A digital instrument reports a finite set of codes. That fact describes the instrument's readout. Deciding whether the underlying physical variable is continuous, discrete or something else requires an additional argument and discriminating evidence.
From a range of values to a code
Consider an ideal three-bit converter covering the interval from 0 to 1 volt. Define eight equal bins of width q = 1/8 volt and let code k represent the bin from kq to (k+1)q. Use the midpoint (k+1/2)q as the reconstructed voltage. The upper endpoint is assigned to the final bin.
An input of 0.30 V produces code 2 and reconstructed value 0.3125 V. An input of 0.34 V produces the same code. Nothing in that record distinguishes the two inputs. The digital output has not shown that the input itself jumps between eight values; the quantizer deliberately groups a continuum of possible inputs.
With four bits, q becomes 1/16 V. The same two inputs give codes 4 and 5, reconstructed as 0.28125 V and 0.34375 V. A change of resolution exposes a distinction that the earlier record could not retain. Real devices also introduce offset, gain error, nonlinearities and noise.
A bound and a probability model are different claims
For the ideal midpoint quantizer within its stated range, the reconstruction error has magnitude at most q/2. That deterministic bound follows from the bin geometry. To claim an error variance q²/12, one must additionally model the error as uniformly distributed across the bin.
A constant input need not explore all positions in a bin. A periodic input can have errors correlated with its phase and the sampling schedule. Consequently, quantization error is not automatically independent white noise. The familiar root-mean-square estimate q/√12 is useful when its assumptions approximate the experiment.
For the three-bit example q/√12 is approximately 0.0361 V. This is a model-based uncertainty contribution, not a measured error for every individual sample. Values outside the instrument's input range can saturate and violate the within-range error bound altogether.
Quantizing amplitude is not sampling time
A device can use fine amplitude resolution while sampling rarely, or coarse amplitude resolution while sampling rapidly. The two limitations lose different distinctions. Increasing the number of bits per sample does not recover a short event that occurred entirely between samples.
The existing sampling and exposure demonstration shows how different continuous signals can produce identical point samples. Its lesson complements the voltage example: a readout map can identify many underlying histories even when each retained number is represented with high precision.
Reconstruction becomes possible only with additional restrictions, such as a justified bandwidth model, and a compatible sampling procedure. The restriction is part of the inference. It is not discovered merely from the existence of the sampled data.
Discrete models can have continuous limits
A discrete recurrence can approximate a differential equation as its step size decreases. For example, repeatedly multiplying a positive quantity by 1 − γΔt gives, after n = t/Δt steps, a factor (1 − γΔt)t/Δt, which tends to e−γt as Δt tends to zero at fixed t and γ. The limiting equation can be useful even when a simulation executes discrete operations.
The reverse inference also fails: a successful continuous approximation does not prove that no finer discrete structure exists. What matters empirically is whether competing models predict different accessible outcomes, and whether those differences survive instrumental limitations and uncertainty.
A finite lattice model, a numerical grid and a fundamental discretization are therefore distinct commitments. Their equations may resemble one another while serving different explanatory purposes.
Reading discreteness in the books
The Single Monad Model's discussion of discrete time connects renewal and perceived continuity within its interpretation of Ibn al-Arabi. The expanded Duality of Time discussion of renewal, persistence and a continuous limit provides another route into the question. Neither discussion should be reduced to the observation that a camera has frames or a converter has codes.
A physical SMM/DTT prediction of discreteness needs a specified scale, a calibrated observable and a departure from a relevant continuous alternative. It must also distinguish a proposed effect from sampling, quantization and numerical artifacts. Instrumental discreteness is a constraint on evidence; fundamental discreteness is a claim about the modelled world.
Sources and further reading
- W. Kester, MT-001: Taking the Mystery out of the Infamous Formula, “SNR = 6.02N + 1.76dB,” and Why You Should Care, Analog Devices. The tutorial explains ideal quantization and the limitations of its noise model.
- Compression and the distinctions we lose: why a grouped record can preserve one task while losing another.