Tensions in Practice: Low-amplitude detail in tension with extreme-value coverage¶
Signal encoding · fixed two-bit indexes
A signal encoder has four two-bit codes. Each code decodes to one declared number, and every input from 0 to 9 maps to its nearest level. Levels 0, 3, 6 and 9 spread the budget across the range. Levels 0, 1, 2 and 3 distinguish small values more finely, but large inputs all end up at 3. The bit budget is unchanged; the placement of its representatives changes.
Resolve low amplitudes
Spend the limited code levels near the values of most interest.
Represent large excursions
Retain representatives near the high end of the admissible range.
Why these aims pull against each other
Concentrating a fixed number of levels improves some local errors by leaving other values farther from their nearest representative. There is no source-independent best placement.
Choose an arrangement to see what changes and what remains difficult.
Finite illustrative comparisons. Text states carry the meaning; color is not a measured score or universal preference.
What this choice protects
What it costs
When it fits
Compare the arrangements
Cover the full range
Decode the four indexes to 0, 3, 6 and 9; choose the closest for every input.
| Input | Decoded | Error | |
|---|---|---|---|
| Low 1 | 0.4 | 0 | 0.4 |
| Low 2 | 1.2 | 0 | 1.2 |
| Low 3 | 2.2 | 3 | 0.8 |
| High | 8.8 | 9 | 0.2 |
- What it protects
- The selected high input 8.8 decodes to 9 with error 0.2.
- What it costs
- The selected low input 1.2 decodes to 0 with error 1.2.
- When it fits
- Fits when high excursions matter enough to spend scarce levels across the whole range.
Illustration note: The finite setting and values are editorial assumptions, not measured effects or recommended operating settings. Error is the absolute difference between input and reconstruction; ties choose the lower level.
Concentrate near zero
Decode indexes to 0, 1, 2 and 3; inputs above the upper decision cell still decode to 3.
| Input | Decoded | Error | |
|---|---|---|---|
| Low 1 | 0.4 | 0 | 0.4 |
| Low 2 | 1.2 | 1 | 0.2 |
| Low 3 | 2.2 | 2 | 0.2 |
| High | 8.8 | 3 | 5.8Clipped |
- What it protects
- The selected 1.2 and 2.2 inputs each have error 0.2.
- What it costs
- The same 8.8 input clips to 3, with error 5.8.
- When it fits
- Fits only when low-amplitude precision matters more and that loss on high excursions is acceptable or separately detected.
Illustration note: The finite setting and values are editorial assumptions, not measured effects or recommended operating settings. Both maps are defined on the same admissible 0–9 range; no extra bits or sample times are added.
What this illustration does—and does not—establish
Signal Quantization: Typical-value precision versus extreme-value coverage directly supplies fixed-budget placement versus overload. The example narrows the broad discrete/continuous theme to explicit decision and reconstruction maps.
- Four displayed inputs do not establish a source distribution or expected distortion ranking.
- This is value quantization, not temporal sampling, physical energy quantization or a complete compression codec.
- Every code uses two bits; headers, codebook transport and other conversion errors are outside the toy.
Source entries
Signal Quantization
Signal Quantization: Typical-value precision versus extreme-value coverage supplies the conflict examined here.
Typical-value precision versus extreme-value coverage
Concentrating levels where a source is common improves typical reproduction but may leave overload or saturation error on rare inputs. Extending coverage can weaken typical resolution at a fixed level budget.
Core Idea
A rule assigns an input to a cell or code index; a reproduction rule associates that index with a level or vector.