KR20110078419A - A pomdp-based adaptive sensing scheduling scheme for cognitive radio systems - Google Patents

Abstract

PURPOSE: An adaptation sensing scheduling method for a recognition wireless system is provided to offer very high maximum channel utilization at a wireless recognition node. CONSTITUTION: A non-occupancy probability of a channel is detected by time variable information that is obtained through a channel sensing. An energy detection number that is performed before a channel switching is controlled according to the non-occupancy probability of a detected channel. The adaptation sensing operates. The time variable information includes an extent of a noise imprecision and a priority user signal power.

Description

POMDP-based Adaptive Sensing Scheduling Scheme for Cognitive Radio Systems

The present invention relates to an adaptive sensing scheduling method based on Partially Observable Markov Decision Process (hereinafter abbreviated as " POMDP ") in Cognitive Radio Systems. The present invention relates to a POMDP-based adaptive sensing scheduling method for an cognitive radio system that provides a high maximum channel utilization rate to a wireless cognitive node while effectively protecting a user in a changing network environment.

In general, radio recognition (CR) has attracted great interest from those skilled in the art as one of the most feasible solutions to the spectrum shortage problem in wireless communication. In a radio aware system, a node searches for a spectral band called a spectral hole that is not being used by an authorized first user (PU) and uses some (or all) of the spectral hall as its communication channel. Since the radio aware system should not interfere with the PU, the radio aware system should move out of the channel as soon as possible when the PU is activated. In order to release the spectrum immediately, the CR system spends a lot of time sensing the spectrum to see if there is a PU in the channel.

During sensing of a particular spectral band, all CR nodes in the network typically stop transmitting on the channel to avoid unclear decisions. This silence duration is called the "quiet period". Although this static period is necessary for the CR system, this static period needs to be minimized as long as the requirements for protecting the PU are met, since it causes performance degradation of the CR system. Therefore, the CR system must effectively manage this static period so that not only good protection of the PU but also high maximum channel utilization can be achieved, that is, efficient spectrum sensing scheduling is required for improving the performance of the CR system.

Several spectrum sensing scheduling schemes for CR systems have been proposed in the literature. Note that the spectrum sensing scheduling algorithm typically determines when, how long and what type the CR node performs the sensing.

References Y-C Liang, Y. Zang, E. Peh, and A. T. Hoang, "Sensing-throughput tradeoff for cognitive radio networks." IEEE Trans. Wireless Commun., Vol. 7, no, 4, Apr, 2008., based on energy detection, addresses the problem of designing an optimal sensing time (ie, length of static time) to maximize the attainable throughput of the CR system while fully protecting the PU. Studied.

See also A. T. Hoang and Y.-C. Liang, "daptive scheduling of spectrum sensing periods in cognitive radio networks," in Proc. IEEE GLOBECOM 2007, Nov. 2007. proposes two schemes to jointly schedule sensing and data transmission on the channel, one scheme to make the decision based on the channel condition to maximize system throughput, and the other scheme to lose data Considering the amount of data and channel state waiting for scheduling to minimize this.

Further, references D. Datla, R. Rajbanshi, A. M. Wyglinski, and G. J. Minden, "Parametric adaptive spectrum sensing framework for dynamic spectrum access networks," in Proc. IEEE DySPAN 2007, Apr. 2007. proposes a sensing scheduling structure to adaptively adjust the sensing interval of a channel based on the PU occupancy of the channel.

The IEEE 802.22 Wireless Local Area Network (WRAN), on the other hand, employs a two-stage sensing mechanism: high-speed sensing and precision sensing. With this mechanism, the detection algorithm schedules fast or precise detections in each static period. Although the detection structure for high speed and precision detection is not explicitly mentioned in the IEEE 802.22 draft, it is used for precision detection because of the short detection time, energy detection is used for high speed detection, and the ability to recognize signal source is important for precision detection. It is anticipated that detection will be employed. In IEEE 802.22 WRAN, when there is noise uncertainty and interference from other CR systems, references H. Kim and K. G. Shin, "n-band spectrum sensing in cognitive radio networks: energy detection or feature detection", Proc. Mobicom 2008, Sept. 2008. vol. 7, no. 2, Feb. 2008. determines which of the energy detection and feature detection is better at each signal-to-noise ratio (SNR) level and below that threshold finds a threshold for the received PU signal power for which feature detection is preferred.

References W. S. Jeon, D. G. Jeong, J. A. Han, G. Ko, and M. S. Song, "An efficient quiet period management scheme for cognitive radio systems", IEEE Trans. Wireless Commun., Vol. 7, no. 2, pp. 505-509, Feb. 2008. proposes a detection scheduling structure that reduces unnecessary feature detection and channel switching by performing feature detection only after a PU detection alarm has been issued N consecutive times from energy detection. This structure effectively protects the PU and provides very high channel utilization for the CR system, but this structure has the disadvantage that the PU signal power and / or noise uncertainty cannot adapt well to time varying network conditions where time varies.

Accordingly, the present invention is that the detection scheduling method to reduce the feature detection and channel switching applied to the existing radio recognition system as described above, is not adaptive in the first time, the time variable network conditions in which the user signal power and / or noise uncertainty changes over time Is proposed to solve various problems,

The problem to be solved by the present invention is to provide a POMDP-based adaptive sensing scheduling method for cognitive radio system to provide a very high maximum channel utilization to a wireless cognitive node while effectively protecting the user first in a network environment that changes over time. have.

Another problem to be solved by the present invention is to perform adaptive detection by adaptively adjusting the number of energy detections performed before feature detection or channel switching according to the received first user signal power and the degree of noise uncertainty. It is to provide a POMDP based adaptive sensing scheduling method for an cognitive radio system.

According to the present invention for solving the above problems the "POMDP-based adaptive sensing scheduling method for cognitive radio system",

A scheduling method for adaptively detecting a channel in a cognitive radio system including a plurality of radio recognition nodes (CR nodes) that perform channel detection, traffic transmission / reception, and channel switching on a channel,

Detecting the non-occupancy probability of the channel with the time variable information obtained through the channel detection, and adaptively performing the adaptive detection by adaptively adjusting the number of energy detections performed before feature detection or channel switching according to the non-occupancy probability of the detected channel. It is characterized by.

The time varying information is characterized by first including the user signal power and the degree of noise uncertainty.

The non-occupancy probability is calculated based on the partially observable Markov decision method (POMDP).

The non-occupancy probability is determined in consideration of the number of active CR nodes.

The CR node detects the non-occupancy probability by calculating a confidence vector through joint tracking of the initial distribution of the channel state and the last action and observation value in each determination interval, and presets the non-occupancy probability and the predetermined operation mode for determining the detected non-occupancy probability. The decision threshold value is compared, and the operation mode is determined by the magnitude.

The operation mode includes an energy detection mode, a feature detection mode, and a channel switching mode.

According to the present invention, the adaptive detection is performed by adaptively adjusting the number of energy detections performed before feature detection or channel switching according to the received user signal power and the degree of noise uncertainty, thereby effectively performing the PU in a network environment that changes over time. This provides the advantage of providing very high maximum channel utilization to the CR node while protecting.

Hereinafter, described in detail with reference to the accompanying drawings a preferred embodiment of the present invention. In the following description, well-known functions or constructions are not described in detail since they would obscure the invention in unnecessary detail.

The present invention provides a method of performing adaptive sensing by adaptively adjusting the number of energy detections performed before feature detection or channel switching according to the received priority user signal power and the degree of noise uncertainty.

In the present invention, since the CR system selects the next mode of operation (traffic transmission, energy detection, feature detection, channel switching) based on the probability that the channel is empty (non-occupancy probability) in each decision interval, a kind of decision algorithm ( decision making algorithm).

It is generally noted that the spectral sensing results do not exactly indicate empty channels due to sensing errors. As a result, the CR system simply uses the detection result to guess that the channel is empty. Therefore, the CR system should select an operating mode that is optimally possible with opaque data. Considering that POMDP is well suited for modeling decision problems in the absence of ambiguity, we propose an adaptive sensing scheduling (ie, operation mode selection) structure based on the POMDP configuration.

Recently, several decision problems of CR networks have been formulated into POMDP.

References Q. Zhao, L. Tong, A. Swami, and Y. Chen, "Decentralized cognitive MAC for opportunistic spectrum access in ad-hoc networks" A POMDP framework, IEEE J. Sel. Areas Commun., Vol. 25, no. 3, pp. 589 ", 99, Apr. 2007. proposes a distributed MAC protocol that allows CR nodes to independently search for spectrum opportunities without a central coordinator or a dedicated control channel. The CR nodes with data must first follow the POMDP model. Detect the selected channel, and if the channel is empty, the node sends data on the channel.

References R.-T. Ma, Y.-P. Hsu, and K.-T. Feng, "POMDP-based spectrum handoff protocol for partially observable cognitive radio networks" w, in Proc. IEEE WCNC 2009, Apr. 2009. proposes a POMDP based spectrum handoff structure. In this structure, when the PU appears on the channel on which traffic is being sent, the transmitter determines to hand off the channel by applying POMDP.

References Y. Chen, Q. Zhao, and A. Swami, "Distributed spectrum sensing and access in cognitive radio networks with energy constraint, IEEE Trans.Signal Proc., Vol. 57, no. 2, pp. 783-797, Mar. 2009 is designing a distributed spectrum detection and access strategy for CR-specific networks with energy constraints based on POMDP, which allows the system to determine whether the PU is in sleep or performing detection. When sensing is selected, the node selects a channel to detect and then decides whether to access the channel based on the sensing result, while the above structures focus primarily on channel selection for opportunity access, while the present invention There is a difference in handling discovery scheduling.

Hereinafter, the specific operation of the present invention will be described.

There is a channel pool including a plurality of channels. The present invention contemplates a CR system operating on a channel looking for an empty channel in the pool until the presence of the PU is detected. Upon detecting the PU on the operating channel, the CR system switches to another empty channel.

The CR system under consideration consists of a plurality of CR nodes, one of which acts as the central controller (eg, master, access point or base station) of the system. The term “base station” hereafter refers to all types of central controllers. The base station acts not only as the central controller itself, but also as the CR node. The CR node performs normal operation on the channel under the control of the base station, namely channel sensing, traffic transmission / reception and channel switching.

References H. Urkowitz, "energy detection of unknown deterministic signals," in Proc. IEEE, vol. 55, pp. 523-531, Apr. 1967. and W. A. Gardner and C. M. Spooner, "signal interception: performance advantages of cyclic-feature detectors," IEEE Trans. Commun., Vol. 40, no. 1, pp. 149-159, Jan. 1992. and P. D. Sutton, K. E. Nolan, and L. E. Doyle, "Cyclostationary signatures in practical cognitive radio applications," IEEE J. Sel. Areas Commun., Vol. 26, no. 1, pp. 13-14, Jan. 2008 describes energy detection and feature detection and is well known to be able to compensate for each of the weaknesses.

Since the energy detector only checks the energy level of the signal in the channel, the energy detector can detect the PU signal in a relatively short sensing time compared to the feature detection. However, energy detection is unreliable in noise ambiguity and weakly received PU signals, and cannot identify sources of energy among PU, noise, and other CR systems.

On the other hand, since feature detection searches for signal sources in more detail, feature detection consumes more detection time than energy detection. However, feature detection can recognize signal sources and is robust to noise uncertainty and interference in the channel. Considering the energy detection and feature detection complement each other's weaknesses, the proposed CR system uses both detection schemes for spectrum detection. Furthermore, during channel sensing, all nodes in the system stop transmitting / receiving traffic, i.e., the node only performs channel sensing without other actions.

In the CR system under consideration, the channel time is divided into frames having a fixed length, and the base station schedules transmission on the channel in units of frames. As mentioned above, the normal operations that the CR node performs on the channel are the transmission / reception of traffic, energy detection, feature detection, and channel switching. The detection time for energy detection is typically frame length (in most existing wireless systems with frame-based channel structures, the frame length is several milliseconds or tens of milliseconds, while the detection time for energy detection is tens to hundreds of microseconds. Considering much shorter than that, the present invention designs the system such that the transmission / reception of the traffic and the energy detection are scheduled within the same frame. That is, a small portion of the frame is allocated for energy detection and the remaining portion is used for transmitting / receiving traffic. Therefore, system operation is classified into three types: traffic transmission / reception and energy detection (T / R-ED), feature detection (FD) and channel switching (CS).

1 illustrates a POMDP based adaptive sensing scheduling method for a wireless cognitive system according to the present invention.

Each radio-aware node checks scheduling information for a frame having a corresponding fixed length, including a traffic transmission schedule and a start time (and length) of energy detection (S10).

The radio recognition node transmits / receives data traffic, senses a channel for energy detection (S20), and the detected channel detection information is transmitted to the base station.

)을 비교하게 된다(S40).The base station detects the non-occupancy probability of the POMDP-based channel based on the sensed information (S30), and the first and second determination thresholds for determining the non-occupancy probability and the operation mode of the detected channel (

) Will be compared (S40).

When the non-occupancy probability of the channel is greater than the first determination threshold value, the energy detection mode is determined, and the number of times of energy detection is adjusted according to the non-occupancy probability (S50 and S60).

If the non-occupancy probability is less than the second decision threshold, the channel switching mode is determined. If the non-occupancy probability is greater than the second decision threshold but less than the first decision threshold, the feature detection mode is determined. It is made (S70).

Hereinafter, each process as described above will be described in more detail.

2 illustrates an example of system operation over time. As shown in the figure, the durations of the three modes of operation are different and are represented by l _T , l _F and l _C for the modes T / R-ED, FD and CS, respectively. In the proposed system, since the T / R-ED operation is performed for one frame, ℓ _T is equal to the length of the frame. It is noted that the energy detection time, denoted by 1 _E , is much smaller than the length of the frame (ie 1 _E <1 _T ).

System operation is controlled by the base station selecting the operating mode. When the T / R-ED mode is selected, the base station informs the CR node of the scheduling information for the corresponding frame, including the traffic transmission schedule and the start time (and length) of the energy detection. At this time, the node transmits / receives data traffic, performs energy detection and reports the detection result to the base station according to a schedule. While in the FD mode, ℓ _F, each node performs feature detection, and reports the detection result to the base station.

On the other hand, if the CS mode is selected, all nodes in the system complete the channel switching during ℓ _C and the base station is ready to select the operating mode for the new channel.

The base station selects the next operating mode of the CR system based on the most recent sensing result from the CR node immediately after the current mode ends. In the present invention, the time point at which the base station selects an operation mode will be referred to as a "decision epoch." As shown in the example of FIG. 1, the interval between the two determination intervals may vary depending on the mode of operation performed by the network.

In order to increase the channel efficiency of the CR system while adequately protecting the PU, the base station must make an optimal decision when selecting an operation mode. If the presence of a PU on a channel is expected with a high probability, the base station should decide to switch the channel. However, if the channel is more likely to be empty, the base station can select the T / R-ED mode to allow the CR node to transmit traffic. On the other hand, if the base station cannot be assured of the presence of the PU only from the results from the energy detection, it may be the best decision to select the feature detection mode to more accurately investigate the signal source.

In general, based on the detection result, the CR system cannot accurately determine whether the PU exists or not, but can only infer the existence of the PU. This is because most of the physical sensing techniques used by the CR system have detection errors due to the uncertainty and complex nature of the PU signal and noise.

On the other hand, it is well known that POMDP is a very suitable mathematical tool for modeling decision problems that allow for clarity. In the CR system under consideration, the mode selection of the base station should be considered an ambiguity decision problem since the operation mode should be determined based on the estimation of the PU presence. Therefore, the present invention designs an optimal mode selection structure that adaptively selects an operation mode according to the detection result by applying the POMDP configuration.

로 지수적으로 분포되는 것으로 가정한다. 즉, ON에서부터 OFF로 의 상태 전이율은 λ이며, OFF로부터 ON으로의 상태 전이율은 μ로 주어진다.It is assumed that the channel has only a single PU. In addition, the present invention assumes that all channels in the channel pool have the same PU activity model. If the PU is in the channel, it is said to be in the ON state, and if the channel is empty, the state of the PU is said to be OFF. PU alternates between ON and OFF. Average period of OFF (ON) state

Suppose that it is distributed exponentially. That is, the state transition rate from ON to OFF is lambda, and the state transition rate from OFF to ON is given by μ.

When the base station makes a request to the CR node to detect the operating channel (in T / R-ED or FD mode), each CR node performs channel detection individually, and if the node detects a PU signal, the node Send a detection alarm to the base station.

Due to a detection error, the CR node may falsely generate a PU detection alarm for an empty channel (false alarm), and the CR node may also fail to detect a PU that actually exists on the channel (false detection). In the present invention, false alarms of energy detection and feature detection are denoted by P _f (E) and P _f (F), respectively. Further, false detection probabilities of energy detection and feature detection are represented by P _m (E) and P _m (F), respectively.

The POMDP model for selecting the operating mode of the CR system is represented by a set of six components ( S, A, Ω, P, Φ, R ), where S is the non-occupied operation channel whether it is occupied by the PU. The state space of the core process that indicates if it is. S = {0, 1}, where state 1 indicates the presence of a PU on the channel, and state 0 indicates an empty channel. In the proposed architecture, since the base station checks the channel state in each decision interval for operating mode selection, the state of the core process is estimated only in the decision interval. Let s _t represent the state of the core process in the t-th decision interval (t = 0, 1, ...). Since the channel state in the (t + 1) -th period depends only on the channel state in the t-th period independently of the past state, the core process is the Markov process.

A is a space of actions corresponding to the operation mode of the CR system. For simplicity, the present invention will denote each action with one alphabetic character, that is, traffic transmission / reception and energy detection (T), feature detection (F), and channel switching (C). Thus, A = { T, F, S }. In the present invention, the action taken in the t-th determination section is denoted by A _t .

Ω is the state space of the observation process that represents the number of CR nodes that issue a PU detection alarm in T / R-ED or FD mode. Ω = {0,1, ..., M} when the CR system consists of M CR nodes.

. t번째 구간에서 액션(a)(∈ {T, F, C})이 취해질 때, t번째 구간과 (t+1) 번째 구간 사이의 간격은 ℓ_a이다. 그러므로 PU 액션 모델에 기초하여, a ∈ {T, F, C}에 대해, P is the state transition matrix of the core process. Let p _{i , j} (a) be the probability that the core process (channel) will transition from state (i) to state (j) when action (a) is taken. In other words,

. When the action (a) (∈ {T, F, C}) is taken in the t th section, the interval between the t th section and the (t + 1) th section is l _a . Therefore, based on the PU action model, for a ∈ {T, F, C},

Φ is a set of probabilities describing the interrelationship between the state of the core process and the state of the observation process. When action (a) is taken in the (t-1) -th section, let φ _{i, j} (a) be the probability that the observed value in the t-th section is j under the state of the channel state (i). If Ω _t represents the state of the observation process (ie, the observed value) just before the t-th interval, φ _{i, j} (a) = Pr {Ω _t = j | s _t = i, A _{t −1} = a}. It is noted that the observed values are obtained only in actions T and F. The observed value of j for the channel in state 0 (that is, an empty channel) means that j nodes generated false alarms, and the observed value (j) and channel state (1) indicate that only j nodes are channels. Indicates the presence of a PU on the phase. Thus, for action T,

Similarly, if action F is selected,

R is a compensation function. r _{i , j} (a) is referred to as reception compensation obtained by taking action a (X A ) in channel state i, which causes the CR system to transition to state j. At this time, the total reward associated with the action (a) in the state (i),

.

Let ξ (a) and ρ (a) be the advantages and penalties given to the CR system for selecting action (a), respectively. If the CR system transmits on the channel where the PU is activated, the CR system must be penalized. However, when transmitting on an empty channel, the CR system can benefit from using the channel without distributing the PU. Therefore, in the present invention, r _0,0 (T) = ξ (T), and r _{0 , 1} (T) = r _{1 , 0} (T) = r _{1 , 1} (T) = ρ (T) do. In addition, systems that move unnecessarily out of an empty channel are likely to receive penalties as switching channels unnecessarily degrades the performance of the CR system. Therefore, we say r _0,0 (C) = r _{1 , 0} (C) = ρ (C), and r _{0 , 1} (C) = r _{1 , 1} (C) = ξ (C).

On the other hand, if the CR system is not sure whether the PU exists or not, it is recommended that the CR system perform feature detection, and it is useless to detect without transmitting on an empty channel. Therefore, in the present invention, r _0,1 (F) = r _1,0 (F) = = r _1,1 (F) = ξ (F), and r _0,0 (F) = ρ (F). do.

The channel state s _t can be inferred only from the observed values {Ω ₁ , Ω ₂ , ..., Ω _t }, not directly observable values. Let π _i (0) be the probability that the channel state is zero at the initial time (ie π _i (0) = Pr {s _o = i}). Then, π (0) = (π ₀ (0), π ₁ (0)) is the initial distribution of the channel state. The distribution of channel states in the t-th interval is Π (0) and the joint trace of the action and observed values, (A ₀ , Ω ₁ , A ₁ , Ω ₂ , ..., A _{t -1} , Ω _t Can be obtained from In the present invention, it is defined that π _i (t) = Pr {s _t = i│Π (0), A ₀ , Ω ₁ , ..., A _{t -1} , Ω _t }. Then, π (t) = (π ₀ (t), π ₁ (t)) is called a confidence vector in the t-th decision interval. Note that it is a summary of all the information needed to make π (t) in the t-th interval. Therefore, the base station must calculate a confidence vector in each decision interval.

Based on the action taken in the t-th interval and the observed values measured in the (t + 1) th interval, the transformation of the confidence vector from the tth interval to the (t + 1) th interval Π (t + 1) = (Γ ₀ ( Π (t), A _t , Ω _{t + 1} ), Γ ₁ (Π (t), A _t , Ω _{t + 1} )) are specified using Bayes' equation as follows. When A _t = T or F,

Using equations (1) through (5) and (7), the base station can calculate π (t + 1) from Π (t).

On the other hand, there is no observed value when A _t = C. When moving to a new channel, the system attempts to select an empty channel from the channel pool. To this end, most of the CR system maintains an empty list of channels for switching through out-of-band sensing.

These candidate channels are likely to be empty, but some of them may actually be occupied by the PU due to sensing errors. Let P _vac be the probability that the new channel is actually empty in the t-th decision interval. Since the present invention assumes that all channels in the channel pool have the same PU action model, the probability that the state of the new channel is i in the (t + 1) th period in which channel switching is completed is expressed by Equation (8) below. same.

Since the confidence vector summarizes all the information needed to make a decision in each decision interval, the base station updates this confidence vector using the last action and the most recent observation and selects an action based on that. Policy δ: π → A is a function that maps from a trust vector to an action. There can be many possible policies. Of these, the base station is responsible for maximizing the total expected discounted sum of rewards, which is the sum of the rewards (immediate rewards) obtained immediately at decision time by taking action selected according to the policy and the rewards expected to be obtained in the future. δ ^* ) shall be adopted. It is noted that since future rewards are generally valued as not as valuable as instant rewards, these future rewards are discounted by a coefficient β (0 ≦ β ≦ 1).

In order to find the optimal policy, the present invention first takes the optimal value function (V ^* (Π)) and takes the system into account, starting with the confidence vector π = (π ₀ , π ₁ ) when the optimal policy is to be applied. It is defined as the total expected discounted sum of possible rewards. Let v _a ^* (Π) be the maximum possible expected discounted sum of the reward starting from the confidence vector Π when action (a) is taken. when a = T or F,

이다.here

to be.

In Equation 9, the first term is an instant reward and the second term is a discounted future reward. On the other hand, when the action (C) is taken,

At this time, for a given confidence vector Π, optimal value function, i.e., V ^* (Π) and optimal policy δ ^* (Π),

,

,

In the present invention, an optimal value function and an optimal policy can be calculated using a value iteration method based on dynamic programming.

3 shows the optimal value function of the three actions according to the probability that the channel is empty π ₀ . It is noted that π ₀ is the first element of the confidence vector π = (π ₀ , π ₁ ). The parameter values used to obtain the drawings are as follows. The number of CR nodes is M = 10, the discount factor for future compensation is β = 0.9, and the probability that the new channel selected for switching is actually empty is P _vac = 1.0. The durations of the three operating modes are set at l _F = 20 ms, l _C = 250 ms, l _T = 10 ms and l _E = 50 ms. And [lambda] = [mu] = 1/3600 per second, which means that the ON and OFF durations of the PU on the channel are 1 hour each. Furthermore, the advantages and penalties for each action are ξ (T) = 1, ρ (T) =-10, ξ (F) = 1, ρ (F) =-1 and ξ (C) = 10, ρ (C ) =-10. With this value, the present invention can use Equation 6 to calculate the compensation of each action for a given channel state. On the other hand, the probability of false alarm for energy detection and feature detection is set to 0.1%, that is, P _f (E) = P _f (F) = 0.001. And the probability of PU misdetection is P _m (E) = 0.173 at the base station, P _m (F) = 0.282 × 10 ⁻⁴ , P _m (E) = 0.893 at the other CR nodes, and P _m (F) = 0.035 to be. (These probability of false detection was obtained using a simulation model for the physical spectral sensing described below).

In the present invention, the total compensation obtained by selecting the T / R-ED mode is increased when the channel is more likely to be empty in FIG. It can be seen that it is much larger than the reduction rate in the FD mode. As can be seen from this figure, the higher the probability that the channel is empty, the more preferred the T / R-ED mode, and the lower, the more preferred the CS mode.

보다 더 큰 π₀에 대해 최적의 액션은 T/R-ED 모드이다. 그 결과 베이스 스테이션은, 신뢰 벡터(π₀, 1-π₀)를 계산하고 2개의 결정 임계값(κ _CS 및 κ _FD )과 π₀를 비교하여 CR 시스템의 동작 모드를 간단히 선택할 수 있다.Note that the goal in the present invention is to find the best policy that provides the best action for a given trust vector. 3 shows that the optimal policy in the proposed POMDP configuration can simply be represented by two thresholds ( κ _CS and κ _FD ). In the example of this figure, when π ₀ ≤ κ _CS , the optimal action is channel switching, and for κ _CS <π ₀ ≤ κ _FD , feature detection becomes the optimal action. And

The optimal action for larger π ₀ is T / R-ED mode. As a result, the base station can simply select the operating mode of the CR system by calculating the confidence vectors π ₀ , 1-π ₀ and comparing the two decision thresholds κ _CS and κ _FD with π ₀ .

The selection structure of the proposed operation mode optimally selects the operation mode of the CR system based on the POMDP configuration. Algorithm 1 described below describes the operation of the base station in the proposed architecture. In Algorithm 1, κ _FD and κ _cs are system parameters calculated using the proposed POMDP model.

4 is an example illustrating the adaptability of the proposed structure. The parameter value used is the same as that of FIG. If the PU is activated for the operating channel of the CR system, the number of PU detection alarms generated by the CR node in T / R-ED mode increases, which results in a decrease of π ₀ . The amount of decrease in π ₀ depends on the number of PU detection alarms. The more the node reports the detection alarm, the lower the value of π ₀ . It is noted that the number of PU detection alarms (ie, observed values) depends on time varying coefficients such as noise power and received PU signal power at each CR node. Due to the random nature of this observation, the successive operations performed until channel switching is determined vary over time. For example, in the scenario of FIG. 4, if a PU appears on channel 1 at 370 ms, the CR system operating on channel 1 performs two additional energy detections (in T / R-ED mode) and one feature detection. Go to channel 2. On the other hand, if the PU is activated for channel 2 at 100 300 ms, the CR system receives channel 3 alarms from as many nodes as the base station is assured that the presence of the PU is clear, so channel 3 is based on the results from two additional energy detections without feature detection. Determine the switching to As observed from this example, the proposed structure adaptively selects an operation mode according to the number of CR nodes generating a PU detection alarm used as a metric for channel empty.

Algorithm1: Operation Mode Selection

while CR system is active do

        j ← the number of PU detection alarms reported from CR nodes;

Calculate

Calculate

    end

    else

Calculate

    end

if

    end

else if

    end

    else

    end

    t ← t + 1;

end

The present invention uses the following simulation model proposed as a simulation methodology for evaluating the spectral sensing proposal of IEEE 802.22 WRAN. In this model, the PU is an ATSC DTV station operating at 617 MHz and using 1 MW transmit power. The antenna height of the DTV station is 500m and the channel bandwidth is 6MHz. The CR network consists of M CR nodes, namely one base station and (M-1) CPE stations. CPE nodes are randomly located in a circle with a radius of 16.7 km centered on the base station.

The base station and the CPE station each have an omnidirectional receive antenna having a height of 75m for the base station and a height of 10m for the CPE station. In the present invention, the received power at 617 MHz is obtained by adding -133 dB to the field strength, and the power received at the CR node using the distance between the CR node and the DTV transmitter based on the field strength for the F (50,90) curve. Calculate

5 shows the PU signal power received at the base station and the CPE station according to distance from the PU. In the present invention, when calculating the PU signal power received at the CR node, not only the path loss of FIG. 5 but also lognormal shadow fading with 5.5 dB standard deviation is considered.

이다. 이 모델에서 검출 임계값이

이고, 노드에서 수신된 PU 신호 전력이 P_R일 때, 이 노드에서 전력 검출기의 오 검출 확률은 다음 수식으로 주어진다.In the present invention, the test statistic value is an energy detection model of received signal power. References S. Shellhammer, S. Shankar N., R. Tandra, and J. Tomcik, "Performance of power detector sensors of DTV signals in IEEE 802.22 WRANs" , "in Proc. ACM TAPAS 2006, Aug. Uses power detection in 2006. Two hypotheses, that is, the hypothesis (H _o ) that the channel is empty, and the hypothesis (H ₁ ) that the channel is filled with PU are tested. When B is the channel bandwidth, N _o is the noise power spectral density, n _s is the number of samples, and the test statistic is a Gaussian with an average N _o B and variance (N _o B) ² / n _s at H _o . Random variable. And at H1, the test statistic is a Gaussian random variable with mean P _R + N _o B and variance (P _R + N _o B) ² / n _s , where P _R is the PU signal power received at the CR node . In the model used, the received signal is sampled at the Nyquist ratio, so the number of samples is at an energy detection time of ℓ _E.

to be. In this model, the detection threshold is

When the PU signal power received at the node is P _R , the false detection probability of the power detector at this node is given by the following equation.

는 0.1%의 오 알람 확률을 야기하는 값으로 설정되며, 이에 따라 P_f(E) = 0.001이다. 그리고 ℓ_E = 50㎲, B = 6MHz, N_o = -163dBm/Hz이다.In the simulation

Is set to a value that causes a false alarm probability of 0.1%, whereby P _f (E) = 0.001. And ℓ _E = 50㎲, B = 6MHz, N _o = -163 dBm / Hz.

As shown in Equation 13, the false detection probability greatly depends on the received signal power of the CR node, that is, the position of the node in the network. In order to mitigate the effect of the network type on the false detection probability, the present invention performs 100,000 simulation runs. In each simulation run, CR nodes are randomly located within a circle of 16.7 km radius, and the PU power received at each node is determined taking into account the log normal shadow with the path loss of FIG. 5 and the standard deviation of 5.5 dB. At this time, the instant of the false detection probability at each node from the simulation execution is calculated using [Equation 13]. Finally, the present invention averages false detection probabilities from 100,000 simulation runs as P _m (E) at the corresponding node.

와 테스트 통계 값 을 비교하여 PU가 존재하는지 비 존재하는지를 결정한다. 얻어진 테스트 통계 값이

보다 더 크다면, CR 노드에서 특징 검출기는 ATSC DTV 신호가 존재하는 것으로 결정한다. 그렇지 않으면 채널이 비어 있는 것으로 간주된다. 본 발명에서는 1.4의 조절 계수를 가지고 상기 참고문헌에 있는 [수식 35]를 사용하여

를 계산한 결과치인 0.1%의 목표 오 알람 확률 P_f(F)을 충족시키기 위해

를 취한다.

를 CR 노드에서 수신된 DTV 파일럿 신호 전력이 P_R이라 할 때 특징 검출기의 오 검출 확률을 나타낸다고 하자. 에너지 검출에서와 같이, 각 노드에 대해 본 발명에서는 시뮬레이션 실행으로부터

을 얻고 100,000번의 실행으로부터 이들 값을 평균하여 오 검출 확률 P_m(F)을 계산한다.In the present invention, reference is made to the reference H.-S. Chen, W. Gao, and DG Daut, "Spectrum sensing using cyclostationary properties and application to IEEE 802.22 WRAN," in Proc. IEEE GLOBECOM 2007, Nov. In 2007, we perform a simulation to evaluate false alarms and false detection probability of feature detectors with a cyclostationary feature detection model. In the model used, the PU (ATSC DTV station) transmits a pilot tone signal with periodic normal characteristics. Therefore, the present invention can use the period normal of the pilot tone signal to detect the presence of the ATSC DTV system. References H.-S. Chen, W. Gao, and DG Daut, "Spectrum sensing using cyclostationary properties and application to IEEE 802.22 WRAN," in Proc. IEEE GLOBECOM 2007, Nov. According to 2007., the pilot tone signal of an ATSC DTV is a simple cosine wave with a carrier frequency of f ₀ . It is assumed that additional white Gaussian noise is received with the pilot tone signal. The periodic normal feature detector uses the cyclic spectrum of the received signal to determine the presence of the PU. In order to obtain a cyclic spectrum, the present invention applies a fast Fourier transform (FFT) based discrete frequency smoothing method to a pilot tone signal. References H.-S. Chen, W. Gao, and DG Daut, "Spectrum sensing using cyclostationary properties and application to IEEE 802.22 WRAN," in Proc. IEEE GLOBECOM 2007, Nov. As in 2007, the present invention considers the case where the FFT size is 2048, the smoothing coefficient is 5, f ₀ is 2.69 MHz, and the time sampling increment is 0.18578 ms. In theory, the resulting periodic spectrum for the pilot tone signal has a nonzero value when the frequency is zero and the periodic frequency is 2f ₀ . Therefore, the period normal feature detector uses the absolute value of the periodic spectrum as a test statistic when the frequency is zero and the periodic frequency is 2f ₀ . Feature detector at the node detects the threshold

The test statistics are compared to determine whether the PU exists or not. The test statistics obtained

If greater, the feature detector at the CR node determines that an ATSC DTV signal is present. Otherwise the channel is considered empty. In the present invention, using an equation (35) in the above reference with a control factor of 1.4

To satisfy the target false alarm probability P _f (F) of 0.1%.

Take

Let s denote the probability of false detection of the feature detector when the power of the DTV pilot signal received at the CR node is P _R. As with energy detection, for each node in the present invention, from the simulation run

The error detection probability P _m (F) is calculated by averaging these values from 100,000 runs.

In the simulation, the present invention considers the case where noise uncertainty exists. Has the noise power spectral density N _o, if the noise uncertainty of η present in dB, the actual noise power spectral density is given by N _o + e, where e is an average of zero and a standard deviation of η Gaussian random variable.

Maximum channel usage and PU detection delay of the CR system are used as a measure for performance evaluation. The maximum channel utilization of the CR system refers to the percentage of time that the CR system can use for transmission, excluding the channel time used for operations derived from radiocognitive characteristics such as channel sensing and channel switching, and the PU detection delay is PU Is the period from the time of activation of the time immediately before the CR system starts channel switching.

Unless otherwise noted, the simulation parameter values are as follows. L _T = 10 ms, L _F = 20 ms, L _C = 250 ms, L _E = 50 μs, M = 10, λ = μ = (per second) 1/3600, P _vac = 1.0, β = 0.9, ξ (T) = 1, ρ (F) = -10, ξ (F) = 1, ρ (F) -1, ξ (C) = 10 and ρ (C) =- 10.

6 shows the influence of noise uncertainty η and the received PU signal power on the performance of the proposed structure. As can be seen in Figure 5, the received PU signal power of -98 dBm at the base station means that the base station is about 258 km away from the DTV station (PU), where a CPE station located at the same distance from this PU The average received PU signal power is approximately -116 dBm (according to IEEE 802.22-05 / 0007r48, "Functional requirements for the 802.22 WRAN standard," Nov. 2006. It should be noted that power should be able to be detected, ie -116 dBm is the maximum weakly received PU signal power required to detect. If the received PU signal power is weak or there is noise uncertainty, even if the base station receives a PU detection alarm from the CR node in T / R-ED mode, there is a very high probability that the alarm of the energy detector will go wrong with the PU on the channel. Not sure that it exists. In this case, the base station should command the CR node to perform feature detection for more accurate investigation.

As can be seen in FIG. 6 (a), in the proposed POMDP infrastructure, the value of the decision threshold κ _FD for selecting the FD mode becomes higher as the received PU signal power becomes lower as noise uncertainty increases. do. This allows the CR system to perform feature detection earlier and as a result the CR system can reduce the PU detection delay. Furthermore, in the present invention, it can be seen from Fig. 6 (a) that the decision threshold κ _CD for selecting the CS mode is hardly affected by the noise uncertainty and the received PU signal power. This is because there is noise uncertainty and feature detection still produces reliable detection results even if the PU signal is weaker.

On the other hand, as shown in Fig. 6 (b), the proposed structure provides high channel utilization even if the received PU signal power is very weak at high noise ambiguity. This is because a CR system with the proposed structure can avoid unnecessary feature detection caused by false alarms from energy detection by performing energy detection several hours before entering feature detection. Since the feature detection time is generally much longer than the energy detection time, it is noted that frequent feature detection in the absence of a PU significantly degrades channel utilization performance. In the proposed structure, since the confidence vector is updated according to the number of alarms, the number of energy detections between PU activity and feature detection is adaptively adjusted according to the PU signal power and the degree of noise ambiguity. Therefore, weaker PU signal power and / or higher noise uncertainty requires more energy detection prior to feature detection, resulting in longer PU detection delay. However, as shown in Figure 6 (b), the time taken to detect a very weak PU signal at high noise uncertainty is also defined as the maximum allowable PU detection delay and is usually shorter than the detection delay timeout given by 2 seconds in IEEE 802.22 WRAN. .

Figure 7 shows the performance of the proposed structure according to the energy detection time when η = 0 dB. Longer energy detection time is natural to improve the detection performance of the energy detector. However, since energy detection belongs to the T / R-ED mode which is the default mode, longer energy detection time further lowers the maximum channel utilization as shown in the figure. When the received PU signal power is very weak (P _R = -98 dBm), the number of PU detection alarms increases due to improved detection performance due to longer energy detection time, which means that the value of π ₀ is increased to κ _FD . To reach more quickly. As a result, feature detection is performed earlier, which results in shorter PU detection delay. That is, for the weak PU signal, there is a tradeoff between the PU detection delay and the maximum channel utilization depending on the energy detection time. However, when the received PU signal power is strong (P _R = -62 dBm), the number of PU detection alarms does not increase significantly even if the energy detection time becomes long. This is because, for this received power, the PU false detection probability of the energy detector is not high even with a short detection time. Therefore, when the CR system receives a strong PU signal, the longer energy detection time simply degrades channel utilization performance without increasing the detection delay.

8 shows the effect of feature detection time on the performance of the proposed structure when η = 0 dB. Assume that the CR network receives a strong signal from the PU. The various CR nodes in the network then detect the presence of the PU and report an alarm. Since the base station receives alarms from many nodes, it determines the PU appearance and determines channel switching without further detection. That is, the CR system determines whether to switch the channel only with the sensing result from one energy detection. Therefore, when the CR system receives a strong PU signal, the presence of the PU can be assured without feature detection, so the PU detection delay is not affected at all by the feature detection. However, when the received PU signal is very weak, the base station typically determines channel switching after performing several energy and feature detections. Therefore, in the case of a weak PU signal, the PU detection delay increases as the feature detection time becomes longer. On the other hand, it is noted that the proposed structure rarely performs feature detection when the channel is empty or the PU signal is strong. In other words, feature detection is primarily performed when there is a PU and the system receives a weak signal from the PU. As a result, the system only performs a few feature detections, so the influence of feature detection time on channel usage while staying on the operating channel cannot be detected.

For a performance comparison, references W. S. Jeon, D. G. Jeong, J. A. Han, G. Ko, and M. S. Song, "An efficient quiet period management scheme for cognitive radio systems," IEEE Trans. Wireless Commun., Vol. 7, no. 2, pp. 505-509, Feb. The discovery scheduling structure in 2008. is considered. In the present invention, this structure will be referred to as a continuous energy detection (ED) structure. In a continuous ED structure, the CR system periodically performs energy detection and performs feature detection after a system-wide PU detection alarm has occurred consecutively N times from the energy detection. Channel switching is then triggered by system-wide alarms from feature detection. In this structure, a system-wide alarm occurs whenever one or more CR nodes report a PU detection alarm to the base station, where the value of N is fixed to a predetermined constant.

9A-9C illustrate the effect of parameter N on the performance of the continuous ED structure when the PU signal power received at the base station is about -79.4 dBm. The received PU signal power of -79.4 dBm at the base station is described in S. Shellhammer, V. Tawil, G. Chouinard, M. Muterspaugh, and M. Ghosh, "Spectrum sensing simulation model," IEEE 802.22-06 / 0028r10, Sept. It is the default value proposed in 2006. It is noted that this corresponds to a distance of about 150.3 km between the base station and the PU. As can be seen from the figure, the performance of the continuous ED structure is greatly influenced by N, and the value of N, which provides good performance, is highly dependent on the noise uncertainty level. In the present invention, the value of N is set to 2, 8, and 32 for noise uncertainty of 0 dB, 0.1 dB, and 1.0 dB, respectively, considering the tradeoff between channel utilization and PU detection delay performance in a continuous ED structure.

In the present invention, while the POMDP-based structure proposed from FIGS. 10A to 10C provides a high channel utilization regardless of the noise uncertainty level, it can be seen that the channel utilization performance of the continuous ED structure becomes worse as the noise uncertainty becomes higher. . The reason for this is as follows. Consider an empty channel. The probability of false alarm from energy detection increases with higher noise uncertainty. The continuous ED structure performs feature detection unconditionally when N consecutive alarms occur. In addition, whenever one or more nodes falsely generate an alarm from feature detection, this structure switches the channel uselessly. Therefore, as noise uncertainty increases, the likelihood of unnecessary feature detection and channel switching in continuous ED structures increases, which degrades the channel utilization of the CR system. Meanwhile, as already mentioned in FIG. 6 (b), the proposed POMDP infrastructure adaptively controls a confidence vector for selecting an operation mode according to the number of alarms so that unnecessary feature detection / channel switching rarely occurs. Thus, the proposed structure provides very high channel utilization performance regardless of the noise uncertainty level.

Next, let us examine the PU detection delay performance for the two structures according to the received PU signal power. In the continuous ED structure, since the successive N system area energy detection alarms are a prerequisite for performing the feature detection, the PU detection delay is mainly dependent on the value of N regardless of the reliability of the energy detection alarm (FIGS. 10A-FIG. 10c). It is noted that the results detected from energy detection are more reliable for stronger PU signals. As a result, a continuous ED structure with a fixed N cannot be adapted in a time varying environment where the signal power and / or noise uncertainty level received from the PU changes over time. In contrast to the continuous ED structure, as already described in FIG. 6 (b), the proposed structure can adaptively adjust the number of energy detections according to time-varying network conditions. In other words, the weaker the PU signal, the higher the noise uncertainty, the more energy detection is performed before feature detection and channel switching. Therefore, as seen from FIGS. 10A to 10C, the PU detection delay of the proposed structure increases with weaker PU signal and higher noise uncertainty.

11 shows the maximum channel utilization and PU detection delay for two structures depending on the number of nodes in the CR network. In this figure, the noise uncertainty is set to η = 1.0 dB, and the base station is located 150.3 km away from the PU, thus the average received PU signal power at the base station is about -79.4 dBm. In the present invention, it can be seen that the proposed POMDP-based structure outperforms the continuous ED structure in terms of channel utilization and PU detection delay. The performance difference between the two structures becomes larger as the number of nodes increases.

Now, let's examine the performance characteristics of each structure according to the number of nodes. As shown in Fig. 11, the proposed architecture suggests very high channel utilization regardless of the number of nodes in the network. This is because a system with the proposed structure adaptively selects an operation mode according to observation values reflecting time-varying network conditions to avoid unnecessary feature detection and channel switching. On the other hand, in the proposed structure, the observed value (ie the number of PU detection alarms) is statistically more reliable when the system is configured with more nodes. Therefore, a base station with the proposed structure tends to perform feature detection after a smaller number of energy detections as the number of nodes increases, which results in shorter PU detection delay. Let's examine the continuous ED structure. In this architecture, the base station generates a system area alarm whenever at least one node reports an alarm, so the system area false alarm probability is higher as the number of nodes increases. As a result, unnecessary feature detection and channel switching occur more frequently as the number of nodes increases, which degrades channel utilization. When the PU is activated, if there are more nodes in the network, it is likely that more energy detection alarms have occurred due to the higher system area false alarm probability. Thus, the number of remaining energy detection alarms required for feature detection decreases with more nodes in the system. As a result, the PU detection delay becomes shorter as the number of nodes increases as shown in FIG.

In the present invention, the adaptive operation mode selection structure for the CR network with the central controller according to the POMDP structure is designed and its performance is evaluated. The proposed structure can be regarded as a kind of spectral sensing scheduling structure because its main role is to determine the type of sensing to be performed (ie energy detection or feature detection). The base station with the proposed structure estimates the probability that the channel is empty and based on this probability, the base station selects the operation mode of the CR network from among traffic transmission, energy detection, feature detection, and channel switching. Since the optimal policy for mapping a given probability of an empty channel to an operation mode is summarized by two thresholds in the proposed POMDP model, the base station simply compares these decision thresholds with the probability of an empty channel to simply define the operation mode. As a result, the proposed structure can be easily implemented without large computational overhead. In addition, the proposed scheme is designed so that unnecessary feature detection and channel switching rarely occur, so that the scheme can achieve very high channel utilization in a time varying environment where the number of active nodes, PU signal power, or noise uncertainty varies over time. Provide a short PU detection delay. Considering that the CR system should effectively protect the PU without degrading the performance of the CR node, in the present invention, the proposed structure is effective for detecting spectrum in current and future CR systems due to its excellent performance and high adaptability to time-varying network conditions. It is expected to be used for scheduling.

The present invention is not limited to the above-described specific preferred embodiments, and various modifications can be made by any person having ordinary skill in the art without departing from the gist of the present invention claimed in the claims. Of course, such changes will fall within the scope of the claims.

1 is a flowchart illustrating a POMDP based adaptive sensing scheduling method for a cognitive radio system according to an exemplary embodiment of the present invention.

2 is an example of a structure for selecting an operation mode in the present invention.

3 is a graph of the optimal value function of each action according to the probability that a channel is empty.

4 is an example of a confidence vector and an action over time.

5 is a diagram of received DTV signal power in relation to distance from a DTV transmitter.

6A and 6B are graphs of the effect of noise uncertainty on the performance of the proposed structure when M = 10.

7 is a performance graph of the proposed structure versus energy detection time when η = 0 dB and M = 10.

8 is a performance graph of the proposed structure versus feature detection time when η = 0 dB and M = 10.

9A-9C are graphs of the effect of noise ambiguity on the performance of a continuous ED structure when M = 10 and P _R = -79.

10a to 10c show performance comparisons between the proposed structure and the continuous ED structure according to the received PU signal power when M = 10.

FIG. 11 is a comparison of performance between the proposed structure and the continuous ED structure according to the number of CR nodes when η = 1.0 dB and P _R = -79.4 dBm. FIG.

Claims (11)

In the sensing scheduling method in a wireless cognitive system including a plurality of radio nodes (CR node) and a base station for performing channel detection, traffic transmission / reception, channel switching on a channel, Detecting the non-occupancy probability of the channel with the time variable information obtained through the channel detection, and adaptively performing the adaptive detection by adaptively adjusting the number of energy detections performed before feature detection or channel switching according to the non-occupancy probability of the detected channel. POMDP-based adaptive sensing scheduling method for cognitive radio system, characterized in that. 2. The method of claim 1, wherein the time varying information comprises first user signal power and a degree of noise uncertainty. 3. The method of claim 2, wherein the time varying information further comprises a number of active CR nodes that first generate a user detection alarm. The method of claim 1, wherein the non-occupancy probability is calculated based on a partially observable Markov decision method (POMDP). The cognitive radio system of claim 1, wherein the base station detects the non-occupancy probability by calculating a confidence vector through joint tracking of the initial distribution of the channel state, the last action, and the observed value in each determination section. POMDP based Adaptive Sensing Scheduling. 6. The method of claim 5, wherein the base station is characterized in that each wireless recognition node for accurate priority user detection even if the first user received power is weak or there is noise uncertainty, even if the first user detection alarm is received from the wireless or node. POMDP-based adaptive sensing scheduling method for a cognitive radio system characterized in that the command to detect. The method of claim 5, wherein the base station compares the detected non-occupancy probability with a preset first and second decision thresholds for determining an operation mode, and determines an operation mode based on the magnitude. POMDP-based Adaptive Sensing Scheduling Method for Radio Systems. The method of claim 7, wherein the operation mode includes an energy detection mode, a feature detection mode, and a channel switching mode. 8. The method of claim 7, wherein the determination of the operation mode determines the traffic transmission and reception and energy detection modes when the non-occupancy probability of the channel is greater than the first determination threshold, and wherein the non-occupancy probability of the channel is the second. In case of less than the decision threshold, the channel switching mode is determined. If the probability of occupancy of the channel is greater than the second decision threshold but less than the first decision threshold, the cognitive radio is determined. POMDP-based Adaptive Sensing Scheduling Method for System. A sensing scheduling method in a cognitive radio system including a plurality of radio recognition nodes (CR nodes) that perform channel sensing, traffic transmission / reception, and channel switching on a channel, Each radio aware node identifying scheduling information for a frame having a corresponding fixed length, including a traffic transmission schedule and a start time (and length) of energy detection; The radio aware node transmitting / receiving data traffic, sensing a channel for energy detection, and transmitting the detected channel sensing information to a base station; Detecting, by the base station, a non-occupancy probability of a POMDP-based channel based on the sensing information; Comparing the non-occupancy probability of the detected channel with first and second decision thresholds for determining an operation mode; If the occupancy probability of the channel is greater than the first determination threshold, determining an energy detection mode and adjusting the number of energy detections according to the occupancy probability; If the non-occupancy probability is less than the second determination threshold, the channel switching mode is determined. If the non-occupancy probability of the channel is greater than the second determination threshold but less than the first determination threshold, the feature detection mode is selected. POMDP-based adaptive sensing scheduling method for a cognitive radio system comprising the step of determining. 11. The method of claim 10, wherein the non-occupancy probability of the channel is first calculated using time varying information including user signal power and a degree of noise uncertainty.

Images