Venturing Beyond MCUs: Crafting a LoRa Gateway with an Application Processor

[tyrreus]

Hello everyone!

I've accumulated some experience in MCU PCB design and in the realm of digital electronics more broadly. I have successfully developed several ESP and STM boards, programmed them, and everything has turned out well. Now, I'm eager to challenge myself further, particularly with application processors.

My goal is to develop a LoRa gateway device that incorporates a LoRa modem, LTE, Ethernet, and WiFi. It will also feature some ADCs for monitoring the power supply and several GPIOs for connecting external devices like relays, as well as potentially other digital signals such as an I2C OLED.

Embarking on a PCB project for an application processor is new territory for me. I've heard about the complexities associated with high-frequency DDR connections that require trace length matching, and the challenges of soldering BGAs using a hotplate. Many have suggested that such tasks are overly complex and best left to professionals. However, I'm not easily discouraged by these cautionary tales and am determined to design this board using KiCad (since Altium is beyond my budget), manufacture the PCBs through JLC or PCBWAY, and then assemble everything using the hot plate on my desk.

I'm reaching out for your advice. I'm in the midst of selecting a CPU for my design and am currently considering the following options:

The SAM9X60 is my leading candidate, primarily because it's available as a SiP with integrated DDR, which eliminates the need for anything beyond a NAND chip. It boasts robust support in mainline U-Boot and the kernel, is straightforward to program via USB using SAM-BA, and I have significant experience with this processor, especially concerning bootloaders.

The iMX 6ULL is another option, offering a slight edge in performance and energy efficiency. It might be easier to power due to its integrated regulators. Although I'm less familiar with this processor, it appears to be favored in industrial applications.

Are there any better alternatives for a hobbyist? What should be my key considerations for a project of this scale? Thanks for all your help!

[uer166]

Considering the extremely minimal bandwidth of LoRa, the choice of a full-blown applications processor is curious. Is there a reason why you're over-provisioning the compute by 10-100x?

[SiliconWizard]

Considering the extremely minimal bandwidth of LoRa, the choice of a full-blown applications processor is curious. Is there a reason why you're over-provisioning the compute by 10-100x?

The OP seems to want to build a LoRa gateway, not just some simple LoRa node.
That means potentially a lot more to handle.
If I were them, I would first prototype that around some off-the-shelf SBC, and if the result is good, maybe go for a full design, if there's any interest in that as a product or at least as an open-source project.

[tyrreus]

The Original Poster seems to want to build a LoRa gateway, not just some simple LoRa node.

That's exactly right. Our project aims for a greater level of complexity, with a few key concepts in mind:

- SSH for easy applications update.
- A more versatile approach to application development, free from the constraints of specific programming languages or memory management, thanks to the MMU.
- The ability for edge computing and the creation of a local database for storing collected data.
- The option to host a web server for accessing processed and collected data, beneficial in situations without LTE access, to reduce costs, or simply for those who prefer not to store data in the cloud.

I would first prototype that around some off-the-shelf SBC,

That's precisely what we're doing. We've started with a BeagleBone Black, using temporary connections for initial setup. Our goal, however, is to eventually solder all components, including the LTE and LoRa modems, directly onto the board to bypass any issues with connectors.

I'm collaborating with a local university (to share knowledge about embedded systems) and we are focusing on gathering data from wild bee hives scattered throughout the forest (https://github.com/BuzzVerse). We are also developing a LoRa nodes that monitor temperature and humidity within the hives, utilizing ESP processors. While MSP processors might offer better energy efficiency, the ESPs are dirty cheap. Moving forward, we plan to integrate CO sensors to facilitate early fire detection in the forest.

All this stuff is ongoing but I just started to collect info to prepare for the next step that will be a custom Linux board - so this is why I'm asking for your guidance.