Bandwidth, Signal Encoding & Network Interface Considerations
How fast can this link go? What does “1 Gbps” actually measure? And how do voltages on a wire carry ones and zeros without the receiver losing the beat? This lesson gives the exact vocabulary — bandwidth, throughput, latency, goodput, encoding, baud, duplex — and shows how a NIC turns frames into signals.
Bandwidth, throughput, latency, goodput
These four words are used loosely in daily speech but mean precise things in networking.
| Term | Meaning | Units |
|---|---|---|
| Bandwidth | Maximum capacity of the link. | bits per second (bps) |
| Throughput | The rate actually achieved in practice. | bits per second (bps) |
| Latency | Delay for a bit to reach the destination (propagation, processing, queuing). | milliseconds (ms) |
| Goodput | Useful application-payload rate after headers and retransmits. | bits per second (bps) |
Throughput is almost always lower than bandwidth. A 1 Gbps link might deliver 940 Mbps of throughput because the Ethernet preamble, inter-frame gap and headers consume capacity; a busy network can drop lower still due to congestion and retransmissions. Latency is not a rate at all — it is a time. A link can have high bandwidth and still feel slow if its latency is high (a classic problem on satellite links).
Units: bps and the b-versus-B trap
Network speed is measured in bits per second (bps), and the prefixes step by factors of 1000.
| Unit | Represents | Example |
|---|---|---|
| b/s (bps) | bits per second | a dial-up modem: 56 kbps |
| kbps | 1 000 bps | DSL upload |
| Mbps | 1 000 000 bps | Fast Ethernet: 100 Mbps |
| Gbps | 1 000 000 000 bps | Gigabit Ethernet: 1 Gbps |
The single most common mistake is confusing b (bit) with B (byte). There are 8 bits in one byte, so a 100 Mbps link transfers only about 12.5 MB/s (100 ÷ 8). A storage spec quoted in MB (megabytes) uses a different unit and a different meaning from a network spec quoted in Mb (megabits).
Mb = megabit (network speed, decimal ×1000). MB = megabyte (file size, often binary ×1024). When you estimate transfer time, convert to the same unit first: time = size ÷ rate, then fix the units.Signalling: bits into signals
The physical layer carries bits by varying a physical property of the medium. On copper it varies the voltage; on fibre the intensity or wavelength of light; in wireless the power, frequency or phase of radio waves. The sender and receiver must agree on which variation means one and which means zero — that agreement is the signalling method.
Some media encode the signal directly onto the medium (baseband), as Ethernet does. Others modulate a higher-frequency carrier wave, which is how a single phone line carries both voice and ADSL data at once.
Encoding and the NRZ example
Encoding is the rule that maps a group of bits to a pattern of signal elements. The simplest rule is NRZ (non-return to zero): transmit the high level for a 1 and the low level for a 0, changing only when the bit changes. Below, the bits 1 0 1 1 0 are drawn both ways — switch schemes to compare them.
As Lesson 1 introduced, any encoding must define four things, and NRZ illustrates all of them:
- Voltage level / change. NRZ uses a level (high or low), not a transition.
- Bit timing. Each level is held for exactly one bit interval; the receiver samples mid-interval.
- Mode. NRZ here is a baseband, two-level scheme on the wire.
- Purpose. NRZ is simple and efficient, but it does not guarantee transitions, so it needs help (block coding such as 4B/5B) to keep the clocks synchronised.
Baud versus bits per second
A symbol is one signal element. The number of symbols sent per second is the baud rate (also called the signalling rate). The number of bits sent per second is the bit rate. When each symbol carries exactly one bit the two are equal — but with multi-level encoding a symbol can carry several bits at once.
| Levels (symbols) | Bits per symbol | If baud = 1 000 |
|---|---|---|
| 2 (binary) | 1 | 1 000 bps |
| 4 | 2 | 2 000 bps |
| 8 | 3 | 3 000 bps |
| 16 | 4 | 4 000 bps |
Worked example: Gigabit Ethernet's 1000BASE-T uses PAM-5 (five voltage levels) across four pairs. It runs at 125 MBd per pair and encodes 2 bits per symbol per pair, giving 250 Mbps per pair. Multiply by four pairs and you reach 1 Gbps. This is why bit rate and baud rate are not the same number.
Half duplex versus full duplex
Duplex describes whether a device can send and receive at the same time on the medium.
| Mode | Behaviour | Consequence |
|---|---|---|
| Simplex | One direction only. | e.g. a legacy broadcast link |
| Half duplex | Both directions, but not at once. | Collisions possible; needs CSMA/CD |
| Full duplex | Both directions at the same time. | No collisions; doubles usable capacity |
Modern switched Ethernet is full duplex: each device has its own transmit and receive path (separate wire pairs in twisted-pair cable), so collisions cannot occur and CSMA/CD is disabled. A duplex mismatch — one end half, the other full — is a classic cause of a slow, erroring link.
The NIC and its interfaces
The network interface card (NIC) is the hardware that connects a host to the network. It implements two layers: the physical layer (turning bits into signals, handling speed and duplex) and the data link layer (building Ethernet frames and holding the device's MAC address, burned into the card).
- Wired NICs present an RJ-45 (8P8C) port for twisted-pair copper, or an SFP/LC cage for fibre.
- Wireless NICs use a radio and antenna, operating on Wi-Fi (IEEE 802.11) frequencies.
A NIC has a rated interface speed (for example 10/100/1000 Mbps). The interface the host actually uses is determined by the port on the switch and the cable between them.
Speed autonegotiation
Autonegotiation lets two connected Ethernet devices advertise the speeds and duplex modes they support, then automatically pick the best common setting. A port that supports 10/100/1000 Mbps and negotiates with a 100 Mbps partner settles on 100 Mbps full duplex.
The rule to remember: the link always runs at the highest speed and duplex both ends support. If negotiation fails, some devices fall back to speed sensing, which can leave a duplex mismatch — a subtle fault that slows the link and inflates errors without taking it down.
Review
Check what you have learned with these review questions.
Sources & references
- IEEE. IEEE 802.3 — defines Ethernet PHYs, encoding (PAM-5 for 1000BASE-T) and full-duplex operation. standards.ieee.org/ieee/802.3.
- TIA. ANSI/TIA-568.2-D — cabling categories, bandwidth (MHz) and the 100 m channel. tiaonline.org.
- Cisco. CCNA (200-301) Exam Topics, Domain 1.6 (physical interfaces and cabling). cisco.com.
- Peterson, L. & Davie, B. Computer Networks: A Systems Approach, Section 2.2 (Encoding). book.systemsapproach.org.
Target exams: CCNA 200-301 v1.1 · CyberOps Associate 200-201. Last content verification: 2026-10-11.