LIMA
With God's help
עברית

System Properties

Imagine a perfect automatic coffee machine in a café – it is our "system", pressing the button is the input, and the drink in the cup is the output. A memoryless machine makes the coffee only according to the button you pressed right now, with no connection at all to what the previous customer ordered; it is causal because it will never pour coffee before you have actually pressed the button; and it is invertible if it is enough to look at the finished cup to know for sure which button was pressed. The machine is stable if a reasonable press produces a reasonable amount of coffee (and does not flood the café), time-invariant if the same button produces exactly the same espresso both at eight in the morning and at five in the evening, and linear if a double order gives exactly double the coffee, and ordering an espresso + milk together gives exactly the sum of the two separately.

The Theory in Brief

In signal and system analysis, we model a system as an operator \(T\) that turns an input signal \(x(t)\) into an output signal \(y(t) = T\{x(t)\}\). To know which mathematical tools we are allowed to apply to the system, we classify it according to six basic properties (which hold equally in discrete time \(n\)):

  • Memoryless: the output at every moment \(t_0\) depends only on the input at exactly the same moment, \(x(t_0)\) (for example an ideal resistor \(v(t) = R\cdot i(t)\)). If the output depends on past or future values (like a capacitor or an integrator), the system has memory.
  • Invertibility: every two different input signals produce two different output signals. In that case there exists an inverse system \(T^{-1}\) that recovers the input exactly: \(T^{-1}\{y(t)\} = x(t)\).
  • Causality: the output at moment \(t_0\) depends only on present and past inputs (\(x(t)\) for \(t \le t_0\)) and never on the future – a necessary condition for any physical system operating in real time.
  • BIBO Stability: every bounded input yields a bounded output. Formally, if there exists \(M_x < \infty\) such that \(|x(t)| \le M_x\) for every \(t\), then there exists \(M_y < \infty\) such that \(|y(t)| \le M_y\) for every \(t\).
  • Time-Invariance: the behavior of the system does not change over time – shifting the input in time by \(t_0\) simply shifts the output by exactly the same amount of time: \[T\{x(t - t_0)\} = y(t - t_0)\]
  • Linearity: the system satisfies the superposition principle (additivity and homogeneity), that is, for all constants \(a, b\) and signals \(x_1(t), x_2(t)\): \[T\{a\cdot x_1(t) + b\cdot x_2(t)\} = a\cdot T\{x_1(t)\} + b\cdot T\{x_2(t)\}\]

The last two properties – linearity and time-invariance – are the most important in the course, since a system that satisfies both is an LTI system, in which every output can be computed using convolution alone.

x(t)y(t) T{·} Memoryless Invertibility Causality Stability Time-Invariance Linearity LTI → convolution

How to Read the Lab

The input signal \(x(t)\)

The left graph (on mobile: the top one), in glowing cyan. In every test it shows the inputs being tested.

T{·}

The black box \(T\{\cdot\}\)

In the middle, with the formula of the selected system. Points of light flow from the input into the box and from the box to the output.

The output signal \(y(t)\)

The right graph (on mobile: the bottom one), in orange-gold.

The row of 6 diagnostic lamps

At the top of the display: green = the property holds, red = it fails. Clicking a lamp turns the workbench into its stress test, and the active lamp is highlighted with a frame.

The laser beam (memoryless)

A purple vertical marker of \(t_0\) on both graphs, which you can drag. Threads of light stretch from \(y(t_0)\) to the points in \(x\) that affect it: white = present, light blue = past, red = future. A \(-\infty\) arrow marks a dependence on the whole past.

The forbidden zone of the future (causality)

Left of the marker a soft "past and present" fill, right of it diagonal "future" stripes. "Launch a step pulse" runs a step that arrives at \(t_0\), and the dashed gray line is the output without the step. An early response paints the zone red with "Peeking into the future!".

The bound walls (stability)

Dashed lines \(\pm M_x\) on the input and \(\pm M_y\) on the output. The time axis expands and the camera zooms out: the walls are green while the output is caged, and red when it breaks through ("Diverges to infinity").

The shift-overlap test (time-invariance)

Track A in cyan: shift and then the system, \(T\{x(t-t_0)\}\). Track B in magenta: the system and then shift, \(y(t-t_0)\). Welded = one white line; separated = a red gap between them.

a·x₁+b·x₂

The superposition test (linearity)

At the input \(x_1\) (cyan), \(x_2\) (magenta) and the combination \(a\,x_1+b\,x_2\) (white). At the output \(T\{a\,x_1+b\,x_2\}\) in cyan versus \(a\,T\{x_1\}+b\,T\{x_2\}\) in magenta.

The collision test (invertibility)

Two different inputs, \(x_1\) and \(x_2\). If the outputs merge into one white line: "Collision! Information lost". For an invertible system, "Connect inverse system \(T^{-1}\)" adds a green box and a reconstruction graph \(\hat{x}\).

The system bank

"Exam traps": \(2x\), \(x+1\), \(x(2t)\), \(x^2\), \(t\cdot x\); and systems with memory: integrator, moving average and the advance \(x(t+1)\). Choosing a system lights up all 6 lamps at once.

Show me a counter-example

When the active lamp is red: sets the controls and the input signal to the edge case that breaks the property, with one explanatory sentence below the graphs.

Continuous/discrete switch \(t \leftrightarrow n\)

Switches between continuous curves and stem plots ("lollipops", Stem plot) over \(n\) from \(-8\) to \(8\). The lamps are set for each mode separately.

The input signal and the pace

A step, a pulse or a sine (all bounded). The speed slider and the pause button control only the pace of the animations: the step launch, the zoom and the reconstruction.

The readouts

The system, the mode and the lamps, and for each test its own values: what \(y(t_0)\) depends on, the response time, \(M_x\) and \(\max|y|\), the gap between the tracks, and the reconstruction error.

To the lab ↓
Loading the lab...
1

Experiments

  1. The classic linearity trap \((y(t) = x(t) + 1)\). Choose \(y(t) = x(t) + 1\) in the system menu. Notice that all the lamps are green except linearity. Click the linearity lamp to activate the "superposition test", and move the slider \(a\) from \(1\) to \(3\), and then to \(0\). Follow the red gap that opens between the two lines in the output graph.
    This is the most common trap in exams – the student sees the equation of a straight line and assumes the system is linear. The experiment exposes the mistake at once: when you moved \(a\) to \(0\) (zero input), the output still came out \(1\) and not \(0\), and when you multiplied the input by \(3\), the constant \(+1\) was not multiplied by \(3\). A linear system must pass the "zero test" (zero input \(\rightarrow\) zero output) and be fully homogeneous – adding a constant (a DC offset) breaks both.
  2. Time compression – the silent enemy of causality and time-invariance \((y(t) = x(2t))\). Choose the time-compression system \(y(t) = x(2t)\). First click the causality lamp, drag the vertical marker to \(t_0 = 1.5\), and watch the "Peeking into the future!" alert. Then switch to the time-invariance lamp and drag the shift slider \(t_0\) to the right and to the left to see the cyan track and the magenta track separate from each other.
    It is easy to think that "speeding up a film 2×" is an innocent operation, but the simulation shows why it is impossible in real time: to compute the output at moment \(t=1.5\), the system needs to know what the input signal will be at moment \(t=3\) – that is, to peek into the future! In addition, the system is not time-invariant because the order of operations changes the result: shifting and then compressing gives \(x(2t - t_0)\), while compressing and then shifting gives \[x(2(t - t_0)) = x(2t - 2t_0)\] – two completely different shifts on the time axis.
  3. Information loss in the collision test \((y(t) = x^2(t))\). Choose the system \(y(t) = x^2(t)\). Notice that four lamps are green (memoryless, causality, stability and time-invariance), but invertibility is red. Click the invertibility lamp to activate the "collision test" and watch the two opposite signals \(x(t)\) and \(-x(t)\) merge into exactly the same output. Then switch to \(y(t) = 2x(t)\) and click "Connect inverse system \(T^{-1}\)".
    Squaring "erases" the plus/minus sign of the input signal. The moment two different input signals produce exactly the same output, information has been lost that cannot be recovered – no system \(T^{-1}\) will be able to look at the output and guess whether the input was positive or negative. In contrast, in the system \(y(t)=2x(t)\) the relation is one-to-one, and so adding a \(T^{-1}\) box (division by 2) recovers the original signal in full.
  4. When time itself steps in – memoryless but unstable \((y(t) = t \cdot x(t))\). Choose the system \(y(t) = t \cdot x(t)\). First click the memoryless lamp and drag the marker \(t_0\) to see that the laser beam always stays straight (only one point has an effect). Now click the stability lamp (BIBO) and watch the camera do a dramatic zoom-out as the output breaks through the bound walls in red.
    This experiment breaks a common confusion: a system can be linear and memoryless, and still be unstable and not time-invariant. It is memoryless because at moment \(t\) it needs only the current value \(x(t)\), but since its "amplifier" is the time \(t\) itself, it grows to infinity. Therefore, even if we feed in a bounded and quiet input signal (like a step of height \(1\)), the output will grow without bound and break through any finite ceiling \(\pm M_y\).
  5. The fine difference between memory and causality. Turn on the memoryless lamp for a system with past memory (like an integrator) and look at the "glowing fan" of dependency threads. Then switch to the causality lamp and notice on which side of the marker \(t_0\) the dependency threads fall. Now repeat the same check for the advance system \(y(t) = x(t+1)\) and compare the two fans.
    Many students confuse "memoryless" with "causal". The difference becomes clear right before your eyes: every memoryless system (a single straight laser beam) is automatically causal, but a system with memory can certainly be causal – as long as its fan points only to the left (to the past and the present). The moment even one dependency thread crosses to the right into the hatched "future" zone, the system loses its causality.