cognitive radio – a necessity for spectrum poolingfriedrich k. jondral paris, march28, 2006...
TRANSCRIPT
-
Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
Friedrich K. Jondral
Paris, March 28, 2006
Cognitive Radio –A Necessity forSpectrum Pooling
-
2Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
Spectrum Utilization Measurements(550-1000MHz)
density of the
time between
arrivals
dBs-1
density of the
time between
arrivals
dBs-1
electric
field strength
dBµµµµV/m
electric
field strength
dBµµµµV/m
Lichtenau (Germany), September 2001
-
3Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
Synopsis
• Spectrum Pooling
• The Licensed User (LU) System
• HIPERLAN/2 System Overview
• Rental User (RU) System: Physical Layer Issues
• Cognitive Radio
• Rental User (RU) System: Detection and Signaling
-
4Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
Spectrum Pooling
Vision:
Usage of free capacities in licensed frequency bands:
Licensed Users (LUs) lease out spectrum to Rental Users (RUs)
FDMA / TDMA LU system and HIPERLAN/2 RU system:
Two different radio systems →→→→ Coexistence in the same frequency region ?!
time
frequency
-
5Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
• European Wireless Local Area Network Standard
HIPERLAN/2 System Overview
Frame CHannelFCH
Long transport CHannelLCH
Short transport CHannelSCH
Random CHannelRCH
Protocol Data UnitPDU
Access feedback CHannelACH
Broadcast CHannelBCH
Medium Access ControlMAC
MAC-FrameMAC-Frame MAC-Frame
2 ms
flexible Lengths
BroadcastPhase
DownlinkPhase
UplinkPhase
RandomAccess
Pream
ble
Preamble
Pream
ble
Preamble
SCHs LCHsSCHs LCHs SCHsLCHsSCHsLCHs SCHsLCHsSCHs LCHs . . .
Slotted Aloha
Pream
ble
BCH ACHDownlink
PDU StreamsUplink
PDU StreamsFCH RCH
-
6Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
HIPERLAN/2 System Overview
• Physical Layer: OFDM
Transmitter structure and spectrum
serial-to-parallel conversion
parallel-to-
conversion
serial
Cod
Cod
Cod
IDFT (FFT)
D/ARF-Mod.
1
1
1
m
m
m
d (l)1
d (l)r
x (t)MT
x (t)RF
d (l)N
bitsequence
. . .
. . .
. . .
. . .
. . .
. . . . . .
∆f
ff−3 f3f−2 f2f−1 f0 f1
-
7Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
HIPERLAN/2 System Overview
4 µs (= TU + TG)OFDM Symbol
16 TA = 0.8 µs Guard Interval TG
64 TA = 3.2 µs (= 1/ ∆f)Usable Part of an OFDM Symbol TU
50 ns (= 1/20 MHz)Time Distance Between two FFT Input Samples
TA
312,5 kHz (= 20 MHz / 64)OFDM Subcarriers‘ Distance ∆f
20 MHzFFT Bandwidth
52 (= KD + KP)Total Number of OFDM Subcarriers Used K
4Number of Pilots KP
48Number of Data Subcarriers KD
64Total Number of Subcarriers
-
8Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
The Licensed User System
1. Embedding a RU cell into a LU cell
(Hot Spot Scenario)
RUcell
LU system
2. Channel pattern and Occupancy Vector (OV)
LU channel of width 4∆f
carrier distance ∆f
OV = ( 0 1 0 0 1 0 0 0 1 1 0 0 0 0 1 0 )
LU LU LU LU LU
-
9Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
The Licensed User System
4. No Carrier Sense Medium Access (CSMA) in the LU system.
3. No LU signaling channel →→→→ LU detection necessary for the RU system
• hidden / exposed station problem
� The detection has to cover the whole sphere of influence of the RU cell
� Access delay for the LUs
� Waiting period
frequency f2
frequency f1
LU1 LU2 RUdetection area of the RU
-
10Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
RU Physical Layer Issues
• RU system synchronization
– Necessity of a new preambel concept
– Optimum positioning of pilots?
• Interference reducing measures
– Disturbances to the LU system caused by the sin(x)/x-shaped OFDM spectrum– Disturbances to the RU system by FFT leakage
caused by the non orthogonality of the LU signals
� LU detection and signaling
• Optimum detection?
• Quality of detection necessary for coexistence?
• OV transmission calls for a new protocol
-
11Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
Detection and Signaling
The engaged / idle decision has to be done by the RU system for each
LU channel within the pool
���� The frequency resolution is realized by the anyhow existing FFT:
1. Sampling of the signal s(k) band limited to the pool width
2. FFT for 64 samples at a time. The process is repeated n times.
3. The spectrum values belonging to one LU channel are integrated
into a vector z .
4. Decision based on z and on an optimality criterion.
-
12Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
Signal and Detector Model
Structure of the detector in the rental user’s stationStructure of the detector in the rental user’s station
Bandwidth of one licensed user's subband (a·∆f )
-
13Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
Joint Conditional PDFs
NLOS assumption and central limit theorem yield:NLOS assumption and central limit theorem yield:
Only noise in case of licensed user’s absence:Only noise in case of licensed user’s absence:
Likelihood ratio test according to the NEYMAN-PEARSON-criterion:Likelihood ratio test according to the NEYMAN-PEARSON-criterion:
-
14Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
Detector for Uncorrelated Samples
Under the condition , the PDF simplifies to:Under the condition , the PDF simplifies to:
With these conditional PDFs the likelihood ratio results in:With these conditional PDFs the likelihood ratio results in:
Considering the likelihood ratio as a transformation of random variables:Considering the likelihood ratio as a transformation of random variables:
-
15Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
Worst Case Consideration
Maximum deviation from the model, if real/imaginary parts are fully correlated:Maximum deviation from the model, if real/imaginary parts are fully correlated:
Hence, all are identical and can be represented by :Hence, all are identical and can be represented by :
can be interpreted as a concatenated random variable:can be interpreted as a concatenated random variable:
-
16Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
Detection Probability
is the marginal PDF with respect to a randomized distribution:is the marginal PDF with respect to a randomized distribution:
The detection probability can be calculated as:The detection probability can be calculated as:
-
17Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
Receiver Operating Characteristic (ROC)
Worst-case ROC for
varying .
Worst-case ROC for
varying .
-
18Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
Distributed Detection: Single Detection
Every rental user station
conducts individual detection
Every rental user station
conducts individual detection
-
19Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
Distributed Detection: Collection
Rental user stations transmit
their results to the AP
Rental user stations transmit
their results to the AP
-
20Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
Distributed Detection: Broadcast
After an OR-combination, the AP
broadcasts the new allocation vector
After an OR-combination, the AP
broadcasts the new allocation vector
-
21Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
Gain by Distributed Detection
Overall Probability of detection if at least one of the N stations detects the LU access:Overall Probability of detection if at least one of the N stations detects the LU access:
Behavior of the overall false alarm probability if is
predetermined:
Behavior of the overall false alarm probability if is
predetermined:
-
22Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
Spectrum Efficiency
• Impact of the false alarm probability PF on the RU system efficiency
relative spectrumutilizationbyLUs
as detectedbytheRU system
relative spectrum utilization by LUs
-
23Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
Regulation
� Today “spectrum“ is regulated by governmental agencies, e.g. the
American Federal Communications Commission (FCC) or the German
Regulierungsbehörde für Telekommunikation und Post (RegTP)
� “Spectrum“ is assigned to users or licensed to them on a long term basis
normally for huge regions like whole countries
� Doing so, resources are wasted
� Vision: Resources are assigned where and as long as they are needed,
spectrum access is organized by the network (i.e. by the end users)
-
24Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
Self Regulation
� Wireless LANs (IEEE 802.11x)
ISM band: 2400 – 2483.5 MHz
WLAN band: 5150 – 5350 MHz and 5470 – 5725 MHz
� Ultra Wide Band−40
−55
−70
−45
−60
100
101
−50
−65
Frequency in GHz
UWB EIRP Emission Level in dBm
fc greaterthan 3.1 GHz
fc less than960 MHz
Part 15 Limit
-
25Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
Advanced Spectrum Management
� Spectrum reallocation: The reallocation of bandwidth from government or other
long-standing users to new services such as mobile communications, broadband
internet access, and video distribution.
� Spectrum leases: The relaxation of the technical and commercial limitations on
existing licensees to use their spectrum for new or hybrid (for example, satellite and
terrestrial) services and granting most mobile radio licensees the right to lease their
spectrum to third parties.
� Spectrum sharing: The allocation of an unprecedented amount of spectrum that
could be used for unlicensed or shared services.
According to
G. Staple, K. Werbach: The End of Spectrum Scarcity. IEEE Spectrum, March 2004, pp. 41-44
-
26Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
Cognitive Radio
� A Cognitive Radio (CR) is an SDR that additionally senses its
environment, tracks changes and reacts upon its findings.
� A CR is an autonomous unit in a communications environment. In
order to use the spectral resource most efficiently, it has to
- be aware of its location
- be interference sensitive
- comply with some communications etiquette
- be fair against other users
- keep its owner informed
-
27Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
Cognition Cycle
A necessary condition for highest flexibility in mobile communications is a general rethinking in spectrum allocation: Open access
In order to make open access feasible Cognitive Radios are necessary.
Immediate Urgent Normal
ACT
Outside
World
New
States
Prior
States
OBSERVE LEARN
DECIDE
PLAN
Generate
Alternatives
Evaluate
Alternatives
ORIENT
Establish Priority
Receive a Message
Read Buttons
Send a MessageInitiate Process(es)
Register to
Current Time
Pre-Process
Parse
Save Global States
Allocate Resources
Set Display
Infer on Context
Hierarchie
Joseph Mitola III: Cognitive Radio – An Integrated
Agent Architecture for Software Defined Radio.
KTH Stockholm, 2000
-
28Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
CR Properties
Mitola's cognition cycle is very general. The properties of cognitiv radios may be divided into two groups
user centric properties (support functions like finding an appropriate restaurant, recommendation of a travel route, supervision of apointments, . . .)
technology centric properties
- spectrum monotoring
- localization
- awareness of processing capabilities (partitioning and scheduling of processes)
- information and knowledge processing
-
29Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
Technologies to be Implemented
� A CR carries location (e.g. GPS or Galileo) sensors.
� It has to monitor its spectral environment, for example by using a real time
broadband FFT.
� In order to track its location or the spectral environment‘s development, it has to
implement learning and reasoning algorithms.
� When complying with a communications etiquette it has to listen before talk as well
as to prevent the disturbance of hidden stations.
� In order to be fair it has to compromise its own demands with the demands of other
users, most probably making decisions in a competitive environment using the
results of game theory.
� It has to contact its owner via a highly sophisticated man-machine-interface.
-
30Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
Technology Centric CR
Information & Knowledge Processing
ControlSpectrum
MonitoringLocalizationCore
-
31Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
TeC-CR
Station BStation A
Transmission
Channel
measurement
and modeling
Reception
Monitoring
Spectral
Environment
Available
Channel Capacity
Transmission Power
and Spectrum
Management
Interference
Temperature
Spectrum Holes
Noise Statistics
Traffic Volume
RF Signals
Channel
measurement
and modeling
Reception
Monitoring
Spectral
Environment
Available
Channel Capacity
Transmission Power
and Spectrum
Management
Interference
Temperature
RF Signals
Spectrum Holes
Noise Statistics
Traffic Volume
-
32Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
Conclusions
1. Advanced spectrum management will be a hot topic in future
research on wireless networks.
2. Unlicensed (ISM, WLAN) as well as secondary (UWB) spectrum
usage are already under way.
3. First spectrum sharing strategies (e.g. spectrum pooling) are
under investigation.
4. Advanced spectrum management calls for new developments in
networking and terminal devices:
Intelligent Networks & Cognitive Radios
-
33Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
Questions ?
-
34Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
SPECTRUM UTILIZATIONMEASUREMENTS (50-550 MHz)
density of the
time between
arrivals
dBs-1
density of the
time between
arrivals
dBs-1
electric
field strength
dBµµµµV/m
electric
field strength
dBµµµµV/m
Lichtenau (Germany), September 2001
-
35Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
SPECTRUM UTILIZATION(50 MHz-1GHz)
Lichtenau (Germany), September 2001
dB
µµ µµV/m
-
36Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
BOOSTING PROTOCOL (1)
� Signaling RUs + BooSs →→→→ AP ?
• Boosting Protocol
Presently valid OV:LULU LU LU
New constellation
(after the detection
phase) :LULU LU LU
newly occupiedLU channel
releasedLU channel
-
37Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
BOOSTING PROTOCOL (2)
Signaling of newly
occupied channels
Mapping phase
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
LU LU LU LU
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
LU LU LULU
2 2 29 15 15 1510 10 109 9
LU LU LU
Boosting
-
38Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
BOOSTING PROTOCOL (3)
Signaling
of further occupied
LU channels
New OV
(to be signalized
by the AP) LU LU LULU
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
2 2 29 15 15 1510 10 109 9
not to be considered
-
39Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
SUMMARY (1)
� Divided detection combined with a boosting protocol and robust Occupancy
Vector signaling solves the LU detection problem and leads to a common
base of the Physical Layer (PHY).
Parametrized PHY (OV),
number of usable carriers is known
to the RU system‘s MAC
Parametrized PHY (OV),
number of usable carriers is known
to the RU system‘s MAC
MAC frame MAC frame MAC frame
detection
phase
boosting
phase
broadcast
phaseP P
2 ms
-
40Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
SUMMARY (2)
� The RUs system‘s efficiency is mainly determined by the interference reducing
measures!
additionaly deactivated carrieradditionaly deactivated carrier
LU LU LU LU LUf
Power spectrum
density Sxx (f)of a single
OFDM-carrier
Power spectrum
density Sxx (f)of a single
OFDM-carrierOFDM channel ∆∆∆∆f
interference to the
neighboring channel
-
41Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
OUTLOOK
• Integration of the results into our OMNeT++ software demonstrator.
• Simulation with respect to all effects and with realistic channel models
→→→→ Tuning of the free parameters
• Adaptive modulation for optimal use of disturbed channels(FFT leakage)
• MAC layer: Investigation of scheduling algorithms particulary resistent
against bandwidth variations
-
42Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
ACKNOWLEDGEMENT
• Our work on spectrum pooling has been supported by the
German Federal Ministry of Research and Technology
under grant No. IT-KT-ITKT01338200-01BU152.
• The work was mainly done by the INT staff members
- Dipl.-Ing. Timo Weiß
- Dipl.-Ing. Fatih Capar
- Ihan Martoyo, M.Sc.
and by our graduate students
- Jörg Hillenbrand
- Albert Krohn
-
43Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
MULTIDIMENSIONALGAUSSIAN DENSITY
� There is No Line of Sight ( NLOS ) from the LU to be detected to the measuring RU
-
44Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
TOTAL DENSITY
• Noise component density (AWGN)
f zNN
n
T
N
bg c h=1
2 exp -
2πσ σz z2 2
FHG
IKJ
• Resulting density
Convolution of the single densities
f fR N no LU no LUz zc h bg=fR n
SS N
SS N LU LU
zC E
z C E zTc h b g c h c h=
+− +FHG
IKJ
−1
2
1
22 22 1
π σσ
detexp
Derivation of an optimal estimator
-
45Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
ESTIMATOR
• Neyman-Pearson criterion: Maximization of the detection probability PD for a givenfalse alarm probability PF
Optimal estimator:
z C E E zC E
TSS N N
SS N
N
n+ −+F
HGG
IKJJ
− −σ σ λ
σ
σ2 1 2 1
0
2
2c h c he j bg c hc h < -2
LU
lndet
P fD R= z LU LU dz zR1
c h
P fF R= z no LU no LU dz zR1
c h
Likelihood ratio: f
fR
R
LU
no LU
LU
no LU
z
z
c hc h >
LU
λ0
-
46Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
UNKNOWN COVARIANCE MATRIX
� Problem: For the optimal estimator CSS must be known
• Uncorrelated spectrum values for the
LU receiving process:
• Completely correlated real parts and
imaginary parts of the spectrum values:
→ mismatched estimator!optimal detection Receiver Operating Characteristic
(ROC)
for σσσσ² /σσσσ² = 3dB and n = 3
optimal detection Receiver
Operating Characteristic
(ROC)
for σσσσ² /σσσσ² = 3dB and n = 3NS
completely correlated LU spectrum values
uncorrelated LU spectrum values
PF
PD
-
47Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
SNRD ESTIMATION• For which average power of a received LU signal must the LU channel be
classified as used?
→→→→ depends on the permissible interferences on the LUs
AP Access PointAP
LU r
RU
RU
-
48Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
DETERMINATION OF SNRD
Higher SNRD ����lower PF ����enhanced efficiency
high ∆∆∆∆P is advantageous for detectionhigh ∆∆∆∆P is advantageous for detection
-
49Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
RECEIVER OPERATING CHARACTERISTICS (ROCs)• Simulation results: worst case consideration
���� maximal mismatched estimator
ROC,
worst case,
n = 5
ROC,
worst case,
n = 5
PD
ROC,
worst case,
n = 15
ROC,
worst case,
n = 15
PD
-
50Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
ROCs
• best case consideration: Uncorrelated spectrum values
Reality is somewhere between best und worst case
→→→→ Choose n for the worst case !?
ROC,
best case,
n = 15
ROC,
best case,
n = 15
-
51Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
DIVERSITY SOLUTION
Solution:
distributed detection (diversity)
all mobile terminals and three additional Boosting Stations
take measurements jointly
PDbecomes realizable for moderate PF !
Problems:
multipaths (fading)
too high false alarm probability PFPD cannot be realized with only one measurement station!
Given detection probability : PD = 0,999
-
52Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
DISTRIBUTED DETECTION
Licensed UserLU
Rental UserRU
Boosting StationBooS
Access PointAP
AP
LU
RU
BooS
BooS
RU
BooS
RU cell
-
53Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
SIGNALING (1)
• RUs + BooSs →→→→ (AP): Boosting protocol
AP computes the elementwise „exclusive or“ of all OVs
AP
LU
RU
BooS
BooS
RU
BooS
RU cell
-
54Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
SIGNALING (2)
• AP →→→→ RUs + BooSs: Robust time-frequency broadcast
AP
LU
RU
BooS
BooS
RU
BooS
RU cell
Licensed UserLU
Rental UserRU
Boosting StationBooS
Access PointAP
-
55Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
DIVERSITY-GAIN
The individual detection results are statistically independent.
If the receiving conditions at the (m) measuring stations (RUs and BooSs)
are similar, we get :
The individual detection results are statistically independent.
If the receiving conditions at the (m) measuring stations (RUs and BooSs)
are similar, we get :
���� PF decreasesPD for a specific RU may be reduced
���� no fadingdiversity
What is the gain of divided detection ?
≈≈≈≈ 0≈≈≈≈ 00.29220
0.0100.0010.49910
0.3440.1000.8224
0.6480.2940.9003
0.8860.6620.9682
0.9820.9820.9991
PZFPFPDm
-
56Universität Karlsruhe (TH)Research University · founded 1825
Communications Engineering LabINTINT
LU TRANSMISSION MODEL• Signal model ⊕s kf xS b g
∈C
, y
n k
f xN b g∈C
, y
r k
f xRb g∈C
, y
• Transition to n FFT repetitions
f f x x x y y y fS S n n ST
x y z z x y, , , , , , , , ,b g b g bg b g= = =1 2 1 2K K where f x y fS s
n, :b g b g⇒ →x, y R R 2 2
n realparts
n imaginaryparts