Hard Decision Fusion

Each sensing node independently performs a local hypothesis test and reduces its observation to a single bit: 1 (primary user present) or 0 (spectrum vacant). This extreme quantization minimizes the data transmitted over the reporting channel, making it highly bandwidth-efficient. The local decision is typically generated by comparing a test statistic, such as energy, against a pre-defined threshold. The tradeoff is the loss of soft information about signal confidence.

The primary advantage of hard decision fusion is its minimal reporting channel overhead. Transmitting a single bit per sensing node per sensing interval requires far less bandwidth than sending raw energy measurements or quantized test statistics. This makes hard fusion ideal for bandwidth-constrained control channels in large-scale cooperative networks. The tradeoff is a measurable loss in detection sensitivity compared to soft decision fusion, particularly when nodes have disparate signal-to-noise ratios.

Hard decision fusion exhibits inherent robustness against certain reporting channel impairments. A single bit flip due to channel noise has a bounded impact on the global decision, whereas a corrupted analog value in soft combining can disproportionately skew a weighted sum. Fusion rules can be designed to account for known reporting error probabilities. For example, the optimal K value can be adjusted to compensate for a known bit error rate on the reporting links.

The simplicity of hard decision fusion makes it susceptible to Spectrum Sensing Data Falsification (SSDF) attacks, also known as Byzantine attacks. A malicious node can strategically flip its reported bit to manipulate the global K-out-of-N count. Countermeasures include:

• Reputation management: Assigning trust scores based on historical reporting consistency.
• Sequential probability ratio testing: Detecting anomalous reporting patterns over time.
• Robust fusion rules: Modifying the voting threshold to tolerate a known fraction of compromised nodes.

The performance of hard decision fusion depends critically on the local decision threshold at each sensing node and the global K value at the fusion center. These parameters are jointly optimized under the Neyman-Pearson criterion: maximize the global probability of detection subject to a constraint on the global probability of false alarm. When nodes experience independent and identically distributed fading, closed-form expressions exist for the optimal K. In heterogeneous networks with varying node SNRs, the optimal K shifts to favor higher-SNR nodes.

A Byzantine attack targeting the fragility of binary decisions. A malicious node flips its reported bit to corrupt the global fusion outcome.

• Always-Yes Attack: Malicious node constantly reports '1' to trigger false alarms and starve secondary access.
• Always-No Attack: Malicious node constantly reports '0' to cause interference with the primary user.
• Random Falsification: Node flips its true decision with a specific probability, making detection harder.

A defense mechanism that assigns a dynamic trust score to each sensing node based on the historical consistency of its binary reports with the final global decision.

• Nodes with low reputation are down-weighted or excluded from the K-out-of-N count.
• Mitigates SSDF attacks by isolating consistently malicious actors.
• Requires a warm-up period to establish baseline trust levels.

The primary alternative to hard decision fusion. Instead of a single bit, nodes transmit quantized test statistics (e.g., energy levels) to the fusion center.

• Preserves more information, enabling weighted combining algorithms like Maximal Ratio Combining (MRC).
• Superior detection performance at the cost of higher reporting channel bandwidth.
• Hard fusion is a special case of soft fusion with 1-bit quantization.

Hard decisions are robust to moderate reporting channel noise because a flipped bit requires significant interference. However, systematic errors degrade performance.

• Bit Error Rate (BER) on the reporting channel directly impacts the effective probability of detection and false alarm at the fusion center.
• Correlated reporting errors among geographically clustered nodes can defeat the spatial diversity gain of cooperative sensing.

A hybrid approach where a sensing node uses two thresholds to create a 'no decision' region. If the test statistic falls between the thresholds, the node abstains from reporting.

• Reduces reporting overhead by censoring unreliable local observations.
• The fusion center applies the K-out-of-N rule only to the subset of nodes that transmitted a decision.
• Balances bandwidth efficiency with detection reliability.