Skip to content
Tutorials Center
My learning

Topics

Lesson 10 of 79 ยท Physical Layer

Multiplexing and Physical Devices

Learn how one link carries many conversations with TDM, FDM, WDM and statistical multiplexing, and what hubs, repeaters, modems and network cables do.

Laying a cable between two cities is expensive. Laying one cable per phone call would be absurd. Multiplexing is the family of techniques that lets many separate signals share one physical link, and a demultiplexer at the far end splits them apart again. The idea appears at every scale, from a telephone trunk carrying thousands of calls to a single Wi-Fi channel shared by your laptop and phone.

This lesson covers the four main ways to share a link, then the simple devices that live at the physical layer: repeaters, hubs and modems, plus the cables you plug into them. Everything here works with raw signals, not frames or addresses. Those arrive in the data link layer. The previous lesson, Line Coding and Modulation, explained how a single stream of bits becomes a signal.

flowchart LR
    A["Source 1"] --> M["Multiplexer"]
    B["Source 2"] --> M
    C["Source 3"] --> M
    M -->|"one shared link"| D["Demultiplexer"]
    D --> X["Dest 1"]
    D --> Y["Dest 2"]
    D --> Z["Dest 3"]

Many inputs, one link, many outputs. The techniques differ in how they divide the link.

Frequency-division multiplexing (FDM)

FDM gives each signal its own band of frequencies, all transmitted at the same time. Radio and television work this way: each FM station owns a 200 kHz slice of the spectrum, and your radio's tuner picks one slice. Between bands, engineers leave small unused gaps called guard bands so neighbours do not interfere. Cable television, the telephone local loop with DSL (which uses many narrow subcarriers) and Wi-Fi channels are all FDM at heart.

Time-division multiplexing (TDM)

TDM gives each signal the whole link for a short slot, taking turns in a repeating cycle called a frame. In synchronous TDM slot number 3 always belongs to source 3, even when source 3 has nothing to say, so capacity is wasted on idle sources. The classic example is digital telephony: a voice is sampled 8,000 times per second at 8 bits per sample, which is 64 kbit/s, and a T1 link interleaves 24 of those channels. Because the receiver knows each slot's position, no addresses are needed.

Wavelength-division multiplexing (WDM)

WDM is FDM for light. A single optical fiber carries many beams of different colours (wavelengths), each running its own data stream. Coarse WDM (CWDM) spaces channels widely, up to 18 of them 20 nm apart, which allows cheap lasers. Dense WDM (DWDM) packs 40, 80 or more channels into the 1530 to 1565 nm "C band" where fiber loses the least light, with 0.4 to 0.8 nm between them. Each wavelength can carry 100 Gbit/s or more, so one fiber pair can carry many terabits per second. This is how continents are linked.

Statistical multiplexing

Synchronous TDM wastes slots when sources are bursty, and most computer traffic is bursty: a web page arrives in a rush, then nothing for seconds. Statistical (asynchronous) multiplexing allocates capacity on demand: whoever has data gets the next slot. Because the receiver can no longer tell from position who sent what, every chunk needs a label, a header with an address or channel number. This is exactly the principle of packet switching. The price is a buffer and possible queuing delay when many sources burst at once.

Technique Divides the link by Idle sources waste capacity? Needs labels?
FDM frequency yes (band stays reserved) no
Synchronous TDM time slot position yes (slot stays reserved) no
WDM light wavelength yes no
Statistical multiplexing demand no yes

The simulation below compares the two time-based approaches. Four bursty sources share a link that carries one packet per tick. Each source produces a packet with probability 0.15 per tick (total load 60%). Both runs see exactly the same arrivals.

Python Synchronous TDM versus statistical multiplexing

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

Both modes deliver almost the same number of packets, because the load is below capacity. The difference is delay: a TDM source whose packet arrives just after its slot must wait for the cycle to come around, even if the link is empty, while the statistical multiplexer sends it at once. That is the central trade-off. TDM gives each user a guaranteed, predictable share (ideal for voice), and statistical multiplexing gives better use of the link and lower average delay (ideal for data).

The digital hierarchy

Telephone engineers built TDM into standard ladders of rates. A single voice channel is 64 kbit/s (8,000 samples times 8 bits). A T1 frame holds one byte from each of 24 channels plus one framing bit, and frames repeat 8,000 times per second. The European E1 uses 32 slots of 8 bits, 30 for voice and 2 for signalling and synchronization. Optical networks use SONET (North America) and SDH (elsewhere), where each level is a multiple of 51.84 Mbit/s.

Python Arithmetic of the TDM hierarchy

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

The number to remember is 125 microseconds, one frame time (1/8000 s). Every voice sample from every channel has to be delivered within that heartbeat, which is why TDM networks need tightly synchronized clocks.

Physical-layer devices

Repeaters and hubs

A signal weakens (attenuates) and picks up noise as it travels. A repeater receives the weak signal, cleans it, and retransmits a fresh copy, extending the maximum cable length. It works on bits and knows nothing about frames. A hub is a multiport repeater: whatever arrives on one port is copied out to every other port.

That makes a hub cheap, but a hub-based network is one big collision domain. Only one device can talk at a time, and every device sees everyone's traffic, which is also a privacy and performance problem. Hubs are essentially obsolete today. Switches replaced them by forwarding frames only to the right port.

An amplifier, by contrast, boosts the signal and the noise together. A repeater regenerates a digital signal, so noise does not accumulate along the way.

flowchart TB
    subgraph H["Hub: one collision domain"]
        h(("Hub")) --- a1["PC A"]
        h --- b1["PC B"]
        h --- c1["PC C"]
    end
    subgraph S["Switch: one collision domain per port"]
        s(("Switch")) --- a2["PC A"]
        s --- b2["PC B"]
        s --- c2["PC C"]
    end

A hub repeats every bit to every port. A switch gives each port its own collision domain.

Modems and other signal converters

A modem (modulator-demodulator) converts digital data to a signal suited to a particular medium and back. The old dial-up modem squeezed data into the 300 to 3,400 Hz voice band of a telephone line, topping out at 56 kbit/s. A DSL modem uses the unused high frequencies on the same copper pair, splitting them into hundreds of narrow subcarriers and measuring each one's noise (an FDM technique called discrete multitone). A cable modem (DOCSIS) uses channels on the TV coax. Fiber needs an optical network terminal, and a media converter or SFP transceiver swaps between copper and fiber. Your home "router" is usually a bundle: modem, router, switch and Wi-Fi access point in one box.

A network interface card

A NIC (network interface card, or the chip in your laptop) is where the physical layer meets the data link layer. It generates the line code, drives the signal onto the wire, and hands received bits upward.

Beyond the physical layer: access points, brouters and network types

A few more devices and terms are worth placing next to the hubs and modems above, because they sit just above the physical layer.

Access points

A wireless access point (AP) connects Wi-Fi clients to a wired network. It converts between the radio medium and Ethernet frames and works at the data link layer, like a bridge (Wi-Fi basics). An AP is not a router: it does not decide IP routes, and it does not give out addresses by itself. Your home "Wi-Fi router" is a bundle of four boxes: a modem or ONT toward the ISP, a router with NAT and DHCP, a small switch for the LAN ports and an access point. In an office, many APs hang off switches and one controller or cloud service manages them. A range extender (repeater) is simpler and repeats a weak signal, usually halving throughput; a mesh system uses several APs that coordinate on one network name instead.

Brouters

A brouter is a bridge and a router in one device. For protocols it can route (such as IP) it forwards by network address; for others it falls back to bridging by MAC address. The idea mattered in the 1990s, when networks carried several layer 3 protocols at once (IP, IPX, AppleTalk). Today everything is IP, and a multilayer switch or ordinary router covers the case, so the word is mostly exam vocabulary.

More network types

The scale list (PAN, LAN, MAN, WAN) leaves out two terms you will meet:

  • A campus area network (CAN) joins the LANs of several buildings belonging to one organization (a university, a factory site, a hospital) within a few kilometers, usually over fiber run through private ground. It sits between a LAN and a MAN, and the tree topology is its natural shape.
  • An extranet is a controlled slice of one organization's private network (its intranet) opened to outsiders such as suppliers or partners, typically through a VPN or a login (firewalls and VPNs). It is defined by who may use it, not by distance.

Run the lookup below to compare devices by the layer they work at and the network types by reach.

Python Devices by layer, networks by reach

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

The rule of thumb: a device that only handles signals is layer 1, one that reads MAC addresses is layer 2, one that reads IP addresses is layer 3, and the brouter straddles the last two.

Cables and standards

Wired Ethernet uses standards that name the speed, the signalling and the medium. The pattern is speed + BASE + medium, so 1000BASE-T means 1,000 Mbit/s, baseband signalling, twisted pair (T).

Medium Typical use Notes
Cat 5e twisted pair 100 Mbit/s and 1 Gbit/s The long-time default; 100 m channel limit
Cat 6 twisted pair 1 Gbit/s to 100 m; 10 Gbit/s to about 55 m Better shielding against crosstalk
Cat 6a twisted pair 10 Gbit/s to 100 m Thicker and stiffer
Cat 8 twisted pair 25 and 40 Gbit/s to about 30 m Data center racks
Multimode fiber 10 to 100 Gbit/s over tens to hundreds of metres Cheap LED/VCSEL optics, inside buildings and data centers
Single-mode fiber 10 Gbit/s and far above over kilometres Long distance, campus links, WDM
Direct-attach copper (DAC) 10 to 100 Gbit/s over a few metres Within a rack

Copper Ethernet uses an RJ-45 connector with four twisted pairs and two accepted colour orders, T568A and T568B. A straight-through cable uses the same order at both ends and connects a computer to a switch. The older crossover cable swapped pairs to connect like devices directly, but modern ports negotiate the pair roles automatically (Auto-MDI/MDI-X), so one cable type works nearly everywhere. Fiber links use two strands, one for each direction, with LC or SC connectors.

The 100 m limit on twisted pair (usually 90 m of permanent cabling plus patch cords) comes from attenuation and crosstalk. Beyond it, you add a switch or use fiber. Old coaxial Ethernet, 10BASE2 and 10BASE5, shared one cable among all machines. It is history, but it explains why early Ethernet needed the collision rules covered in Medium Access Control.

Fiber also loses light. If a laser sends 0 dBm and a receiver needs at least -14 dBm, you have 14 dB of budget. Single-mode fiber at 1550 nm loses roughly 0.2 dB per kilometre, and each connector or splice costs a few tenths of a dB, so a link of tens of kilometres works without any repeaters. Amplifiers (erbium-doped fiber amplifiers) sit every 60 to 100 km on long-haul routes and boost all WDM channels at once.

FDM and WDM channel plans

Both FDM and WDM are just careful bookkeeping of slices. The program below builds an FM broadcast plan and a DWDM plan, converting between frequency and wavelength with c = 299,792,458 m/s.

Python FDM and DWDM channel plans

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

Notice that higher frequency means shorter wavelength: stepping up 0.1 THz moves the wavelength down about 0.8 nm. That is the figure quoted for 100 GHz DWDM spacing.

Multiplexing divides one link among many signals. Do not confuse it with switching, which decides where each signal goes. A network usually does both: a multiplexed trunk carries many conversations between two switches, and the switches decide which conversation goes out of which port.

Check your understanding

Quick check ยท 5 questions

of 5 answered You scored / 5 ยท perfect!

  1. 1. What is the main drawback of synchronous TDM for bursty computer traffic?

    Slots are assigned by position whether or not the owner has data. Bursty sources leave many slots empty, which statistical multiplexing avoids.
  2. 2. Why does statistical multiplexing need a header on each chunk of data?

    Capacity is given to whoever has data, so the position in the stream carries no meaning. A label identifies the conversation.
  3. 3. Which technique is used to put many data streams on one optical fiber?

    WDM sends each stream on a different wavelength (colour) of light, like FDM for optics.
  4. 4. What is the difference between a hub and a switch?

    A hub is a multiport repeater with a single collision domain. A switch reads the destination address and sends the frame only where it needs to go.
  5. 5. How long does one frame of a T1 or E1 line last, and why?

    Each frame carries one sample from every channel, and there are 8,000 samples per second, so a frame repeats every 1/8000 s = 125 microseconds.

Summary

  • Multiplexing lets many signals share one link; demultiplexing splits them at the other end.
  • FDM divides by frequency, TDM by time slot, WDM by wavelength of light, and statistical multiplexing by demand.
  • Synchronous TDM is predictable but wastes idle slots; statistical multiplexing uses the link better but needs labels and buffers.
  • The telephone hierarchy (64 kbit/s voice, T1 1.544 Mbit/s, E1 2.048 Mbit/s, SONET OC-n) is built on 8,000 frames per second.
  • Repeaters regenerate signals; hubs are multiport repeaters with one collision domain and have been replaced by switches.
  • Modems convert between digital data and the signal of a particular medium; cable standards (Cat 5e/6/6a, multimode and single-mode fiber) set distance and speed.
  • Access points bridge wireless clients onto a wired LAN, a brouter combines bridging and routing, and a modem converts signals for a medium; the network types CAN and extranet fill the gap between LAN, MAN and WAN.

Next: Framing and Error Detection