Week 3 of the MITProfessionalX IOTx Internet of Things: Roadmap to a Connected World course concentrated on the Technologies module, specifically:
Network Connectivity for IoT (Hari Balakrishnan)
A simplified IoT network architecture
Room/body-are networks: Bluetooth Low Energy
Extending communication range
Data Processing and Storage (Sam Madden)
Managing high rate sensor data
Processing data streams
Data consistency in an intermittently connected or disconnected environment
Identifying outliers and anomalies
Localization (Daniela Rus)
Localization for mobile systems
And this is where things (no pun intended) start to get complicated. There is the complexity of IoT networking options for example. Why can’t we just use the wireless technologies that we have for the internet, our cellphones to build IoT systems? Can’t we just use cylinder networks and Wi-Fi technologies? Why do we need something new? The answers aren’t immediately obvious but when you think about it cellular networks are limited by the battery life of, for example, your mobile devices (aka gateways) and are expensive. Wi-Fi networks are limited by their range, the fundamental problem of power consumption is why cellular and Wi-Fi technologies are not applicable to a wide range of IoT scenarios.
Basically, IoT is about unusual events. Well, more specifically, data is at the core of those events. Consider applications in the space of infrastructure monitoring, like home monitoring, or monitoring pipes or other industrial equipment, or medical device monitoring. This is really about understanding when something interesting happens in these monitored devices. And the interesting thing that happens is fundamentally conveyed in data. For example, you might want to know that the temperature in your home went below some threshold, and the pipes are about to burst. Or with a medical patient, you might want to see some signal, like a brain or heart signal, that is showing some sort of anomalous value. Data starts from the sensors, it flows through the phones and base stations, and then ultimately ends up in a cloud-based infrastructure. Then there is the issue of missing and noisy data. These sensors, because they are sampling the real world, have periods of time that are not covered by the data itself. Also, the data that is coming from these sensors and these applications often has anomalies in it, things that are unusual or outliers. And so one of the real challenge is how do we detect and correct those kinds of outliers and anomalies? The classification-based method and frequent itemsets of course! Classification, for those of you like me who weren’t aware of it, is basically a way of giving outliers and anomalies a data set and dividing them into multiple classes. Frequent itemset mining basically compares the frequency of different sets of outliers to the frequency of the sets that occur in the inliers. What are the common things that occur in the outliers? In the frequent itemset mining world its about support. Support means detecting the elements that occur in one of set of data with more than some sort of frequency, ie more than two times.
However, the most complicated part is localization. Devices will have to instantaneously localize themselves. They will have to have a sense of identity and they will have to have a sense of the surrounding world. How does a device compute its position and its heading in the world? Range-based localization and bearing-based localization of course! Unfortunately, this is where trigonometry and algorithms begin to play their part. I never thought I’d be uttering the words robust quadrilateral but that’s just the rabbit hole that this course is taking me down and I have to admit I’m thoroughly enjoying it.
This week the Technologies module concludes with Security in IoT, HCI (Human Computer Interaction) in an IoT World and Robotics and Autonomous Vehicles.