CognitiveRadio.ppt
-
Upload
krish-ramesh -
Category
Documents
-
view
218 -
download
0
Transcript of CognitiveRadio.ppt
-
8/10/2019 CognitiveRadio.ppt
1/27
AI Technologies for the
Cognitive RadioDipti Deodhare
Centre for Artificial Intelligence andRobotics
-
8/10/2019 CognitiveRadio.ppt
2/27
Cognitive Radio = Software Defined Radio
+
Cognition Engine
-
8/10/2019 CognitiveRadio.ppt
3/27
Cognitive Capabilities
Awareness: should be aware of its
own abilities, the regulating policies that govern it, itsneighbours and their abilities etc.
-
8/10/2019 CognitiveRadio.ppt
4/27
Cognitive Capabilities
Perception: should be able to
sense its environment
-
8/10/2019 CognitiveRadio.ppt
5/27
Cognitive Capabilities
Learning: consequent to
perception it should be able tolearn about the generalcharacteristics of its environment
and trends
-
8/10/2019 CognitiveRadio.ppt
6/27
Cognitive Capabilities
Reasoning: relationships between
the various entities should beunderstood and sound decisionsinferred, obviating the
overwhelming, and perhapsimpossible, task of enumeratingever single alternative
-
8/10/2019 CognitiveRadio.ppt
7/27
Cognitive Capabilities
Memory: should demonstrate
improved performance afteroperating in the same environmentover an e!tended period of time
-
8/10/2019 CognitiveRadio.ppt
8/27
Architecture
"oftware Radio(perception)
Cognition #ngine $ %earning #ngine (learning)
$ Reasoning #ngine (reasoning)
$ &nowledge 'ase(awareness, memory (
"oftwareRadio
Cognition #ngine
Knowledge Base
Reasoning Engine
Learning Engine
-
8/10/2019 CognitiveRadio.ppt
9/27
)eters and &nobs Cognitive capabilities are
predicated on availabilitof suitable meters and knobs
in the "DR* )eters to continuousl
monitor the performanceof the radio*
&nobs to reconfigure the radio so that the performance of the radio ismaintained at an acceptable level satisf ing various conte!t drivenconstraints*
The mandate of the cognitive engine is to +read the meters andappropriatel +tune the knobs*
Cognition #ngine
Knowledge Base
Reasoning Engine
Learning Engine
Read meters
Set knobs
"DR
Application
Transport
-etwork Data %ink
. MAC+LLC (
/01
"ecurit
-
8/10/2019 CognitiveRadio.ppt
10/27
%a er2wise distribution of )eters
and &nobs Contd*)#T#R"
Interference,
'#R, receivedsignal power,noise power,"-R, fadingstatistics,doppler spread,dela spread,angle of arrival,d namic range
&-3'" /ower, frequenc band
of operation, carriermodulation t pe,
baseband modulationt pe, pulse shaping,data rate, number ofchannels, bandwidth,equali4ation, antennatuning, antennasteering, antenna heightad5ustment, t pe ofantenna .if more thanone t pe of antennaavailable(
Physical Layer
Data Link Layer(MA ! LL )
"etwork
#ransport
$ec%rity
Application
-
8/10/2019 CognitiveRadio.ppt
11/27
%a er2wise distribution of )eters
and &nobs Contd*)#T#R
6rame error rate&-3'"
6rame format,6rame si4e,multiple access,duple!ing, 6#C,AR7.enable8disable(
Physical Layer
Data Link Layer(MA ! LL )
"etwork
#ransport
$ec%rity
Application
-
8/10/2019 CognitiveRadio.ppt
12/27
%earning #ngine
The %earning #ngine $ Collection of classifiers
developed using supervised and unsupervisedlearning techniques*
-umeric data " mbolic data
%earning.collection of
classifiers(
)odulation,Carrier 6req*,'#R, "-R,/wr, Coding
Classify intoqualitative classes such as good, bad high, average, low
-
8/10/2019 CognitiveRadio.ppt
13/27
)T9 "i4e vs ThroughputRef: Anna Calveras Auge;, ??@
6or each '#R value,an optimal )T9e!ists
'#R: meter at the
ph sical la er* )T9: knob at the network la er*
-eural -etworkClassifier
'#R Throughput
)T9
6igure is a graph for /oint to /oint /rotocol .///( anddeterministic errors*Braphs would be different for other protocols like 6rameRela , As nchronous Transfer )ode .AT)(, #thernet etc*
The would also var depending on the error t pe, namel ,deterministic and burst*It is proposed that the graphs can be +learnt b a )ulti2%a er /erceptron .)%/( or some other suitable mechanism*Reinforcement %earning mechanisms are also potentiallrelevant*
-
8/10/2019 CognitiveRadio.ppt
14/27
36D) Transmitter
Modulator (QAMQPSK BPSK)
Parallelto
serialincomingbits
C 1
C 2
C 3
C n
IFF
! 1
! 2
! 3
! n
36D) transmission offers opportunitfor adaptation
-
8/10/2019 CognitiveRadio.ppt
15/27
36D) Receiver
Serial to"arallel
#e$modulator
incomingbits
C 2
C 3
C n
IFF
! 1
! 2
! 3
! n
C 1
-
8/10/2019 CognitiveRadio.ppt
16/27
Capacit ma!imi4ation in a non2A=B- channelcan be modeled as a Reinforcement %earning
problem* M modulation t pes: M > , M C, , M M *
N coding t pes: C > , C C, , C N *
M i can transmit d i data bits per s mbol and has a probabilit of bit error ei.S ( for signal to noiseratio S *
Coding t pe C has rate r and can correct c bit
errors per block of si4e b bits* ! S : s mbol time, a constant* " : corrected bit error rate coming out of the
decoderE measured b the radio*
-
8/10/2019 CognitiveRadio.ppt
17/27
6or modulation t pe M i and coding t pe C ,the resulting capacit is given b :
C i# F .. d ir (8# $ ( .>2 " (
C i# can act as a performance measure* 3b5ective function:
ma$ f%M i #C #"&' d ir .>2 " (
-
8/10/2019 CognitiveRadio.ppt
18/27
-
8/10/2019 CognitiveRadio.ppt
19/27
&nowledge Representation
#ach node or relation is associated to anontolog that defines the concept* An
ontolog consists of slots representingvarious attributes of that concept* 6or e.(.here is the +Device ontolog created in
3=% .=eb 3ntolog %anguage(* %)e uset*e S) , editor to create and edit ontolo(iesand R- /ravity to visuali0e t*em.&
-
8/10/2019 CognitiveRadio.ppt
20/27
Device Ontology
-
8/10/2019 CognitiveRadio.ppt
21/27
Reasoning
Reasoning involves traversing the semanticgraphs to obtain relevant conclusions*
"ome ontologies for the cognitive radio: radio,channel, spectrum, power, coding, modulation,etc*
"ome inferences: frequency f c is sparsely used from time t 1 to time t 23 for c*annel c# capacity isma$imi0ed 4it* modulation type m i and codin(met*od c and so on.
-
8/10/2019 CognitiveRadio.ppt
22/27
A "imple #!ample using
/redicate %ogic .papers include authors from 9" DoD( "oftware Radio ."R( e!ports predicates to the
knowledge2base regarding detected signals s> , sC,, s
N#of the form
si(nal req%s i #f i &G si(nal") . s i #) i &
Boal: To find some f c and ) that does not overlapan detected signal, while ma!imi4ing ) and
hence the radio;s capacit * Define, not verlap%f c #)#s i &
' . f i+) i 526f c7)52 ( H . f i7) i 528f c+)52 (
-
8/10/2019 CognitiveRadio.ppt
23/27
A "imple #!ample using
/redicate %ogic Define predicate:action: moveBand & old ,' old , & new , ' new !
precond: i 6' N : not verlap . f ne4 #) ne4 #si( postcond: 2. center req . f old ( center req . ) old ((
. center req . f ne4 ( center req . ) ne4 (( Define the +Reasoning ob5ective function:
f R .center req . f c( band4idt* . ) (( F ) This will give a polic 2based cognitive radio that will
search out the largest continuous piece of bandwidth forcommunication*
H
GG
G
G
-
8/10/2019 CognitiveRadio.ppt
24/27
)emor
Case2based Reasoning Tools are useful "warm Intelligence technologies
0))s Reinforcement %earningCan all pla a role in searching large knowledge
banks to enable quick responses*
-
8/10/2019 CognitiveRadio.ppt
25/27
" of fre#$encies %ammed t&is is
'ess t&an t&et&res&o'd!
(op rate )EC Data Rate
Increases to reduce 5amming
J8@Reduces to meet
'#R requirements
K)ore robust totolerate errors
)a! L threshold
/reliminar Anal sis for6requenc 0opping "olutions
using D#A% Inputs
-
8/10/2019 CognitiveRadio.ppt
26/27
#stablishing a Test2bed
"imulation in )atlab B-9 radio $ local vendors available
-
8/10/2019 CognitiveRadio.ppt
27/27
Thank 1ou