Sensing-Throughput Tradeoff

The required Probability of Detection (Pd) is often mandated by regulators (e.g., 99.9% for TV white spaces) to protect primary users. A higher target Pd demands longer sensing times or more complex algorithms like cyclostationary feature detection, directly reducing the time available for secondary data transmission. This creates a hard ceiling on achievable throughput.

The physical layer frame is divided into a sensing slot and a transmission slot. A longer sensing slot improves detection reliability but shrinks the transmission slot, reducing throughput. Conversely, a short sensing slot maximizes throughput but risks missing the primary user. Periodic sensing structures, where sensing occurs in dedicated quiet periods, inherently cap the maximum channel utilization.

While cooperative spectrum sensing (CSS) mitigates the hidden node problem, it introduces a reporting overhead. The time spent by secondary nodes transmitting local hard or soft decisions to a fusion center consumes bandwidth that could otherwise be used for data. The choice between hard decision fusion (low overhead, less accurate) and soft decision fusion (high overhead, more accurate) directly impacts the net throughput gain from cooperation.

The statistical behavior of the primary user dictates the optimal sensing-throughput strategy. For a primary user with long ON/OFF periods, infrequent but highly accurate sensing is optimal. For a rapidly intermittent primary user, a short, frequent sensing strategy is necessary to avoid collisions, even at the cost of throughput. Spectrum occupancy prediction models can preemptively adjust the tradeoff based on learned traffic patterns.

Below a certain SNR wall, caused by noise uncertainty, no amount of sensing time can guarantee reliable detection using an energy detector. In this regime, increasing the sensing duration yields diminishing returns, and the throughput collapses to zero if a stringent Pd must be maintained. Escaping the SNR wall requires feature-based detection methods, which have their own computational complexity costs.

A high Probability of False Alarm (Pfa) causes the secondary user to declare the channel busy when it is actually idle, leading to a missed transmission opportunity and a direct loss in throughput. The Neyman-Pearson criterion frames the tradeoff as maximizing Pd for a given Pfa constraint. A poorly chosen sensing threshold that inflates Pfa can be more detrimental to throughput than a short sensing window.

The statistical likelihood that a spectrum sensing algorithm correctly identifies a primary user signal when it is actively transmitting. This is the primary metric for primary user protection and is typically constrained by regulators to be above 0.9.

• Directly competes with throughput: longer sensing increases Pd
• A missed detection causes a harmful collision with the primary user
• The complement (1 - Pd) is the probability of misdetection

The likelihood that a sensing algorithm incorrectly declares a frequency band occupied when it is actually vacant. Each false alarm represents a lost transmission opportunity for the secondary user.

• Directly reduces secondary user throughput
• Governed by the Constant False Alarm Rate (CFAR) threshold
• The tradeoff with Pd is visualized on the Receiver Operating Characteristic (ROC) curve

A graphical plot that illustrates the diagnostic ability of a binary classifier by mapping probability of detection against probability of false alarm for varying detection thresholds. The ROC curve is the primary evaluation tool for sensing algorithms.

• The area under the curve (AUC) quantifies overall sensing performance
• A perfect detector achieves Pd=1 at Pfa=0
• Used to compare energy detection, cyclostationary detection, and matched filter detection

An optimal detection framework that maximizes the probability of detection subject to an upper bound constraint on the probability of false alarm. This forms the theoretical basis for designing sensing algorithms that operate within the sensing-throughput tradeoff.

• Does not require prior probabilities of primary user activity
• Implemented via the Likelihood Ratio Test (LRT)
• Practical implementations often use energy detection as a low-complexity approximation