies occur
Leverage sparse anomalies and low-rank traffic
Dynamic network delay and traffic cartography
Map network state via limited measurements
Monitor network health
![Page 6: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/6.jpg)
6
General context: NetSci analytics
General tools: process, analyze, and learn from large pools of network data
Clean energy and grid analytics Online social media Internet
Square kilometer array telescope Robot and sensor networks
Biological networks
![Page 7: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/7.jpg)
Dynamic network delay cartography Kriged Kalman filter predictor Optimal network sampling Empirical validation: Internet2 and NZ-AMP data
Unveiling network anomalies via sparsity and low rank
Network-wide link count prediction RF cartography for cognition at the PHY Conclusions and future research directions
7
Roadmap
![Page 8: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/8.jpg)
8
Why monitor delays? Motivating reasons Assess network health Fault diagnosis Network planning
Application domains Old 8-second rule for WWW Content delivery networks Peer-to-peer networks Multiuser games Dynamic server selection
Low delay variability
High delay variability
![Page 9: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/9.jpg)
Desiderata: infer delays from a limited number of end-to-end measurements only!
9 1Cooperative Association for Internet Data Analysis. [Online]. www.caida.org
Sprint
Qwest
AT&T
UUNet
C&W
Level 3
PSINet
Research issues and goal Few tools are widely supported, e.g., traceroute, ping Additional tools from CAIDA1 Require software installation at intermediate routers Useless if intermediate routers not accessible
![Page 10: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/10.jpg)
Inference task Measure on subset Predict on remaining paths
Problem statement Consider a network graph with links, nodes, and paths
10
Challenges Overhead: # paths ( ) ~ # nodes Heavily congested routers may drop packets
Q: Can fewer measurements suffice? Most paths tend to share a lot of links [Chua’06]
![Page 11: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/11.jpg)
11 D. B. Chua, E. D. Kolaczyk, and M. Crovella, “Network kriging,” IEEE J. Sel. Areas Communications, vol. 24, no. 12, pp. 2263–2272, Dec. 2006.
Network Kriging prediction Given , , universal Kriging:
To obtain , adopt a linear model for path delays
Sampling matrix S known (selected via heuristic algorithms)
![Page 12: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/12.jpg)
Wavelet-based approach [Coates’07] Diffusion wavelet matrix constructed using network topology Can capture temporal correlations, but for time slots High complexity ( ) cannot have
12 M. Coates, Y. Pointurier, and M. Rabbat, “Compressed network monitoring for IP and all-optical networks,” in Proc. ACM Internet Measurement Conf., San Diego, CA, Oct. 2007.
Spatio-temporal prediction
Q: Should the same set of paths be measured per time slot? Load balancing? Measurement on random paths?
Prior art does not jointly offer Spatio-temporal inference with online path selection, at low complexity
![Page 13: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/13.jpg)
Delay measured on path
13
Measurement noise i.i.d. over paths and time with known variance
Component due to traffic queuing: random-walk with noise cov.
Component due to processing, transmission, propagation: Traffic independent, temporally white, w/ cov.
Simple delay model
K. Rajawat, E. Dall’Anese, and G. B. Giannakis, “Dynamic network delay cartography,” IEEE Transactions on Information Theory, 2013.
![Page 14: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/14.jpg)
Goal: Given history find
14
Kriged Kalman Filter: Formulation Path measured on subset
KKF:
![Page 15: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/15.jpg)
15
KKF updates State and covariance recursions
KKF gain
Kriging predictor
![Page 16: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/16.jpg)
17
Which paths to measure? KKF can model and track network-wide delays
Error covariance matrix
Online experimental design: minimize
Log-det: D-optimal design (entropy of a Gaussian r. v.)
Practical sampling of paths? Optimal measurements? Criterion?
![Page 17: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/17.jpg)
Greedy algorithm
Submodular + monotonic greedy solution optimal [Nemhauser’78]
Operational complexity can be reduced further [Krause’11] Increments can be evaluated efficiently: with
Repeat S times
Algorithm
18 A. Krause, C. Guestrin. “Submodularity and its Applications in Optimized Information Gathering: An Introduction”, ACM Transactions on Intelligent Systems and Technology, vol. 2 (4), July 2011
Can be modified to handle cases when Each node measures delay on all paths – which S nodes to choose? All nodes measure delay on only one path – which path to choose?
![Page 18: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/18.jpg)
Measurements every minute for 3 days in July 2011 ~ 4500 samples
Empirical estimates; see e.g., [Myers’76] Techniques modified to handle measurements on subset of paths First 1000 samples used for training; 50 random paths used for training
Internet2 backbone 72 paths Lightly loaded
One-way delay measurements using OWAMP
Training phase employed to estimate ,
Empirical validation: Internet2
19 Data: http://internet2.edu/observatory/archive/data-collections.html
![Page 19: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/19.jpg)
20
True Kriging
Wavelet KKF
Network delay cartography (Internet2)
![Page 20: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/20.jpg)
21
KKF
Kriging
Wavelets
Normalized MSPE (Internet2)
“Optimal” paths
Random paths
Normalized MSPE
![Page 21: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/21.jpg)
22
Empirical validation: NZ-AMP Delays measured on NZ-AMP, part of NLANR project 186 paths, heavily loaded network
Measurements every 10 minutes during August 2011 ~ 4500 samples Round-trip times measured using ICMP, paths via scamper
Data: http://erg.wand.net.nz
![Page 22: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/22.jpg)
23
Random path selection
“Optimal” path selection
Normalized MSPE (NZ-AMP)
![Page 23: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/23.jpg)
Kriging
Wavelets
KKF
Scatter plots (NZ-AMP)
24
vs.
![Page 24: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/24.jpg)
25
Takeaways
Spatio-temporal inference useful for network health monitoring
Dynamic network delay cartography via Kriged Kalman filtering
Near-optimal path selection by utilizing submodularity
Empirical validation on Internet2 and NZ-AMP datasets
![Page 25: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/25.jpg)
Dynamic network delay cartography
Unveiling network anomalies via sparsity and low rank Traffic modeling and identifiability (De-) centralized and online algorithms Numerical tests
Network-wide link count prediction RF cartography for cognition at the PHY Conclusions and future research directions
26
Roadmap
![Page 26: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/26.jpg)
27
Traffic anomalies
Backbone of IP networks Traffic anomalies: changes in origin-destination (OD) flows
Motivation: Anomalies congestion limits end-user QoS provisioning
Objective: Measuring superimposed OD flows per link, identify anomalies by leveraging sparsity of anomalies and low-rank of traffic.
Failures, transient congestions, DoS attacks, intrusions, flooding
Time
Traf
fic V
ol.
![Page 27: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/27.jpg)
28
Model Graph G(N, L) with N nodes, L links, and F flows (F >> L) (as) Single-path per OD flow zf,t
є {0,1}
Anomaly
LxT LxF
Packet counts per link l and time slot t
Matrix model across T time slots:
0 0.2 0.4 0.6 0.8 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
f1
f2
l
![Page 28: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/28.jpg)
29
Low rank of traffic matrix
Z: traffic matrix has low rank, e.g., [Lakhina et al‘04]
Data: http://math.bu.edu/people/kolaczyk/datasets.html
![Page 29: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/29.jpg)
30
Sparsity of anomaly matrix
A: anomaly matrix is sparse across both time and flows
0 200 400 600 800 10000
2
4x 108
Time index(t)
|af,t
|
0 50 1000
2
4x 108
Flow index(f)
|af,t
|
Time
Flows
![Page 30: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/30.jpg)
31
Problem statement
Given and routing matrix , identify sparse when is low rank
fat but still low rank
(P1)
Low-rank sparse vector of SVs nuclear norm and norm
![Page 31: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/31.jpg)
32
Anomaly identification Change detection on per-link time series [Brutlag’00], [Casas et al’10] Spatial PCA [Lakhina et al’04] Network anomography [Zhang et al’05]
Prior art
Rank minimization with the nuclear norm, e.g., [Recht-Fazel-Parrilo’10]
Matrix decomposition [Candes et al’10], [Chandrasekaran et al’11]
Principal Component Pursuit
(PCP)
Observed Low rank Sparse
![Page 32: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/32.jpg)
33
Challenges and importance
not necessarily sparse and fat PCP not applicable
Important special cases
R = I : matrix decomposition with PCP [Candes et al’10] X = 0 : compressive sampling with basis pursuit [Chen et al’01] X = CLxρW’ρxT and A = 0 : PCA [Pearson 1901] X = 0, R = D unknown: dictionary learning [Olshausen’97]
LT + FT >> LT
X A Y
STRUCTURE
![Page 33: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/33.jpg)
34
Exact recovery Noise-free case
M. Mardani, G. Mateos, and G. B. Giannakis,``Recovery of low-rank plus compressed sparse matrices with application to unveiling traffic anomalies," IEEE Trans. Information Theory, 2013.
(P0)
Theorem: Given and , assume every row and column of has at most k<s non-zero entries, and has full row rank. If C1)-C2) hold, then with (P0) exactly recovers
C1)
C2)
Q: Can one recover sparse and low-rank exactly? A: Yes! Under certain conditions on
![Page 34: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/34.jpg)
35
Intuition
Exact recovery conditions satisfied if
r and s are sufficiently small
Nonzero entries of A0 are “sufficiently spread out”
Incoherent rank and sparsity- preserving subspaces
R satisfies a restricted isometry property
Remarks Amplitude of non-zero entries of A0 irrelevant Conditions satisfied for certain random ensembles w.h.p.
![Page 35: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/35.jpg)
36
Numerical validation Setup L=105, F=210, T = 420 R ~ Bernoulli(1/2) Xo = RPQ’, P, Q ~ N(0, 1/FT) aij ϵ {-1,0,1} w.p. {π/2, 1-π, π/2}
Relative recovery error
% non-zero entries (ρ)
rank
(XR
) (r)
0.1 2.5 4.5 6.5 8.5 10.5 12.5
10
20
30
40
50
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
rank
(X0)
[r]
[(s/FT)%]
![Page 36: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/36.jpg)
37
In-network processing Spatially-distributed link count data
Goal: Given local link counts per agent, unveil anomalies in a distributed fashion by leveraging low-rank of the nominal data matrix and sparsity of the outliers.
Challenge: not separable across rows (links/agents)
n
Centralized: Decentralized:
Agent 1
Agent N
Local processing and single-hop communications
![Page 37: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/37.jpg)
38
Separable regularization Key property
Lxρ ≥rank[X]
V’
W’ C
Separable formulation equivalent to (P1)
(P2)
Nonconvex; less variables:
Proposition 3: If stat. pt. of (P2) and , then is a global optimum of (P1).
![Page 38: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/38.jpg)
39
Distributed algorithm
M. Mardani, G. Mateos, and G. B. Giannakis, “In-network sparsity regularized rank minimization: Algorithms and applications," IEEE Transactions on Signal Processing, 2013.
Alternating-direction method of multipliers (ADMM) solver for (P2) Method [Glowinski-Marrocco’75], [Gabay-Mercier’76] Learning over networks [Schizas-Ribeiro-Giannakis’07]
Consensus-based optimization Attains centralized performance
![Page 39: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/39.jpg)
40
Benchmark: PCA-based methods Idea: anomalies increase considerably rank(Y)
Algorithm i) Form subspace via r-dominant left singular vectors of Y (resp. )
Assumes knowledge of r:=rank(X)
σi(X)
Index i
-----Y -----X
ii) Infer anomalies from
[Lakhina et al’04] For t = 1,…,T [Zhang et al’05] Sparse anomalies
![Page 40: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/40.jpg)
41
Synthetic data Random network topology
N=20, L=108, F=360, T=760 Minimum hop-count routing
0 0.2 0.4 0.6 0.8 10
0.2
0.4
0.6
0.8
1
False alarm probability
Det
ectio
n pr
obab
ility
PCA-based method, r=5PCA-based method, r=7PCA-based method, r=9Proposed method, per time and flow
0 0.2 0.4 0.6 0.8 1
0
0.2
0.4
0.6
0.8
1
---- True ---- Estimated
Pf=10-4
Pd = 0.97
![Page 41: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/41.jpg)
42
Internet2 data Real network data
Dec. 8-28, 2008 N=11, L=41, F=121, T=504
0 0.2 0.4 0.6 0.8 10
0.2
0.4
0.6
0.8
1
False alarm probability
Det
ectio
n pr
obab
ility
[Lakhina04], rank=1[Lakhina04], rank=2[Lakhina04], rank=3Proposed method[Zhang05], rank=1[Zhang05], rank=2[Zhang05], rank=3
Data: http://www.cs.bu.edu/~crovella/links.html
0100
200300
400500
0
50
100
0
1
2
3
4
5
6
Time
Pfa = 0.03 Pd = 0.92
---- True ---- Estimated
Flows
Ano
mal
y vo
lum
e
![Page 42: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/42.jpg)
43
Dynamic anomalography
M. Mardani, G. Mateos, and G. B. Giannakis, "Dynamic anomalography: Tracking network anomalies via sparsity and low rank," IEEE Journal of Selected Topics in Signal Processing, pp. 50-66, Feb. 2013.
Construct an estimated map of anomalies in real time
Streaming data model:
(Robust) subspace tracking Projection approximation (PAST) [Yang’95] Missing data: GROUSE [Balzano et al’10], PETRELS [Chi et al’12] Outliers: [Mateos-Giannakis’10], GRASTA [He et al’11]
Compressed “outliers” challenge identifiability
Goal: Given estimate online when is in a low-dimensional space and is sparse
![Page 43: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/43.jpg)
0
2
4CHIN--ATLA
0
20
40
Anom
aly am
plitu
de
WASH--STTL
0 1000 2000 3000 4000 5000 60000
10
20
30
Time index (t)
WASH--WASH
0
5ATLA--HSTN
0
10
20
Link
traff
ic lev
el DNVR--KSCY
0
10
20
Time index (t)
HSTN--ATLA
44
Online estimator Challenge: not separable across columns (time)
Approach: regularized exponentially-weighted LS formulation
---- estimated ---- real
o---- estimated ---- real
![Page 44: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/44.jpg)
45
Delay cartography
Internet2 data (Aug 18-22,2011) End-to-end latency matrix N=9, L=T=N; 20% missing data
Network distance prediction [Liau et al’12]
Data: http://internet2.edu/observatory/archive/data-collections.html
Relative error: 10%
Approach: distributed low-rank matrix completion
![Page 45: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/45.jpg)
Unveiling network traffic anomalies via convex optimization Leveraging sparsity and low rank
46
Takeaways
Reveal when and where anomalies occur
Exact recovery of low-rank plus compressed sparse matrices
Distributed/online algorithms with guaranteed performance
![Page 46: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/46.jpg)
Dynamic network delay cartography
Unveiling network anomalies via sparsity and low rank
Network-wide link count prediction Semi-supervised learning for traffic maps Batch and online processing Empirical validation: Internet2 data
RF cartography for cognition at the PHY Conclusions and future research directions
47
Roadmap
![Page 47: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/47.jpg)
48
A commuting conundrum Objective: map a “good” route for packet delivery
Application domains Transportation networks [Gastner-Newman’04] Communication networks [Soule et al’05] Sensor networks [Abrams et al’04]
Measure traffic at few roads/links only
![Page 48: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/48.jpg)
49
Model Graph G(N, L) with N nodes, L links, and F flows (F >> L) (as) Single-path per OD flow zf,t Packet counts per link l and time slot t
Incomplete, noisy measurements on a subset of links
0 0.2 0.4 0.6 0.8 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
f1
f2
l
![Page 49: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/49.jpg)
50
Problem statement
Goal: Given and historical data , find
Impact Ability to handle missing data Online prediction capturing spatio-temporal correlations Computationally-efficient link traffic prediction
P. A. Forero, K. Rajawat, and G. B. Giannakis, “Prediction of partially observed dynamical processes over networks via dictionary learning,” J. Machine Learning Research, 2013.
Prior art Traffic estimation [Zhang et al’05] Kriging [Chua et al’06], plus traffic modeling [Vaughn et al’10] Topology-driven basis expansion [Crovella-Kolaczyk’03], [Coates et al’07]
![Page 50: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/50.jpg)
51
Data-driven model of link counts Sparse representation of link counts
Notation:
Dictionary Learning (DL) [Olshausen-Field’97] Given , find dictionary (basis) and sparse
Q: How about DL from incomplete data ?
![Page 51: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/51.jpg)
52
Capturing spatial link dependence Auxiliary graph with vertices = links in G Edge weights = number of OD flows common to links Adjacency matrix: , graph Laplacian
1 1
1 1
1
G
Regularizers effect sparsity and smoothness over
Cost function to learn D
![Page 52: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/52.jpg)
53
Semi-supervised DL Semi-supervised Dictionary Learning (SSDL)
Given , find dictionary (basis) and sparse
SSDL biconvex, block-coordinate descent (BCD) solver Update via parallel entry-wise soft-thresholding Update each via QP + projection onto the Euclidean ball
Proposition: BCD’s iterates converge to a stationary point of SSDL
γ -γ
![Page 53: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/53.jpg)
54
Link load prediction Given and learnt dictionary , solve
Captures sparsity of and smoothness of link loads over
Predict based on
Scaling factor reduces bias in [Zou-Hastie’05]
![Page 54: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/54.jpg)
55
Batch processing summary
![Page 55: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/55.jpg)
56
Test case: Internet2 Internet2 measurement archive
Prediction improves as link load increases
Training phase – 30 links measured Operational phase – 30 links measured
L=54, T=2000
![Page 56: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/56.jpg)
57
Prediction error (Internet2) Normalized prediction error:
Q = number of columns of D; t0=2000
Gravity-based [Zhang et al’05]; Diffusion wavelets [Coifman-Maggioni’07]
SSDL outperforms competing alternatives
Training with 30 links Training with 50 links
![Page 57: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/57.jpg)
58
Online processing Capture temporal correlations on
Given and dictionary , solve
Predict based on
Dictionary update
![Page 58: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/58.jpg)
59
Real-time prediction (Internet2) Q=60, different values of the forgetting factor Measure traffic at 30 links only
SSDL-based tracker outperforms diffusion wavelets
![Page 59: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/59.jpg)
Prediction of network processes from incomplete observations
60
Takeaways
Spatial correlation of link counts via Laplacian regularization
Online algorithms capturing temporal correlations
Semi-supervised learning
Link count prediction based on dictionary learning
![Page 60: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/60.jpg)
Dynamic network delay cartography
Unveiling network anomalies via sparsity and low rank
Network-wide link count prediction RF cartography for cognition at the PHY Interference spectrum cartography Channel gain cartography
Conclusions and future research directions
61
Roadmap
![Page 61: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/61.jpg)
62
Fixed radio Policy-based: parameters set by
operators
Software-defined radio (SDR) Programmable: can adjust
parameters to intended link
Cognitive radio (CR) Intelligent: sense the environment
& learn to adapt [Mitola’00]
RX
TX
CR
Dynamic Resource Allocation
RF environment
- sensing - learning
- adapting to spectrum
What is a cognitive radio?
Cognizant transceiver: sensing Agile transmitter: adaptation Intelligent DRA: decision making Radio reconfiguration decisions Spectrum access decisions
![Page 62: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/62.jpg)
63
US FCC
Inefficient occupancy
0 1 2 3 4 5 6GHz
PS
D
Spectrum scarcity problem
Fixed spectrum access policies Useful radio spectrum pre-assigned
![Page 63: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/63.jpg)
64
PSD
f
SU
PU1
PU2 PU3
noise floor
PSD
f
SU PU1
PU2
PU3
Dynamical access under user hierarchy
Spectrum underlay Restriction on transmit power levels Operation over ultra wide bandwidths
Spectrum overlay Constraints on when and where to transmit Avoid interference to Pus via sensing and adaptive allocation
Primary users (PUs) versus secondary users (SUs/CRs)
Spectrum underlay Spectrum overlay
![Page 64: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/64.jpg)
65
Source: Office of Communications (UK)
Cooperative sensing for efficient sharing Multiple CRs jointly detect the spectrum [Ganesan-Li’06][Ghasemi-Sousa’07]
Benefits of cooperation Spatial diversity gain mitigates multipath fading/shadowing Reduced sensing time and local processing Ability to cope with hidden terminal problem
Limitation: existing approaches do not exploit space-time dimensions
![Page 65: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/65.jpg)
66 J. A. Bazerque and G. B. Giannakis, “Distributed spectrum sensing for cognitive radio networks by exploiting sparsity,'’ IEEE Transactions on Signal Processing, pp. 1847-1862, March 2010.
Cooperative PSD cartography Idea: CRs collaborate to form a spatial map of the RF spectrum
Goal: Find PSD map across space and frequency
Specifications: coarse approx. suffices
Approach: basis expansion of
![Page 66: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/66.jpg)
67
Modeling Transmitters
Sensing CRs
Frequency bases
Sensed frequencies
Sparsity present in space and frequency
![Page 67: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/67.jpg)
Data Rx-power at cognitive radio
Space-frequency basis expansion Find Tx-power of source s over frequency band
68
Estimate sparse to find PSD at
Sparsity-promoting regularization
![Page 68: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/68.jpg)
69
Exchange of local estimates
Scalability
Robustness Lack of infrastructure
Decentralized Ad-hoc
Centralized Fusion
center
Distributed recursive implementation
Consensus-based approach Solve locally
Constrained optimization using ADMM
![Page 69: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/69.jpg)
RF spectrum cartography sources
70
NNLS Lasso
As a byproduct, Lasso localizes all sources via variable selection
candidate locations, CRs
![Page 70: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/70.jpg)
71
Centralized sensing No fading Ns=25
4 CR Rxs
2 CR Txs
Simulated test: PSD map estimation
![Page 71: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/71.jpg)
72
“True” Tx spectrum
Sensed at the consensus step
Distributed consensus with fading
Starting from a local estimate, sensors reach consensus
![Page 72: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/72.jpg)
73 J. A. Bazerque, G. Mateos, and G. B. Giannakis, ``Group-Lasso on Splines for Spectrum Cartography,’’ IEEE Transactions on Signal Processing,’’ pp. 4648-4663, October 2011.
Spline-based PSD cartography Q: How about shadowing?
: unknown dependence on spatial variable x
Path-loss Shadowing
A: Basis expansion with coefficient functions
![Page 73: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/73.jpg)
74
Frequency basis expansion PSD of Tx source is
Basis expansion in frequency
Basis functions Accommodate prior knowledge raised-cosine Sharp transitions (regulatory masks) rectangular, non-overlapping Overcomplete basis set (large ) robustness
![Page 74: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/74.jpg)
75
Spatial PSD model Spatial loss function Unknown
Per sub-band factorization in space and frequency (indep. of )
BEM:
Goal: estimate PSD atlas as
![Page 75: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/75.jpg)
Nonparametric basis pursuit
Twofold regularization of variational LS estimator
(P1)
Available data: location of CRs measured frequencies
Observations
76
Avoid overfitting by promoting smoothness
Nonparametric basis selection ( not selected)
![Page 76: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/76.jpg)
77
Thin-plate splines solution
Unique, closed-form, finitely-parameterized minimizers!
Proposition 1: Estimates in (P1) are thin-plate splines [Duchon’77]
where is the radial basis function , and
Q2: How does (P1) perform basis selection?
Q1: How to estimate based on ?
![Page 77: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/77.jpg)
78
Lassoing bases (P1) equivalent to group Lasso estimator [Yuan-Lin’06]
Matrices ( and dependent)
i) ii) iii)
Group Lasso encourages sparse factors Full-rank mapping:
Proposition 2:
as
w/
Minimizers of (P1) are fully determined by
![Page 78: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/78.jpg)
79
Simulated test
S P E C T R U M M A P
Basis index Frequency (Mhz)
sources; raised cosine pulses sensing CRs, sampling frequencies bases; (roll off x center frequency x bandwidth)
Original Estimated
![Page 79: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/79.jpg)
80
Numerical test IEEE 802.11 PUs
PSD cartography
Maps estimated under fading + shadowing + overlapping bases
CRs
Original Estimated
Channel 6
Channel 11
![Page 80: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/80.jpg)
81
Real RF data
Frequency bases identified Maps recovered and extrapolated
IEEE 802.11 WLAN activity sensed
CRs
1
2
3
4
5
6
7
8
9
10
11
12
13
14
-50 -60 -40 -30 -20 -10 (dBi)
![Page 81: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/81.jpg)
82
Semi-supervised DL for PSD maps
S.-J. Kim and G. B. Giannakis, "Cognitive Radio Spectrum Prediction using Dictionary Learning," Proc. of Globecom Conf., Atlanta, GA, 2013.
Signal model
Rx-power measured by a few CRs
Batch formulation
Online algorithm via exponentially weighted criterion
![Page 82: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/82.jpg)
83
Numerical tests
![Page 83: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/83.jpg)
84
Recap: PHY sensing via RF cartography
S.-J. Kim, E. Dall’Anese, and G. B. Giannakis,“Cooperative Spectrum Sensing for Cognitive Radios using Kriged Kalman Filtering,” IEEE J. Selected Topics in Signal Processing, pp. 24-36, Feb. 2011.
Power spectral density (PSD) maps
Capture ambient power in space-time-frequency
Can identify “crowded” regions to be avoided
Channel gain (CG) maps
Time-frequency channel from any-to-any point
CRs adjust Tx power to min. PU disruption
![Page 84: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/84.jpg)
E. Dall’Anese, S.-J. Kim, and G. B. Giannakis, “Channel Gain Map Tracking via Distributed Kriging,” IEEE Trans. on Vehicular Technology, pp. 1205-1211, March 2011. 85
Kalman filtering
Kriging interpolation
Approach: spatial LMMSE interpolation (Kriging) + KF for tracking channel dynamics
Payoffs: tracking PU activities; accurate interference models; efficient resource allocation
Outlook: jointly optimal PHY CR sensing and access
Channel gain cartography CG after averaging small-scale fading (dB)
State-space model for shadowing
![Page 85: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/85.jpg)
86
Approach: low-rank matrix completion
Payoffs: global view of any-to-any CGs; real-time propagation metrics; efficient resource allocation
Outlook: kernel-based extrapolator for missing CR-to-PU measurements, or future time intervals
Any-to-any CG estimation Shadowing model-free approach
Slow variations in shadow fading
Low-rank any-to-any CG matrix
Estimated CG map
S.-J. Kim and G. B. Giannakis, “Dynamic Network Learning for Cognitive Radio Spectrum Sensing,” Proc. of Intl. Workshop on Comp. Advances in Multi-Sensor Adaptive Process., Saint Martin, 2013.
![Page 86: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/86.jpg)
87
Approach: blind dictionary learning
Payoffs: tracking PU activities; efficient resource allocation
Outlook: missing data due to limited sensing; distributed and robust algorithms
PU power and CR-PU link learning Reduce overhead in any-to-any CG mapping
Learn CGs only between CRs and PUs
Detection of PU activity
Estimated CG
Online detection of active PU transmitters
S.-J. Kim, N. Jain, and G. B. Giannakis, “Joint Link Learning and Cognitive Radio Sensing," in Proc. of Asilomar Conf. on Signals, Systems, and Computers, Pacific Grove, CA, Nov. 2011.
![Page 87: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/87.jpg)
88
Takeaways PHY layer spatiotemporal sensing via RF cartography Space-time-frequency view of interference and channel gains
PU/source localization and tracking
Parsimony via sparsity and distribution via consensus Lasso, group Lasso on splines, and method of multipliers
Identify idle bands across space Aid dynamic spectrum access policies
![Page 88: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/88.jpg)
Dynamic network delay cartography
Unveiling network anomalies via sparsity and low rank
Network-wide link count prediction RF cartography for cognition at the PHY Conclusions and future research directions
89
Roadmap
![Page 89: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/89.jpg)
90
The big picture ahead… Network path delay maps
0100
200300
400500
0
50
100
0
1
2
3
4
5
6
Time
Network traffic anomaly maps
Flows
Ano
mal
y vo
lum
e
RF interference cartography
Channel gain maps Coverage maps
Network cartography: succinct depiction of the network state
Vision: use atlas to enable spatial re-use, hand-off, localization, Tx-power tracking, resource allocation, health monitoring, and routing
![Page 90: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/90.jpg)
91
Concluding summary Dynamic network cartography
Global state mapping from incomplete and corrupted data
Statistical SP toolbox Sparsity-cognizant learning, low-rank modeling
Framework to construct maps of the dynamic network state Real-time, distributed scalable algorithms for large-scale networks
Path delay and link traffic maps Prompt and accurate identification of traffic anomalies PHY layer sensing in wireless CR networks via RF cartography
Kriged Kalman filtering of dynamical processes over networks Semi-supervised dictionary learning Distributed optimization via the ADMM Thank you!
![Page 91: Cartography for Cognitive Networks · Source: CISCO Visual Networking Index Global Mobile Data Traffic Forecast Update, 2012 -2017 “Smart” devices multiply traffic Large-scale](https://reader034.vdocuments.us/reader034/viewer/2022051822/5fec4d2f5302010a736ebbb3/html5/thumbnails/91.jpg)
92
Questions?
http://spincom.umn.edu University of Minnesota
Dr. J. A. Bazerque UofM
Dr. E. Dall’Anese UofM
Dr. P. A. Forero SPAWAR
Dr. S. J. Kim UofM
M. Mardani UofM
Prof. K. Rajawat IIT Kanpur