Burst Onset Detection

A probabilistic framework that models the burst onset as a changepoint in the statistical properties of the received signal. By maintaining a running posterior distribution over possible changepoint locations, this method provides both a detection decision and a confidence interval for the onset time. The algorithm computes the probability that a change in variance or mean has occurred given all prior observations, making it highly robust to non-stationary noise floors and dynamic channel conditions. Unlike threshold-based methods, Bayesian approaches naturally incorporate prior knowledge about expected burst durations and inter-arrival times.

The most computationally efficient onset detection method, computing the instantaneous power of the received signal via a moving average of squared magnitude samples. A burst onset is declared when the energy envelope crosses a pre-defined threshold above the estimated noise floor. Key design parameters include:

• Threshold level: Typically set 6-10 dB above the noise floor to balance sensitivity against false alarms
• Averaging window length: Must be shorter than the expected rise time to avoid smearing the transient edge
• Hysteresis: Prevents chattering detections when the signal hovers near the threshold This method is ideal for real-time FPGA implementations but struggles in low-SNR environments.

The optimal linear detector for a known transient shape in additive white Gaussian noise. A template of the expected ramp-up signature is correlated with the incoming signal stream; the onset is declared at the time index producing the maximum correlation peak. This technique maximizes the signal-to-noise ratio at the decision point and is particularly effective when the transmitter's amplitude ramp profile is known a priori from enrollment. Implementation requires:

• A pre-characterized transient template from the target device
• Continuous cross-correlation of the template against the IQ sample stream
• Peak detection with interpolation for sub-sample timing accuracy

Leverages the multi-scale decomposition properties of the discrete wavelet transform to identify burst onsets across different time-frequency resolutions simultaneously. The transient's sharp discontinuity generates large wavelet coefficients at fine scales, while the steady-state signal energy concentrates at coarser scales. By tracking the modulus maxima of wavelet coefficients across decomposition levels, the algorithm can distinguish true burst edges from noise spikes. This approach excels at detecting transients with varying rise times and is inherently robust to colored noise and narrowband interference that would trigger false alarms in simpler energy detectors.

A sequential hypothesis testing framework that processes samples one at a time, accumulating evidence for the presence or absence of a burst until a decision boundary is crossed. Unlike fixed-sample detectors, SPRT minimizes the average detection time for a given false alarm and miss rate by continuously evaluating the log-likelihood ratio. The algorithm maintains two hypotheses:

• H₀: Only noise is present
• H₁: Signal plus noise is present When the cumulative log-likelihood ratio exceeds an upper threshold, a burst is declared. This method is optimal for low-latency applications where minimizing detection delay is critical.

Exploits the phase discontinuity that occurs at the moment a transmitter's oscillator and modulator are energized. Rather than monitoring amplitude changes, this technique tracks the instantaneous phase derivative (frequency) and identifies the onset by detecting a sudden deviation from the expected phase trajectory. The method is particularly effective when:

• The transmitter exhibits a transient carrier feedthrough spike
• The PLL settling transient causes a rapid frequency sweep
• Amplitude-based detection is compromised by fading or interference Phase-based detection can identify the onset even when the amplitude change is gradual, as the phase discontinuity is often more abrupt and hardware-specific.