Key takeaway
Two different continuous signals can produce the same stored samples. Calibration evidence cannot recover frequency information lost at acquisition.
More stored rows do not settle the acquisition question
A buyer sees a long, regular time series and wants to identify an oscillation. The instrument has a calibration record, units are documented and the numbers look smooth. Those facts do not establish which original frequencies the acquisition preserved. The system may have sampled or reduced a higher-frequency signal into a misleading lower-frequency pattern.
National Instruments’ acquisition guidance explains aliasing and the role of bandwidth and anti-alias filtering. This review asks a narrower question than whether the sensor is accurate: does the retained acquisition evidence support the particular frequency or waveform conclusion proposed for these samples?
Compare two signals with identical sample values
Consider this hypothetical, uniformly sampled fixture. One continuous signal is cos(2π × 2 × t); another is cos(2π × 8 × t), with time t in seconds. Both are sampled at ten samples per second, at t = n/10 for integer n. Their first five values, rounded to three decimals, are shown below.
The equality is mathematical, not a sensor performance claim. At those sample instants, the eight-hertz phase differs from the negative two-hertz phase by whole turns, and cosine has the same value for positive and negative phase. The author also computed twenty points locally; the largest numerical difference was about 1.1 × 10⁻¹⁴, attributable to finite arithmetic in the calculation.
The stored samples alone therefore cannot decide whether the illustrated original tone was two or eight hertz. A fitted two-hertz curve does not prove that the original process oscillated at two hertz. It is one signal consistent with the retained observations.
| Sample instant | 2 Hz cosine | 8 Hz cosine |
|---|---|---|
| 0.0 s | 1.000 | 1.000 |
| 0.1 s | 0.309 | 0.309 |
| 0.2 s | -0.809 | -0.809 |
| 0.3 s | -0.809 | -0.809 |
| 0.4 s | 0.309 | 0.309 |
Establish the band before interpreting the spectrum
At ten samples per second, half the sample rate is five hertz. The eight-hertz tone lies above that boundary. The sampling condition discussed by NI requires the signal to be band-limited below half the sampling rate for the stated reconstruction result. The example deliberately violates that condition.
A practical review needs more than a nominal sample-rate setting. NI describes analog anti-alias filtering before the sampler and explains that real filters have a transition band rather than an ideal vertical cutoff. Ask for the relevant front-end bandwidth, filter response/configuration and actual sampling path. Do not reduce those facts to an unexplained rule that twice the desired frequency is always sufficient.
If values were later downsampled, identify the earlier rate, reduction factor and filtering applied before that reduction. Once unwanted content is already aliased into the band of interest, a later ordinary low-pass filter cannot distinguish it from genuine content at the same apparent frequency.
Complete an acquisition evidence register
For a hypothetical equipment archive, the record owner can identify four distinct cases. Series A has its acquisition settings and relevant filter evidence. Series B retains only a nominal export cadence. Series C was decimated, but the reduction/filter record is missing. Series D contains a documented gap. These are different limitations; none is resolved merely by increasing the number of historical rows supplied.
The register names the measured quantity and units, device/channel, native sample rate, actual timing evidence, front-end bandwidth, anti-alias filter, decimation history and intended inference band. Timing gaps or irregular intervals require a model appropriate to that acquisition; the uniform fixture above must not be transplanted unchanged.
A specialist must decide whether the documented response is adequate for the proposed use. Calibration traceability, frequency coverage, amplitude fidelity and the meaning of a derived failure label remain separate questions. The goal is a bounded acceptance statement, not a certificate that every inference from the series is reliable.
Narrow the claim when acquisition evidence is missing
The completed fixture leaves the original two-versus-eight-hertz question unresolved. Its next request is for acquisition and filtering evidence, not a cleaner plot or a more confident model label. If that evidence cannot be recovered, the recipient may need a narrower permitted analysis that does not require distinguishing the lost frequency information.
Bring the acquisition register and proposed inference band to the instrumentation owner. Specify the missing setting, filter response or decimation record that could change the decision. The synthetic demonstration establishes a mathematical ambiguity; it has not inspected a real instrument or diagnosed its operating condition.