Using Neighbor’s State Cross-correlation to Accelerate Adaptation in Docitive WSN

ISSN 2349-7815
International Journal of Recent Research in Electrical and Electronics Engineering (IJRREEE)
Vol. 4, Issue 1, pp: (38-48), Month: January - March 2017, Available at: www.paperpublications.org
Page | 38
Paper Publications
Using Neighbor’s State Cross-correlation to
Accelerate Adaptation in Docitive WSN
Dr. Charbel Nicolas
Computer department, AUL University, AUL University Campus, Dekwaneh, Lebanon
Abstract: In WSN, sensor nodes have limited energy budget therefore this paper mainly focus on power saving by
using the docition paradigm. Docition is a new teacher-student paradigm proposed to improve cognitive radio.
Although it improves the infrastructure based networks it has a weakness in case of ad-hoc mobile networks. The
energy constraints and the total mobility of the network complicate the selection of the appropriate teacher for a
student. By selecting the wrong teacher, there is a high probability that the taught information may be faulty, and
thus the student radio diverges from the best state. This causes a high amount of energy loss, though the most
important concern in ad-hoc networks is energy limitation. In this paper, we propose a dynamic docition for
teacher selection based on the auto-correlation degree of the teacher’s candidate environment and the cross-
correlation degree between the teacher candidate and the student environments. We validate our approach in the
context of coexistence between WSN and WiFi. The WSN detects, models and exploits the unused time slots in the
electromagnetic spectrum, left by WiFi, using dynamic docition. The simulation results show that the use of
dynamic docition outperforms the existing docition in mobile networks. The improvements are shown through the
low link overhead percentage (20% less overhead) and the low packet loss ratio (30% improvement).
Keywords: Docitive; Online Prediction Problem; WSN; pareto model; IEEE802.11 b/g;cognitive radio.
I. INTRODUCTION
Now a days the radio spectrum congestion is the major problem that wireless sensor networks are facing [12][11].
Recently, in the domain of cooperative and distributed networks, to mitigate congestion and radio problems, a new
paradigm has emerged called docitive networks [1]. Docitive network is a wireless network, where the docitive radio is an
upgrade of cognitive radio [2]. Cognitive radio uses artificial intelligence and machine learning algorithms to process
local channel observations. In addition to local observations, docitive radio includes information from the neighboring
nodes to this process. The purpose of this information exchange is to accelerate and improve the decision process.
Docition from the Latin docere is to teach, which relate to radios that teach other radios [1]. From this view docition is the
teacher-student paradigm. It is generally known that intelligence is impacted by the degree of observation. Still as
cognition and learning have received a significant attention from various communities in the past, the process of
knowledge transfer, the teaching, over wireless medium has received fairly little attention to date. Similarly to distributed
cognitive radio, docitive radio needs to cooperate with neighboring radios to have docition. The difference from regular
distributed cognitive networks is that in docition the student teacher paradigm does not impose the use of the exchanged
information: the teacher does not teach end results but proposes elements of the methods for getting there.
In the concept of docitive network, there was no indication about the circumstances to use docition, implying that it is
always useful to use docition. Specifically when the authors in [1] defined the degree of docition they just cited the
degrees as static choices, hence the node do not have dynamic behavior as to choose its own degree of docition. The
degrees of docition ranged from no docition to perfect docition.
During docition, exchanging the Q-learning [3] table is essential as it contains the information of docition. As this table
contains the information related to the observed state of the teacher it is crucial that a student node have the similar
environment state as the teacher node. We claim that although docition provides fast convergence to the learning node,

ISSN 2349-7815
Page | 39
Paper Publications
the assumptions taken by the paradigm impose coherence between the states of the teacher and the student. The reason is
there are no measures taken in the docitive paradigm to insure that the teacher and student nodes have coherent states. In
our study we are interested in mobile networks with no infrastructure such as MANET and WSN. We note that in such
networks, nodes may not have coherent states, moreover the same weakness appear in the case of an unlicensed densely
populated frequency band as the ISM band. For the rest of the paper we refer to the previously presented docition as
classic docition (CD).
As a solution, in this paper, we present dynamic docition (DD) that enables the teacher and the student nodes, in case of
mobility or incoherent channel state between them, to specify the level of docition dynamically based on the level of
environment predictability. We add the environment predictability probe (EPP) as a new element to classic docition. This
element estimates the level of coherence between the teacher and student nodes, and, based on the result, if the
environment state is stable enough and there is a certain level of coherence, it is acceptable to apply docition otherwise no
docition is applied.
To evaluate the DD performance, we apply it on the case of a coexistence situation between IEEE802.15.4 (wireless
sensor) and IEEE802.11b/g (WiFi) technologies, where sensor nodes are mobile. To escape the WiFi interference, the
sensors adopt the White Space (WS) modeling technique and the Pareto model proposed in [4]. The behavior of the nodes
in this use case scenario is similar to the cognitive radio that adapt an unlicensed user (known as secondary user) to
coexist with a licensed user (known as primary user) [5]. The use of such scenario in our case is to provide to weak
technologies as the technologies using IEEE802.15.4 standard (limited power supply, low power, low processing and
transmission rate) the adaptation means to coexist with dominant technologies like IEEE802.11 (unlimited power supply,
high transmission power and rate, and high traffic generation). Applying DD in this scenario improves the efficiency of
the power consumption and provides faster convergence time.
Intuitively, we expect that as the system becomes more uncorrelated and more dynamic, the use of CD becomes
disadvantageous. In this paper, we quantify this claim by simulations, we propose DD, and we chose to apply it in the
case of startup docition on WSN as it has the least energy cost relatively to other degrees of docition and validate its
performance by the simulations.
II. DYNAMIC DOCITION
CD is composed of four elements: acquisition, intelligent decision, action, docition. The main contribution of DD on
mobile networks is that it provides to the student an efficient manner to select a teacher and to know dynamically when to
apply docition or not. In mobile networks, the teacher is the most important actor as it should give credible teachings. To
determine the teacher existence DD is used. The DD adds a new element on CD which is the EPP to be applied before the
docition element. Some models characterizing the spectrum activity can always be assumed (cf. [6], [7]). These models
predict the spectrum occupation, and the duration of each recurring event, therefore based on these assumptions the EPP is
used to construct a local channel model and estimate the level of similarity between teachers and students for present and
future states.
The predictability of the environment is detected using correlation functions and stochastic modeling. Due to the mobility
in ad-hoc networks there are some conditions related to the model integrity that are assured by the EPP. If the model
validation test executed by a node is done before it changed its location then no teaching can be made by the node (GPS or
the method in [10] can detect a location change). On the other hand if the node did not change its location after validating
the model then the node is accepted as a teacher candidate. Although in [8] it was stated that if two nodes are able to
communicate implies a minimum level of cross correlation between their environments, the EPP must assure the existence
of such minimum level of cross correlation between the teacher and the student environments.
III. DYNAMIC DOCITION FOR WHITE SPACE DETECTION
To illustrate the application of dynamic docition, we use the white space detection in a coexistence environment between
a wireless sensor network (WSN) and a WiFi network. In this case of coexistence, a white space is the inter-arrival time
between two consecutive WiFi bursts. The problem in such coexistence scenarios is that the WiFi nodes do not detect the
WSN nodes transmissions. Therefore the WiFi APs do not backoff when there are transmissions from the WSN. The aim
is that each sensor node (SN) tries to model the white spaces that are not used by WiFi and based on the model the packet

ISSN 2349-7815
Page | 40
Paper Publications
length or the number of packets to be sent is determined. The goal of applying DD is to optimize the convergence time
which impacts the energy cost for a newly arriving SN to the WSN. Each of the existing nodes will send the white space
model parameters to the new node along with the gain (the duration of applying the same model parameters) that was
provoked by using the model. The new node chooses from all the neighboring teacher candidates the best suited model to
use as its own startup model.
To construct the model, the teacher candidate will estimate the gain using the online prediction problem. In the evaluation
of our proposal we use the modeling mechanism proposed in [4] as a use-case to illustrate the improvements made by DD.
As our aim is mobile network with no infrastructure, in this type of networks with limited energy supply, the mission of
DD is to swap between startup docition and no docition. We adopt startup docition for white space detection. In startup
docition, the initial concept, docitive radios teach their policies to any newcomers at the beginning when the nodes are
going ON. We must emphasize that for the initial concept, the aim was static networks, and a newcomer is a base station
that is going from OFF state to ON state. Based on this assumption Gains are expected due to the highly correlated
environment and due to the fact that teaching nodes have an updated environment model where they have applied their
cognitive actions. In the case of ad-hoc networks none of the above assumptions exists, therefore DD should be applied.
The Pareto model is declared by [4] as an acceptable model for WiFi white space modeling. Therefore each node
will try to construct such model to be able to communicate with other nodes in the WSN.
A. Mecanism of the Teaching Node:
The teaching node must apply these steps:
1- Sampling the channel, and calculating white spaces durations ( ) (inter-channel occupation time).
2- Modeling for future prediction by detecting P( , ) using maximum likelihood estimation (MLE) [9]:
∑ ; is the minimum[ ]; is the number of WS samples.
3- Saving the time (duration) of the application of P ( , ).
4- Applying online learning problem (on the model) to estimate the model efficiency. A newly calculated Pareto
exponent β’ is adopted only if it does not belong to this interval . If β’ is adopted then the time is
reset (duration of the model application), go to step 2. If the level of resemblance between the old and the new model
(gain) is high, the time (duration) is not reset. k is the acceptable percentage of deviation between .
5- Setting itself as teaching candidate if it did not change its location after validating the model and the gain is high.
6- Sending its teaching credentials (model parameter, application & testing duration) if it is a candidate teaching node
and it detects a new node.
Remarks: 1- The Gain in (4) is the goodness of fit between the model and the new samples. This method for determining
the efficiency of the chosen model is better than comparing the throughput level before and after a model change, because
a sender does not need to cooperate with the receiver to know the throughput. 2- EPP covers from step 3 to 6 at the
teaching node and all the steps at the learning node mechanism.
B. Mecanism of the Learning Node:
1- If the node detects a location change, it sends a beacon.
2- The node receives the credentials of the candidate teaching neighbors.
3- A minimum level of consensus between the teaching candidates should exist. If the correlation among them is low no
docition is applied, else the node goes to step 4; the Average of the received E( ) is calculated and the deviation
percentage from this average is analyzed. If the deviation is higher than K (ex: K=10%) then there is no correlation. The
acceptable teacher candidates are the nodes that have the existing in the interval [ ] where
; N is the number of teacher candidates, is the received Pareto shape, K is the ratio that restricts the accepted
correlation level between the neighboring teacher candidates. Moreover, it is used to filter the teacher candidate nodes.

ISSN 2349-7815
Page | 41
Paper Publications
4- The node selects as teacher the one with the highest gain and that has a β that falls in the accepted deviation. If there is
no acceptable candidate, no docition is applied.
5- After a time of stability, the node changes its state from learning node to teacher node.
Step 3 explanation: the main idea is to prove that CD is unsuitable for Ad-hoc networks. If the neighborhood has diverse
states or has high deviation from an average β there is no docition. Otherwise, if there is a quasi-total agreement to one
choice of β then apply docition. In case of DD the deviation percentage determines the restriction on applying docition. If
there is a loose interval for β deviation the docition may be applied most of the time with a consequence on the Packet
Loss Rate (PLR) (cf. Fig 2, Fig 3).
IV. SIMULATION
We used Matlab simulator to compare our proposed DD protocol against CD performance. On a surface we set a grid of
overlapping WiFi access points (APs) and wireless sensors nodes. Each AP has a coverage area (πR2
) and a permissive
overlap area π (R2
-R12
). The permissive area usually exists to allow an overlap between multiple APs and a certain level
of interference. This is to enable the hand over between mobile nodes and APs. The sensor nodes are distributed using the
Poisson point process (PPP) distribution and submitted to the interference of the WiFi APs.
Each AP is assigned a WiFi traffic generator. The sensor nodes sample the durations of the channel vacancies between
WiFi transmissions (the White spaces “WS”) and use the MLE to reconstruct their model. From [4], the sensors adopt the
Pareto model construction methods and steps. The Pareto model used has two parameters; α and β. α is determined as the
minimum acceptable WS length “1ms” and β is found using the MLE and the WS samples. To sample the WS samples
we adopted the same parameter values used in [4] as they were optimized using empirical results: 100ms windows are
used for sampling the channel state. As a consequence the maximum white space length can reach 100ms. The sampling
rate used is 200Hz.
For the teaching node, we fixed at 10% the acceptable deviation from the old beta. Henceforth, by the acceptable
deviation we refer to the K value at the learning node mechanism.
The simulation is divided into two parts. The first part deals with the energy cost of the signaling stage of docition. The
second part compares the performance of implementing DD and CD on a WSN.
Energy Cost of The Signaling Stage Of Docition:
The sensors can construct the WS Pareto model using two methods: the direct one based on the clear channel assessment
(CCA) samples and the indirect one based on docition information received in the packets header. A comparison between
the energy cost of both is done in fig 1. First we calculate the energy cost for a CCA sample and the energy cost for
adding the docition information in the header, then for the case of the CCA sampling we multiply this cost with a number
of samples and for the case of docition information we multiply it by the number of teacher candidates. The energy cost in
the case of docition information exchange is related to the transmission and reception duration of the number of bits.
Therefore for docition the added cost is equal to the cost of the extra bits sent, we have in total 16 bits sent:
-2 bits are used to indicate the duration of the model usage (the gain) (cf. Table 1):
Table 1: duration mapping
00 01 10 11
t(ms)< 100 t>=100 & t<200 t>=200 & t<500 t>= 500
-14 bits are used as the float value of beta:
* 4 bits are used for the integer part (max value =15)
* 10 bits are used for the decimal part (max value =1024)
Due to the addition of the docition information to the beacons and to insure that the cost will not exist during all the
beacon exchange but just when there are docition, the packet length indicates if the 16bits are added or not so the nodes
do not always execute the additional 16 bit listening.

ISSN 2349-7815
Page | 42
Paper Publications
The energy cost is E=U*I*t.
T:duration(Seconds) U: voltage(Volt) I: current(Amp)
The CCA is performed by sampling 8 symbols the duration is128microseconds.
The duration for transmitting or receiving one bit is 64 microseconds and it needs 0.0174Amp for transmission and
0,0197Amp for reception, the voltage used is 3V.
Figure 1: the energy cost in function of the number of the 16bits exchanged in case of docition and the number of CCA samples
in case of sampling.
In fig 1 the docition signaling (middle curve) is less costly than the sampling (highest curve) and at the same time it
distributes the energy between the sender and receiver. The cost on the receiver side is represented by the lowest curve.
System Simulation:
In what follows we simulate the total system of mobile sensors and fixed WiFi AP. A grid of 36 WiFi APs with a
transmission radius of 100m is covering a surface of 106
m2
. The sensor nodes coexist with these APs on the same surface
and have a transmission radius of 50m.
The rate of selecting valid information from the neighbors is monitored at section a), since docition based methods main
objective is to obtain information on the environment faster than self-acquiring of such information. We compare three
methods of selecting the beta value: CD that is based on choosing the β value randomly from the neighbors, DD that is
based on EPP and choosing the β value that occurs most often among the neighbors due to cross-correlated wireless
spectrum state. The use of docition to obtain information is precisely to use them to improve the network performance.
Therefore, in section b), we examine the PLR and the percentage of overhead introduced by docition.
Since we are interested in “startup docition”, the simulation consists of bringing outside nodes and placing them evenly in
a plane which already contain distributed sensors based on PPP and WiFi APs. This simulation is repeated enough times
to get good confidence intervals. As we study the “startup docition”, the worst case scenario is chosen for the mobile
nodes where they are imported from an environment where there is very little WiFi traffic, therefore, they have big WiFi
white spaces compared to the network nodes.
We performed four types of simulations. The first two observe the effect of the parameter K used by the student to filter
the received values of β. During these two, 30 nodes are imported and the results are observed depending on the number
of nodes in the network. In the third simulation, the nodes density is increased but the proportion of new neighbors for a
node coming from outside (student) remain unchanged. Finally, in the fourth simulation, the total number of nodes in the
network remains constant while the percentage of new neighbors increases.
While the first two scenarios correspond to the case where the existing sensor nodes are fixed and the only movement is
the arrival of new nodes, the latter two allow observing the behavior of the sensor network in a mobile situation. In fact,
the mobility has the effect that new nodes arriving from outside find in their new neighborhood, nodes that have a good
estimate of beta as they have had time to get it, and other nodes newly arriving from elsewhere in the network without

ISSN 2349-7815
Page | 43
Paper Publications
having consistent information about the environment. Therefore, by increasing the number of nodes arriving from outside,
we can observe the mobility effect of sensors already in the network on the percentage of correct selection of information
by docition. We call the first two scenarios "quasi-static scenario" and the last two "mobile scenario."
a) Testing the efficiency of the methods used in docition:
Quasi static scenario:
For each node is assigned the β of the traffic generated by the AP that covers it. The probability of choosing the right β
from the neighbors is tested. In the static scenario all the neighbor nodes are supposed static. Therefore they are all
teacher candidates because by construction of the scenario, they all have a good estimation of beta. The acceptable
deviation percentage is set to 20% and R1=75% of R. In fig 2 the results show that for the CD, the probability of selecting
the correct neighbor is relatively unchanging (~67%) in function of the number of nodes covering the region. In case the
selection of β is based on the maximum number of neighbors that have adopted its value, the percentage of correct
selection increments in function of the number of existing nodes. It presents a better estimation than the CD. Although,
the latter performed better than the CD, the DD based on the deviation percentage from the average (middle column in Fig
2 and Fig 3) shows the best performance between the 3 methods.
Figure 2: percentage of the correct selection of the true beta value in function of the number of nodes that existed before the
arrival of the student nodes.
We restrict to 10% the acceptable deviation percentage for DD. A better performance is shown in Fig 3. The drawback of
additional restriction on the β selection results in more restrictions on the use of docition.

ISSN 2349-7815
Page | 44
Paper Publications
Mobile scenario:
In the mobile scenario, the mobility of the neighbors eliminates their candidature to become teacher nodes in the case of
DD. For the CD there is no distinction between selecting such nodes or selecting a teacher candidate node. Fig 4 shows
the degradation of the classic docition. By contrast the results of DD are improving. Specifically, the mobile non teacher
candidates cause a high diversity in the β values. Therefore, in the CD there is a high probability to select the wrong β
value. The results shown in Fig 4 are influenced by forcing 70% of the new nodes neighbors to be imported from random
locations.
For the experiment of Fig 5, 300 sensor nodes are set as existing nodes. The non expert neighbors’ percentage is adjusted
at the same time the percentage of the correct β selection is monitored. It is shown that the non expert nodes do not
influence the nodes that are using DD, but they affect the CD because there are no distinction between the expert teachers
and the non expert ones.
Figure 5: percentage of the correct selection of the true beta value in function of the percentage of inexperienced neighbors.
b) Using channel sampling and MLE to determine the suitable WiFi WS model for traffic generation
Previously, in subsection a) quasi static scenario, the student algorithm used in DD has proven to be highly efficient rela-
tively to classic docition. For this part, a noise is indirectly introduced through the fact that the β values are not assigned
directly to the nodes, but rather the nodes use channel sampling and the MLE to determine the APs WS model. In the
subsection b) quasi static scenario, the DD and CD should have similar performance due to the correlated neighbors’
environment, in contrast to what is going to happen in the mobile scenario where the performance of the teacher algorithm
is highlighted. In this part, we compare the efficiency of 3 methods used to construct the Pareto distribution that models

ISSN 2349-7815
Page | 45
Paper Publications
the WiFi WSs. The methods are: direct channel sampling with the use of MLE, CD and DD. The effect of the deviation
from the WiFi WS model is determined. The deviation has an effect on the PLR and on the link overhead percentage.
Quasi static scenario:
Henceforth, the WSN nodes transmit packets with a length of 2ms. The WSN nodes predict WiFi WS lengths by finding
the best β value as the appropriate Pareto shape. For the rest of the experiments we set the acceptable deviation percentage
for DD to 10%.
The PLR and the link overhead ratio are monitored, in Fig 6 the traffic performance is shown. CD uses random β selection
from the neighbors. The same performance tests are done for the DD. The traffic performance when using the β found by
channel sampling and using the MLE represents the best performance that can be reached, because by executing the WS
sampling the WSN nodes have the exact WiFi WS model that governs the wireless channel. R1 is set to 50% of R. In the
case of static neighborhood, all the types of docitions along to the model construction using sampling and MLE have
approximately the same performance level. The compared results show convergence to the best link characteristics. The
overhead percentage reaches 38 % and the packet loss is 22% as shown in Fig 6. Although, we have similar performance
outcome for the 3 methods, the DD showed a slightly better performance than CD.
In Fig 7 the same experiment as in Fig 6 is executed but more surface intersection is allowed between the WiFi APs (R1
equal 70% of R) to add more diversity for the neighboring β values. We suppose that by doing so, there will be a higher
probability for a false β selection.
Figure 6: the upper graph represents the overhead % and the lower graph represents the PLR both graphs are in function of
the number of nodes that existed before the arrival of the student nodes.
The higher PLR in Fig 7 relatively to the PLR in Fig 6 is caused by the small WS length that remains from the incre-
mented percentage of superposed WiFi APs coverage area. All the docition methods have similar high performance, due

ISSN 2349-7815
Page | 46
Paper Publications
to the stationary environment. The only cause to have a wrong β selection is by selecting a neighbor that exists in a
different region of coverage. Although it has a low probability, its effect is shown by the weaker performance of classic
docition.
the number of nodes that existed before the arrival of the student nodes.
Mobile scenario:
In the following, R1 is set to 70% of R. By introducing the mobility aspect (50% of the neighbors are not teacher
candidate), Fig 8 shows a high deterioration in the CD performance. In contrast, the resilience and stability of DD main-
tained a low PLR while keeping a minimum overhead percentage. Using the MLE as reference to estimate the
performance of DD and CD, DD maintained a PLR of 20% the same as MLE. On the other hand, CD’s PLR diverges and
reaches 45%. During the simulation, the overhead percentage did not vary for CD and DD maintaining the value of 20%
for CD and 30% for DD. Similar to fig 8, in Fig 9 the overhead percentage stabilizes on 17% for CD and 30% for DD.
Due to space limitation we only show the overhead percentage for the case of fig 10 as it shows a deviations in the
overhead percentage ranging between 14% and 28%.
Figure 8: the PLR in function of the number of nodes that existed before the arrival of the student nodes.

ISSN 2349-7815
Page | 47
Paper Publications
In Fig 9, the percentage of the neighbors that are none teacher candidate is raised to 70 %. More degradation of the CD
performance is observed. The PLR increases in function of the number of existing nodes, till it reaches 50%. In contrast
the DD performance is not influenced by the added inexperienced neighbors.
Next, 500 existing nodes are used. Fig 10 is the result of varying the percentage of the inexperienced neighbors. The CD
has an incrementing PLR with the percentage of inexperienced nodes. The DD proved to be the best suited for mobility,
as it prevents the inexperienced nodes from poisoning the learning nodes.
Figure 9: the PLR in function of the number of nodes that existed before the arrival of the student nodes.
What is presented is an added intelligence on when to apply docition. At the same time the selectivity when choosing a
teacher provoke an improvement on the decision process even though a simple method for the teacher selection is used.
More intelligent methods can be proposed, but the main idea in this approach is not to apply randomly the docition in case
of a mobile network. The results show that the CD and DD achieve more or less the same performance in a static
environment. In contrast, it is clearly seen the destructive effect of mobility on CD as the number of mobile inexperienced
nodes increases. There is practically no effect on the DD as it filters the non expert mobile nodes.
the number of nodes that existed before the arrival of the student nodes

ISSN 2349-7815
Page | 48
Paper Publications
V. CONCLUSION
Based on cognitive radio, docition provides a mean to improve the methods of processing and reaction to link variations. In
our work, we added the environment predictability aspects for docition, to improve the learning, optimize the energy
consumption and to catalyze fast convergence of the wireless sensor nodes. By proving there is environment stability for
the teaching node EPP proves that there is some state information that is worth teaching. Otherwise, if the environment
state is chaotic, it is difficult to find useful information to teach. The EPP test the cross correlation between the student and
the teacher to indicate the existence of useful information for the student among the taught data. Compared to CD, the
simulation showed that DD provides an improvement of more than 50% of correct beta value’s selection. An improvement
is also shown on the PLR. Moreover the added energy cost on the signaling is estimated and it shows that exchanging the
docition information has a lower energy cost than allowing a node to do its own channel sampling.
ACKNOWLEDGEMENT
The authors want to thank Dr. Michel Marot for his help useful comments on this paper.
REFERENCES
[1] L. Giupponi, A. G-Serrano, P. Blasco, M.Dohler, “Docitive Networks – An Emerging Paradigm for Dynamic
Spectrum Management”, IEEE Wireless Communications, 2010.
[2] S. Haykin "Cognitive Radio: Brain-Empowered Wireless Commun-ications"in IEEE J. Select. Areas Commun,
2005.
[3] M. E. Harmon and S. S. Harmon, “Reinforcement Learning: A Tutorial,” 2000.
[4] J.Huang, G.Xing, G. Zhou, R. Zhou, “Beyond Co-existence: Exploiting WiFi white space for zigbee performance
assurance”, 18th IEEE International on Network Protocols ICNP, 2010.
[5] J. Mitola, “Cognitive Radio,” Licentiate proposal, KTH, Stockh-olm,Sweden, December 1998.
[6] Y. Xu, W.Lee, “Exploring Special Correlation for Link Quality Estimation in Wireless Sensor Networks” in PerCom
2006.
[7] R. K. Guha, S. Sarka, “Characterizing Temporal SNR Variation in 802.11 Networks” in IEEE transaction on
vehicular technology 2008.
[8] S.A. Jafar, S. Srinivasa, “Capacity Limits of Cognitive Radio With Distributed and Dynamic Spectral Activity” ICC
2006.
[9] J. L. Petersen, “Estimating the Parameters of a Pareto Distribution Introducing a Quantile Regression Method”.
Technical report 2000.
[10] C.Nicolas, M.Marot, M.Becker, "A Self-Organization Mechanism for a Cold Chain Monitoring System", VTC
Spring, 2011.
[11] Y. Wu, K. S. Liu, J. A. Stankovic, T. He, S. Lin, " Efficient Multichannel Communications in Wireless Sensor
Networks", ACM Transactions on Sensor Networks (TOSN) Volume 12 Issue 1, March 2016.
[12] M. R. Mundada, P. B. Desai, “A survey of congestion in Wireless Sensor Networks”, 2016 International Conference
on Advances in Human Machine Interaction

Using Neighbor’s State Cross-correlation to Accelerate Adaptation in Docitive WSN

More Related Content

What's hot (19)

Similar to Using Neighbor’s State Cross-correlation to Accelerate Adaptation in Docitive WSN (20)

Recently uploaded (20)

Using Neighbor’s State Cross-correlation to Accelerate Adaptation in Docitive WSN