A timestamp's displayed precision is only one part of temporal resolution. Sampling intervals, exposure windows, noise and the assumed signal family all affect what an instrument can distinguish. A simple oscillating signal shows how missed variation can look like slower motion or no motion at all.
Different continuous signals can share sampled values
Take x(t) = cos(2πft) and record it at tn = n/fs. For instantaneous sampling, adding any integer multiple of fs to f leaves every sampled value unchanged. Reflection of frequency also leaves this cosine sequence unchanged.
At eight samples per second, signals at 6 Hz and 2 Hz give identical values. A signal at 8 Hz gives the constant sequence one. The signal has changed between readings even though the discrete record suggests no change.
This is aliasing. Increasing the number of decimal places in the recorded values does not resolve it. The ambiguity arises from the observation procedure and the allowed signal family.
Change the sampling and exposure
IDEAL COSINE · ONE-SECOND DISPLAY · NO NOISE
Green: original continuous signal. Purple dashed: a curve matching the recorded samples. Gold dots: readings averaged over centred exposures. The dashed curve is not a reconstruction guaranteed to be the original signal.
At 8 samples per second, this 6 Hz cosine shares its instantaneous samples with a 2 Hz cosine.
An exposure averages rather than freezes
For a centred rectangular exposure of width w, the recorded value is the average of x over tn − w/2 to tn + w/2. Direct integration gives A cos(2πftn), where A = sin(πfw)/(πfw), with A = 1 at w = 0.
The display sets w to a chosen fraction of 1/fs. Its dashed curve is A cos(2πfat), where fa is f folded into the range zero to fs/2. That curve matches the sample values, including exposure attenuation.
It is a sample-matching interpolation, not necessarily the response of another unit-amplitude physical input. Known input amplitude and a calibrated exposure response can provide additional information. Unknown amplitudes leave further ambiguity.
At f = fs with a full-interval exposure, A = 0: the detector averages one whole cycle and records zero. With instantaneous exposures the same setting records one. Sampling and averaging are distinct operations.
Recovery requires restrictions on the signal
Sampling theory explains when a continuous signal can be recovered from an ideal sample sequence under a bandlimit assumption. A safe elementary condition is that the sampling rate exceed twice the highest permitted frequency, with appropriate idealizations about the full sequence.
A finite, noisy experiment adds boundaries, calibration error and limited observation duration. A handful of points cannot establish that an unknown signal contains no higher-frequency component. That restriction must come from physical knowledge, filtering or additional measurements.
The one-second graph is therefore an aliasing illustration, not a claim to reconstruct arbitrary signals. The display includes both endpoints, so eight sample intervals have nine plotted readings. It assumes the cosine continues beyond the displayed interval when computing centred exposures at the endpoints.
Read completion through an explicit detector
The book's clock calibration and sampled-history discussion distinguish records from a presumed continuous process. In SMM/DTT, a completion rule must identify which distinctions become accessible to later observations.
The demonstration shows why discrete records alone cannot establish discrete physical time. A stronger proposal predicts how the record changes when the sampling or detector coupling changes, using a common underlying rule.
Sources and further reading
- C. E. Shannon, Communication in the Presence of Noise, Proceedings of the IRE 37, 10–21 (1949), especially the sampling theorem. The exposure demonstration is an original teaching calculation.
- Continue with continuous and discrete descriptions and what an image records.

