Edge AI Evangelist’s Thoughts Vol.15: Bluetooth Microcomputers - Part 2

November 16, 2021
AI Lab, Articles, Data Science Insights

Hello everyone, this is Haruyuki Tago, Edge Evangelist at HACARUS’ Tokyo R&D center.

In this series of articles, I will share some insights from my decades of experience in the semiconductor industry and I will comment on various AI industry-related topics from my unique perspective.

In this volume, I will continue my discussion about microcomputers. In particular, I would like to talk about its Bluetooth functionality for data collection in an edge AI system.

Bluetooth Enabled Microcomputers

In part 1 of this two-part series, I gave a general overview of a Bluetooth-enabled microcomputer. I also showed three demonstrations using Silicon Lab’s BGM220P Wireless Gecko Module Starter Kit. In these demonstrations, we investigated how the communication distance affected several other parameters.

In this volume, I will talk about the development of a Bluetooth low-power thermometer program using the Thunderboard and its battery life expectancy [3].

Experiment Preparations

For this experiment, the Thunderboard, manufactured by Silicon Labs, was used. The Thunderboard is a small-scale board that is 30.4 mm wide and 45.4 tall. The board also includes the CR2032 coin cell battery holder shown in Figure 1, SoC, five types of sensors, and a debug interface. The sensors include humidity, temperature, 6-axis acceleration, and illumination sensors. Other components include a 2-channel digital microphone and an internal temperature sensor inside the SoC.

The System on Chip is the EFR32BG22 Wireless Gecko with a 76.8 MHz ARM Cortex-M33 processor. This processor comes with a DSP instruction and floating-point unit, 512bK Flash ROM, 32kB Ram, various I/Os, and Bluetooth wireless functionality.

The left image shows all of the various components that make up the Thunderboard. The right image shows the dimensions of the board along with the locations of the debugger and battery holder.

Figure 1 Silicon Labs社 Thunderboard [3]

For this experiment, I used Simplicity Studio 5, which was the same software development tool (SDK) used in the previous projects [4][5]. Simplicity Studio 5 came with an advanced energy monitoring (AEM) function. Additionally, it came with software development and debugging functions. It could also dynamically measure the current consumption, supply, voltage, power consumption, and other parameters of the Bluetooth microcomputer [6]. Figure 2 shows the EFR Connect application [7], by Silicon Labs, that was installed on the central device and received data from the Thunderboard.

The image shows a setup for testing the average currents using a computer equipped with Simplicity Studio 5 and the Thunderboard which is attached to a wireless started kit and an iPhone 12.

Figure 2. Program Development

The program shown in Figure 2 is a Bluetooth low-power thermometer that was slightly modified [8][9][10]. For the following projects, only the internal temperature sensor in the SoC was used. Turning to the workshop materials, the program was originally developed in an environment where the USB terminal of the Thunderboard was directly connected to the USB terminal of the PC. This didn’t cause any issues with developing the program itself, but there were some issues when it came to using the AEM function (current and voltage measurements). In order to use these functions in Simplicity Studio 5, it was necessary to connect it to the Pc via the BRD4001A Wireless Starter Kit Mainboard shown in Figure 2.

Since the program puts the SoC into a sleep state during inactive periods (when it isn’t reading or sending sensor data) we refer to it as a “low power” thermometer. This is done to minimize power consumption and extend the battery life. In practicality, the average current consumption during the active period was around several MA, while it droped to only several uA during inactive periods. This is a difference of more than 1000 times. This function is especially suitable for applications with long periods between measurement intervals.

Advertising & Connected Modes

In Bluetooth communication, the first step is for the peripheral (the Thunderboard) to start advertising. The central device then catches the signal and sends a scan request/response. Finally, a connection request is sent and the device enters connected mode. At this point, it is possible to send and receive temperature data. In this paper, we focused mostly on the advertising mode.

The image shows a computer screen that shows the output for current measurements along with indicators that show the time intervals of the advertising and connected modes.

Figure 3. Advertising & Connected Mode Waveforms

Looking at Figure 3, we can see an example of current consumption measurements for the Thunderboard. The horizontal axis in Figure 3 shows the time, with the right showing the current time (0s) and the left showing 12 seconds into the past. The vertical axis shows the current consumption, and it is important to note that it is displayed on a logarithmic scale. The minimum value recorded was 1nA (1e-9A) and the maximum value was 100mA (1e-2A). This created a range of 10million nA. Figure 3 illustrates this range, where the bottom section of the current waveform showed a reading of about 1uA (1e-6A). At its peak, the current reached a value of several hundred uA in advertising mode. This created a pulse-like pattern in semi-regular intervals.

The left image shows a detailed waveform of a single pulse and the time interval for the pulse. The right image shows a more general waveform along with the total advertising interval.

Figure 4. Advertising mode waveform (Right), Detailed waveform with expanded x-axis for a single pulse (Left)

The current waveform during advertising mode is also shown in Figure 4. In this experiment, the advertising interval was set to 0.2 seconds within the program, and the measurement results were almost the same (Figure 4, right). When the time axis was changed to a scale of 1.25ms, the waveform was expanded as shown on the left. The resulting waveform was approximated as a rectangle where the width was about 3ms with a peak value of about 3.5mA.

Electric Current Model and Battery Life Estimation

Next, I want to show an example that demonstrates the electric current model during Advertising mode. Looking at the left side of image 5, we can see the current output model, and on the right, we can see a simple rectangle approximation of this current. For this experiment, the number of advertising channels was fixed to 3, and the transmit power was fixed at 0dBm. Using the values shown on the right, we could calculate the average current using the following formula:

Average current formula

From this formula, I found that the average current can only be expressed in terms of the advertising interval.

The left image shows a general waveform of the current model while the right image shows a model drawing that depicts a rectangle approximation for the current.

Figure 5. Average current model drawings

For the next step of the experiment, I began to vary the advertising interval between 0.02 and 10.24 seconds according to the Bluetooth specifications. I then measured the average current using the AEM. The experimental results were then compared with the theoretical model using the graphs shown on the left of Figure 6. As we can see, the two average current lines were almost identical. This proved that the rectangular waveform model above was a good approximation for the average current.

Moving on to battery life, it was expressed as battery capacity (mAh) and average current (mA). One example was the CR2032 coin cell battery, used in the Thunderboard, which was 220mAh [11]. Looking at the right graph in Figure 6, we can see the expected battery life for different advertising intervals. If, for example, the advertising interval is set to 1 second, the battery will last for 764 days ( approximately two years and one month). Of course, this also depends on the working environment and condition of the battery.

The right image shows a linear graph for the battery life expectancy that increases as the advertising interval also increases. The left image shows a graph that both the average current model and collected measurements transposed on top of each other.

Figure 6. Advertising interval & average current relationship (Left), Advertising interval and battery life expectancy (Right)

Summary

With edge AI applications in mind, a Bluetooth Low-power Thermometer program was developed using Silicon Labs Thunderboard. This program was used to measure and model current consumption and battery life using the internal temperature sensor in the SoC.
The average current consumption in Advertising mode consisted of a cycle of active and sleep periods. For the active periods, a few mA pulsed for several ms, and several uA pulsed for several ms during the sleep period.
The battery life of a coin cell battery varied greatly depending on the advertising interval. The battery life was found to be anywhere from several weeks to several years. Using the Thunderboard as an example, with an advertising interval of 1 second, the battery life was estimated to be about 764 days (approximately 2 years and one month).

Thank you for reading this volume of my Edge AI Evangelist’s Thoughts series. I always aim to provide interesting and stimulating content. I hope that you have learned something useful today and I will work hard to release the next volume shortly.