Spectrum Sensing

A non-coherent detection method that measures the total received signal power within a target frequency band over a defined observation period and compares it against a pre-calculated noise threshold. Energy detection is computationally simple and does not require any prior knowledge of the primary user's signal structure, making it a universal sensing technique. However, its performance degrades significantly under low signal-to-noise ratio (SNR) conditions and it cannot differentiate between primary user transmissions and other sources of interference.

• Key Limitation: Susceptible to the SNR wall phenomenon, below which reliable detection becomes impossible regardless of observation time.
• Common Application: Used as a coarse, first-pass sensing stage in hierarchical detection architectures.

An optimal coherent detection technique that correlates the received signal with a known copy of the primary user's transmitted waveform to maximize the output SNR. Matched filtering requires perfect knowledge of the signal's structure, including its modulation type, pulse shape, and packet format. It achieves superior detection performance with minimal observation time but demands dedicated circuitry for each signal type to be detected.

• Key Advantage: Achieves the theoretical maximum SNR at its output, enabling detection at very low input power levels.
• Key Disadvantage: Requires precise synchronization and demodulation, consuming significant computational resources.

A robust detection method that exploits the periodic statistical properties inherent in modulated signals, which are absent in stationary noise. By analyzing the spectral correlation function, this technique can distinguish between different modulation types and separate overlapping signals. It is highly resilient to noise uncertainty but requires significant computational resources to compute the cyclic autocorrelation function over multiple cyclic frequencies.

• Key Advantage: Can identify the specific modulation scheme of the primary user while detecting its presence.
• Key Disadvantage: High computational cost of O(N²) makes it unsuitable for real-time wideband sensing.

A blind detection method that analyzes the eigenvalues of the sample covariance matrix computed from multiple received signal samples. Eigenvalue-based sensing leverages the fact that the covariance matrix of a signal-plus-noise mixture has different eigenvalue properties than that of pure noise. Techniques like the Maximum-Minimum Eigenvalue (MME) detector do not require noise variance knowledge, making them immune to the noise uncertainty problem that plagues energy detection.

• Key Advantage: Robust against noise uncertainty without requiring primary user signal knowledge.
• Key Application: Effective in multi-antenna cognitive radio systems where spatial diversity is available.

A coherent detection technique specifically designed for signals containing known patterns, such as preambles, pilot tones, or spreading sequences. Waveform-based sensing correlates the received signal against a locally stored copy of the known pattern, providing reliable detection at very low SNRs. It is commonly used in standards-compliant systems like IEEE 802.22 WRAN where the primary user signal structure is well-defined.

• Key Advantage: Achieves high detection probability with short sensing durations due to pattern correlation gain.
• Key Limitation: Applicability is restricted to signals with standardized, known repetitive patterns.

An emerging class of sensing techniques that employs deep neural networks to learn complex, non-linear signal features directly from raw I/Q samples or spectrograms. AI-driven sensing models, such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs), can jointly perform signal detection and automatic modulation classification without explicit feature engineering. These models excel in low-SNR environments and can adapt to evolving signal types through continuous learning.

• Key Advantage: Superior performance in low-SNR and complex interference environments compared to model-based methods.
• Key Challenge: Requires substantial labeled training data and computational resources for inference on edge devices.

The intelligent core that consumes spectrum sensing data to build a radio environmental map and decide on transmission parameters. It implements the observe-orient-decide-act (OODA) loop, using the sensing subsystem as its primary environmental input. The engine applies machine learning models to classify spectrum occupancy patterns and predict future availability, enabling proactive rather than reactive frequency hopping.

A critical sensing failure mode where a cognitive radio is shadowed by terrain or buildings from a primary transmitter. The radio falsely detects a spectrum hole and begins transmitting, causing harmful interference to a nearby primary receiver it cannot hear. This vulnerability drives the need for cooperative sensing architectures where multiple spatially distributed nodes share local observations to overcome individual sensing blind spots.

A denial-of-service attack where a malicious actor mimics the signal characteristics of a licensed primary user—such as a TV broadcast pilot tone or radar pulse pattern—to trick spectrum sensors into marking a band as occupied. This forces legitimate secondary users to vacate perfectly usable spectrum. Defenses include RF fingerprinting to identify unique hardware-level transmitter imperfections and location verification using angle-of-arrival estimation.

An architecture where multiple spatially distributed cognitive radios share local sensing decisions with a fusion center to improve global detection reliability. Key combining rules include:

• OR Rule: Declare presence if any single node detects the primary user—maximizes sensitivity but increases false alarms
• AND Rule: Require unanimous agreement—minimizes false alarms but risks missed detection
• Soft Combining: Share raw energy measurements rather than binary decisions for optimal statistical performance

The central processing node in a cooperative sensing network that aggregates local observations from distributed cognitive radios and applies a combining algorithm to make a global decision about primary user presence. The fusion center must account for correlated shadowing between nearby nodes and reporting channel errors where sensing data is corrupted during transmission. Modern implementations use deep learning fusion to learn optimal combining weights from historical data.

A frequency band that is allocated to a primary user but temporarily vacant in a specific geographic location and time window. Spectrum sensing must distinguish between three states:

• White Space: Completely unoccupied, safe for secondary access
• Gray Space: Partially occupied by low-power signals, requires careful power control
• Black Space: Actively occupied by a primary user, strictly off-limits The temporal duration of a spectrum hole determines its suitability for different traffic types—voice requires persistent holes while bursty data can exploit fleeting opportunities.