Sensor Networks

Sensor networks will play an important role in many IoT solutions. As shown in the figure below, sensor networks usually rely on intelligent gateways (1, 4) to ensure backend communication (7) and to integrate local networks of sensors or sensor nodes. Multiple sensors can either be integrated into a multi-sensor device or physically distributed as sensor nodes that use local wireless communication to send data back to a local gateway for further processing. A multi-sensor device would typically use a specialized board for sensor integration (2), such as Arduino, FPGAs, etc. Multiple sensors can be directly connected through one or multiple local bus systems (3). The wireless sensor network connects multiple sensor nodes (6) via local wireless communication (5).

AIA for sensor networks

AIA for sensor networks

So as we can see, most sensor networks are built around a local component that combines the features of an IoT gateway (as described above) with local business logic. The figure below provides an overview of the key components of this type of intelligent sensor gateway.

Key features of intelligent sensor gateway

Key features of an intelligent sensor gateway (Source: Bosch Connected Devices and Solutions)

The selection of technologies depends primarily on the concrete requirements, some of which might even contradict each other. For example, the requirements for system accuracy might lead to the selection of an expensive sensor element, but the budget may not allow for this. In this case, either the best possible compromise must be found or the requirements must be adjusted accordingly. Depending on the task at hand, a one-off investment (especially in software development) could help to reduce the production cost per unit, for instance by working around the limitations of sensor elements chosen on cost grounds by means of data fusion and very good algorithms.

Naturally, sensor connectivity plays an important role. We have already discussed and compared the general details of close-range wireless technologies in the section on IoT gateways above. The following table provides an overview of the connectivity technologies that are typically used for wireless sensor networks, grouped by area.

Sensor connectivity

Sensor connectivity (Source: Bosch Connected Devices and Solutions)

Sensors and Sensor Categories

Most sensors comprise one or more sensor elements (depending on the metrics required). For large production volumes (>10,000), ASICs (application-specific integrated circuits) are used. Both parts are housed in a package with electrical and electronic interfaces to the outside. This package is used for mounting (usually by means of soldering to a circuit board).

In large-volume production environments like the automotive industry, the ASIC controls the sensor element in accordance with the configuration provided by the surrounding system (the sensor node). It also preprocesses the data from the sensor element, for example by linearizing measured values, correcting deviations, and so on. The sensor usually communicates outwards (with the surrounding system) using digital interfaces such as I2C or SPI. The ASIC often supports different operating modes for different applications. For example, modern acceleration sensors support modes in which the sensor runs using very little power and waits for a movement. If motion is detected (because acceleration occurs), the sensor sends a message to inform the surrounding system. As a result, the surrounding system can remain in sleep mode for as long as possible, and only becomes active when an interesting event occurs.

Modern sensors contain an increasing number of sensor elements. Sensors that can measure rotation, acceleration, and magnetic fields on all three spatial axes are standard today. Each axis and each type of measurement has a corresponding Degree of Freedom (DoF). Thus, a highly integrated sensor would be a 9DoF sensor.

Stefan Schuster, Sensor Network expert from Bosch Connected Devices and Solutions, describes the key sensor categories for us below.

Many IoT applications require motion and orientation sensors, in particular gyroscopes and acceleration sensors. Gyroscopes measure the rotation (gyration) of a system along the three spatial axes. Unlike highly specialized sensors, which only measure a single axis, sensors in consumer devices usually measure all three axes. Typical measurement ranges for sensors in consumer devices are up to several 100°/s. Smartphones represent one field of application for these sensors. The information obtained can be used to run games, for example. In the automotive industry, gyroscopes are used to identify the yaw of a vehicle and correct it if necessary (ESP). Information about the rotation is merged with information about the gravitational acceleration (cf. acceleration sensors) to determine the position of a device in space.

Acceleration sensors measure acceleration along the three spatial axes. This includes gravitational acceleration (1g in the direction of the center of the earth). Since this metric is also determined, acceleration sensors on three axes can also identify which way is “down”. This is used in smartphones and tablets, for example, to set the display orientation. Typical measurement ranges for this type of sensor in consumer devices are +/- 16 G on all three axes. One example of an industrial application is accident detection (shock sensors). Furthermore, acceleration sensors are often used for positioning applications. They are usually installed in consumer devices in the form of a MEMS element. Although vibrations can distort measurements, they can also be detected (within the limits of sensor capabilities).

Another important category of sensors relates to magnetic field sensors. These sensors measure the magnetic field in all three spatial axes, depending on the design. Typical measurement ranges go from a few uTesla up to 1000 uTesla. By way of comparison, consider the following scenario. The geomagnetic field in the Stuttgart region is around 50 uTesla. Based on magnetic field data, a compass can be implemented. To achieve greater precision, however, this data is combined with data about gravitational acceleration and the rotation of the device.

Possible applications for these sensors include compass function, the detection of magnetic fields that arise due to electrical current flow (to determine if a device is switched on or off, for example), or the registration of magnetic fingerprints to detect a change in direction or position.

The next category concerns environmental sensors. These sensors detect temperature, air humidity and pressure, illumination level, or noise. Gases or substances in the air can also be detected. To increase accuracy, the different sensor data is often merged and compared. For example, a precise temperature reading is a prerequisite for calculating precise relative air humidity.

These sensors are used in the area of smart homes, for instance, to determine the conditions in a home. Light sensors can also be used as proximity sensors. If a sensor is suddenly obscured, for example, this can indicate that a smartphone is being held up to an ear.

The microphone can register more than just noise: it can also be used as a simple sensor to determine the sound levels in a particular location. However, with increasing demands on the information provided by the metrics, increased processing power or corresponding functional units in the microprocessor will soon be required for frequency analysis.

Microphones and light sensors can still frequently be found as analog sensors, with no digital interfaces. The data must then be processed elsewhere accordingly.

Sensor Fusion

Modern sensor networks often rely on a technique known as sensor fusion, which fuses inputs from multiple sensors into a single digital model. One example of advanced sensor fusion discussed in this book is autonomous driving (see IoT Application Domains, “Connected Vehicle” section). In this use case, data from multiple sensors like cameras, radar, and LIDAR is combined to create a 3D model of a vehicle’s close surroundings, including objects like other vehicles, pedestrians, and cyclists.

Sensor Network Application Areas

Stefan Schuster, Sensor Network expert from Bosch Connected Devices and Solutions, has compiled the following short overview of the typical usage profiles of sensor networks in the areas of smart homes, smart wearables, transport and logistics, and Industry 4.0. These usage profiles can be used as a starting point for a development that is typical to a particular area. It is important to point out that these usage profiles are constantly evolving, because of new technologies or new requirements. In addition, the manufacturer’s platform strategy can often lead to sensor nodes being built from predefined components, although the requirements may allow for further optimization in individual cases. Reasons for this include economies of scale for purchasing, manufacture, and reuse of know-how or software parts, which help to reduce development time, risk, and cost.

Sensor Nodes in Smart Homes

The area of smart homes includes solutions that monitor the condition of doors, windows, or even lights and heating, and that control them accordingly. Information is also forwarded to the user, who can intervene as required by configuring the system or using direct control commands.

The requirements that sensor nodes must meet in this area can be roughly classified as follows:

  • In the consumer sector, which is our focus here, cost targets are very tight, and the number of units is medium to high. Product lifecycles are short, which means it must be possible to update the software. The sensor nodes must be integrated into the home as unobtrusively as possible, so they must often be small. Users generally expect them to be easy to install, so communication must be wireless. Customers do not want heavy maintenance requirements, so battery life must be long (usually two years). Batteries must also be readily available, so a standard battery type should be used. Depending on the manufacturer’s strategy, interoperability with systems from other manufacturers may be important, so standard connectivity may be required.
  • The system must work as promised; however, the imperative for functional reliability is not comparable to that of automotive components, for example. “Good enough” performance is often the goal.

Based on these requirements, the following profile can be defined as a starting point:

  • The data the sensor elements must record is determined by the intended use of the sensor node. It is often a good idea to provide additional sensors that may not be needed just yet, but that will allow for future functionality enhancements. One example would be to include acceleration sensors in a sensor node that measures temperature. In a future software enhancement, the customer may then implement a function to have the sensor react to shaking, for instance by taking a measurement outside of an otherwise active cycle. The main priority for the sensors is to maintain the required level of accuracy with minimal power consumption. Software can often be used to compensate for reductions in accuracy.
  • Minimal power consumption does not necessarily mean low consumption during active operation. Instead, this often involves supporting operation that allows the sensor to remain in sleep mode for as long as possible, only becoming fully active when changes occur to the physical metrics being registered (motion-activated wake-up mode in gyroscopes and acceleration sensors, for example).
  • The microprocessor must meet low to medium demands for processing power and memory. Typical processors are based on a Cortex M3 or M4 core. The M4 has a Digital Signal Processing (DSP) unit and is therefore the preferred choice for tasks that involve signal processing – for example, analyzing audio signals from a microphone. Typical memory sizes are 16-128 KB RAM and 64-512 KB flash memory. Energy efficiency is also important in this context. This is generally achieved not only through low power consumption during operation, but also by supporting modes in which only very small parts of the microprocessor are operational. These parts preprocess data and signals and then bring the entire microprocessor into active mode when required. Modern microprocessors with Cortex M3 have three, four, or more of these modes. The time during which the microprocessor must be fully active to complete its tasks should be kept as low as possible. The wake-up time of the microprocessor must also be considered. This should be as short as possible to ensure short cycle times.
  • The algorithm must of course meet the functional requirements, but must also be optimized for energy efficiency (to allow the electronic components to remain switched off for as long and as often as possible), enabling the microprocessor selected to be as small and as cost-efficient as possible.
  • Power supply to the system is determined on the basis of the target costs for manufacture and the required battery life. DC/DC transformers are typically used, which are very efficient. If the budget is very tight, linear regulators can be used. These are not as efficient, but are much cheaper – at the expense of the battery life of the system. Batteries are usually selected from the types readily available on the market (i.e. AA, AAA) and are limited by the space available.

The wireless technologies used in the smart home sector vary widely. Typical technologies are presented in the wireless connectivity table above. Data transfer rates are generally low. For energy efficiency reasons, as little data as possible is sent as seldom as possible. This has repercussions for the algorithm, which must evaluate the sensor data as far as possible in the device, and then send high-level information only. (Consider the following example: The algorithm receives data about the acceleration and rotation of a door at one-second intervals. However, it only transmits the event “door open”, which it extracts from the data. This event occurs much less frequently and can even be transmitted in a single bit.)

Wi-Fi is not really feasible for battery operation if a battery life of two years is required, and is used only in exceptional cases. Connection to the sensor node is often via Bluetooth (particularly the low-energy “LE” variant) or ZigBee. With ZigBee, the standard is sufficient to specify events such as “door open/closed” (cf. ZigBee Home Automation), whereas with Bluetooth LE, the specification does not reach this high level. Range is usually not critical – however, problems may arise in large houses, or houses with very thick concrete (or similar) walls. In this case, an alternative technology can be selected or intermediate stations can be used. (ZigBee supports mesh networking, in which each node communicates data between other nodes, thus increasing overall coverage.)

Depending on the interoperability requirements, the protocol used can either be LWM2M, proprietary protocols, or the higher layers specified by ZigBee. LWM2M and end-to-end IP connectivity are particularly suitable for ensuring maximum independence from individual transfer technologies.

Smart Wearables

This area includes sensor nodes that are built into sports devices, for example, or worn directly on the body.

Demand for these devices comes from the consumer sector, just like in the area of smart homes. The cost targets per item are very tight, and the number of units ranges from medium to very high. The space requirements of the components are a critical factor. Because devices usually run on a battery, energy requirements are also very important. However, users often accept the fact that battery life is limited, and this is compensated for by using rechargeable batteries. Communication requirements are usually characterized by short transfer paths (to a smartphone); similarly, connectivity is also characterized by connection to a smartphone, and to the wireless standards and protocols available in the phone. From a functional point of view, requirements are often prioritized so that minor inaccuracies in data capture are accepted in favor of very small size and low costs (tolerance in recording movements with a wrist band during the day, for example). Update capabilities are often limited; new functions are not implemented in the sensor nodes, but rather are added in the smartphone (as an app) or in the backend. However, there is a growing trend towards over-the-air updates for sensor nodes.

Available space and power requirements are important criteria for selecting sensor elements. As a rule, the only sensors included are those that are necessary for capturing the required data. Because of space limitations, it can be a good idea to use very highly integrated sensor elements (9 DoF), even though these are more difficult to evaluate due to cross-coupling effects, and despite the fact that higher tolerances may arise in data capturing.

The processing power requirements that the microprocessor must meet are low to medium. Available space and battery requirements are decisive factors. SoC (system-on-chip) systems offer significant advantages in this case, as they combine microprocessor and wireless connectivity in one integrated circuit (IC). Typical memory sizes are 4-64 KB RAM and 16-256 KB flash memory.

Since the microprocessor selected is often very small due to space, cost, and energy usage requirements, the algorithm can be designed in such a way that large amounts of data are transferred in cycles. (The power consumption of the radio transmission range is not particularly critical, as the required range is generally low and batteries can be recharged.) Data is then evaluated in the smartphone or the backend. The more data is transferred with little preprocessing, the more new functions can be implemented in the smartphone or backend via updates. However, this depends to a large extent on the specific use case. For example, a step counter will reliably identify steps in the sensor node, and only report the event “step” to the smartphone.

Recall that the Asset Integration Architecture defines several IoT architecture patterns. In the AIA, most cases in this area can be covered by the M2M pattern, with the smartphone playing the role of the gateway and mapping the business logic. If there are multiple sensors, they can be linked to each other (body area network), in which case the architecture would be based on the Device2Device pattern.

Since the battery is often very small and peak current is limited, both power requirements over time and peak currents must be considered for all components during the design phase. The power supply may need to provide for appropriate buffers.

Bluetooth LE is currently the dominant wireless connectivity technology in use, thanks to its interoperability with a wide range of smartphones, its low power requirements, and its sufficient range.

Transport and Logistics

This area encompasses sensor nodes that are used to monitor assets like containers and crates while they are being transported or stored. It includes the monitoring of location (indoor/outdoor), vibrations, temperature, and so on. Depending on the use case, the recorded data is either transferred wirelessly upon request or cyclically, or else it is recorded persistently and read directly (in a data logger use case). If the data is read directly (via USB or similar), then a wireless connection is not necessary.  Power is usually provided by a battery. Reliability requirements are generally high. Depending on the asset to be monitored, a decision will be made in favor of either a low cost per unit with a high number of units, or a medium cost per unit with a low number of units. Service life using a battery depends greatly on the use case and especially on the length of time for which the asset must be monitored. Space requirements are also highly dependent on the relevant asset to be monitored, but are usually of medium importance. Requirements due to environmental influences can be high, depending on the specific environment (for example, one with extreme temperatures).

The selection of sensor elements is dependent on the data to be monitored – it is not possible to make a general statement about necessary accuracy levels or other requirements. Depending on the asset to be monitored, measurement ranges usually extend beyond those found in typical consumer applications (for example, monitoring goods in a cooling chamber).

The requirements the microprocessor must meet are low to medium – these are usually applications for recording and transferring data. For designs in which data is recorded for a long time before being read, the memory requirements are higher. Additional memory is often used in these cases. Depending on how long the asset needs to be monitored for, the power requirements are a decisive factor and also influence the choice of microprocessor.

In many cases, the algorithms are comparatively simple threshold triggers. For example, if a monitored asset exceeds a certain temperature for a specific length of time, an alarm is triggered. However, depending on the use case (such as motion detection), the algorithms can also be complex.

Data transfer is highly dependent on the use case. Three general cases can be distinguished:

If the device needs to be monitored within a large area (such as outdoors) and if data transfer during monitoring is a priority, technologies such as GSM-based data transfer are useful. These make sensor nodes very expensive, consume a large amount of energy, and have high peak currents, resulting in high demands on the power supply. In the AIA, this would be the Device2Backend pattern.

If the asset is in a controlled, manageable space, and if data is transferred during monitoring, technologies such as Wi-Fi (which has low energy efficiency), BLE, LoRa, or proprietary wireless standards can be used – usually depending on the existing infrastructure. A gateway is also frequently used. In the AIA, this would be the Enterprise IoT pattern.

If data is not transferred during monitoring, but rather is read as required, one option would be to use cable technologies (for instance USB, which can also be used to simultaneously recharge the sensor node). Alternatively, short-range wireless technologies can also be a good option. BLE is particularly suitable where tablets, smartphones, or PCs are used to read the data, as it is widely supported by these devices.

Industry 4.0

This area is linked to the field of transport and logistics. It concerns the monitoring of machines and equipment as well as the materials and components used. Reliability requirements are high. Cost pressure per number of units is not as high as for consumer products in the areas of smart homes and smart wearables. The need for minimal installation space depends on the location in which the sensor node will be installed, but we will classify the space requirements as medium here.  Depending on the application, power may be available via cable. However, a battery is commonly used to minimize the effort required to retrofit a sensor node of this type. Requirements due to environmental influences can be high, depending on the specific environment (including extreme temperatures, oil mist, or vibrations, for example).

When selecting a sensor, high levels of reliability and accuracy are usually the main deciding factors. Machines have a wide range of in-built sensors. Depending on the use case and the connectivity of these sensors, the data they capture can be tapped into so that the entire solution can be achieved without dedicated sensor nodes. For retrofitting solutions, however, a sensor node to be installed later would be useful.

The microprocessor must meet medium to high requirements – algorithms may become quite complex, and high reliability requires complex software with correspondingly high memory and processing power requirements. As with the sensor element, cost pressure is not as important a consideration with a low number of units as in consumer applications with a high number of units.

Energy efficiency is important if the solution (especially a retrofit solution) needs to be completely wireless. If sensors are installed in concealed locations, changing a battery is a high cost factor due to the required down time and impact on working times. Depending on where the sensor is installed, a long service life can be achieved by choosing large batteries.

In terms of the algorithms, a lot depends on the energy efficiency requirements and available energy. If these requirements are very tight, it is a good idea to process data on the sensor node so that little communication with the superordinate system is required. For example, an event is only transferred when a specific level of wear and tear has been reached. If flexibility in data evaluation is a priority, and if other information, such as vibration data, must be sent to the backend, data transfer must take place more often – at the expense of battery life.

Wireless connection is critical in the industrial sector, because in many cases, the environment does not offer favorable conditions for wireless communication. Technologies such as LoRa can be useful here, but a wireless standard has not yet been established. Most machine operators and manufacturers would like a connection to the field bus systems used – which are usually proprietary. This can be done using appropriate gateways which are then developed specifically for the application.