Edge computing
Edge computing moves data processing out to where the data originates. That means faster response, less data traffic and the ability to work without a constant cloud connection. At Move we develop edge solutions that can analyse, filter and store data locally and only pass on what creates value.
Edge computing
Processing data where it originates
In many IoT solutions it makes no sense to send all the raw data straight to the cloud. The volume of data can be too large, the connection too slow or unstable, and some decisions have to be made faster than a cloud connection can respond. That is where edge computing comes in.
Faster response
Decisions made locally in real time
Less data traffic
Only what creates value
The right compute
From MCU to AI accelerator
Local storage
A buffer when the connection drops
Processing data close to the sensors
Edge computing means that part of the data processing takes place locally, in the product, the gateway or a computer close to the sensors. Instead of just collecting data and passing everything on, the edge device can analyse the information itself and act on it.
It can be as simple as calculating the minimum, maximum and average of a sensor value over a given period. It can also mean detecting whether a value exceeds a defined threshold and only sending an alarm when something deviates from the norm. In more advanced systems, the edge device can perform signal processing, computer vision or AI inference and make decisions locally.
Response in real time
The main advantage is response time. If a machine has to be stopped because its vibrations suddenly change, it is not necessarily sensible to send the data to a cloud first, wait for the analysis and then send a command back. With edge computing, the system can respond locally, in real time or close to it.
Less data and less energy
There is also a big advantage when it comes to data volumes. A sensor may generate thousands of measurements, but the central system only needs the trend over time, statistical values or information about deviations. By processing data locally, the volume of data can be reduced considerably before anything is passed on.
That saves both bandwidth and energy. It is particularly important in battery-powered IoT products, where radio communication often uses considerably more energy than the data processing itself. If the product can analyse a hundred measurements locally and send the result just once, it can make a significant improvement to battery life.
Video, sound and AI at the edge
With video and sound, edge computing becomes even more interesting. Cameras and microphones generate large amounts of data that can be expensive and slow to send continuously. Instead, an AI model can, for example, analyse video locally and pass on only an event, a result or relevant metadata. At the same time, this can support privacy by design, because raw video or audio can stay locally in the product when the application allows it.
The right compute platform
An important part of the edge architecture is choosing the right compute platform. Today there are many different compute modules, from small microcontrollers and embedded Linux modules to powerful multicore processors, GPUs, NPUs and dedicated AI accelerators. The choice has to be based on the real processing need, not on a wish for as much computing power as possible.
Neither too much nor too little
Too much compute power can be just as unsuitable as too little. An oversized module costs more, uses more energy and can generate considerably more heat, even if the application only uses a small part of its capacity. That in turn can lead to a larger power supply, more advanced thermal design, heat sinks, ventilation or a larger enclosure. That is why we try to size the compute platform to fit the actual workload, with the necessary margin for future features.
Conversely, the solution should not sit too close to the limit either. If the processor constantly runs close to maximum capacity, it can cause problems with latency, peak loads and later software extensions. That is why we look at CPU load, memory, storage, accelerators, interface needs, power consumption and thermal conditions, among other things, when the compute platform is chosen.
Local storage
Local storage is often a natural part of an edge solution. It can be important to keep raw data for a limited period, even if only processed data is normally sent to the cloud. If an alarm occurs, you can then choose to retrieve the relevant data from around the event for closer analysis.
Storage as a buffer
Storage can also be used as a buffer when connectivity is unstable. If the LTE connection drops, the product does not have to lose data. The measurements can be stored locally and passed on when the connection is re-established. The same principle is particularly relevant for satellite communication, where the connection can be expensive, have high latency or only be available in certain time windows.
Requirements for the architecture
That places requirements on the architecture. How much storage should the product have? How long should data be kept? Which data has the highest priority? What happens when the storage is full? Should old data be overwritten, compressed or deleted? And how is the data protected if the product is physically out in the field?
Edge computing is designed as one system
Edge computing is therefore not just about choosing a more powerful processor. Compute module, memory, storage, power consumption, thermal design, software, connectivity and cloud architecture have to be designed as one system. In some products, a small MCU is enough for local filtering and statistics. Others need Linux, multicore processors, a GPU, an NPU or dedicated AI accelerators.
At Move we work with the whole chain, from sensors and data acquisition to embedded software, edge computing, connectivity and cloud. The goal is to find the right balance between what should happen locally and what should be passed on. The earlier that architecture is defined, the better we can optimise the product for response time, data volume, energy consumption, heat, hardware cost and operating economics.
Frequently asked questions
What is edge computing?
Edge computing means that data is processed close to where it originates, for example directly in the product or in a local gateway, rather than all the processing taking place in the cloud.
Why not just send all the data to the cloud?
It can take unnecessary bandwidth, energy and cloud capacity. Edge computing makes it possible to pass on only relevant or processed data.
How do you choose the right compute platform?
It depends on workload, real-time requirements, memory, storage, interfaces, energy consumption and thermal conditions, among other things. The aim is to choose enough compute power without paying for, or cooling, capacity the product does not use.
Can edge computing be used for AI?
Yes. AI models can, for example, analyse images, sound or sensor data locally and pass on only results, events or alarms.
Can data be stored locally?
Yes. Local storage can be used for historical data, fault analysis or as a buffer if the connection to the cloud is temporarily lost.
Is edge computing relevant when connectivity is poor?
Yes. It is particularly relevant with unstable LTE coverage, remote installations or satellite communication. The product can carry on measuring and processing data locally and synchronise when the connection comes back.
Does edge computing save power?
It can. If local data processing reduces the number and size of transmissions, the energy used for communication can be reduced considerably, especially in battery-powered IoT products.
Shall we turn your idea into reality?
Technical development takes more than good ideas — it takes the right skills. We have gathered hardware, software and engineering expertise under one roof, so you get from prototype to production quickly and safely.
Reach out to us →