Abstract
Detecting occupied subbands is a key task for wireless applications such as unlicensed spectrum access. Recently, detection methods were proposed that extract per-subband features from sub-Nyquist baseband samples and then apply thresholding mechanisms based on held-out data. Such existing solutions can only provide guarantees in terms of false negative rate (FNR) in the asymptotic regime of large held-out data sets. In contrast, this work proposes a threshold mechanism-based conformal risk control (CRC), a method recently introduced in statistics. The proposed CRC-based thresholding technique formally meets user-specified FNR constraints, irrespective of the size of the held-out data set. By applying the proposed CRC-based framework to both reconstruction-based and classification-based sub-Nyquist spectrum sensing techniques, it is verified via experimental results that CRC not only provides theoretical guarantees on the FNR but also offers competitive true negative rate (TNR) performance.
Introduction
The rapid growth of wireless demand necessitates efficient spectrum sharing, where secondary users access idle frequency bands without interfering with primary users. A critical component of this is spectrum sensing—the ability to accurately detect occupied subbands. Missing an occupied band, known as a "False Negative," leads to harmful interference. To minimize power consumption and hardware complexity, sub-Nyquist sampling is often employed to scan wide bandwidths. However, this approach complicates the detection process, as the resulting signals are undersampled and noisy.
Current spectrum sensing methods typically use a "held-out" calibration dataset to determine a detection threshold. A significant limitation of these methods is that their reliability guarantees are often asymptotic, requiring very large datasets to ensure the False Negative Rate (FNR) stays within safe bounds. In practical scenarios with limited data, these methods can fail to meet safety constraints.
Methodology
This work introduces a thresholding framework based on Conformal Risk Control (CRC) to provide finite-sample reliability guarantees. The overall system architecture is shown in Figure 1.
Sub-Nyquist Sampling
The system utilizes Multicoset Sampling (MCS) to capture wideband signals at sub-Nyquist rates. As depicted in Figure 2, MCS employs $P$ parallel channels, each applying a unique time delay $c_p T$ before sampling at a lower rate $1/LT$. This allows for the reconstruction of the wideband spectrum from a reduced number of samples.
Feature Extraction and CRC Thresholding
Following sampling, features are extracted from the baseband samples $\mathbf{Y}$. Two approaches are considered:
- PSD-based: Reconstructing the Power Spectral Density using compressive sensing.
- Logit-based: Using a Convolutional Neural Network (CNN) to estimate occupancy probabilities for each subband.
The core innovation is the CRC Predictor. Unlike standard thresholding, CRC applies a statistical correction that treats the calibration data conservatively. By incorporating a mathematical "buffer," CRC ensures that the expected FNR remains below a user-defined level $\alpha$, regardless of the calibration set size or the underlying signal distribution.
Experimental Results
The framework was evaluated using simulated Binary Phase Shift Keying (BPSK) signals across 40 subbands with a total bandwidth of 25 MHz. Key findings include:
- Reliability Guarantees: The CRC method consistently maintained the FNR below the target threshold ($\alpha = 0.1$) across various Signal-to-Noise Ratios (SNRs) and sampling rates.
- Robustness to Data Scarcity: Even with small calibration sets (e.g., 50 samples), CRC provided the promised safety guarantees, whereas traditional parametric and non-parametric baselines frequently violated the FNR constraints.
- Spectral Efficiency: Despite its conservative nature, the CRC-based approach achieved competitive True Negative Rates (TNR), ensuring that "holes" in the spectrum are still identified efficiently.
Conclusion
Conformal Risk Control provides a robust, distribution-free "wrapper" for sub-Nyquist spectrum sensing, offering formal safety guarantees that are critical for unlicensed spectrum access. Future research will explore "CRC-aware" training for neural networks to further optimize the trade-off between safety and spectral efficiency, as well as testing the framework in highly dynamic real-world environments using Software Defined Radios.
Summary updated