AIoT Power Supplies
Nov 10, 2023
AIoT or the amalgamation of Artificial Intelligence (AI) with the Internet of Things (IoT) is one of the most promising new developments in the electronics industry. According to one IC manufacturer, it will become the largest segment in the microcontroller market within the next ten years.[1]
The advent of IPv6 (Internet Protocol version 6) opened up an almost unimaginably large address space for IoT devices – enough unique addresses to individually identify every grain of sand on the earth many times over – meaning there is no longer any practical limitation on the number of IoT devices.
With the advent of new concepts such as smart cities, smart industry, smart transportation and smart homes, the field is now wide open for billions of IoT devices. And therein lies the problem. A billion is also a number difficult to imagine. A million seconds is 12 days, but a billion seconds is 31 years. Consider billions of sensors streaming data or requesting instructions. The information flow would overwhelm any remote data server or SCADA (Supervisory Control and Data Acquisition) system.
With the advent of new concepts such as smart cities, smart industry, smart transportation and smart homes, the field is now wide open for billions of IoT devices. And therein lies the problem. A billion is also a number difficult to imagine. A million seconds is 12 days, but a billion seconds is 31 years. Consider billions of sensors streaming data or requesting instructions. The information flow would overwhelm any remote data server or SCADA (Supervisory Control and Data Acquisition) system.
This is where AI steps in. AIoT has the simple aim of making big data IoT systems manageable. Intelligent, self-organizing systems create local processing loops where data from a multitude of sensors is collected, analyzed and acted upon without the need for top level intervention (Figure 1).
The collection and communication of the data is the function of the IoT network, the analysis or “insight” into the data is the function of the machine-learning or AI system which can then independently recognize patterns, make predictions, and take the appropriate action.
The advantages of AIoT are improved scalability and reduced data traffic (more sensors can be added without overloading the system), real-time pattern recognition using so-called ‘edge computing’ (the heavy “number crunching” work is done locally at the ‘edge’ of the sensor network and only the results and system overview datasets need to be communicated through the gateway), faster reaction times (fractions of a second instead of several seconds), greatly improved fault tolerance (machine learning can identify incorrect or missing data and use IoT’s inherent fault detection protocols to bypass defective nodes) and, maybe most important of all, fewer human errors.
These advantages will lead to across-the-board innovations in many ‘smart’ systems, from smart cities that will continuously monitor and analyze real-time traffic flow data to recognize accidents, prioritize emergency vehicles and optimize public transport, to smart grids that can optimize grid balancing, load sharing and the integration of renewable energy and energy storage systems into the system, to smart healthcare that uses wearables to monitor and predict medical emergencies, to smart industry, with effective just-in-time supply chain management, optimized production lines and condition-based maintenance, to name just a few buzzwords.
As with any complex, interactive, large-scale systems, operational reliability becomes paramount. The machine-learning analysis and inherently fault-tolerant network of IoT communications are essential to the AIoT concept, but a reliable power supply for the electronics must not be overlooked.
Some have claimed that the IoT sensors and actuators can be powered from long-life batteries or via energy harvesting – the idea that it is possible to extract sufficient energy from solar cells, thermal TEG generators or vibration transducers to charge a supercapacitor or directly power an IoT transponder. Having built and researched several energy harvesting projects, I can say from personal experience that this is a pipe dream. There is simply not enough reliable ambient energy available to be harvested to power the vast majority of IoT sensors and actuators. Batteries are also a no-go. With a conservative figure of one million IoT sensors, ten-year life batteries would need to be replaced at the rate of 275 per day, or about one battery change every two minutes for normal working hours.
An AC/DC power supply unit is not just a secure source of electrical energy for the sensors, actuators, processors, and gateways, by adding a digital communication interface it becomes a key component of the whole AIoT system. Fortunately, there already exists a successful and proven digital power supply communication and control system architecture – the Power Management or PM-bus. The PM-bus is an extension of the popular I²C (Inter-Integrated Circuit) protocol which is used for on-board communication between digital ICs. Like the I²C, the PM-bus is a two-wire serial interface, requiring only data (SDA) and clock (SCL) signals, which makes it low-cost and simple to implement, an important factor when implementing a large-scale system. However, it has extra commands specifically designed for power supply monitoring and control as well as PEC (Packet Error Checking) to confirm that the data transfer has not been corrupted.
The PM-bus protocol is ideally suited for AIoT applications because not only can the digital power supply be remotely switched on or off or placed into standby to conserve power, the additional command set also allows the output voltage to be remotely adjusted, the current and power limits to be set, the AC input line to be monitored and the temperature of the power supply to be monitored, and the internal memory to be interrogated for present and previous error codes and for inventory reasons. Combined with the machine learning algorithms, a PM-bus enabled power supply becomes an active part of the entire AIoT system, delivering useful monitoring and alarm data (Figure 2).
Fig. 2: RACM1200-V power supply monitoring signals and timing
One of the most powerful features of the PM-bus protocol is the on-the-fly configuration commands. For example, the load limits before the power supply switches from constant voltage to constant current or constant power mode can be set or forced by external commands. For example, if the machine learning AI algorithm decides that a load peak is about to occur, it can preset the power supply to best cope with the temporary overload without shutting down. Usually, the smart fan function usually means that the power supply controls the fan speed, switching off the fan during low loads or colder temperatures to conserve energy, reduce acoustical noise and extend fan life, or gradually increasing the fan speed as the load increases.
If the AI recognizes a certain load pattern, for example, a battery charging station that is used on regular basis with only short periods of time in between full-load charging conditions, it might decide to turn off the fan during the rest periods to keep the temperature high and reduce the extent of thermal cycling in the power supply. A stable high operating temperature is usually less stressful to the electronic components in a power supply than a continuous thermal cycling caused by repeated heating up and cooling down periods.
As mentioned before, the PM-bus evolved from the I²C bus. This introduces a major drawback. The controller only pulls the signal lines down so that the power supplies have the option of also pulling down the clock or data lines to take over the bus to send information back to the controller. This means that the PM-Bus relies on pull-up resistors to bring the serial lines back up high when none of the units are active (figure 3). Although the PM-Bus protocol allows up to 127 individual addresses on one control line, if this number of devices were to be connected to a single bus, the combined capacitance would slow down the rise-time of the signals to the point of uselessness.
Fig. 3: PM-bus signal. Each line can only be pulled down, relying on resistors to pull up the signal lines back up to VDD.
With higher bus capacitances, the rise time of the signal gets longer.
If a larger number of PM-bus enabled power supplies were to share a single bus, then they would need to be arranged into groups of, say, 16 smart power supplies that communicate via a low-cost PM-bus repeater ICs. These buffer ICs have a very low input capacitance of around 6pF allowing the full PM-bus address space to be occupied without overloading the bus, even with longer cable runs. They can be powered from one of the target power supplies ‘always on’ outputs, a low power 5V output that remains active even when the main power supply stage has been powered down (Figure 4). The “always on” output can also deliver sufficient current to power other devices, such as, for example, a long range (LoRa) transceiver circuit to allow remote control over kilometer-wide distances, even in heavily built-up inner-city areas or in noisy industrial factory locations.
Fig. 4: PM-Bus repeater IC powered from the ‘always on’ 5V auxiliary output.
The collection and communication of the data is the function of the IoT network, the analysis or “insight” into the data is the function of the machine-learning or AI system which can then independently recognize patterns, make predictions, and take the appropriate action.
The advantages of AIoT are improved scalability and reduced data traffic (more sensors can be added without overloading the system), real-time pattern recognition using so-called ‘edge computing’ (the heavy “number crunching” work is done locally at the ‘edge’ of the sensor network and only the results and system overview datasets need to be communicated through the gateway), faster reaction times (fractions of a second instead of several seconds), greatly improved fault tolerance (machine learning can identify incorrect or missing data and use IoT’s inherent fault detection protocols to bypass defective nodes) and, maybe most important of all, fewer human errors.
These advantages will lead to across-the-board innovations in many ‘smart’ systems, from smart cities that will continuously monitor and analyze real-time traffic flow data to recognize accidents, prioritize emergency vehicles and optimize public transport, to smart grids that can optimize grid balancing, load sharing and the integration of renewable energy and energy storage systems into the system, to smart healthcare that uses wearables to monitor and predict medical emergencies, to smart industry, with effective just-in-time supply chain management, optimized production lines and condition-based maintenance, to name just a few buzzwords.
As with any complex, interactive, large-scale systems, operational reliability becomes paramount. The machine-learning analysis and inherently fault-tolerant network of IoT communications are essential to the AIoT concept, but a reliable power supply for the electronics must not be overlooked.
Some have claimed that the IoT sensors and actuators can be powered from long-life batteries or via energy harvesting – the idea that it is possible to extract sufficient energy from solar cells, thermal TEG generators or vibration transducers to charge a supercapacitor or directly power an IoT transponder. Having built and researched several energy harvesting projects, I can say from personal experience that this is a pipe dream. There is simply not enough reliable ambient energy available to be harvested to power the vast majority of IoT sensors and actuators. Batteries are also a no-go. With a conservative figure of one million IoT sensors, ten-year life batteries would need to be replaced at the rate of 275 per day, or about one battery change every two minutes for normal working hours.
An AC/DC power supply unit is not just a secure source of electrical energy for the sensors, actuators, processors, and gateways, by adding a digital communication interface it becomes a key component of the whole AIoT system. Fortunately, there already exists a successful and proven digital power supply communication and control system architecture – the Power Management or PM-bus. The PM-bus is an extension of the popular I²C (Inter-Integrated Circuit) protocol which is used for on-board communication between digital ICs. Like the I²C, the PM-bus is a two-wire serial interface, requiring only data (SDA) and clock (SCL) signals, which makes it low-cost and simple to implement, an important factor when implementing a large-scale system. However, it has extra commands specifically designed for power supply monitoring and control as well as PEC (Packet Error Checking) to confirm that the data transfer has not been corrupted.
The PM-bus protocol is ideally suited for AIoT applications because not only can the digital power supply be remotely switched on or off or placed into standby to conserve power, the additional command set also allows the output voltage to be remotely adjusted, the current and power limits to be set, the AC input line to be monitored and the temperature of the power supply to be monitored, and the internal memory to be interrogated for present and previous error codes and for inventory reasons. Combined with the machine learning algorithms, a PM-bus enabled power supply becomes an active part of the entire AIoT system, delivering useful monitoring and alarm data (Figure 2).
Fig. 2: RACM1200-V power supply monitoring signals and timing
One of the most powerful features of the PM-bus protocol is the on-the-fly configuration commands. For example, the load limits before the power supply switches from constant voltage to constant current or constant power mode can be set or forced by external commands. For example, if the machine learning AI algorithm decides that a load peak is about to occur, it can preset the power supply to best cope with the temporary overload without shutting down. Usually, the smart fan function usually means that the power supply controls the fan speed, switching off the fan during low loads or colder temperatures to conserve energy, reduce acoustical noise and extend fan life, or gradually increasing the fan speed as the load increases.
If the AI recognizes a certain load pattern, for example, a battery charging station that is used on regular basis with only short periods of time in between full-load charging conditions, it might decide to turn off the fan during the rest periods to keep the temperature high and reduce the extent of thermal cycling in the power supply. A stable high operating temperature is usually less stressful to the electronic components in a power supply than a continuous thermal cycling caused by repeated heating up and cooling down periods.
As mentioned before, the PM-bus evolved from the I²C bus. This introduces a major drawback. The controller only pulls the signal lines down so that the power supplies have the option of also pulling down the clock or data lines to take over the bus to send information back to the controller. This means that the PM-Bus relies on pull-up resistors to bring the serial lines back up high when none of the units are active (figure 3). Although the PM-Bus protocol allows up to 127 individual addresses on one control line, if this number of devices were to be connected to a single bus, the combined capacitance would slow down the rise-time of the signals to the point of uselessness.
Fig. 3: PM-bus signal. Each line can only be pulled down, relying on resistors to pull up the signal lines back up to VDD.
With higher bus capacitances, the rise time of the signal gets longer.
If a larger number of PM-bus enabled power supplies were to share a single bus, then they would need to be arranged into groups of, say, 16 smart power supplies that communicate via a low-cost PM-bus repeater ICs. These buffer ICs have a very low input capacitance of around 6pF allowing the full PM-bus address space to be occupied without overloading the bus, even with longer cable runs. They can be powered from one of the target power supplies ‘always on’ outputs, a low power 5V output that remains active even when the main power supply stage has been powered down (Figure 4). The “always on” output can also deliver sufficient current to power other devices, such as, for example, a long range (LoRa) transceiver circuit to allow remote control over kilometer-wide distances, even in heavily built-up inner-city areas or in noisy industrial factory locations.
Fig. 4: PM-Bus repeater IC powered from the ‘always on’ 5V auxiliary output.
Bus-controllable power supplies feature an interface connector as well as the main DC output power terminals (Figure 5). The advantage of the digital interface is that a more comprehensive alert, fault and alarm signaling is possible to aid the AI in its decision-making. For example, the RACM1200-V has a built-in multicolor status LED that indicates AC-input failure, DC-output low, over-temperature imminent, over-temperature shutdown, overload and DC fail conditions, but the digital bus interface also adds programmable pre-warning limits on the output voltage range, load current monitoring and overload conditions – giving the AI system sufficient time to react to extreme temperature or fault conditions before the power supply goes into shutdown to protect itself.
In summary, AIoT is an exciting new enabling technology that will benefit us all over the next decade, improving our lives in many different subtle ways. As one of the leading global manufacturers of AC/DC and DC/DC converters, RECOM is already gearing up to support this innovation by designing and manufacturing power supplies with digital PM-Bus interfaces for seamless integration into AIoT systems.
[1] Remi El-Ouazzane „A Tsunami of TinyML Devices is Coming”, EE Times, 07.28.2023
In summary, AIoT is an exciting new enabling technology that will benefit us all over the next decade, improving our lives in many different subtle ways. As one of the leading global manufacturers of AC/DC and DC/DC converters, RECOM is already gearing up to support this innovation by designing and manufacturing power supplies with digital PM-Bus interfaces for seamless integration into AIoT systems.
[1] Remi El-Ouazzane „A Tsunami of TinyML Devices is Coming”, EE Times, 07.28.2023
Applications
| Series | |||||
|---|---|---|---|---|---|
| 1 |
|
RACM1200-V
Focus
|
|
|