1. Introduction to Cognitive Radio
The concept of Cognitive Radio (CR) originated from the foundational work of Dr. Joseph Mitola in 1999. Through learning and understanding, CR can adaptively adjust its internal communication mechanisms and change specific wireless operating parameters in real time—such as power, carrier modulation, and coding—to adapt to the external wireless environment and autonomously find and use idle spectrum. It can help users select the best and most suitable service for wireless transmission and can even delay or proactively initiate transmissions based on existing or soon-to-be-available wireless resources.
Joseph Mitola's definition of cognitive radio emphasizes the ability to learn. A cognitive radio system needs to consider every possible parameter in the communication environment and then make decisions. Compared with Mitola's definition, the definition provided by the U.S. Federal Communications Commission (FCC) for spectrum-efficient allocation is more widely accepted by the industry. In a notice in December 2003, the FCC defined cognitive radio as a radio that can interact with its communication environment and change its transmission parameters based on the interaction results. The FCC definition is mainly based on spectrum resource allocation and management issues. Currently, the planning and use of wireless spectrum resources are formulated by governments, and wireless communication devices require government permission to use spectrum. Fixed spectrum allocation policies have led to inefficient spectrum utilization. In addition to Joseph Mitola and the FCC, many scholars have defined cognitive radio. For example, Simon Haykin, combining the views of Mitola and the FCC, defined cognitive radio as an intelligent wireless communication system that can sense the external environment, learn from it using artificial intelligence technology, and change system parameters such as transmission power, carrier frequency, and modulation mode in real time so that the system adapts to environmental changes, thereby achieving high spectrum utilization and optimal communication performance.
2. Characteristics of Cognitive Radio
1. Environmental sensing capability: This is the premise of CR technology. Only on the basis of environmental sensing and detection can spectrum resources be used. The main function of spectrum sensing is to monitor a certain range of frequency bands and detect spectrum holes and other information.
2. Learning ability and adaptability to environmental changes: This reflects the intelligence of CR technology. When a primary user signal appears, CR can proactively back off as soon as possible and switch freely among spectrum holes.
3. High reliability of communication quality: The system is required to achieve highly reliable communication at any time and any place, accurately determine the time, location, frequency band, and other information of primary user signals, adjust its own parameters in a timely manner, and improve communication quality.
4. Reconfigurability of system function modules: CR devices can be dynamically programmed according to the spectrum environment and can also support different transceiver technologies through hardware design. Reconfigurable parameters include operating frequency, modulation mode, transmit power, and communication protocols.
3. Key Technologies of Cognitive Radio
3.1 Spectrum Sensing Technology
Spectrum sensing technology is the foundation and premise of CR applications, a core CR technology, and a prerequisite for ensuring efficient spectrum allocation. Spectrum sensing monitors available frequency bands in real time and performs related analysis without interfering with authorized users, thereby discovering spectrum holes. Spectrum sensing technology must ensure good detection performance. If the detection probability is too low, it will interfere with the normal communication of authorized users; if the false alarm probability is too high, cognitive users will be unable to properly access idle spectrum, reducing spectrum utilization. Spectrum sensing is divided into single-node sensing and multi-node cooperative sensing.
Single-node spectrum sensing means a single user makes an independent decision without involving complex system structures or data fusion, so it is relatively simple. However, sensing performance improvement cannot break through physical limitations. Against this background, cooperative spectrum sensing was proposed, effectively overcoming the physical limitations of single nodes, improving spectrum sensing performance, and better adapting to lower signal-to-noise ratio environments. Multi-node cooperation achieves the system-required detection threshold through cooperation among detection nodes, thereby reducing the requirements for individual detection nodes and lowering the burden on each node.
Classic spectrum sensing methods include energy detection algorithms, matched filter detection algorithms, and cyclostationary detection algorithms. Energy detection has low computational complexity, is simple to implement, and does not require prior knowledge of the primary user; however, it is susceptible to noise uncertainty, and detection performance drops sharply at low signal-to-noise ratios. In recent years, methods based on random matrix theory have gradually been applied to spectrum sensing. These methods use the eigenvalues of the covariance matrix as the test statistic for the signal and then compare statistical features with a threshold to achieve spectrum sensing. However, this method requires computing eigenvalues and accurately estimating the threshold, and the accuracy of the threshold estimate strongly affects spectrum sensing performance. Therefore, spectrum sensing algorithms based on machine learning have gradually become a research focus.
The purpose of spectrum sensing is to monitor a frequency band and, by analyzing the components of the received signal, determine whether the band is occupied by a Primary User (PU). From a mathematical perspective, spectrum sensing is essentially a binary classification problem:
- H0: The test statistic is below the detection threshold, so the PU is considered absent, and the Secondary User (SU) can access the authorized band.
- H1: When the test statistic exceeds the detection threshold, the PU is determined to be present, and the SU cannot access the authorized band.
The mathematical model is shown below:

Here, s(n) is the primary user signal, x(n) is the signal received by the secondary user at time n (n = 1, 2, ..., N), w(n) is the noise signal, and h(n) is the channel gain.
Machine learning can currently solve binary classification problems well. Therefore, traditional machine learning algorithms such as K-Means clustering, Support Vector Machine (SVM), or Gaussian Mixture Model (GMM) can be used, as can artificial neural networks such as Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), or Long Short-Term Memory (LSTM) for classification.
3.2 Modulation Recognition Technology
Modulation recognition (or modulation classification) of communication signals refers to processing received signals under noisy and multi-signal-type conditions and using certain specific signal parameters to automatically determine the signal modulation type. It is an indispensable step in signal detection, analysis, and demodulation.
Communication signal modulation recognition has two classifications. One is intra-recognition, which refers to classification and recognition among several major signal categories, such as MASK, MFSK, and MPSK. The other is inter-recognition, which refers to classification and recognition among signals within a major category, such as 2ASK and 4ASK signals. Modulation recognition of communication signals mainly includes three basic steps: signal preprocessing, feature extraction and selection, and classification recognition. The basic structure of the recognition process is shown in the figure.

The signal preprocessing part mainly provides suitable processed data for feature extraction and recognition, isolates each signal to ensure that only one signal enters subsequent stages at a time, and segments the signal to eliminate fading effects caused by the wireless channel. Signal preprocessing tasks generally include frequency downconversion, intermediate frequency filtering, signal separation, carrier synchronization, and carrier frequency component removal. Therefore, signal preprocessing is very useful in transmitter-source environments. The main work of feature extraction and selection is to extract suitable parameter features for signal modulation recognition from preprocessed data. Classification recognition is mainly used to select and determine effective recognition methods, rules, and classifier types.
4. Current Status and Trends of Cognitive Radio
At present, cognitive radio technology has received widespread attention from academia and industry. Many famous scholars and research institutions have devoted themselves to research on CR-related technologies and have launched many important CR research projects. Examples include the Spectrum Pooling system proposed by Professor F. K. Jondral of the University of Karlsruhe in Germany, the COVUS system developed by Professor R. W. Brodersen's research group at the University of California, Berkeley, the OCRA project proposed by Professor Ian F. Akyildiz and others at the Broadband and Wireless Networking Laboratory of Georgia Institute of Technology, the U.S. military DARPA XG project, and the EU E2R project. Driven by these projects, achievements have been made in basic theory, spectrum sensing, data transmission, network architecture and protocols, integration with existing wireless communication systems, and prototype development. IEEE has organized two important international annual conferences, IEEE CrownCom and IEEE DySPAN, to exchange results in this area, and many important international academic journals have published special issues on cognitive radio. Currently, the most notable effort is the work of the IEEE 802.22 working group, which is developing technical standards for broadband wireless access using idle TV bands. This is the first IEEE standardization activity to introduce the concept of cognitive radio.
Based on the current status of cognitive radio technology, it is expected that cognitive radio will develop along the following directions in the future:
1. Research on basic theory and related applications to lay a solid foundation for large-scale application. Important topics include the information-theoretic foundations of cognitive radio and CR network technologies such as spectrum resource management and cross-layer joint optimization.
2. Development of experimental validation systems. Multiple experimental validation systems are currently under development. Their successful development will provide testbeds for validating the basic theories and key technologies of cognitive radio and promote large-scale application.
3. Integration with existing systems. Although it is currently believed that CR applications should not require authorized users to make any changes, if authorized users and CR users work cooperatively, implementation will be easier and efficiency will improve. Some research is already considering methods to integrate cognitive radio into existing wireless communication systems, and preliminary results have been achieved. Substantial demand is expected in this area in the future.