Houston Texas Appliance Parts: Where Sensor Fusion and Sensor Processors Stand in IoT

Houston Texas Appliance Parts

Where Sensor Fusion and Sensor Processors Stand in IoT

by Houston Texas Appliance Parts on Thursday 12 January 2023 10:22 AM UTC-05

Sensors have become synonymous with the Internet of Things (IoT), and some industry watchers even call IoT the Internet of Sensors. So, while the sensor's role in the IoT bandwagon is indisputable, how will it impact major design considerations like interface, signal conditioning and compensation, and software algorithms?

To understand a sensor's impact on IoT design, you must first understand how IoT designs—and the sensors included in those designs—have evolved over the years.

In the first generation of IoT devices, the amount of data processing was limited since initial IoT devices weren't very complex. First-gen devices were conduits of the data and relied heavily on cloud computing platforms for processing. These cloud computing platforms could be characterized by an almost infinite amount of sensor data processing.

Fast forward to today, a time when processing is a highly desired feature at the IoT edge. Modern IoT devices have a good balance of processing capabilities that ensures the device not only gets the job done fast enough, but also accurately and with a low power budget.

Figure 1: A multitude of sensors is embedded into a variety of IoT applications. (Source: Texas Instruments)

In other words, while sensor data is in the analog domain, there's always some conversion of this data into the digital domain to ensure sensor data remains useful in a larger system. "The conversion and processing of analog must be done fairly quickly, accurately and with the lowest power possible, since most of these IoT devices are battery powered," said Albert Lee, technical marketing director for smart sensing and displays at Synaptics.

Lee also pointed to the importance of a flexible analog front-end (AFE) that can support many sensor-input types, such as capacitive, inductive and magnetic sensors. "That provides component area savings and BOM cost savings by eliminating the need for additional controllers for various sensors."

That clearly hints toward greater sensor integration around processors. Before we delve into this premise, however, it's worth revisiting another important sensor technology known as sensor fusion, and how it's being reinvigorated by artificial intelligence (AI) and machine learning (ML) algorithms.

Past and present of sensor fusion

Sensor fusion, a topic discussed years ago, is finally seeing implementation in complex sensing applications, such as context awareness. It combines multiple sensors to help deliver a complete picture of what's occurring in an environment, helping overcome the individual weak spots of different sensing technologies.

Ron Lowman, strategic marketing manager for IoT at Synopsys, noted some important trends regarding sensor fusion during an interview with EE Times. "Microcontrollers are being integrated into sensors, and more sensor companies are integrating processing and intelligence into sensors to add more value," he said. "We also see the trend of multiple sensors being integrated into different solutions."

Lowman presented the example of smartphones, which went from having a couple of sensors to dozens in a few years—but now designers must figure out how to miniaturize them. There are also other design challenges related to addressing voltage and integrating emerging technologies, such as silicon carbide (SiC) and gallium nitride (GaN), into sensors. "Still, we've seen a lot of progress along the way and expect the push for intelligent sensors to continue," Lowman added.

Figure 2: Sensor fusion comprises a complete hardware and software solution optimized for a wide range of ultra-low-power IoT applications. (Source: Synopsys)

While it's no secret that AI and ML algorithms are enhancing sensor fusion, it's still at a nascent stage. The fundamental challenge remains to be the underlying software. "Designers still need to figure out where to run their software and how to navigate complex algorithms and concepts to achieve end-to-end implementation, while also accounting for miniaturization," Lowman said.

Giovanni Campanella, general manager for building automation at Texas Instruments (TI), expressed similar views while acknowledging the role of AI and ML algorithms in interpreting the mountains of data coming from the sensors. "As more and more sensors are added to the system, the algorithm needs to be refined and improved so that the overall decision process is improved, and the correct actions can be taken to solve a problem or overcome a situation identified by the sensors."

For example, LiDAR tech isn't enough to enable autonomous navigation in a robot. Adding other sensors like vision and radar, and then implementing AI and ML algorithms, will allow the robot to recognize and learn from new situations and quickly adapt to them.

"Sophisticated algorithms are needed to make something out of the data acquired by one or multiple sensors," Campanella said. "These algorithms also need to learn about recurring situations in order to refine future decisions."

While acknowledging the profound impact of both AI and ML, Synaptics' Lee noted another crucial aspect of these software algorithms. "We see a continuous, but incremental, migration of AI/ML from cloud-based solutions to edge-based solutions."

He also pointed toward the hard requirements for IoT devices at the edge for low-latency, low-power operation and accurate processing. Though Lee recognizes that edge IoT devices will never replace the processing power of cloud-based solutions, moving forward, he sees a pragmatic partition between edge- and cloud-based solutions.

Advent of sensor processor

Sensor fusion or not, the number of sensors continues to grow, and that calls for innovative new solutions. Companies like Synaptics are combining multiple discrete sensor controllers into one controller. "Such a device can support capacitive, inductive and Hall effect sensors simultaneously," Lee said. "In the future, we hope to have compatibility established with certain types of MEMS sensors like force and inertial sensors."

A sensor processor captures and intelligently handles input from up to four sensors in a tiny, ultra-low-power form factor. The FlexSense sensor processor from Synaptics incorporates a microcontroller that connects to two proprietary, low-power AFE engines, which sense and digitize data from the capacitive and inductive elements on the touch surfaces of an IoT product.

Figure 3: Sensor processors intelligently fuse multiple sensors in a single processor while employing proprietary algorithms. (Source: Synaptics)

TI's Campanella has a more cautious perspective on sensor processors. Depending on the application, the sensor processor approach could be more suitable than a discrete one, but not always the right solution for an IoT design, according to Campanella. "Having a solution that integrates a sensing element, analog front-end and processor, similar to TI's mmWave radar sensors, is useful for space-constrained applications where critical decisions need to be made at the edge, like medical or robotics applications."

As with any semiconductor architecture, sensor-related designs are expected to change iteratively, and there will likely be continued development over technology generations.

The post Where Sensor Fusion and Sensor Processors Stand in IoT appeared first on EE Times.