Next generation computing for IoT applications
Prof. Brejesh Lall
Head, Bharti School of Telecom Technology and Management - IIT Delhi
Internet of Things (IoT) is defined as the inter-networking of physical devices, vehicles, buildings & other items embedded with electronics, software, sensors, actuators & network connectivity which enable these devices to collect & exchange data. It further goes on to state that the definition of the IoT has evolved due to the convergence of multiple technologies, real-time analytics machine learning, commodity sensors, and embedded systems.
The term IoT can be attributed to Kevin Ashton in 1997 for his work at P&G on RFID tags to manage supply chains & later at MIT, starting the Auto-ID research centre. International Data Corporation, IDC, forecasts Worldwide Spending on the IoT to reach $745 Billion in 2019, led by the Manufacturing, Consumer, Transportation, and Utilities Sectors. The industries that are forecast to spend the most on IoT solutions in 2019 are discrete manufacturing, process manufacturing, transportation, and utilities. IoT spending among manufacturers will be largely focused on solutions that support manufacturing operations and production asset management. Consumer IoT spending will make it high on the list. The leading consumer use cases will be related to the smart home, personal wellness, and connected vehicle infotainment.
Industrial IoT is one of the fastest & and largest segments in the overall IoT space by the sheer numbers and the value that these services bring to industry & manufacturing. Consumer IoT, healthcare, transportation & logistics, agricultural & environmental, energy, smart city, government & military are considered to be the major sectors for adopting IoT applications. The economic impact of IoT, along with new revenue streams, will be created like factory automation, reducing time to market, improved supply chain logistics, green energy solutions & increased productivity.
IoT, by its very nature, enables the collection of data via its sensors. Millions of sensors & external factors produce vast amounts of data points which are stored in cost effective, easy retrieval storage systems for developing applications in real use cases by applying machine learning (ML) techniques to the data set. ML enables identifying correlations between them extract meaningful information from variables and transport to storage for further analysis. Applying ML to IoT data enables automatic improvement of its algorithms & ultimately more accurate predictions.
IoT is a great enabler for Artificial Intelligence (AI). Since the 1960s, AI is generally considered a buzzword, with little reality. According to research by McKinsey & Co., AI & ML are being developed in 60% of IoT activities. Three major changes have helped increase of use of AI: The convergence of algorithmic advances, data proliferation, & tremendous increases in power & storage capabilities at a lower cost. Business leaders expect the adoption of AI & ML to outpace other technologies. Bharti School, IIT Delhi being a visionary with ML & AI as focus areas is on the path of developing applications in several technology areas.
The bar graph shows technology adoption in IoT (McKinsey & Company, Jan 2019 report). For AI and ML to scale, production-grade data platforms are needed. Business leaders expect that to happen, with the adoption of AI and ML expected to outpace other technologies.
5G for IoTs
5G will be a transforming enabler for IoT applications. Although 5G derives from 4G but is a new set of technologies. 5G is the fifth generation of cellular technology. It is designed to reduce latency, increase bandwidth, have massive connectivity, and improve the flexibility of wireless services. 5G technology has a theoretical peak speed of 20 Gbps, while the peak speed of 4G is only 1 Gbps. This, along with Software Defined services, SDN, will open up a whole new era of services, automation & customer experience.
ETSI, European Telecommunications standard, lists the diverse applications required to be supported on 5G in the given triangle. As can be seen, each of the applications has a requirement of unique combinations of high reliability & ultra low latency, capacity enhancement in terms of throughput and massive IoTs in terms of sheer numbers.
The main usage scenarios of a 5G are: (i) Enhanced Mobile Broadband (1 to 10 Gbps connections to UE/endpoints). (ii) Ultra-Reliable & Low-Latency communications (URLLC) - < 1ms round trip end-to-end latency, 99.999% availability. (iii) Massive Machine type communication (mMTC) – 1 million nodes in 1 sq km, up to a 10 point battery life on endpoint IoT nodes. Moreover, 5G promising lower latency, can improve the performance of real time business applications as well as other digital experiences such as online gaming, video conferencing, and self-driving cars. 5G involves two major sets of changes to wireless cellular connectivity that includes (a) Improvements to IoT applications that use Low Power Wide Area Network, LPWAN, network technologies will be incremental & evolutionary. (b) Deployment of mmWave spectrum (Millimeter wave spectrum is the band of spectrum between 30 GHz and 300 GHz, this spectrum can be used for high-speed wireless communications as seen with the latest 802.11ad Wi-Fi standard, operating at 60 GHz)and Software Defined Network/Network Function Virtualization services on the 5G new Network will revolutionize applications deployment. This will support the development of factory automation, autonomous driving, and broadband applications like AR/VR, near real time robotic applications.
IIT Delhi moving with pace for IoT applications
IITD has set up various Labs focused on developing technology on 5G, in various fields of research aided by Government funds. There is also a 5G COE test bed with implemented with industry support. This program has been conceptualized to fast-track realization of Digital India initiatives and aid application development for Indian start-ups and industries. The 5G ecosystem being developed at Bharti school enables research and development to explore how some of the country’s key challenges can be addressed with advanced mobile technologies. The main focus of the ecosystem is deployment of LTE-A (4.5G) solution with advanced test cases including use cases for IoT and then 5G NR deployment and testing the use cases catering to 5G RF characteristics and key use cases like beam tracking as well as other use cases like security on Device to Device (D2D), Energy optimization on IoT devices& futuristic end devices connectivity on LiFi as well as Edge Computing. 4.5G system is already up and running in the Centre of Excellence and uses cases like air pollution &water monitoring has already been demonstrated. Going forward the emphasis is to conceptualize more use cases and work with startups to run analytics from the cellular IOT use case data which will ensure seamless connectivity of connected devices, machines, and things, supporting consumer, business, and industrial applications. Apart from 5G, IoT use cases, IITD is also exploring the key research agendas and challenges in ML, NFV, SDN and Security domain.
Three years back Bharti School at IIT Delhi started an IoT Lab with Industry support. Today there are a number of ongoing research projects on different aspects of IoT like deployment of network of sensor nodes for air pollution monitoring, water quality monitoring system, security for IoT, energy harvesting on IoT edge device, edge computing& ML for healthcare, vision. There are a number of Government &Industry funded projects where development work is ongoing in these fields.
The 5G CoE has already been used for demonstrating various use cases which has been developed across different departments at IITD. On one hand some of the existing use cases (Already Tested in 2018) are : (i) Autonomous Vehicle Testing, (ii) Disaster management cases involving Drones, (iii) Smart Agriculture Monitoring Use cases, (iv) Smart Cities Application e.g., Smart bins and other IoT cases, (v) Smart Aquaponics related applications, (vi) Development of Water Quality Monitoring Systems, (vi) Development of Air Quality Monitoring Systems. While on the other hand some of the planned use cases (under development) are : (i) Vehicle to Vehicle Communication Applications, (ii) Vehicle to Pedestrian Communication Devices and Applications, (iii) Remote Water Quality Monitoring (Rivers in India), (iv) Image processing - Live map (terrestrial) tracking for Drone deliveries, (v) 360 degree, 4K streaming videos, (vi) Advanced (Predictive) Air quality index monitoring
IoT Use Cases
Air Pollution monitoring
The typical workflow of an IoT project starts with the identification of use case generally in terms of its social & environmental impact. Since an IoT use case generally requires both software & hardware/ embedded systems skills, research scholars & undergraduate students are accordingly selected. Requirements of the use case are defined; the sensors identified &either fabricated or procured. For faster development the first prototype is generally created on an off the shelf Arduino/ Raspberry Pi/ similar board, depending on the type of application& the coding done in Python, C++. In case the requirement is to have a device which will function in energy constrained areas or require low latency, the PCB is fabricated & optimum MCU chipset & peripheral ICs accordingly selected& mounted/soldered. For lower requirements of energy & latency, as in Base station L1 layer, the Field Programmable Gateway Array, FPGA technology is preferred & the FPGA programmed as per algorithms developed n the Labs. The FPGA is Field Programmable Gate Array. It is a type of device that is widely used in electronic circuits. FPGAs are semiconductor devices which contain programmable logic blocks and interconnection circuits. It can be programmed or reprogrammed to the required functionality after manufacturing.
The FPGA allows researchers to ‘burn’ connections between the logic modules to make up complex processing logic, similar to CPU instructions. A CPU can be built out of an FPGA since they are both based on logic gates. FPGA can solve current IoT problems on power efficiency & concurrency. The parallel execution of instructions is a major plus point of FPGA. FPGAs tend to be more power efficient for IoT devices which may be connected on a tree or placed deep under the soil. There are a number of FPGA developments planned in different 5G components, including IoT.
For similar power constrained devices network connectivity is another major point which needs addressing in the beginning during requirements/design stage. There are a number of LPWAN networks which may be considered like NB-IoT (Narrow Band IoT), LoRa (Long Range, Low power), Sigfox. NB-IoT is a 3GPP standardized protocol. This kind of requirements needs to be detailed in the requirements stage as they are required for deciding the final deployment architecture/platform.
The IoT devices like air pollution monitoring sensor boards under next-gen development at IIT all the time are connected on a heterogeneous wireless network & monitored from a central console. The transmitted data is sent from the sensor devices to the data collection server/storage which stores the measured/calculated data, GPS coordinates, timestamp of measurement on a dedicated database, thus providing input for statistical analysis & finally predictive analysis.
Machine Learning (ML) techniques are being developed for a number of other applications at IIT Delhi, like the preliminary diagnosis of TB in the primary health centres, maize disease diagnosis, fish detection & movement of schools of fish, vision, Augmented Reality, communications.
Autonomous Car at IIT Delhi
One of the first use case of autonomous car was developed& tested at Bharti School on a 5G pre-standard test bed sponsored by Ericsson. Autonomous cars have not yet reached their full potential of being completely self driven. A car demo was set up at IMC 2018 with control centre at Bharti School, IITD. The test was executed meeting level 2 conditions, to a large extent, & partially level 3.
The Standard of Automotive Engineers (SAE) has defined following automation levels.
Use cases for level 2 testing will generally have the following components:
- Video cameras (forward, rear, side) — detect traffic lights and signals of other cars; read road signs; detect pedestrians, cyclists and obstacles; provide 360 degrees of visibility around the car;
- Ultrasonic sensors — complement the vision, detecting hard and soft objects, and measuring the position of objects close to the vehicle;
- Position sensors —built in the wheels to sense their movements and detect the car position on the map;
- GPS navigation — catches signals from satellites to provide more accurate positioning;
- Central computer — provides analytics of the gathered data and influences decision-making (this is in the development stage at IIT Delhi).
In higher levels of automation, the software and algorithms for autonomous cars must be extremely sophisticated in order to process the info from the extended network of sensors and make the right decisions. Vehicles should literally learn communication skills to be able to see hear and interpret any situation on the road. Moreover, the system should be able to predict the behavior of people both in and out of the car and correct it for the sake of safety.
Proposed Multi-access Edge Computing development
Multi-access Edge Computing (MEC) is an enabler for IoT applications, which will optimize the deployment of IoT use cases requiring lower latency & lower bandwidth. Use case is defined as a specific situation in which a product or service could potentially be used. Use cases where some kind of assessment or computation at edge is required for lower latency to the edge devices or have is a requirement for lower bandwidth on the cellular network, MEC is a great enabler. As edge computing is closer to the edge devices, a whole new class of cloud native applications can be developed. The IoT Lab at IIT Delhi is developing the MEC platform for use cases like the video analytics for surveillance and safety& the autonomous car.
The MEC implementations will be software entity on a virtualization infrastructure. ETSI defines the MEC framework grouped into system level, host level & network level entities. Typical MEC architecture for video analytics use case aligned to ETSI framework& being developed at IITD is detailed below.
A number of video edge devices will connect to IoT gateway on local wireless network. There will be basic edge computing algorithms coded in the Edge Compute device which will analyse the received data & send only interesting data to the MEC host over the cellular network. Near real time inference analysis can be performed at this stage, reducing latency &next hop network bandwidths. At the data analytics server in the core, the data will be prepared, the model trained & image predicted.