1. Internet of Things
Amr El Mougy
Gina Maher

2. Data Collection in Sensor Networks

3. Data Collection from BLE Sensors
BLE sensors are standalone. Data collection has to be done from each sensor individually
Mainly two methods for BLE data collection:
Smartphones (public sensing)
Gateway
Communication is typically one way
Configuration parameters are allowed in the reverse direction

4. Comparison of Gateways and Smartphones
Fixed data collection point  reliable
Reliability depends on the smartphone density
Medium cost
Low cost
Requires an additional device to configure the sensors
Can send configuration commands from the phone
Depending on the type, it can support WiFi, cellular connection, or both
Always supports WiFi and cellular connections
Always participates in the data collection
Participation depends on the user’s willingness
Dedicated for data collection
Data collection competes with other applications
Typically has stable power source
Battery-powered

5. Structure of BLE Gateway
Processor/Application Host
API
BLE
Stack Profiles
TCP/IP
BLE
Physical/Link Layers
Ethernet Stack
Cellular Stack
WiFi Stack

6. Gateway Types Payload Extraction Packet Encapsulation (Data Pipe)
BLE Attribute
WiFi Packet
Handle
UUID
Value
WiFi - H
IP-H
TCP-H
Value
BLE Attribute
Handle
UUID
Value
WiFi Packet
WiFi - H
IP-H
TCP-H
Handle
UUID
Value

7. Data Collection from ZigBee Sensors
Smartphones generally do not support ZigBee
All ZigBee networks have a PAN coordinator where all data is collected
From there the data can uploaded to the Internet
Thus, the coordinator can act as a gateway

8. Why is Data Collection Challenging?
Reliability, automation and fault-tolerance
WSNs are required to remain operational for long durations without human intervention
All nodes have dual function: sensing and relaying
Transmissions typically consume the most energy
Nodes go to low power (sleep) state to conserve energy
This is called duty cycling. Low duty cycles mean nodes remain in sleep state for longer durations to conserve energy
Generated data needs a path to the PAN coordinator
Sleeping nodes will not participate in forwarding  need to plan sleep schedules carefully
Sensor data is many-to-one. Control data is one-to-many
Control traffic is minor but often more critical. Reliability is often imposed
Energy efficiency is at the core of all challenges!!

9. The Three Stages of Data Collection
Node Deployment
Control Message Disseminat-ion
Data Delivery
Area Coverage
Location Coverage
Flooding-Based
Gossiping-Based
QoS Metrics
(Latency, reliability, energy)

10. The Node Deployment Problem
Problem Statement: What is the optimum placement of sensor nodes in order to satisfy the requirements of a certain application?
This placement needs to ensure that all required attributes are sensed and all nodes have a path to the PAN coordinator (the coverage and connectivity problems)
Sensors are modeled as a circular disk with sensing range Rs metres and communication range Rc metres Rc
Rs is determined by the sensing hardware while Rc is determined by transceiver Rs

11. The Node Deployment Problem
Nodes go into sleep mode periodically and their batteries die
Conclusion  not all deployed nodes may be active at all time
Thus, redundancy is required in the network (deploy more nodes than needed)
The problem now is known as the k-coverage and k-connectivity problems:
every point must be covered by at least k sensors
every sensor must have a path to the gateway even if k – 1 paths fail

12. 1-coverage, 1-connectivity
Examples
1-coverage, 1-connectivity
Strip pattern used as building block for 1-coverage and 1-connectivity network
** F. Wang and J. Liu, “Networked Wireless Sensor Data Collection: Issues, Challenges, and Approaches,” IEEE Communication Surveys and Tutorials, Vol. 13, No. 4, 2011.

13. Node Deployment for Area Coverage
Goal is to cover every point in the area with at least k sensors
Research has discovered the following pattern to be the universal element pattern for 1-coverage and k-connectivity (k ≤ 6)
Where d1 and d2 are parameters that depend on Rc and Rs
** F. Wang and J. Liu, “Networked Wireless Sensor Data Collection: Issues, Challenges, and Approaches,” IEEE Communication Surveys and Tutorials, Vol. 13, No. 4, 2011.

14. Examples
** F. Wang and J. Liu, “Networked Wireless Sensor Data Collection: Issues, Challenges, and Approaches,” IEEE Communication Surveys and Tutorials, Vol. 13, No. 4, 2011.

15. ** F. Wang and J. Liu, “Networked Wireless Sensor Data Collection: Issues, Challenges, and Approaches,” IEEE Communication Surveys and Tutorials, Vol. 13, No. 4, 2011.

16. Node Deployment for Location Coverage
Goal is to cover specific points in an area with at least k sensors
Coverage is straightforward and is typically manual
These locations may be sparse, thus requiring relay nodes
The challenge is to deploy relay nodes to ensure connectivity
More that one relay node may be needed to ensure k connectivity
** F. Wang and J. Liu, “Networked Wireless Sensor Data Collection: Issues, Challenges, and Approaches,” IEEE Communication Surveys and Tutorials, Vol. 13, No. 4, 2011.

17. Node Deployment for Location Coverage
Start with the deployed sensors
Add a relay node with three edges at the intersection of the medians
Check the lengths of each edge. If greater than communication range, add a relay

18. Traffic-Aware Node Deployment
The preceding algorithm does not consider traffic
Nodes close to the base station relay more traffic
Needs to consider the nature of traffic so as not to over-consume the batteries of certain relay nodes
Example: how to optimally distribute N relay nodes if you have 2 sources S1 producing 60% of the data and S2 producing 40% of the data: S1 S2 S1 S2
1 3 N
1 3 N
1 3 N
1 3 N - Δ V V
1 3 N + Δ
1 3 N S0 S0

19. Data Delivery
After the network is deployed, sensor data has to be relayed to the base station
The network is usually viewed as a tree, rooted at the base station
The tree is either manually defined or autonomously discovered
If autonomously discovered, the process is either central (by the base station), or distributed (every sensor node discovers the topology on its own)
Nodes also go to sleep cycles and eventually die out
Data may be lost due to interference or collisions
** F. Wang and J. Liu, “Networked Wireless Sensor Data Collection: Issues, Challenges, and Approaches,” IEEE Communication Surveys and Tutorials, Vol. 13, No. 4, 2011.

20. Data Delivery Models
Event-driven: data is generated in response to an event. Data from several sensors may be highly correlated. Fusion techniques often employed
Query-driven: network is interactive. Only sends data on demand
Continuous-based: real-time data. Network is always sending data
Time-driven: data is collected periodically from the environment
Transmitted data may be loss-tolerant or not
Internet routing protocols are not suitable for sensor networks since they do not consider energy efficiency

21. Data Delivery Functions
Two main modules:
Topology maintenance: construct topology that guarantees coverage and connectivity while considering network dynamics
Transmission scheduler: determines when packet transmissions should take place to reduce collisions, ensure energy efficiency and consider QoS constraints
** F. Wang and J. Liu, “Networked Wireless Sensor Data Collection: Issues, Challenges, and Approaches,” IEEE Communication Surveys and Tutorials, Vol. 13, No. 4, 2011.

22. Topology Maintenance Two main parameters to control:
Transmit power: to maintain a balance between connectivity and energy efficiency (out of our scope)
Duty cycles: to put the sensors in sleep mode
Challenges in duty cycle management include:
End-to-end delay: nodes waiting for other nodes to wake up
Collision rates: shortening active cycles means more nodes trying to transmit at the same time
Control overhead: mainly for synchronization
** R. Carrano, D. Passos, L. Magalhaes, and C. Albuquerque, “Survey and Taxonomy of Duty Cycling Mechanisms in Wireless Sensor Networks,” IEEE Communication Surveys and Tutorials, Vol. 16, No. 1, 2014.

23. Synchronous Duty Cycling
All nodes keep common time reference. Not necessarily a global time
Synchronization information is exchanged throughout the network to keep a certain degree of alignment
Overhead increases significantly with the number of nodes
Rendezvous: strict synchronization. Challenging to maintain for large networks. Suffers from data forwarding interruption (nodes go to sleep, unaware that a frame is coming
Skewered: deals with data forwarding interruption. Typically a tree is formed and a schedule is set. Needs topology discovery and maintenance
** R. Carrano, D. Passos, L. Magalhaes, and C. Albuquerque, “Survey and Taxonomy of Duty Cycling Mechanisms in Wireless Sensor Networks,” IEEE Communication Surveys and Tutorials, Vol. 16, No. 1, 2014.

24. Semi-Synchronous Duty Cycling
Nodes are clustered and the clusters are synchronized
Inter cluster communication is asynchronous
Main challenge is choosing the cluster head
Two ways of achieving this:
Spontaneous: loose association. Only timestamps are exchanged. Not efficient, nodes may belong to many clusters
Election-based: nodes communicate to elect a manager periodically (typically node with highest remaining power). Extra overhead needed
** R. Carrano, D. Passos, L. Magalhaes, and C. Albuquerque, “Survey and Taxonomy of Duty Cycling Mechanisms in Wireless Sensor Networks,” IEEE Communication Surveys and Tutorials, Vol. 16, No. 1, 2014.

25. Asynchronous Duty Cycling
Preamble: sending nodes transmit a preamble frame that is longer than any active/sleep cycle. Nodes wakeup periodically to check for this frame. If not found they sleep right away
Receiver-initiated: nodes periodically wake up and send a beacon, indicating willingness to receive. Senders send an ACK if they have a packet to send
On demand: relies on the availability of a separate radio that can trigger wake up calls. The second radio should consume less than a tenth of the energy of the primary radio
** R. Carrano, D. Passos, L. Magalhaes, and C. Albuquerque, “Survey and Taxonomy of Duty Cycling Mechanisms in Wireless Sensor Networks,” IEEE Communication Surveys and Tutorials, Vol. 16, No. 1, 2014.

26. Asynchronous Duty Cycling
Random: works in highly dense networks. Nodes go to sleep for a period that is proportional to the number of neighbors
Schedule-based: sender and receiver schedule wake up slots such that they overlap in at least one slot. Duty cycles here are typically high
** R. Carrano, D. Passos, L. Magalhaes, and C. Albuquerque, “Survey and Taxonomy of Duty Cycling Mechanisms in Wireless Sensor Networks,” IEEE Communication Surveys and Tutorials, Vol. 16, No. 1, 2014.

27. Reliability In some applications, reliable packet delivery is required
Traditionally, this is done using sequence numbers and receiver feedback
Hop-by-hop recovery notices missing sequence numbers and sends feedback asking for retransmissions
High packet loss can also occur due to changing network dynamics. In this case end-to-end recovery is also needed at the base station
The base station notices many sequence numbers missing and may instruct the topology maintenance module to react
** F. Wang and J. Liu, “Networked Wireless Sensor Data Collection: Issues, Challenges, and Approaches,” IEEE Communication Surveys and Tutorials, Vol. 13, No. 4, 2011.

28. Latency
A node’s transceiver consumes a decreasing amount of energy for the following states: transmission, reception or idle, and sleep
Thus a node sending to its neighbor may waste time and energy waiting for it to wake up
If sleep cycles are not considered, high packet loss may be incurred
Also need to maximize sleep cycles
Over a path, it is optimal for nodes to wake up only when they need to send/receive, and then go right back to sleep
** F. Wang and J. Liu, “Networked Wireless Sensor Data Collection: Issues, Challenges, and Approaches,” IEEE Communication Surveys and Tutorials, Vol. 13, No. 4, 2011.

29. DMAC Sensors on a path will wake up, receive, send, then sleep
A sensor sets More Data Flag to 1 if it wishes its receives to turn back to receiving mode after transmitting, instead of going back to sleep
A receiver connected to two senders will predict that there are more packets to be received, and will return to receiving after transmission
Nodes use slotted CSMA. After each slot, senders can set the More to send flag to 1 to indicate that there is a retransmission
** F. Wang and J. Liu, “Networked Wireless Sensor Data Collection: Issues, Challenges, and Approaches,” IEEE Communication Surveys and Tutorials, Vol. 13, No. 4, 2011.

30. Throughput
Nodes near the base station are the bottleneck of the network in terms of throughput and energy consumption
Some authors suggest dynamic placement of nodes. Thus, k coverage and connectivity are not uniform throughout the network
Funneling-MAC divides the network into three regions: high intensity, low intensity, and medium intensity
TDMA and more redundancy are used in the high intensity region
CSMA is used as we move away from the sink
** F. Wang and J. Liu, “Networked Wireless Sensor Data Collection: Issues, Challenges, and Approaches,” IEEE Communication Surveys and Tutorials, Vol. 13, No. 4, 2011.

31. Control Message Dissemination
Control messages come from the sink. Thus, it is a one-to-many communication
Even though these are relatively a small number of packets, they are critical to network performance
Two main approaches:
Flooding: each node forwards any control packet it receives until reaches the maximum hop count. Many duplicates and redundancies occur. High energy consumption
Gossiping: every node forwards the message but only according to some predefined probability. If properly defined, there is a high chance that the whole network is covered. Packet loss may lead to some nodes missing important messages