Content

Model: source, transmitter (encoder, modulator), channel, receiver (demodulator, decoder), destination, with noise added in the channel. Analogue signals vary continuously; digital signals use discrete symbols. Digital transmission resists noise (regeneration), allows error correction and encryption, and is the standard today. Bit rate is the number of bits per second; baud rate (symbol rate) is the number of symbols per second: bit rate = baud rate x log2(number of symbol levels).

The electromagnetic spectrum is divided into bands (radio, microwave, infrared, visible); regulators assign frequency bands to services, and the ISM bands are licence-free. Bandwidth B is the range of frequencies used. Nyquist: a noiseless channel of bandwidth B carries at most 2 B symbols per second. Shannon: the capacity of a noisy channel is C = B log2(1 + SNR) bits per second, where SNR is the signal-to-noise power ratio. Decibels: SNR_dB = 10 log10 (P_signal/P_noise); attenuation is also expressed in dB (dB loss adds along a link). Filters (low-pass, high-pass, band-pass) limit the bandwidth and remove noise.

Analogue-to-digital conversion: sampling at more than twice the highest frequency (Nyquist), quantisation into 2^n levels and coding. Multiplexing shares a channel: frequency division (FDMA, each user has a band), time division (TDMA, each user has time slots) and code division (CDMA, users share time and frequency with different codes).

1. Relate bit rate to baud rate. 2. State Shannon's capacity formula. 3. Convert an SNR of 100 to decibels. 4. Compare FDMA and TDMA.

1. Bit rate = baud rate x log2(number of levels). 2. C = B log2(1 + SNR). 3. 20 dB. 4. FDMA assigns each user a separate frequency band; TDMA assigns time slots on the same band.

Layers: perception (sensors, actuators), network (connectivity), processing (edge and cloud) and application. Healthcare uses: remote patient monitoring, wearable devices, smart hospital equipment tracking and chronic disease management. Cloud service models: IaaS (infrastructure), PaaS (platform) and SaaS (software). 5G offers high bandwidth, low latency and many connected devices, enabling telesurgery and real-time monitoring. Challenges: security, privacy, interoperability and battery life.

A biosensor converts a biological quantity into an electrical signal (for example glucose, ECG). Signal conditioning: amplification, filtering and analogue-to-digital conversion. An embedded system is a dedicated microcontroller-based system with sensors, actuators and a radio. Design constraints: low power (energy harvesting or small batteries), small size and reliability. Medical embedded artificial intelligence runs simple models on the device to detect events such as arrhythmias.

1. List the layers of an IoT architecture. 2. Difference between IaaS, PaaS and SaaS? 3. What is signal conditioning? 4. Name two challenges of IoT in healthcare.

1. Perception, network, processing and application. 2. IaaS provides infrastructure, PaaS a development platform, SaaS complete applications. 3. Amplifying, filtering and digitising the sensor signal. 4. Security and privacy, interoperability, power consumption.

Layers: 1 physical (bits on the medium), 2 data link (frames, error detection, medium access), 3 network (addressing and routing), 4 transport (end-to-end delivery), 5 session, 6 presentation, 7 application. The TCP/IP model has four layers (link, internet, transport, application). Each layer adds a header to the data (encapsulation). Topologies: bus, star, ring, tree, mesh. Media: twisted pair, coaxial cable, optical fibre (highest bandwidth, immune to electromagnetic interference) and wireless.

Analogue modulation: AM (amplitude), FM (frequency), PM (phase). Digital modulation: ASK, FSK, PSK and QAM (both amplitude and phase; more bits per symbol but needs higher SNR). Spread spectrum spreads the signal over a wider band for resistance to interference. ISM bands (for example 2.4 GHz) host Wi-Fi, Bluetooth and medical telemetry. Propagation loss in free space follows the Friis equation and increases with the square of distance and frequency.

IEEE 802.11 (Wi-Fi) uses OFDM (orthogonal frequency division multiplexing: many narrowband subcarriers, robust against multipath). The data link layer frames data, detects errors (parity, CRC cyclic redundancy check) and controls medium access (MAC; Ethernet uses CSMA/CD, Wi-Fi uses CSMA/CA). IEEE 802.15.4 is the low-power standard for sensor networks. Network layer: IP addresses (IPv4 32 bits; IPv6 128 bits), routing and congestion control.

TCP: connection-oriented, reliable (acknowledgements, retransmission, flow and congestion control). UDP: connectionless, low overhead, no guarantee, for real-time data. IoT protocols: MQTT (publish-subscribe through a broker, lightweight, over TCP) and CoAP (a lightweight request-response protocol over UDP for constrained devices). HL7 and the Clinical Document Architecture (CDA) define the structure of clinical messages and documents so that different health information systems can exchange data.

1. Name the seven OSI layers. 2. What is the purpose of CRC? 3. Compare TCP and UDP. 4. What is MQTT? 5. Why is OFDM used in Wi-Fi?

1. Physical, data link, network, transport, session, presentation, application. 2. Detecting transmission errors. 3. TCP is reliable and connection-oriented; UDP is connectionless and faster but unreliable. 4. A lightweight publish-subscribe messaging protocol used in IoT. 5. It is robust to multipath fading and uses spectrum efficiently.

Embedded (edge) AI runs trained machine-learning models directly on wearables and medical devices (for example arrhythmia detection or fall detection). Advantages: low latency, works without a connection, less data transmitted and better privacy. Constraints: limited memory, processing power and battery, so models are compressed (quantisation, pruning) and kept small. Validation, safety and regulation are essential in medical applications.

1. Give two advantages of edge AI. 2. Name a technique to shrink a model for a small device.

1. Low latency and better privacy (also lower bandwidth and offline operation). 2. Quantisation or pruning.