Improving Power Efficiency In Ultra-Low Power Designs

Faster data communications in phones and data centers grabs headlines, but many applications don’t require the continuous, high-data-rate communications needed for video streaming or image processing.

In fact, for many devices, designing for better performance results in wasted energy and sharply curtails the time between battery charges. That is especially true for machine-to-machine (M2M) applications such as smart remote gas or water meters, which may only need to be connected once or twice a month to transmit meter readings.

This is the domain of low-power wide-area network (LPWAN) technologies, which includes LoRaWAN, Wi-SUN, NB-IoT, LTE-M, and the latest, Wi-Fi HaLow. For these applications, low-power wide-area networks have many benefits. They provide IoT and sensor connections over long distances, and they support battery life of 10 or more years. This is particularly important if the installations are remote or in places where it is difficult to change out batteries. Long battery life makes devices less expensive to maintain and that, in turn, makes power-efficient LPWAN design especially attractive.

When designing power efficient systems for an LPWAN, there are various considerations and tradeoffs involving topology, communication methods, and sleep mode management. Each of these factors impacts power consumption.

LPWANs use both mesh and star topologies. A mesh topology provides redundancy and hopping capability. When one of the nodes in the mesh network malfunctions, the data packets are re-routed to other nodes to prevent system downtime, which is particularly important in applications such as power grids. Rerouting can keep customers from being affected, even in situations where severe weather events, such as a tornado, have damaged part of the grid. The drawback is that with the nodes in near-continuous listening mode, in case a packet needs rerouting, more power is consumed.

Star topologies are simpler. One node communicates with another node directly within the network. With the star topology, longer idle times or sleep modes are possible, preserving power. So potentially, a star topology could be more power-efficient, but it’s also more limited.

That’s one piece of the puzzle. On top of there are a variety of LPWAN communications technologies, including LoRaWAN and Wi-Fi HaLow, which have low-power features defined in the specifications.

“Wi-Fi HaLow delivers long-range, low-power Wi-Fi that meets the unique requirements of the Internet of Things (IoT) to enable a variety of use cases in industrial, agricultural, smart building, and smart city environments,” said Kevin Robinson, senior vice president of marketing for the Wi-Fi Alliance. “Wi-Fi HaLow delivers low power through multiple energy-saving techniques, including Target Wake Time (TWT), Extended Max Idle Period, Hierarchical Traffic Indication Map (TIM), Short Beacons, and Null Data PHY frames, which are necessary for applications including sensor networks and wearables. These features work in harmony to extend the sleep time of Wi-Fi HaLow devices by years instead of hours, improve the coding efficiency to reduce airtime, and to reduce the packet size and transmission time relative to legacy management and control frames, which all contribute to the low-power benefits of Wi-Fi HaLow.”

Modulation is one of the key low-power methods used in LoRa. “Modulation methods are important to achieve power efficiency,” said Olivier Seller, vice chair of the technical committee for LoRa Alliance. “With a constant envelope for the transmitter, a LoRa chip can transmit +20dBm while consuming only 65mA (45% efficiency), for instance. For the receiver, because no prior synchronization is required, as soon as the receiver is active, it may receive a signal if present. If no signal is present, it may go back to idle mode very quickly — typically, 10ms to 100ms. The protocol uses this instant reception property to set up downlink opportunities in the form of short reception windows. To achieve the longest battery lifetime, devices acting as sensors are placed in ‘receive’ mode for short periods of time only after sending an uplink (Class A). By not listening at all times, they conserve energy.”

The communication methodology is another important factor in power management. Every time a data packet is sent from the device, power will be consumed. Inefficient communication may take more data bits or packets with longer “on-air” radio time and consume more power. Some designs can support an adaptive data rate to successfully shorten on-air time and if possible, avoid retransmissions, which increase the overhead. The most power-efficient communication is asynchronous, with data sent and received in one direction only. While asynchronous communication can stretch battery life, the tradeoffs are lower throughput and higher latency. That may work for some applications, particularly IoT devices that only need to communicate once or twice a month, or even devices that must communicate once every 24 hours.

Fig. 1: Different use models and their impact on battery life. Source: Ingenu

For some market segments, quality-of-service (QoS) is a critical element. To meet QoS goals, cellular LPWANs may employ synchronous communication, where packets travel in both directions. In synchronous communication, a device sends a message to the base station for permission to send data. Although synchronous communication can help keep QoS at the desired level, this method eats up more power than asynchronous communication. Additionally, the LTE-M provides voice support, another power-hungry application. Fewer bits transferred equals power savings.

One of the power-saving techniques used is sleep mode management. A device can be programmed to go to sleep more often, pushing consumption down into the microamp rather than milliamp range. Even in sleep mode, many different considerations come into play. For example, when will the device go to sleep? When would a device be available to wake up to listen for messages? Is there any mandated device wake up? Would it go through the authentication process every time it wakes up to ensure security? In general, the longer the device sleeps, the longer the battery life.

But there are many factors that can impact battery life. The LPWAN choice is only one of them.

“They include the amount of data transmitted, topology type, RF radio design, and the handshaking between the radio tower and the device,” said Mike Willey, vice president of advanced technology at Paragon Innovations. “In the case of the NB-IoT and LTE-M, the radio tower may take up to 70 seconds just to authenticate the IoT devices. And some developers may not be aware that batteries discharge themselves over time. Depending on the chemical content, some lithium-ion batteries may discharge themselves in less than a year. Finally, extreme hot or cold temperatures also will impact how much energy you will lose. Most of the projected length of the battery life is based on calculations and assumptions.”

Selecting the right battery type for industrial IoT is key to long life
Most LPWAN design discussions involve using coin-size cells or AA consumer type non-rechargeable batteries. Based on calculations and certain assumptions of data communication methods, the battery life can achieve 10 or more years. That doesn’t always work out as expected, however. Other factors directly impact battery life, such as battery chemicals, size, discharge, temperature, and the electronics circuit characteristics. Some developers have overlooked the fact that even when the device is in sleep mode, there is leakage current. While small, it still consumes power. In some use cases, the circuit leakage ultimately may drain more than 50% of the battery energy.

For electronic devices, the most popular batteries today use lithium. To achieve long battery life, in excess of 25 years, industrial IoT batteries use metallic lithium combined with other chemicals. In some industrial applications, such as smart meters, the batteries will last as long as the meters themselves. These batteries can be custom designed with a booster capacitor. The battery self-discharge level depends on manufacturing quality. This is why a premium battery for industrial IoT device may cost a great deal more than an equivalent consumer battery. The difference can be 25 years, give or take a few years.

“In selecting batteries for industrial IoT applications, it is important to consider the size, chemicals, circuit design, and its power consumption,” said Isabelle Sourmey, application engineer at Saft. “Most IoT designers may not have the knowledge or information to calculate the leakage, which will directly impact the life of the battery. They may be surprised to find out in a few tests that the batteries only last a fraction of the 10+ years specified. Therefore, it would be helpful to rely on battery manufacturers’ expertise and tools to select the most adequate battery and simulate device behavior during the design cycle.”

During the design phase, it is much easier to address the leakage issue upfront. But to really maximize power efficiency for LPWAN devices it’s necessary to optimize the efficiency of the overall system — sleep mode management, communication, and chip design — while minimizing current leakage.

“The key to achieving 10 years of battery life is the adoption of low-leakage solutions during sleep states,” said Ron Lowman, strategic marketing manager for IoT at Synopsys. “Take clock gating as an example, where all clocking is turned off to the chip except for a small low-power clock/PLL and a ‘keep alive region’ that is kept on. Additionally, you need to use slower restarts for applications that don’t have a fast restart or latency requirement. There are different tools available to achieve these techniques, such as the adoption of Synopsys’ Thick Oxide Libraries, which significantly reduce power during sleep states.”

Designing the overall system with power savings in mind
LPWAN is very application-specific, ranging from HD video surveillance to voice support to simple infrequent gas meter readings. Those applications demanding higher system performance consume more power. The first step is to select the appropriate topology or architecture for the applications. For example, if the application requires voice support, LTE IoT — the same as LTE–M, where M stands for machine — is the only choice. The technology was invented to support mobile devices, including cell phones. It has more features, but draws more power. Alternatives dedicated to M2M do not support video or audio, and are more power-efficient.

Once a topology has been selected, engineers need to think about how to achieve low power consumption, starting at the earliest stages of the design. Obviously, whenever possible, they need to choose low-power processors, components, and configurations, while ensuring the design can deliver the required performance reliably. And they need to consider which of those elements can run at lower voltages.

“One of the key techniques used to reduce the chip power consumption is to lower the supply voltage (P ~ V**2),” said Marc Swinnen, semiconductor product marketing director at Ansys. “But a very low voltage brings two problems. One, it places very high demands on the quality of the power distribution network, leaving no room for voltage drop on the way to the logic gates if the chip is to work. Second, at these low voltages, even a small voltage drop at a gate reduces that gate’s switching speed, so it also lowers the achievable chip speed. Both of these effects require very careful analysis, simulation, and modeling to ensure that the chip meets its operational goals under all possible scenarios of switching activity, temperature, and variation in silicon process parameters.”

Sleep mode management
Putting the system in sleep mode as much as possible is another obvious approach, but it’s not as simple as it sounds. System wake-up consumes energy, making it important to determine when the system sends or receives messages. While a water meter only needs to send data once a month, if the system includes additional functions such as water leakage detection, then shorter time intervals between wake-ups become necessary. You don’t want to wait for a month to discover a leak. Also, a system for an oil and gas platform, for example, could be on call. In that case, the device needs to wake up every so often to check for messages. How often that happens depends on how quickly it needs to respond. When the operator needs to check platform status, does it require a response within an hour, or can it wait for 24 hours?

LTE IoT has built-in power-saving features that can be utilized for this purpose. These including power-save mode (PSM) and extended discontinuous reception (eDRX). PSM enables the IoT device to wake up at fixed-time intervals — instead of being awake all the time, like a mobile phone or surveillance system — in order to transmit data or monitor messages. Then it goes back to sleep. Based on calculations, a device that transmits once a day in PSM mode potentially can achieve 10 years of battery life, subject to other conditions as described in this article. eDRX is similar to PSM, except the network initiates it.

Efficient communication methods
Simple, efficient communication is the goal. Using one-way, asynchronous communication will consume less power unless synchronous communication is absolutely needed. It’s not just system wake-up that consumes power. Handshaking to establish communication with the network, including authentication, is a consideration, as well.

An effective way to save power is to establish pre-authentication. When an IoT device or node establishes connection with a network for the first time, the network will go through the security authentication process. After the initial connection, a pre-authentication can be established so that when the device wakes up, the network doesn’t go through the full authentication cycle again, thus cutting communication time between the device and the network.

Power consumption also depends on traffic patterns and the size of the packet being transmitted. The more bits transmitted, the more power required. One approach to reduce power is to use adaptive bit lengths. Depending on the application, not every transmission requires the same number of bits. An adaptive bit length approach can automatically reduce the packet size to shorten the communication time.

Better integration with IP security
As with other networks, LPWAN needs security. Security IP can be integrated into the chip to achieve compartmental security. That way, even if one segment of the device is attacked, the other segments will not be compromised.

“TrustZone introduces a separate secure address space and processor state to support strong isolation between secure and non-secure applications,” said Mark Knight, director of architecture product management at Arm. “The isolation between these two worlds is achieved by hardware logic present in the Advanced Microcontroller Bus Architecture (AMBA) interconnect, peripherals, and processors. To support this capability, standard trusted platform software has been developed that supports trusted applications. This code typically implements trusted boot, the secure world switch monitor, a secure partition manager, and a small, trusted OS.”

Rambus takes a similar approach with its Root of Trust IP core, which it developed based on a RISC-V processor. By developing this as a discrete, purpose-built component, it can be highly optimized for low power.

“One of the hallmarks of failure is running a security algorithm within an insecure processor,” said Scott Best, technical director at Rambus. “A processor needs to be optimized, like any other circuitry. If you’re optimizing it for power, or you’re optimizing it for performance, or you’re optimizing it for security, to think you’re actually going to get one of those three benefits without focusing on it is recklessly optimistic.”

Customized security can sharply decrease the amount of power required for security. Alongside of that, programmability can keep the device current with new security threats.

“Power-gating functional blocks are key,” said Andy Jaros, vice president of IP sales and marketing at Flex Logix. “In the case of eFPGA in low-power wide area networks, when a snooping block sends an interrupt signal to indicate a communication event needs to occur, the eFPGA can be powered on and programmed within microseconds with whatever function is being targeted to run in the eFPGA. For example, the function could be a state machine that needs to read and prep data for sending, proprietary encryption algorithms, account information, etc. The configuration file to be programmed would be stored in non-volatile memory and can be updated as that function changes over the system’s lifecycle. In these types of applications, the amount of eFPGA would be very small and would not consume a significant amount of power. Additionally, an eFPGA can be ported using UHVT transistors, a method which favors low power in applications where performance is secondary. If there is wake-up time flexibility, the solution that uses the least amount of power would be to power off the eFPGA and re-program it on power up.”

Conclusion
LPWAN technologies claim to provide device battery life of up to 10 and in some cases 20 years, but numerous factors influence battery life, not just the communications.

Factors to take into account include designing the overall system with power savings in mind, sleep mode management, technology topologies, effective communication, better chip design using IP and simulation, and IoT battery selection. Some developers focus on system design, treating IoT battery selection as an afterthought. It is advisable to include battery selection in the design cycle. Finally, there’s a big difference between industrial IoT batteries and consumer AA batteries when it comes to battery life and performance. Considering all the factors involved upfront will ensure the best outcome, as well as shorten design time.

Related
Massive IoT Interop Fuels Protocol Battle
The race to scale communications for smart cities.
Grappling With Smart City Security Issues
Economics and convenience are driving cities to connect services, but they’re also widening the attack surface.
Design For Narrowband IoT
The challenge of creating chips for ultra-low-power applications with long lifetimes and always-on circuitry.