⚡ electrical engineering

GLRT for Reconfigurable Intelligent Surface aided Spectrum Sensing

This paper proposes a Generalized Likelihood Ratio Test (GLRT) framework for Reconfigurable Intelligent Surface (RIS)-aided spectrum sensing under correlated noise conditions, demonstrating through theoretical derivation and numerical results that the proposed method achieves superior detection performance compared to an Energy Detector, particularly when observations are limited.

Original authors: Nikhilsingh Parihar, Praful Mankar, Sachin Chaudhari

Published 2026-07-28

📖 7 min read🧠 Deep dive

CC BY 4.0

Original authors: Nikhilsingh Parihar, Praful Mankar, Sachin Chaudhari

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the air around us is a crowded dance floor, buzzing with invisible signals. Some dancers are the "Primary Users" (PUs)—the VIPs who own the floor and have the right to dance whenever they want. Then there are the "Secondary Users" (SUs), the eager guests who want to join in but can only dance when the VIPs aren't using the space. This is the world of Cognitive Radio, a smart system that lets devices share wireless spectrum efficiently. But here's the catch: the VIPs might be whispering so quietly, or the room might be so echoey and noisy, that the guests can't tell if the VIPs are even there. If the guests step in while the VIPs are dancing, it causes a chaotic crash.

To solve this, scientists use a tool called Spectrum Sensing, which is like a super-sensitive ear trying to hear the VIPs. However, real life is messy. The signals bounce off walls (multipath fading), the background noise isn't just a steady hum but a chaotic, correlated roar, and sometimes the VIPs are hidden behind obstacles. Enter the Reconfigurable Intelligent Surface (RIS). Think of an RIS as a magical, programmable mirror wall made of thousands of tiny tiles. Unlike a normal mirror that just reflects light randomly, an RIS can be programmed to tilt its tiny tiles in perfect unison. It can catch a weak, whispering signal from a VIP and bounce it directly to the guest, amplifying the sound so it's impossible to miss. But to make this mirror work, you first need to know exactly where the VIP is and how loud they are shouting—information that is often a complete mystery.

This paper tackles the tricky problem of how to use this magical mirror wall when you don't know the VIP's location or volume, and the room is filled with messy, correlated noise. The authors propose a clever two-step strategy using a method called the Generalized Likelihood Ratio Test (GLRT). First, they act like detectives: they use a special mathematical technique to guess the VIP's location and volume based on the faint echoes they catch. Once they have these best guesses, they program the RIS mirror to focus all its energy on that specific spot, creating a super-strong signal path. Then, they use their "super-ear" to listen again. The paper simulates this process on a computer and finds that this detective-and-mirror approach is much better at spotting the VIP than the standard "Energy Detector" (which just listens for any loud noise without understanding the signal). The results show that even when the noise is weird and the number of listening samples is small, this new method is far more reliable, especially when the RIS is set up perfectly rather than randomly.

The Story of the Smart Mirror and the Mystery Signal

In the high-stakes game of wireless communication, devices are constantly playing a game of "hide and seek" with the radio spectrum. The Primary User (PU) is the signal owner, and the Secondary User (SU) is the one trying to sneak in a transmission without causing interference. To do this safely, the SU must perform Spectrum Sensing: listening intently to see if the PU is active. But in the real world, this is like trying to hear a whisper in a hurricane. The signal might be blocked by buildings, bounced around by walls (a phenomenon called multipath fading), or drowned out by noise that isn't just random static but is "correlated," meaning the noise in one moment is related to the noise in the next, making it harder to filter out.

Enter the Reconfigurable Intelligent Surface (RIS). Imagine a wall covered in thousands of tiny, programmable mirrors. In a traditional setup, if a signal hits a wall, it scatters everywhere, losing strength. But an RIS can be "tuned." By adjusting the angle of each tiny mirror, the RIS can catch a weak signal and reflect it directly toward the receiver, boosting the signal strength like a spotlight. However, there's a big problem: to tune the mirrors perfectly, you need to know exactly where the signal is coming from and how strong it is. In many real-world scenarios, the SU doesn't know the channel conditions (the path the signal takes) or the PU's transmit power. It's like trying to aim a spotlight in a dark room without knowing where the person is standing.

This paper proposes a solution that combines detective work with a smart mirror. The authors suggest a system where the RIS is divided into groups. Since the RIS might have hundreds or thousands of elements (in their simulation, 512 elements), trying to estimate the signal path for all of them at once is mathematically impossible with the limited number of antennas the receiver has (8 in their setup). So, they use a group-wise approach. They activate one group of mirrors at a time, listen to the signal, and use a statistical method called Maximum Likelihood Estimation (MLE) to guess the signal's path and power for that specific group.

Once they have these guesses, they do something clever: they use the estimated information to configure the RIS. They calculate the perfect angle for the mirrors to maximize the signal gain, essentially creating a "super-highway" for the signal to travel from the PU to the SU. With the mirrors now optimally set, they perform the actual detection. They compare two methods: the standard Energy Detector (ED), which just measures how much total energy is in the air, and their proposed Generalized Likelihood Ratio Test (GLRT). The GLRT is smarter; it uses the estimated channel information and power to build a specific mathematical model of what the signal should look like, then checks if the received data matches that model.

The authors ran computer simulations to test this idea. They set up a scenario with 8 receiving antennas and an RIS with 512 elements, divided into 16 groups. They tested the system under various conditions, including different signal-to-noise ratios (SNR) and levels of noise correlation. The results were clear: the GLRT method, powered by the optimally configured RIS, consistently outperformed the standard Energy Detector.

In one set of simulations, they looked at how well the system could detect a signal when the noise was "correlated" (meaning the noise had a pattern, making it harder to ignore). They found that as the noise correlation increased, the detection probability actually improved for their method. This is because the "whitening" process (a mathematical trick to remove the noise pattern) works better when the noise has a structure to it. Furthermore, they compared an RIS that was set up with the perfect, calculated angles against one set up with random angles. The optimally configured RIS was far superior, proving that the "detective work" of estimating the channel first is crucial.

Interestingly, the paper notes that even if the RIS is set up randomly, the GLRT method still beats the standard Energy Detector with an optimally set RIS. This suggests that the mathematical intelligence of the GLRT is a powerful tool in itself. However, the best performance comes from combining both: the smart math of the GLRT and the perfectly tuned mirrors of the RIS.

The paper also highlights a limitation they had to overcome. Normally, to estimate a channel with 512 elements, you would need at least 512 antennas, which is impractical. By grouping the RIS elements and activating them sequentially, they managed to estimate the channel with only 8 antennas. This "group-wise" strategy makes the technology feasible for real-world devices that can't carry huge antenna arrays.

In conclusion, the authors demonstrate that by treating the RIS not just as a passive reflector but as an active part of the sensing process—using it to first estimate the unknown environment and then to boost the signal—they can achieve much more reliable spectrum sensing. Their simulations show that this approach is particularly effective in challenging environments with correlated noise and limited observation time, offering a robust path forward for future 6G networks where devices need to share the spectrum efficiently and safely. The key takeaway is that knowing how to listen (using GLRT) and where to aim the mirrors (using optimal phase shifts) makes all the difference in hearing the whisper in the hurricane.

Technical Summary: GLRT for Reconfigurable Intelligent Surface Aided Spectrum Sensing

Problem Statement
Spectrum sensing (SS) is fundamental to cognitive radio networks, enabling secondary users (SUs) to detect primary users (PUs) and utilize available spectrum. However, detection reliability is often compromised by unknown propagation factors, including multipath fading, correlated noise, and unknown PU transmit power. Furthermore, in non-line-of-sight (NLoS) or obstructed environments, the direct link between the PU and SU may be severely attenuated, leading to poor detection probabilities even at high Signal-to-Noise Ratios (SNR). While Reconfigurable Intelligent Surfaces (RIS) offer a solution by dynamically controlling the wireless environment, existing RIS-assisted sensing frameworks often rely on impractical assumptions, such as perfectly known channel state information (CSI), uncorrelated white Gaussian noise, or asymptotic approximations that require long sensing durations. Additionally, estimating the high-dimensional PU-RIS channel directly is often infeasible when the number of RIS elements exceeds the number of receiving antennas.

Methodology
The paper proposes a comprehensive spectrum sensing framework for RIS-aided systems operating under correlated noise and unknown channel conditions. The methodology is structured around a group-wise Beyond-Diagonal (BD-RIS) architecture and a Generalised Likelihood Ratio Test (GLRT):

System Model: The system comprises a single-antenna PU, an $N$ -antenna SU, and an $L$ -element RIS. To overcome the dimensionality constraint where $L > N$ (making direct channel estimation impossible), the RIS elements are partitioned into $G$ disjoint groups. These groups are activated sequentially, allowing the SU to estimate the PU-RIS sub-channel vectors ( $h_g$ ) for each group.
Parameter Estimation: The authors derive Maximum Likelihood (ML) estimates for the unknown PU-RIS sub-channel vectors and the PU transmit power ( $\sigma^2_x$ ). The received signal is pre-whitened to handle correlated noise ( $R_n$ ). The estimation involves solving a generalized eigenvalue problem to determine the channel direction and subsequently calculating the power estimate.
RIS Optimization: Using the estimated channel parameters, the RIS phase-shift matrix ( $\Phi$ ) is optimized to maximize the effective channel gain at the SU. A closed-form solution is derived using eigenvalue decompositions, ensuring the unitary constraint of the BD-RIS is satisfied.
Detection Framework:
- GLRT: The proposed GLRT utilizes the estimated channel and power parameters to construct a test statistic based on the maximized log-likelihood ratio. This approach jointly accounts for the estimated channel structure and the unknown transmit power.
- Energy Detector (ED): For comparison, an Energy Detector is also formulated using the optimally configured RIS. The paper derives the detection and false-alarm probabilities for the ED, approximating the test statistic as Gaussian using the Central Limit Theorem.

Key Contributions

Joint Estimation and Optimization: The paper introduces a framework that jointly estimates the PU-RIS channel and transmit power via ML estimation, subsequently using these estimates to optimize the RIS phase shifts. This addresses the challenge of unknown parameters in realistic environments.
Group-Wise BD-RIS Architecture: To handle large RIS sizes where $L > N$ , the authors propose a sequential group-wise activation strategy. This enables reliable channel estimation and sensing without requiring impractical constraints on the number of RIS elements relative to receive antennas.
Correlated Noise Handling: Unlike many prior works assuming white Gaussian noise, this study explicitly models and addresses correlated noise conditions, deriving analytical expressions for detection performance under these constraints.
GLRT vs. ED Analysis: The paper provides a rigorous comparison between the proposed GLRT and the conventional ED, demonstrating that GLRT leverages channel structure knowledge to achieve superior performance, particularly in low SNR and limited observation scenarios.

Results
Numerical simulations validate the proposed framework under various conditions ( $N=8$ , $L=512$ , $G=16$ ):

Channel Estimation: The Mean Squared Error (MSE) of the channel estimate decreases monotonically with increasing SNR and the number of observations, confirming the reliability of the group-wise ML estimator.
Detection Performance: Receiver Operating Characteristic (ROC) curves show that the GLRT consistently outperforms the ED across all SNR levels (0 dB, -4 dB, -8 dB). The GLRT maintains a higher detection probability for a given false-alarm rate.
Impact of Noise Correlation: The study analyzes the effect of the noise correlation coefficient ( $\rho$ ). Results indicate that detection probability increases with higher noise correlation, as whitening effectively reduces noise spread, improving the distinction between hypotheses.
Optimization Impact: Optimal RIS phase configurations significantly outperform random configurations for both GLRT and ED. Notably, the GLRT with random RIS phases still outperforms the ED with optimal RIS phases, highlighting the robustness of the GLRT approach.

Significance
The paper claims that its primary significance lies in bridging the gap between theoretical RIS-assisted sensing and practical deployment constraints. By addressing the combined challenges of unknown channel states, unknown transmit power, and correlated noise within a group-wise RIS architecture, the proposed GLRT framework offers a robust solution for spectrum sensing. The authors assert that this approach enables efficient and reliable detection even when the PU location and channel parameters are unknown, providing a practical pathway for enhancing cognitive radio capabilities in complex, real-world propagation environments. The work demonstrates that combining ML-based parameter estimation with optimal RIS configuration yields significant performance gains over conventional energy detection methods.

The Story of the Smart Mirror and the Mystery Signal

More like this