The company where I work wants to do a proof of concept for our new product, a force-sensing platform.

The platform contains four three-axial load cells, so we have 12 channels, which are to be digitized at 2kHz and 16bits. This gives us 2.000 dataframes per second, with 24 bytes of payload each.

Our proposed architecture is as following:

  • The analog signals from the cells is acquired by a dedicated data-acquisition board;
  • The digital stream is sent to the ATmega on Yún over SPI;
  • The data is then relayed to the Atheros over Serial;
  • Finally, the data is transmitted over UDP (or TCP).

I wonder if it is a sensible way of doing it, with the following questions:

  • How would one normally use the Ethernet functionality? Can I call a function on the sketch, or do I need to put some program or script in the Linux, calling it from the sketch?
  • What would be the maximum transmission speed (in bytes per second) from the ATmega side to the Linux side?
  • What would be the maximum throughput of the Ethernet?

I have calculated 24 bytes * 2000 = 480 bytes per second, or a baud rate of 480 * 8 = 3840bps. Is that right?

I would appreciate any help, it could be just a link to the proper information resources.

1 Answer 1


The Yun is really two different computers (one AVR and on Linux) connected by a serial link. Basically the AVR is connected to a serial console on the Linux machine, so you can "type" commands into Linux by sending strings out the serial port from the AVR.

The serial link normally comes up at 250,000 buad, so that should not be a bottleneck for your application.

Typically, the AVR will "type" a command to run a Python program on the Linux side (called the Bridge) that will then communicate with AVR code using a high level protocol. Unfortunately, this "bridge" connections tends to be slow and flaky.

But you could do what you want very easily by just pumping netcat commands into the Linux machine from the AVR machine every time you wanted to send a UDP packet out. The overhead would only be a few serial bytes per packet (you'd send "nc -u \r" where everything but data is likely the same for every packet).

Here is an example of a program that runs on thew AVR and periodically reads some data from temperature sensors connected via 1-Wire, and then runs a "curl" command on the linux console to record the readings to a server...


In your case, your replace the 1-Wire reading code with SPI reading code, and replace the "curl" that sends HTTP data command with an "nc" command that would send a UDP packet (using the "-u" parameter).

With this technique, I think you could get something working very very quickly and likely easily meet your data rate requirements.

Note that while this is a great technique for getting a proof-of-concept working quickly, I would not use a setup like this in production. Instead consider something like a BeagleBone Black where you could read the SPI data and send the UDP pack in a single C program that ran as a normal Linux process.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.