False Rejection Rate (FRR)

FRR is the ratio of false rejections to the total number of genuine access attempts. It is calculated as:

• Formula: FRR = (Number of False Rejections / Total Genuine Attempts) × 100%
• A false rejection occurs when a system fails to match a legitimate sample against the enrolled template for that identity.
• This metric is a direct measure of a system's failure of convenience, representing frustrated users who are incorrectly locked out.

FRR has a direct, inverse relationship with the False Acceptance Rate (FAR). Adjusting system sensitivity creates a trade-off:

• Lowering the threshold makes the system more tolerant, decreasing FRR (fewer legitimate users blocked) but increasing FAR (more impostors accepted).
• Raising the threshold makes the system stricter, decreasing FAR (higher security) but increasing FRR (more legitimate users blocked).
• The Equal Error Rate (EER) is the point where FRR and FAR are equal, often used as a single benchmark for overall system accuracy.

In the context of Radio Frequency Fingerprinting, FRR is often caused by environmental and hardware variability:

• Channel Fading: Multipath propagation and Doppler shifts distort the transient and steady-state signal features used for identification.
• Temperature Drift: Analog components like power amplifiers and oscillators exhibit behavioral changes with temperature, altering the RF fingerprint.
• Low SNR: In noisy environments, the unique hardware impairments are buried, making it difficult for the neural network to extract a clean embedding.
• Aging: Long-term component degradation causes a slow drift in the device's signature away from its original enrollment template.

FRR is a primary challenge for Few-Shot Device Enrollment systems. With minimal training samples:

• A prototypical network may compute an unrepresentative class prototype, causing it to reject a legitimate query sample that falls outside a tight decision boundary.
• Data augmentation techniques, such as adding synthetic channel impairments, are critical to regularize the model and reduce FRR by exposing it to expected variability.
• The goal is to learn an embedding space where intra-class distance (variations of the same device) is minimized, even with a sparse support set.

In commercial deployments, the acceptable FRR is dictated by the use case's tolerance for user friction:

• High-Security Access: A military radio network may tolerate a higher FRR (e.g., 1-5%) to ensure a near-zero FAR, prioritizing security over convenience.
• Consumer IoT: A smart home device requires a very low FRR (<0.1%) to prevent repeated user frustration and support tickets.
• Continuous Authentication: For passive, ongoing verification, a single false rejection can lock a user mid-session, making a low FRR critical for seamless operation.

FRR is not a static number but a function of the system's operating point:

• Detection Error Tradeoff (DET) Curve: A standard plot with FRR on one axis and FAR on the other, showing the trade-off across all thresholds.
• Score Distribution: Plotting the histogram of similarity scores for genuine and impostor attempts. The overlap area represents errors; FRR is the portion of the genuine distribution falling below the chosen threshold.
• Testing requires a diverse, representative dataset of genuine access attempts across varied channel conditions to get a statistically significant FRR.

The security counterpart to FRR, measuring the likelihood that a system incorrectly authenticates an unauthorized user or device. FAR and FRR are inversely related; tightening the decision threshold to lower FAR inevitably increases FRR. This trade-off is visualized on the Detection Error Trade-off (DET) curve. In high-security physical-layer authentication, FAR represents a successful replay attack or spoofing attempt bypassing the RF fingerprinting system.

The point on the Detection Error Trade-off (DET) curve where the False Acceptance Rate and False Rejection Rate are equal. EER serves as a single, balanced metric for overall biometric system accuracy. A lower EER indicates superior system performance. When evaluating few-shot device enrollment systems, EER provides a concise benchmark for comparing Prototypical Networks against Siamese Networks without needing to specify an arbitrary operating threshold.

The mechanism that directly governs the FRR-FAR trade-off. A confidence score is a probability value output by a classifier indicating its certainty in a prediction. The system's decision threshold determines the minimum score required for acceptance:

• High threshold: Increases security (lowers FAR) but causes more false rejections (raises FRR).
• Low threshold: Improves convenience (lowers FRR) but risks unauthorized access (raises FAR). In few-shot learning, confidence scores from Prototypical Networks are often derived from cosine similarity distances in the embedding space.

A classification paradigm where the model must correctly identify known classes while also rejecting samples from unknown classes not seen during training. This directly impacts FRR in deployed systems, as an Out-of-Distribution (OOD) detection failure can manifest as either a false acceptance or a false rejection. In RF fingerprinting, open set recognition is critical for identifying rogue transmitters that were not part of the initial few-shot enrollment support set.

A security process that constantly verifies a device's identity throughout an entire session based on physical-layer traits, rather than just at initial login. FRR in a continuous authentication context is measured per-transaction or per-time-window. A single false rejection during a session may trigger a re-authentication challenge or session termination. This paradigm amplifies the importance of low FRR, as frequent interruptions due to drift compensation failures or transient channel conditions degrade operational usability.

A technique used to determine if a biometric sample originates from a live, present device rather than a replayed or spoofed source. In RF fingerprinting, liveness detection distinguishes a genuine transmission from a replay attack using recorded signals. A system with overly aggressive liveness checks may increase FRR by rejecting legitimate transmissions that exhibit anomalous but authentic transient signal characteristics due to environmental factors or hardware impairment drift.