Signals, Channels, and Noisy Measurements

A bridge page showing how signals are transformed by channels and sensing systems, and why noise is the common object tying communication and estimation together.
Modified

April 26, 2026

Keywords

signal, channel, noise, measurement, communication

1 Application Snapshot

A large fraction of communication and sensing can be summarized in one sentence:

a useful signal is generated, transformed by a channel or measurement system, and observed with corruption

That sentence already contains the three basic objects:

  • signal
  • channel
  • noise

This page is the shortest bridge from the site’s math modules into the language used in communication systems, sensing pipelines, filtering, and inference.

2 Problem Setting

A signal may be:

  • a time-varying waveform
  • a discrete-time sequence
  • a vector of transmitted symbols
  • an image or sensor field sampled over space

A channel or measurement system then transforms that signal.

At first pass, the picture often looks like

\[ y = Hx + \eta \]

or, in time-indexed form,

\[ y_t = h(x_t) + \eta_t. \]

Here:

  • \(x\) or \(x_t\) is the underlying signal
  • \(H\) or \(h\) describes the channel, system, or sensing map
  • \(\eta\) or \(\eta_t\) is noise or corruption

In communication, \(x\) may be a transmitted message waveform and \(y\) the received waveform.

In sensing, \(x\) may be the underlying state or scene and \(y\) the measured sensor output.

3 Why This Math Appears

This language reuses several math layers already on the site:

  • Linear Algebra: many channels and sensing systems are modeled as linear maps or matrices
  • Probability: noise models describe what corruption or uncertainty does to the signal
  • Signal Processing and Estimation: filtering and denoising try to recover useful structure from corrupted observations
  • Information Theory: channels are also information bottlenecks with rate and reliability limits
  • Numerical Methods: reconstruction and inversion often become computational problems, not only formulas

So the signal-channel-noise picture is not a narrow engineering trick. It is the shared translation layer between communication, sensing, filtering, and inference.

4 Math Objects In Use

  • signal \(x\)
  • received or measured signal \(y\)
  • channel or sensing map \(H\) or \(h\)
  • noise \(\eta\)
  • sometimes bandwidth, spectrum, or sampling rate
  • sometimes an estimator or decoder trying to recover information from \(y\)

At first pass, two very common models are:

\[ y = Hx + \eta \]

and

\[ y = x + \eta. \]

The first emphasizes transformation plus corruption.

The second emphasizes direct noisy observation.

5 A Small Worked Walkthrough

Imagine a transmitter sending a binary symbol sequence through a noisy link.

The ideal transmitted signal is

\[ x \in \{-1,+1\}^n. \]

The receiver observes

\[ y = x + \eta, \]

where \(\eta\) is random noise from the channel.

Now imagine a sensor measuring temperature with a noisy probe.

The underlying signal is the true temperature trajectory \(x_t\), and the sensor outputs

\[ y_t = x_t + \eta_t. \]

These two settings look different on the surface, but the math objects are strikingly similar:

  • there is an underlying signal
  • the observation is corrupted
  • the downstream task is to recover structure, estimate truth, or decode information

That is why communication and sensing reuse so much of the same math.

6 Implementation or Computation Note

Once the signal-channel model is written down, the workflow branches into three practical questions:

  1. Representation What form of the signal should we work with: time domain, frequency domain, symbols, or state variables?

  2. Recovery Are we trying to denoise, estimate hidden structure, detect symbols, or decode a message?

  3. Limits Is the main bottleneck noise, bandwidth, computational cost, or information loss?

Use these pages as the strongest follow-on support:

7 Failure Modes

  • treating the received or measured signal as if it were the underlying truth
  • talking about noise as if it only means “small random error,” when it can also represent model mismatch or interference
  • forgetting that the same observation model can support very different downstream tasks: decoding, denoising, estimation, or detection
  • confusing a channel model with a decoder or estimator
  • ignoring whether the important bottleneck is bandwidth, noise level, or recoverability

8 Paper Bridge

  • 6.011 / Signals, Systems and Inference - First pass - official MIT bridge where signals, systems, and noisy inference appear in one shared language. Checked 2026-04-25.
  • EE376A / Information Theory - Paper bridge - useful once the signal view turns into channel limits, coding, and information bottlenecks. Checked 2026-04-25.

9 Sources and Further Reading

Back to top