1.2. Background on Cognitive Radio
The inefficient usage of the limited frequency spectrum makes it difficult to meet the increasing demand for wireless communication capacity [
5–
7]. Cognitive radio has been introduced as a solution to this problem. Cognitive radio is defined as “a radio that can change its transmitter parameters based on interaction with the environment in which it operates” [
8]. The term “cognitive radio” was first introduced by Mitola in 1999 [
9]. Cognitive radio is an evolved version of software-defined radio that can reconfigure itself, such that it can adapt its waveform parameters to the environment to meet higher-layer user demands for a high quality of service (e.g., voice over IP, video). Cognitive radio implementations fall between two extremes. At one end is the Mitola radio, a radio that collects information about all observable wireless information. As such, it is a theoretical construct that cannot be implemented in practice, but provides an ideal to aim for in cognitive radio research. At the other end is what can be practically implemented, which may be a spectrum-sensing cognitive radio with information about only the frequency spectrum [
10]. In our work, we seek to move towards the Mitola radio ideal by employing more and varied information in our control loop, such that the cognitive radio can make more informed decisions.
One can define the cognitive ability of a radio as capturing and gathering information regarding the state of the environment, processing this information and then determining corrective action based on its findings. This cognitive process is not limited to monitoring the power level in a specific frequency band, but can also include the spatial and temporal variations in the radio environment due to the mobility and time dependency of most wireless devices. Based on the users’ demands, these devices need to access a free and/or unused spectrum band at different times and/or locations. Reconfigurability empowers the radio to dynamically adapt to a changing radio environment [
10]. This means that the cognitive radio will adjust to communicate in an appropriate frequency band and with a suitable waveform (
i.e., modulation type).
Spectral and temporal analysis of the radio spectrum reveals three broad categories of frequency band usage [
11]:
Frequency bands that are predominantly unoccupied;
Frequency bands that are moderately occupied; and
Frequency bands that are heavily occupied.
Note that a channel may be heavily occupied at one period in time, but not at another. Cognitive radio offers opportunistic usage of the frequency spectrum if permitted by primary users who currently “own” that slice of spectrum [
12]. This process is called dynamic spectrum access, which may rely on algorithms and concepts found in game theory and network information theory.
A dynamic system is defined to be cognitive if it employs the perception-action cycle and has memory, attention and intelligence [
13]. In the perception-action cycle depicted in
Figure 1, a perceptor gathers measurements and sends them as feedback information to an actuator that uses this to control the perceptor via the environment. Memory is needed since the environment is nonstationary. The actuator prioritizes the allocation of limited resources, and feedback enables the presence of intelligence in this system by providing the perception of the environment to the actuator.
Three fundamental cognitive radio tasks based on the perception-action cycle are introduced in [
11]. For the receiver, information gathering and analysis must be performed to determine the condition of the dynamic radio environment. For the transmitter, power budgeting and dynamic spectrum access based on information regarding the presence of the primary users must be calculated and executed. Finally, there must be a feedback channel between the transmitter and receiver regarding information about the radio environment. A cognitive radio can more effectively adapt to the radio environment if it can cooperate with other cognitive radios [
14].
A cognitive radio network seeks to serve the individual communication requirements of multiple primary and secondary users. In doing so, three practical challenges arise [
13]. First, the vacancies in spectrum come and go due to temporary usage of the spectrum by a licensed primary user. Finding these vacancies can be accomplished more efficiently in a cooperative manner by the secondary users in a cognitive radio network. Second, for each cognitive radio, information gathered by the receiver components must be processed and sent to the transmitter side. This will induce a delay in the feedback channel. Third, the security of the cognitive radio network can be compromised by malicious users in different locations and time frames. These practical issues need to be solved; in addition, solutions that use game theory to promote a cooperative strategy between the secondary users are also necessary.
1.3. Big Data Framework
As mentioned above, cognitive radio is being explored to address the scarce spectrum problem. The medium access control (MAC) protocol for any realizable system allows wireless users to use the frequency channels based on the current state of the network. For example, in [
15], various MAC protocols are investigated for use in cognitive radio based on different methods for users to start communication. In this article, the usage of data-mining techniques in MAC protocols for cognitive radio is studied.
The term “big data”, often used when referring to data mining, extends beyond the volume of data acquired to include also the velocity (how fast the data are being transmitted), the variety (the different data types included), veracity (the accuracy or truthfulness of the data) and value (the tangible benefits that the data provide). The authors have demonstrated that large-scale social media networks exhibit the five Vs of “big data” and can serve as a viable source of real-time knowledge extraction though data mining [
16,
17].
Using techniques, such as data mining, and including them in the perception-action cycle of a cognitive radio is an emerging field of study. Recently, a vision was presented for the use of “big data” techniques to enhance the performance of a cognitive radio network [
18]. The big data vision is foundational for illustrating cognitive networked sensing, cognitive radar, smart grid and cognitive radio networks. The data employed in [
18], however, are only those concerning the wireless channel, which are then employed to enhance the decision-making process regarding the usage of available wireless channels.
1.4. Main Contributions
The application of data sources other than wireless channel data in cognitive radio has not been significantly studied. These new information sets can be used to predict the traffic, channel condition and other conditions of a wireless network. As an example, we demonstrate herein that data extracted from a social media network by means of data mining can inform a cognitive radio.
Cognitive radios work by collecting information about a statistically varying environment and then applying methods and algorithms that react to this collected information to maximize certain performance goals. In this work, we employ data mining and game-theoretic techniques that employ new environmental data. Specifically, we investigate a new cognitive radio scheme that uses crowd-sourced social media information obtained through data mining of a large-scale social media network. We employ game theoretic algorithms for adaptation and reaction to a varying radio environment. The performance improvements from these adaptations help demonstrate the merit of what we call the “data mining-informed cognitive radio”.
1.5. Proposed Cognitive Radio
The research framework that we employ to develop more advanced cognitive radio networks consists of two thrusts. First, we use sources of information other than wireless channel information and collect relevant information using data-mining techniques to inform cognitive radio networks. The main goal of this thrust is to understand these novel sources of information and to find new methods to collect and analyze data that are not directly correlated to the wireless channel information, but which are relevant to channel usage. The second thrust employs appropriate game theoretic techniques to better utilize the spectrum and to perform resource sharing between the secondary wireless users in a wireless network. In this paper, we introduce smarter cognitive radio nodes and networks by way of improved algorithms, and it is shown that, based on crowd-sourced information, one can increase data transmission throughput.
Wireless channel information is currently the primary feedback information source used in cognitive radio. This information helps the various radio network nodes to adapt their waveforms, e.g., frequency, data transmission rate, modulation, etc. If the channel is used by other primary or secondary users, the user can detect the channel load and back off. However, there are other sources of information that may be helpful in predicting the near-term or future conditions of a wireless channel. For example, information about nearby weather conditions or forecasting a rainy day will help the coordinator and/or secondary users to use more robust modulations compared to more data-rate-efficient modulations. Predicting a sunny day would imply using more data-rate-efficient modulations. Another example is the acquisition of information about an emergency situation, such as a fire at a school, by data mining of social media and other sources. In this case, the moderator or secondary users must prioritize data transmission related to the emergency event over their own data. Gathering these new data will add additional information for better decision making and, thus, improve the performance of a wireless network.