Tensions in Practice: Signal retention in tension with sampling cost¶
Ideal sampled tone · a declared finite bandwidth
Sampling can make different frequencies look identical. In this toy, a one-cycle and a three-cycle cosine produce the same readings on a four-samples-per-unit grid. Two different responses avoid misreading the three-cycle tone: remove it before sampling, or use an eight-sample grid that can distinguish it. The first saves samples by giving up that content; the second keeps the content and records more values.
Keep the wanted fast variation
Retain information carried by the three-cycle component rather than filtering it away.
Bound sampling and storage work
Use a lower sampling rate when the excluded band is not part of the desired information.
Why these aims pull against each other
Filtering reduces the bandwidth presented to the sampler. Faster sampling instead increases the record’s ability to distinguish that bandwidth; neither choice is equivalent to averaging the ambiguous record later.
Choose an arrangement to see what changes and what remains difficult.
Finite illustrative comparisons. Labels carry the meaning; color does not establish a preference or measured effect.
What this choice protects
What it costs
When it fits
Compare the arrangements
Filter then sample at 4
Use an ideal filter whose passband excludes the three-cycle tone, then take four samples per time unit. Invented time unit; source cos(6πt), three cycles per unit. An ideal filter removes this above-band tone before sampling four times per unit. “No sample” marks times outside that sampling grid.
| Source value | Recorded | |
|---|---|---|
| 0/8 | 1 | 0 |
| 1/8 | −√2/2 | No sample |
| 2/8 | 0 | 0 |
| 3/8 | √2/2 | No sample |
| 4/8 | −1 | 0 |
| 5/8 | √2/2 | No sample |
| 6/8 | 0 | 0 |
| 7/8 | −√2/2 | No sample |
- What it protects
- The low-rate record does not turn this removed tone into an apparent one-cycle component.
- What it costs
- The real three-cycle information is absent from the record and cannot be recovered from these zeros.
- When it fits
- Fits a task where that higher-frequency band is unwanted and a practical filter can sufficiently reject it before sampling.
Illustration note: The filter is idealized and removes the toy’s only tone. Real filters have transition bands and finite attenuation; no filter specification is supplied.
Keep it; sample at 8
Sample the declared three-cycle source at eight times per time unit. The same source is sampled eight times per unit with no removal of this three-cycle component. Values are exact; the model contains no content above three cycles per unit.
| Source value | Recorded | |
|---|---|---|
| 0/8 | 1 | 1 |
| 1/8 | −√2/2 | −√2/2 |
| 2/8 | 0 | 0 |
| 3/8 | √2/2 | √2/2 |
| 4/8 | −1 | −1 |
| 5/8 | √2/2 | √2/2 |
| 6/8 | 0 | 0 |
| 7/8 | −√2/2 | −√2/2 |
- What it protects
- The selected grid distinguishes the retained tone from the one-cycle alternative.
- What it costs
- Twice as many values are recorded as under the four-sample option, before accounting for implementation overhead.
- When it fits
- Fits wanted content within the stated bandwidth and an affordable sampling/data budget.
Illustration note: The finite trigonometric values are editorial mathematics, not measured sensor data. The rate is not sufficient for arbitrary unknown higher-frequency content.
What this illustration does—and does not—establish
Aliasing: Bandlimit-Before versus Sample-Faster (Remedy Routing) supplies the two remedies and their different costs. The finite trace makes both the lost component and the extra recorded values visible without inventing an empirical response curve.
- The example specifies exact time points and a bandwidth bound; it does not infer that bound from the samples themselves.
- The source-value column describes the mathematical input, which a real measurement would not know in advance.
- Sampling density and amplitude quantization are different choices; increasing bit depth alone would not separate the aliased frequencies.
Source entries
Aliasing
Aliasing: Bandlimit-Before versus Sample-Faster (Remedy Routing) supplies the conflict examined here.
Bandlimit-Before versus Sample-Faster (Remedy Routing)
Two distinct corrections prevent aliasing — removing high-frequency content *before* sampling (anti-alias filtering) and raising the sampling rate above twice the highest frequency — and they are not interchangeable. The tension is between filtering (which discards real high-frequency content permanently) and oversampling (which raises data volume and cost).