Local application processing

In constrained cellular IoT application designs, it is important to note that using the radio link for communication can be costly on your power budget if it is used too often.

Therefore, you need to consider how to you use your link. A rule of thumb is to process as much as you can on the application processor, and then send only the essential information as rarely as you can to optimize for low power.

To reduce the radio link usage, you can use different power saving techniques. For example, using Power Saving Mode (PSM) or increased Extended Discontinuous Reception (eDRX) intervals depending on how you want to use your radio link. If your device is rarely going to send anything to the cloud, it can be beneficial to use PSM to avoid the power consumption penalty of reattaching before sending compared to turning the modem completely off. Because the nRF9160 hardware is optimized for low power, you can leverage our features by following our Power optimization guide.

We have an Online Power Profiler tool to estimate power consumption for a device. After your initial estimations, we recommend using the Power Profiler Kit II to measure the current consumption. During development, it is important to test your design on live networks to debug and improve power consumption. It is also useful to verify supported network configurations in your location, which will reveal the device configurations needed to extend the battery life of your product as much as possible.

Edge computing means doing the computational processing on the edge and limiting the data sent over the air by sending only the preprocessed data. The introduction of TinyML now enables machine learning models and training running on the constrained devices. Because the nRF9160 SiP is supported on the Edge Impulse platform, it is possible to train the model using the vast computation power of the cloud and then moving the trained model into your application to run it on the edge device. This means that you can capture a lot of data, run it in your machine learning model, and get back the relevant result. For example, you have a scenario where you constantly send accelerometer data (x,y,z data plots) from a sensor over the air, which could draw ~60 mA and then be processed on the cloud. Compare this to locally processing the same data and just sending a small information event over the air when something interesting is happening, which would draw ~3 mA instead. An example of this information sent to the cloud is Pump has stopped working, please inspect locally.

We recommend using the nRF9160 DK for evaluating and developing your applications, and following the Getting started with nRF9160 DK guide.

Some considerations about local application processing are:

See the Improving Energy Efficiency for Mobile IoT white paper for more information on the low-power features of LTE-M and Narrowband Internet of Things (NB-IoT).