Este guia está disponível atualmente em inglês.
Technology / Technical guide
Wind Measurement Filtering: Bandwidth, Delay & Latency
Filtering can suppress rapid variation while also changing timing. Communication can delay arrival without applying the same waveform transformation. These mechanisms can produce similarly late-looking traces, yet they have different causes and implications for interpreting a changing wind quantity.
Smoother values may represent an older part of the signal.
Distinguish amplitude response, frequency-dependent delay and delivery time before comparing two measurement channels.
01 / Amplitude response
A filter changes components unequally.
A filter’s frequency response describes how each frequency component is scaled and phase-shifted under its linear-system model.
A filter’s frequency response describes how each frequency component is scaled and phase-shifted under its linear-system model. A low-pass response commonly attenuates rapid components more strongly than slow ones. The resulting trace can look less variable even where the physical input is unchanged.
A cutoff label alone does not describe all attenuation, transition behaviour or delay. A filtered maximum can differ from the original maximum. That is a change in the represented signal, not necessarily a change in the turbine or sensor’s physical behaviour.
02 / Delay definitions
Phase delay and group delay answer different questions.
Phase delay is negative phase shift divided by angular frequency, τp = −φ/ω for nonzero ω.
Phase delay is negative phase shift divided by angular frequency, τp = −φ/ω for nonzero ω. Group delay is the negative slope of unwrapped phase with angular frequency. MathWorks distinguishes these generally different quantities.
For a continuous-frequency convention with angular frequency in rad/s, that group-delay slope gives seconds. With digital frequency in rad/sample, it gives samples. The frequency convention matters before a numerical delay is attached to an event.
03 / Averaging example
A symmetric moving average has a modelled delay.
Hypothetical filter: a causal five-sample equal-weight moving average is a linear-phase finite impulse response filter of order four.
Hypothetical filter: a causal five-sample equal-weight moving average is a linear-phase finite impulse response filter of order four. Its group delay is two samples where the response is nonzero. With uniform 0.02 s sampling, that delay corresponds to 0.04 s.
The output also averages five different input instants, so shifting its time label does not restore the fluctuations removed by averaging. The invented interval and filter describe a mathematical example, not a MARSTON device, setting or proposed modification.
04 / Event interpretation
A step need not behave like a narrowband component.
A sudden change contains a range of frequencies.
A sudden change contains a range of frequencies. Its visible rise and apparent event time depend on the complete response and the criterion used to identify the change. One group-delay value cannot universally predict every threshold crossing.
Different filters can have similar passband amplitudes but different phase behaviour. Frequency-dependent delay can reshape a waveform rather than merely translate it in time. This is why amplitude bandwidth and timing response should remain conceptually separate.
05 / Transport effects
Arrival latency is a different part of the chain.
A stored observation can arrive later because of transmission or buffering.
A stored observation can arrive later because of transmission or buffering. Variable delivery latency can change arrival order without changing when the sensor observed the value. Filtering, by contrast, changes how the output value relates to a span of input values.
Combining both mechanisms into one unexplained time offset hides their effects. The timestamp guide distinguishes observation and arrival labels; filtering explains why a correctly timed output can still represent delayed signal behaviour.
06 / Interpretation scope
Timing claims belong to a defined transformation.
The sampling guide covers frequency ambiguity, while this page concerns retained information and delay after a defined transformation.
The sampling guide covers frequency ambiguity, while this page concerns retained information and delay after a defined transformation. Sampling faster does not automatically remove sensor dynamics or filter delay.
No actual MARSTON filter, bandwidth, communication latency or synchronisation performance is stated here. These principles support data interpretation without filter design, controller tuning, logging changes or field test instructions.
Continue the review with Wind Waveform Sampling: Nyquist Conditions & Aliasing.
Continue the review with Yaw Feedback: Setpoint, Measurement & Closed-Loop Error.
Continue the review with PI & PID Control: Proportional, Integral & Derivative Roles.
Continue the signal-processing review with notch-filter attenuation and phase and numerical derivatives and noise.
Extend the timing discussion with feedback delay and stability boundaries. For separate state-information and estimation topics, review states and observability and Kalman prediction, innovation and covariance.
General background: MathWorks: Phase delay, group delay and linear-phase FIR interpretation. These references explain industry concepts; they are separate from the applicable MARSTON product and project documents.
Key concepts and conditions.
Read each quantity with its reference, units and stated conditions. A general relation does not establish a model-specific value.
| Concept / quantity | Meaning / relation | Conditions / limits |
|---|---|---|
| Amplitude response | Component gain versus frequency | A linear-system model is specified. |
| Phase delay | Phase shift relative to component frequency | Phase and units need a consistent convention. |
| Group delay | Negative unwrapped phase slope | Seconds or samples follow the frequency variable. |
| Moving average | Five values; order four; two-sample delay | Uniform sampling and nonzero response. |
| Transport latency | Arrival time minus observation time | Delivery effects are distinct from filtering. |
Engineering questions.
Does smoothing prove a physical signal became steadier?
No. The transformation itself can reduce visible variation.
Are phase delay and group delay identical?
Not generally. They use different relationships with phase.
Does a two-sample delay mean two seconds?
Only if the sample interval is one second; units must be translated.
Can a time shift restore averaged-away peaks?
No. Relabelling does not recover the removed waveform information.
Is communication delay the same as group delay?
No. They describe transport and signal transformation respectively.
Does this page select a MARSTON filter?
No. Its moving average is an illustrative mathematical model.
Technical enquiry
Discuss the technical question.
Filtering can suppress rapid variation while also changing timing. Share the model and applicable operating or design information for an engineering discussion.