Matched Filter Detection

The matched filter is the optimal linear filter for detecting a known signal in additive white Gaussian noise (AWGN). Its impulse response is a time-reversed, conjugated copy of the target signal template. By correlating the received waveform with this template, the filter coherently integrates signal energy while averaging out uncorrelated noise. This process produces a peak SNR at the sampling instant equal to 2E/N₀, where E is the signal energy and N₀ is the noise power spectral density. No other linear filter can exceed this output SNR, making it the theoretical gold standard for detection.

A matched filter is mathematically equivalent to a correlation receiver. While the filter performs convolution in the time domain, the correlation receiver multiplies the received signal by a local replica and integrates the product over the symbol period. Both implementations produce identical output statistics at the decision instant. The choice between them is an engineering trade-off:

• Matched filter: Implemented in analog hardware or as a digital FIR filter, suitable for continuous-time processing.
• Correlation receiver: Requires precise synchronization but is more flexible for digital baseband processing with variable template waveforms.

The matched filter's optimality critically depends on precise timing, phase, and frequency synchronization with the incoming signal. A mismatch between the local template and the received waveform causes a rapid degradation in output SNR. In practice, this necessitates:

• Timing recovery loops (e.g., early-late gate synchronizers) to align the sampling instant with the peak of the correlation function.
• Carrier recovery circuits (e.g., Costas loops) to compensate for frequency offset and phase rotation.
• Frame synchronization sequences to identify the start of a transmission burst. Without these, the matched filter's performance collapses toward that of a non-coherent energy detector.

The filter's impulse response is defined by the known transmitted waveform. For a signal s(t) of duration T, the matched filter impulse response is *h(t) = s(T - t)**, where * denotes complex conjugation. This design principle extends to:

• Pulse shaping: Root-raised cosine filters are matched to their transmit counterparts to minimize inter-symbol interference while maximizing SNR.
• Radar chirps: Linear frequency-modulated pulses are compressed using matched filters, achieving range resolution inversely proportional to bandwidth.
• Spread spectrum: Pseudo-noise code sequences are detected via matched filtering, providing processing gain equal to the spreading factor.

The matched filter's detection performance is characterized by the Receiver Operating Characteristic (ROC) curve, which plots probability of detection (P_D) against probability of false alarm (P_FA). For a known signal in AWGN:

• P_D increases monotonically with SNR for a fixed P_FA.
• The deflection coefficient d² = 2E/N₀ quantifies the separability of the signal-present and signal-absent hypotheses.
• In spectrum sensing, the matched filter achieves the shortest sensing time to reach a target (P_D, P_FA) pair compared to energy detection or cyclostationary methods, but requires full knowledge of the primary user's waveform.

Despite its optimality, the matched filter faces practical barriers in cognitive radio spectrum sensing:

• Requires perfect signal knowledge: The cognitive radio must know the primary user's exact modulation format, pulse shape, and framing structure.
• High computational complexity: Dedicated matched filters are needed for each potential primary user waveform type.
• Synchronization overhead: Acquiring timing and frequency lock before detection defeats the purpose of fast spectrum scanning.
• Vulnerability to model mismatch: If the actual received signal deviates from the template due to multipath or hardware impairments, performance degrades sharply. These limitations motivate the use of semi-blind or blind sensing methods like eigenvalue-based detection when signal knowledge is unavailable.

The ROC curve is the standard tool for visualizing the performance of a binary detector like the matched filter. It plots the probability of detection (Pd) against the probability of false alarm (Pfa) as the decision threshold varies. A matched filter operating on a known signal in AWGN achieves the theoretically best possible ROC curve, pushing the curve towards the upper-left corner. The area under the ROC curve (AUC) is a single metric for comparing detectors; the matched filter maximizes this area for a given SNR.

In real-world radar and spectrum sensing, the noise floor is not constant. A fixed detection threshold would cause the false alarm probability to fluctuate wildly. CFAR processors dynamically adapt the threshold by estimating the noise power from adjacent reference cells. Common algorithms include:

• Cell-Averaging CFAR (CA-CFAR): Averages surrounding cells.
• Ordered-Statistic CFAR (OS-CFAR): Selects the k-th value from sorted cells, robust to interfering targets. The matched filter's output is fed into a CFAR block to maintain a stable Pfa in non-stationary noise.

Unlike the matched filter, energy detection is a blind sensing method that requires no prior knowledge of the primary user's signal structure. It simply measures the energy in a band and compares it to a threshold. While computationally simple, it suffers critically from noise uncertainty. There exists an SNR wall below which no amount of sensing time can guarantee reliable detection. A matched filter has no such SNR wall, making it vastly superior for detecting known waveforms at very low SNRs.

Most communication signals exhibit cyclostationarity—periodic statistics in their mean and autocorrelation due to modulation, carrier frequencies, or cyclic prefixes. This detector exploits the spectral correlation function to distinguish signals from stationary noise. Key advantages over matched filtering:

• Can differentiate between signal types (e.g., BPSK vs. QPSK) without demodulation.
• Robust to noise uncertainty. However, it requires significantly higher computational complexity and a longer observation time to resolve the cyclic frequencies.

The matched filter is the optimal coherent detector, meaning it requires exact knowledge of the received signal's phase. In practice, phase synchronization errors degrade its performance. Non-coherent detectors, like the energy detector or quadrature matched filter, discard phase information by operating on the signal envelope. The trade-off is clear:

• Coherent (Matched Filter): Best sensitivity, requires phase lock.
• Non-Coherent: Simpler hardware, but suffers a 1-3 dB SNR penalty in AWGN. The choice hinges on the channel's phase stability.

In a cognitive radio frame, time spent sensing is time not spent transmitting data. The sensing-throughput tradeoff defines the optimal sensing duration that maximizes the secondary user's throughput while protecting the primary user. A matched filter minimizes this tension because it achieves a target Pd and Pfa with the shortest possible sensing time for a known signal. A shorter sensing duration directly increases the time available for data transmission, improving the network's overall spectral efficiency.