Cognitive radio architecture

• An architecture for cognitive radio consists of the

functions, components, and design rules necessary to
support the evolution of cognitive radio.
• The architecture integrates the contributions of
researchers focusing on the specific disciplines of
software radio, network engineering, natural language
processing, and machine learning.
• This architecture minimizes the dependence on
knowledge-engineering through the integration of
machine learning. This chapter defines the cognitive radio
architecture that emerged through the definition of RKRL
Primary Cognitive Radio Functions
The primary radio cognition functions consist of:
1.Recognize user communications context
2.Mediate wireless information services as a function of
context.
• The cognition functions should rely on processing streams
of user interaction with applications as its primary means
of inferring communications context. Only as a last resort
should the cognition functions require user input that has
no purpose other than to assist the cognition functions
• The mediation of wireless information services includes
continually keeping track of the parameters of wireless
networks are present in the environment which include
spectrum occupancy, received signal strength as a function
of time and space, available QoS, and related cost(s).
Behaviours
Waking Behavior:
• The waking behavior is optimized for real-time
interaction with the user, isochronous control of
software radio assets, and real-time sensing of
the environment.
• Thus the awake-state(waking behavior) cognition
maps environment interactions to the current
stimulus-response cases, the dynamic subset of
the embedded srModels(simulation response
Models).
• Incremental machine learning maps these
interactions to integrated knowledge, the
persistent subset of the srModels.
Sleeping Behavior:
• Cognitive PDAs detect conditions that permit or
require sleep. For example, if the PDA predicts or
becomes aware of a long epoch of disuse (e.g.
overnight), then it PDA may autonomously initiate
sleeping behavior.
• Otherwise, it would request permission to enter
sleeping behavior from the user. During the sleep
epoch, the PDA process experience from the waking
behavior using non-incremental machine-learning
algorithms.
• These algorithms map current cases and integrated
knowledge onto integrated knowledge (b).
Prayer Behavior:
• A conflict is a context where the user overrode a PDA
decision about which the PDA had no little or no uncertainty.
Map b may resolve the conflict. If not, then it will place the
conflict on a list of unresolved conflicts (map g).
• Attempts to resolve unresolved conflicts via the mediation of
the PDA’s home network may be called prayer behavior. The
unresolved-conflicts-list is mapped (l) to RKRL XML queries
to the PDA’s home network expressed in KQML. Successful
resolution maps network responses to integrated knowledge
(m).
• Alternatively, the PDA may present the conflict sequence to
the user, requesting the user’s advice during the wake cycle
(map p).
Cognitive Radio Components
• Cognition functions implemented via cognition components.
These include data structures and related processing components
• The data structures include the reinforced hierarchical sequences
Words, Phrases, Dialogs and Scenes of the Observe Phase. Within
each of these sequences, the novel sequences represent the
current stimulus-response cases of the cognitive behavior model.
• The known sequences represent the integrated knowledge of the
cognitive behavior model.
• Known sequences may consist of Radio Knowledge
Representation Language (RKRL) statements embedded in the
PDA or of knowledge acquired through independent machine
learning.
• The Nearest sequence is the known sequence that is closest in
some sense to the novel sequence.
 World Model , S
• The World Model, S, consists primarily of bindings
between a-priori data structures and the current
scene.
• These structures are also associated with the Observe
Phase. Dialog states, action requests, plans, and
actions are additional data structures needed for the
Observe, Orient, Plan, and Act Phases respectively.
• Each internal data structure maps to an RKRL frame
consisting of handle (e.g. set, or stimulus), model,
body (e.g. subset, or response), resources, and context
(RKRL URL, source, time, and place of the frame).
 Processing Components
• The input map θ consists of components that transform external stimuli to the
internal data structure Sensory Data.
• The transformation θ consists of speech recognition, lower-level software radio
waveform interface components, etc., that create streams of primitive reinforced
sequences.
• Reasoning components include the map ρ that identifies the best match of
known sequences to novel sequences. These are bound to scene variable by
projection components, π.
• The maps θ, θ, ρ and π constitute Observe Phase processing. Word and phrase
level bindings are interpreted by the components ɸ to form dialog states. The
components Г create action requests from bindings and dialog states.
• The maps ɸ and Г constitute Orient Phase processing. Scene bindings include
user communications context. Context-sensitive plans are created by the
component ∑ that evaluates action requests in the Plan Phase. The Decision
Phase processing consists of map ∆ that maps plans and scene context to
actions.
• Finally, the map E consists of the effector components that change the PDA’s
internal states, change displays, synthesize speech, and transmit information on
wireless networks using the software radio personalities.
A-Priori Knowledge Taxonomy
• As a minimum, that taxonomy should include
the named sets of RKRL 0.3.
• Minimum concepts to be represented internal
to the cognitive PDA include time, space, RF
propagation, radio networks, external entities,
and user communications contexts.
• In addition, the a-priori knowledge taxonomy
must include a hierarchical model of the
physical world and a model of the internal
structure of the SDR.
Observe-Phase Data Structures
• As illustrated, the speech and/or
text channels may be processed via
natural language facilities with
substantial a-priori models of
language and discourse.
• The use of those models should
entail the use of homeomorphic
mappings among the word, phrase,
dialog, and scene levels of the
observation phase hierarchy and
the encapsulated component(s).
• The Observe Phase components match new
stimuli to known stimuli.
• When an exact match is not possible, the
components may deliver one or more
hypotheses. Hypotheses may consist of best-
match, or a prioritized list of partial matches.
• Bindings may be computed as the interface
from the observation phase hierarchy to the
orient phase.
Radio Procedure Knowledge Encapsulation
• Radio knowledge may be embodied in components
called radio knowledge sources. If so,they are
organized as set-theoretic maps among wake-cycle
phases (observe, orient, plan, decide,act).
• A subset of these (e.g. radio procedure KSs) may
control radio personalities as illustrated in Figure.
These knowledge sources accept a-priori
knowledge from RKRL an associated XML micro-
world.
• Each knowledge source also saves the knowledge
it learns in RKRL format, conforming to RKRL
axioms
 Orient-phase Components
• The Orient-phase determines the novelty of stimuli
represented in the sequence hierarchy.
• The orient phase components bind variables and
structure requests for planning.
• These components may directly invoke decide-phase or
act phase components if the immediate context dictates.
• The orient-phase components may preempt planning if
the stimuli strongly invoke immediate action.
• The orient phase determines when the PDA needs sleep
or prayer behaviors and requests them according to the
communications context.
Plan Phase Components
• The plan calculus may include reasoning over time, space,
user identities, and contexts.
• Each request corresponds to an operation that can be
accomplished by an effector.
• The axiomatization of time permits times to be compared
and time delays to be computed.
• The plan calculus generates and compares plans and
detects conflicts.
• The Plan-phase generates alternatives, including expressing
plans to peers and/or the network to obtain advice.
• A plan element may be a sequence that has been mapped
to [Here, Now]. There may be competing and partially
fulfilled plans.
 Decide-Phase Components

• The decide-phase selects among alternatives generated by

the planning phase. Its knowledge representation depends on
the radio procedure components.
• The decide-phase allocates computational and radio-
resources to subordinate (conventional radio) software,
based on the activation of a plan.
• Autonomous decisions may be taken when the correlation
between the current situation and the reference situation is
high, and when the radio has been in this exact spatial-
temporal context before.
• Tentative decisions may be offered to the user for
confirmation, which makes it more likely to be invoked in the
future. In the case of unresolved conflicts among plans, the
decide phase lets the user do the task, observing and
recording the user’s initiation of low level communications
and information processing tasks.
 Act-Phase Knowledge Representation
• The Act-stage initiates tasks with specified
resources for specified amounts of time.
• The PDA thus provides the user with access to
radio resources (bands, modes).
• The knowledge representation of the act
phase depends on the radio knowledge
sources contributing to the parameters of or
participating in the action.
 Design Rules
1. The cognition functions shall maintain an explicit topological model of the world,
of the environment (including the user, the physical environment, and the radio
networks, and of the internal states of the radio). Context shall be axiomatically
represented using the topological model.
2. The model of the world shall follow an explicit axiomatic treatment of time,
space, radio frequency, radio propagation, and the identity of entities.
3. Models shall be represented in an open architecture radio knowledge
representation language suited to the representation of radio knowledge (e.g.
RKRL 0.3). That language shall support topological properties and axiomatic
models of cognitive radio.
4. The cognition functions shall maintain location awareness, including the sensing of
location from global positioning satellites, and sensing position from local wireless
sensors and networks. Location shall be an element of context.
5. The cognition functions shall maintain awareness of time to the accuracy
necessary to support the control of radio functions. Time shall be an element of
context.
6. The cognition functions shall maintain an awareness of the identity of the PDA, of
its primary user, and of other legitimate users designated by the primary user.
Current user shall be an element of context.

7. The cognition functions shall reliably infer the user's communications context and
apply that knowledge to the provisioning of wireless access by the SDR function.

8. The cognition functions shall model the propagation of its own radio signals with
sufficient fidelity to estimate interference to other spectrum users. The cognition
function shall also assure that interference is within limits specified by the spectrum
use protocols in effect in its location (e.g. in spectrum rental protocols). It shall defer
to the wireless network in contexts where the network manages interference.

9. The cognition functions shall model the domain of applications running on the host
platform, sufficient to infer the parameters needed to support the application.
Parameters modeled include QoS, data rate, probability of link closure (Grade of
Service), and space x time x context domain within which wireless support is needed.

10. The cognition functions shall configure and manage the SDR assets to include
hardware resources, software personalities, and functional capabilities as a function of
network constraints and use context.
11. The cognition functions shall administer the computational resources of the
platform. The management of software radio resources may be delegated to an
appropriate SDR function (e.g. the SDR Forum domain manager). Constraints and
parameters of those SDR assets shall be modeled by the cognition functions. The
cognition functions shall assure that the computational resources allocated to
applications, interfaces, cognition and SDR functions are consistent with the user
communications context.
12. The cognition functions shall represent degree of belief in external stimuli and in
inferences. A certainty calculus shall be employed consistently in reasoning about
uncertain information.

13. The cognition functions shall recognize preemptive actions taken by the network
and/or the user. In case of conflict, the cognition functions shall defer the control
of applications, interfaces, and/or SDR assets to the network or to the primary
user,according to an appropriate operations assurance protocol.