IoT Mock Exam 1 — Short Form

Suggested conditions: closed book, 60–75 minutes. Answer in complete sentences. A correct technology name without a constraint-based justification earns only partial credit. Do not consult the model answers until all questions are complete.

This mock uses different scenarios, numbers, and topic order from the 2021 exam while staying within the supplied lecture extracts and question sheets.

Task 1 — Architecture and NB-IoT

(a) Architecture and gateway selection

A museum has battery-powered environmental sensors using a short-range non-IP radio protocol. The sensors must continue to control a local humidity alarm when the Internet connection is unavailable. A cloud service is used for long-term storage and reporting.

1. Should the system primarily use a cloud-, edge-, or mesh-based architecture?
2. Is a gateway appropriate? Explain its protocol, processing, buffering, and failure-handling roles.
3. State one disadvantage of your choice.

(b) NB-IoT power and reachability

A utility company installs water meters in widely separated basement rooms across a city. Each meter sends one reading per day but may occasionally receive a configuration command.

1. Choose NB-IoT or 802.15.4 and justify the choice.
2. Explain whether eDRX or PSM is more appropriate if the configuration command must eventually be deliverable without the meter initiating an uplink first.
3. State the energy-versus-latency trade-off.

Task 2 — IEEE 802.15.4 and 6LoWPAN

(a) MAC selection

A warehouse contains many battery-powered asset tags. They report at regular intervals, require coordinated sleep periods, and operate in an environment with intermittent narrow-band interference. Some tags are several radio hops from the gateway.

1. Select unslotted CSMA/CA, slotted CSMA/CA with GTS, or TSCH. Justify your choice.
2. Explain which problem CSMA/CA solves and which problem frequency hopping solves.
3. Which function must be provided by a higher-layer protocol for the multi-hop part?

(b) 6LoWPAN fragmentation

A 730-byte IPv6 datagram is sent over 802.15.4 using 6LoWPAN fragmentation. Assume that every fragment except the last has 64 bytes of payload.

1. What is the datagram offset of the sixth fragment? Show the calculation.
2. Name the information used to identify and position the fragments.
3. Explain why this 6LoWPAN fragmentation does not replace IPv6 fragmentation.

Task 3 — MQTT

(a) Topic alias

A publisher repeatedly sends measurements to the long topic:

factory/building-2/line-4/motor-17/vibration

The first MQTT 5 PUBLISH contains the concrete topic and Topic Alias = 4. A later PUBLISH contains an empty topic name, Topic Alias = 4, and payload {"rms":0.83}.

1. What mapping must the broker maintain?
2. Why is the mapping associated with the publisher’s connection rather than being global?
3. How does the broker route the later alias-only message?

(b) Shared subscriptions and normal subscriptions

Three processing clients subscribe to:

$share/anomaly-detectors/factory/+/motor/+/vibration

A dashboard subscribes normally to:

factory/#

A publisher sends one message to:

factory/building-2/motor/motor-17/vibration

1. Will all three processing clients receive the message? Explain.
2. Will the dashboard receive it? Explain.
3. To which topic should the publisher send a message intended for this shared group?

Task 4 — Connected data

(a) Key-value versus linked data

A smart-campus system stores sensors, rooms, buildings, capabilities, maintenance events, and responsible organizations.

1. Give one reason a key-value model is preferable for storing the latest reading of a known sensor.
2. Give one query for which a linked/graph model is preferable.
3. Explain why the graph model represents that query more naturally.

(b) Continuous query and CEP

A stream contains timestamped room-temperature readings. Write pseudo-code for a continuous query that reports the average temperature for each room over a sliding 10-minute window, updated every 2 minutes.

Then describe an ECA/CEP rule that creates a PossibleFire event when a rapid temperature rise and a smoke event occur in the same room within 30 seconds. Identify the atomic events and the complex event.

(c) Update protocol selection

A delivery robot normally moves at a predictable constant velocity, but sometimes encounters slippery ground and wheel slip.

1. Which update protocol is energy-efficient during predictable motion?
2. Why can it become inaccurate on slippery ground?
3. What is the difference between reducing the number of updates and compressing the bytes in each update?

Task 5 — Embedded Machine Learning

(a) Placement and acceleration

A wearable ECG device must detect a dangerous rhythm within milliseconds. The device has a low-power MCU and a small FPGA accelerator. A cloud service is available for long-term analysis and model retraining.

1. Which part of the ML pipeline should run locally and which part can run in the cloud? Justify the placement.
2. State two benefits and two challenges of local inference.
3. Why might an FPGA or other hardware accelerator be useful?

(b) Model optimization

A trained dense neural network is too large and slow for an embedded device.

1. Explain magnitude-based pruning and why removing the smallest-magnitude weights can unexpectedly increase the loss.
2. How does a saliency-based approach such as Optimal Brain Damage improve the pruning decision?
3. Compare quantization with binarization, including their effect on computation and hardware implementation.

Self-check after finishing

Before opening model answers, mark each answer for whether it contains:

• the correct named concept;
• the relevant constraint or mechanism;
• a comparison with the rejected alternative;
• one limitation or trade-off;
• a calculation with units where applicable.

---

Model answers intentionally omitted from this first simulation attempt.