Continuous-time and discrete-time coordinates
Write x(t) for continuous time and x[n] for integer indices. The sequence x[n − n0] delays x[n] for positive integer n0. Reversal x[−n] reflects around zero. Lists need starting indices: shifting an origin changes convolution alignment.
Check yourself: Where does an originally nonzero sample land after the transformation?
Linearity and time invariance
Linearity requires T{ax1 + bx2} = aT{x1} + bT{x2}. Time invariance requires matching input and output delays. Test them independently. Fixed nonzero initial conditions can add a zero-input response, so specify when an equation-based input-output model assumes initial rest.
Check yourself: Does zero input give zero output, and does a delay commute with the system?
LTI convolution and support
For an initially resting discrete-time LTI system, y[n] = Σk x[k]h[n − k]. Only overlapping nonzero samples contribute. If finite sequences occupy a through b and c through d, output is zero outside a + c through b + d; interior cancellation can still create zeros. Continuous-time convolution instead integrates over time.
Check yourself: Have you evaluated h[n − k], rather than multiplying matching list positions?
Causality and bounded-input bounded-output stability
A discrete-time LTI system is causal exactly when h[n] = 0 for n < 0, and BIBO stable exactly when Σn |h[n]| is finite. A finite impulse response with finite coefficients is stable but can be noncausal if negative-index coefficients are nonzero.
Check yourself: Did you inspect negative indices separately from the absolute sum?
Frequency variables and sinusoidal response
In cos(2πft), f is in Hz and ω = 2πf in rad/s. Sampling at fs gives Ω = 2πf/fs in rad/sample, modulo 2π. A stable LTI system scales complex exponentials by its frequency response without changing frequency; this excludes startup transients.
Check yourself: Is your frequency measured per second or per sample?