When enabling CSMA it defaults to 4 retries and sets min BE to 3 and max BE 5. I'm not sure if these are sensible defaults, they are the reset values of the transceiver.

The IEEE 802.15.4 (2011) standard also defines them as standard values. However, if one want to send out data at a precise time, without (unpredictable) delay, BE should be set to 0. Figure 11 (p. 23) in the IEEE 802.15.4 standard shows that the backoff-wait is performed always initially before CCA (and of course later again, if the channel was busy).

Since this behaviour could be of interest for other MAC-protocols and is "hidden" in HW in this Atmel device, we should maybe document that somewhere in the driver. I think e.g. Contiki-MAC would expect a min BE value of 0 to avoid any delays before sending.

The wait-time, after which the packet will be send again, is defined as:

• random(2^(BE–1)) * backoffPeriod
• backoffPeriod = 320µs

Since this behaviour could be of interest for other MAC-protocols and is hidden "hidden" in HW in this Atmel device, we should maybe document that somewhere in the driver. I think e.g. Contiki-MAC would expect a min BE value of 0 to avoid any delays before sending.

That's actually a very good point. I wasn't aware that the random backoff wait is already active before the first attempt to transmitting. As I'm aiming for a MAC layer is even more important to me.

So in the first step it would be ok for me to have minBE=0 as a default, but I guess this should be configurable somehow. While going through all the interfaces involved, I still fail to find a good place for this, as it's really hardware specific and maybe not well placed as a NETCONF option.

What do you say?

yes, I guess making this option configurable is a good idea.

I actually think it should be an ordinary NETCONF option. All devices that implements CSMA/backoff in hardware will be able to set this option. All other devices simply don't implement it and will automatically return -ENOTSUP by design. But we should clarify in the documentation, that this option is explicitly described as hardware-only function. If the option can not be set, the MAC layer has to handle the CSMA/backoff procedure. @haukepetersen do you agree?

To go further, this could even be an indication for the MAC layer to en/disable software CSMA/backoff.

I'm without a computer for the next two and a half weeks - hence, someone else should review this.

@OlegHahm ping

@jremmert-phytec-iot I updated the docs and now _AUTOCCA and _CSMA options exist both in parallel. But there is no "AUTOCCA-emulation" for at86rf2xx yet.

Please rebase.

@daniel-k, willing to provide an extension to the ifconfig shell command to control this? If not, please rebase, squash, and send me a ping so I can test it.

Rebased. Will look into ifconfig now.

@OlegHahm
sc_netif.c has functions for setting u16 und i16 sized netopt options, while my new option NETOPT_CSMA_RETRIES expects uint8_t. I wonder now if I should expand my option to uint16_t or if I should add another function to sc_netif.c handling uint8_t. That would mostly depend on what those NETOPT options should look like in general.

I would suggest to add either a setter for uint8_t to the shell commands or see if you can come up with a unified setter function that take different widths.

@OlegHahm
Shell controls are now implemented for enabling/disabling CSMA and setting retries. Waiting for (N)ACK now.

ACK. Please rebase and squash!

You have to add a string representation of the new netopt types.

@OlegHahm string representations added and squashed again