Skip to content
Tutorials Center
My learning

Topics

Lesson 2 of 79 ยท Getting Started with Networks

Transmission Modes: Simplex, Half-Duplex and Full-Duplex

Learn the three directions a link can carry data, serial versus parallel transmission and asynchronous versus synchronous framing, with runnable models of turnaround and UART frames.

Once two devices are connected, the first question about the link is simple: who is allowed to talk, and when? A cable between two computers can behave like a one-way street, like a narrow bridge that traffic crosses in alternating directions, or like a two-lane road with traffic flowing both ways at once. These three behaviors are the transmission modes. The words come from telephony, but they describe a physical or logical property of every link you will meet, from a keyboard cable to a fiber across an ocean.

Directions are only one of the choices. A link also decides whether it sends bits one at a time or many at once (serial or parallel), and how the receiver knows where each character or block starts (asynchronous or synchronous). This lesson covers all three choices, builds small models of each, and explains which technologies use which. You met LAN, WAN and the other network types in the first lesson, and you will see later how signals and media carry the bits themselves.

Direction: simplex, half-duplex, full-duplex

Mode Directions Analogy Examples
Simplex One way only, always the same sender Radio or TV broadcast, a one-way street Keyboard to computer, a TV antenna, a sensor that only reports
Half-duplex Both ways, but only one at a time Walkie-talkie ("over"), a one-lane bridge Old shared-cable Ethernet, Wi-Fi on one channel, CB radio
Full-duplex Both ways at the same time A telephone call, a two-lane road Switched Ethernet, mobile phone calls, a TCP connection between hosts

Simplex is the cheapest: the sender has a transmitter and the receiver has a receiver, and nothing more is needed. The price is that the receiver can never say anything back, not even "I did not get that". Reliable protocols over a simplex link have to work without feedback, using redundancy such as error-correcting codes, or repeating the data (as TV broadcasts and some satellite links do).

Half-duplex lets both sides send, but the medium is shared between the two directions. Before a device can transmit, it must make sure the other side has stopped, and when the direction changes there is a turnaround time: a radio must switch from receive to transmit, a modem must change its mode, and a shared Ethernet cable must wait for the cable to be quiet. On busy shared media the devices also need rules to avoid talking at the same time, which is the subject of medium access control.

Full-duplex uses separate channels for the two directions, for example one wire pair for each direction, two separate frequencies, or two separate time slots. Both sides can send whenever they like and never wait for each other. Modern switched Ethernet is full-duplex, which is why it has no collisions and no CSMA/CD, as described in the Ethernet lesson. It doubles the aggregate capacity of a link: a 1 Gbit/s full-duplex port can carry 1 Gbit/s in each direction at the same time.

flowchart LR
    subgraph SIMPLEX["Simplex"]
        direction LR
        S1["Sender"] --> R1["Receiver"]
    end
    subgraph HALF["Half-duplex"]
        direction LR
        A2["Station A"] <-->|"one at a time"| B2["Station B"]
    end
    subgraph FULL["Full-duplex"]
        direction LR
        A3["Station A"] -->|"channel 1"| B3["Station B"]
        B3 -->|"channel 2"| A3
    end

Three ways a link can use its direction.

Be careful with two points. First, the protocol can be full-duplex even if the wire is half-duplex: a TCP connection sends data both ways, but over a half-duplex wireless channel the radio still alternates. Second, a duplex mismatch, where one end of an Ethernet link is set to full-duplex and the other is half-duplex, produces strange slow performance with many errors, a classic troubleshooting story.

A model of the cost of turnaround

The first program compares the three modes for a conversation in which two devices exchange the same number of messages. Half-duplex alternates and pays a turnaround time for each change of direction. Full-duplex runs both streams at once, so only the slowest stream counts. The second experiment varies the turnaround time to show why small messages and slow turnarounds make half-duplex links painful.

Python Half-duplex turnaround versus full-duplex

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.

For 50 messages in each direction, full-duplex needs about 0.4 seconds, a quarter of the half-duplex time, because half-duplex does the two directions one after the other (twice the transmission time) and pays for 99 turnarounds. With many short messages, the turnaround dominates: at 20 ms per switch, half-duplex takes more than 50 times as long as full-duplex. That is why interactive protocols suffer on half-duplex media, and why designers of radios work hard to reduce switching time.

Serial and parallel transmission

A second choice is how many bits move at once.

  • Parallel transmission sends several bits at the same time, on separate wires: 8, 16 or 64 wires plus a clock. It looks fast, because a byte moves per clock tick, and it was the standard inside computers and for printers (the old parallel port). But the wires are never exactly equal: signals on different wires arrive at slightly different times (skew), and neighboring wires disturb each other (crosstalk). As clock rates increase, the skew eats the whole clock period, so parallel links are limited to short distances and low speeds.
  • Serial transmission sends the bits one after another on a single path. Every network link between machines is serial, for two reasons: it needs few wires (cheaper and easier over long distances), and one fast signal is easier to keep clean than a bundle of synchronized ones. Modern buses inside computers (PCI Express, SATA, USB) have also moved to fast serial links, sometimes with several serial lanes in parallel that are each self-clocked and can be skewed relative to each other.

The last lines of the second program below show the skew limit: with 2 ns of skew between wires, a parallel bus cannot clock faster than 500 MHz, however many wires it has.

Asynchronous and synchronous transmission

A serial receiver reads bits from a line, and must know where each unit of data begins, and when to sample each bit. There are two answers.

Asynchronous transmission sends each character as an independent frame. The line idles at 1. A start bit (0) announces a character and lets the receiver align its clock to the sender's, and then the data bits follow (least significant first), an optional parity bit for error detection, and one or two stop bits (1) that return the line to idle. The clocks of the two ends need to be only roughly equal, because the receiver resynchronizes at every start bit. This is the format of the classic UART serial port, used by microcontrollers, GPS modules and console connections. It is simple and cheap, but the overhead is large: for 8 data bits, a frame with 1 start bit and 1 stop bit has 10 bits, so 20% is overhead.

Synchronous transmission sends a whole block of data without start and stop bits for each character. A preamble or flag marks the beginning, and the receiver keeps its clock locked to the sender's, either by a separate clock wire or, more usually, by using an encoding with plenty of signal transitions, which the receiver uses to recover the clock. Ethernet, HDLC and SONET are synchronous. The overhead is small (a few bytes per block), so the method suits fast links and large frames.

The framing program builds a UART frame with a parity bit, decodes it, shows a one-bit error caught by the parity check, and compares the overhead of asynchronous and synchronous framing.

Python UART frames and framing overhead

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 bit string 01000001001 for the letter A is the start bit, the eight data bits 10000010 (least significant first), the parity bit and the stop bit. The efficiency table shows why every extra bit counts: a frame with a parity bit and two stop bits wastes one third of the line. For a 1,000-byte block, asynchronous transmission needs 10,000 bits against about 8,000 for synchronous transmission.

Line configuration: point-to-point and multipoint

Direction (simplex, half-duplex, full-duplex) says which way bits may flow. Line configuration is a different question: how many devices share one physical link? There are two answers.

A point-to-point line is a dedicated link between exactly two devices. The whole capacity of the line belongs to those two ends. A leased line between two offices, a cable between a router and a switch, and the link from a TV remote to a TV (using infrared) are point-to-point. The advantages are simple: the full bandwidth is guaranteed, there is no contention, a fault is easy to locate, and privacy is good because nobody else shares the wire. The drawbacks are cost and scale. Connecting every pair among n devices needs n(n-1)/2 dedicated links, installation takes time, and a broken cable leaves the two ends cut off.

A multipoint (multidrop) line connects three or more devices to one shared link, such as a bus with many taps or a radio channel with many receivers. It is cheaper and easier to extend, and one wire can serve many users. The capacity, however, is shared, so the devices need rules for taking turns, and a fault on the shared medium can affect everyone. The shared capacity can be divided in two ways:

  • Spatial sharing: several devices can use the link at the same moment (for example different frequencies or different antenna directions).
  • Temporal (time) sharing: the devices take turns, each using the full link for a short slot. The rules for turns are covered in medium access control.

Multipoint wiring suits a primary-secondary arrangement (one controller polls many terminals), while point-to-point links are the building block of switched networks, where each device gets its own cable to a switch. That is why modern Ethernet is built from point-to-point links even though the old coaxial Ethernet was multipoint.

Python How the number of links grows

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 table shows why a full mesh of dedicated lines stops being practical long before 100 devices, while a bus or a star grows by one connection per device. It also shows the price of a shared multipoint line: each extra device gets a smaller share.

Where each choice shows up

Technology Direction Serial or parallel Synchronization
Keyboard cable (old) Simplex-like, device to host Serial Asynchronous
UART serial console Full-duplex (separate TX and RX) Serial Asynchronous
Old parallel printer port Mostly simplex Parallel Strobe signal
10BASE2 shared coaxial Ethernet Half-duplex Serial Synchronous
Switched Gigabit Ethernet Full-duplex Serial (4 pairs) Synchronous
Wi-Fi Half-duplex per channel Serial Synchronous
TV and radio broadcast Simplex Serial Synchronous

Check your understanding

Quick check ยท 5 questions

of 5 answered You scored / 5 ยท perfect!

  1. 1. A walkie-talkie lets both people speak, but only one at a time. Which mode is it?

    Half-duplex links carry data in both directions, but alternate, which needs a turnaround between the directions.
  2. 2. Why do switched Ethernet links not need CSMA/CD?

    Full-duplex uses a separate path for each direction, so the two devices never compete for the same wire.
  3. 3. Why have fast links moved from parallel to serial transmission?

    Signals on parallel wires arrive at slightly different times and interfere with each other. A single fast serial signal avoids both problems.
  4. 4. An 8N1 UART frame carries 8 data bits. How many bits are on the wire per character, and what is the efficiency?

    One start bit, 8 data bits, no parity and one stop bit make 10 bits, so 8/10 = 80% of the bits are data.
  5. 5. What does the start bit give an asynchronous receiver?

    The falling edge from idle (1) to 0 tells the receiver exactly when the character begins, so the clocks only need to agree for about ten bit times.

Summary

  • Simplex links carry data one way only, half-duplex links carry both ways but alternate (with a turnaround cost), and full-duplex links carry both ways at once on separate channels.
  • Switched Ethernet is full-duplex, which removed collisions. Wi-Fi and old shared Ethernet are half-duplex.
  • Parallel links move several bits per clock on several wires but are limited by skew and crosstalk. Serial links use one path per lane and dominate modern networking and buses.
  • Asynchronous transmission frames every character with start, parity and stop bits and tolerates loose clocks, at a cost of 20% or more of overhead. Synchronous transmission sends whole blocks and recovers the clock from the signal, with very little overhead.

Next, see how a message actually crosses a network in How Data Travels: Packets and Switching.