Field Guide · term

Also known as: preamble detection, sync correlation, correlator detection

Preamble correlation is the technique that detects the start of a packet by sliding a matched filter for a known preamble along the received stream and declaring detection wherever the cross-correlation forms a sharp peak.1 The preamble is a fixed sequence the transmitter sends before the payload; correlating against a stored copy of it is the optimal way to find that sequence buried in noise, and the position of the peak fixes where the packet — and its timing — begins.

received + noise correlator h[n]=preamble* peak = packet start correlation output
The correlator's output stays near noise level until the preamble aligns with its stored template, where it spikes; the peak's location times the packet.

How it works

Correlation computes, at each offset, the sliding dot product of the incoming samples with a time-reversed, conjugated copy of the known preamble — which is exactly what a matched filter does. When the preamble is not present the terms add with random signs and the output hovers near zero; when it aligns, every term reinforces and the output jumps to a large peak. This maximizes the signal-to-noise ratio at the peak, so it is the optimal detector for a known sequence in additive white noise, regardless of how the noise scatters the individual samples.

The quality of detection depends on the preamble’s autocorrelation. An ideal preamble has one tall main lobe and small sidelobes, so the peak is unmistakable and a slight misalignment does not raise a competing bump. Sequences engineered for this — Barker codes, Gold codes, Zadoff–Chu sequences — are chosen precisely because their sidelobes are low. Practical detectors add:

  • Normalization — dividing the correlation by the running input energy, so the threshold works regardless of absolute signal level (a constant-false-alarm-rate detector).
  • Non-coherent operation — taking the magnitude of a complex correlation so detection survives an unknown carrier phase, at a small SNR cost versus a coherent (known-phase) correlator.
  • A threshold balancing missed detections against false alarms, often set from the measured noise floor.

Beyond finding the packet, the peak’s exact sample index gives a coarse timing estimate, and the complex value at the peak carries the carrier phase and frequency offset — useful seeds for clock recovery and carrier correction that follow.

In practice

Correlation is computed one of two ways. A time-domain sliding correlator is cheap for a short preamble and can run continuously, one multiply-accumulate window per sample. For a long preamble it is more efficient to correlate in the frequency domain — an FFT-based fast convolution computes the whole correlation for a block at once, trading latency for far fewer operations, the same overlap-save trick used for long FIR filters. Either way, a carrier frequency offset is the practical enemy: a large offset rotates the samples during the correlation window and shrinks the peak, so wideband systems either search a bank of frequency hypotheses in parallel or use a differential correlation (correlating sample-to-sample phase changes rather than absolute phase) that is inherently offset-tolerant at some SNR cost. The preamble length is chosen to balance these forces: longer gives a taller, sharper peak and better estimates, but costs airtime and widens the frequency-search burden.

Relevance to SDR

Preamble correlation is how nearly every packet radio acquires a burst. Wi-Fi correlates its short/long training fields, Bluetooth and Zigbee correlate an access-address/preamble, LoRa correlates chirp symbols, and LTE/5G correlate Zadoff–Chu synchronization signals. In land-mobile digital voice the same mechanism finds the frame-sync pattern that marks each frame. GopherTrunk relies on correlation-style sync detection inside its protocol decoders: matching the known P25/DMR/NXDN sync sequence against the demodulated symbols is what triggers frame alignment and lets the parsers lock onto a control or voice channel. Because the sync words in these land-mobile formats are short and the symbols are already timing-recovered, the correlation runs in the time domain over the symbol stream rather than through an FFT, and tolerating a few symbol errors lets the detector hold sync through brief fades instead of dropping the channel on every glitch.

Sources

  1. Matched filter — Wikipedia, on the correlation receiver as the optimal detector of a known signal in white noise. ↩

See also