Multi-master RS485

On 3/31/2021 4:39 AM, pozz wrote:
> Another question on 10BASE-T1L. What is the complexity of a single node?

There are lots of ways to evaluate complexity.

Your hardware costs (for the "slaves") will likely increase.  You will
also likely incur the cost of a switch -- which may also impact deployment
logistics and costs (e.g., if you were hoping to daisy chain from one
node to the next NEARBY node, a switch forces duplication of cable).

You'll need an MCU with a MAC and, hopefully, external PHY interface.
Then, the actual PHY, itself.

Your software complexity will likely increase at the driver level.
Instead of a Linux "master" and /ad hoc/ slaves, you will likely run
something *like* Linux on the slaves, as well (to benefit from the
network stack).

OTOH, the whole polling issue becomes moot.  It's just as easy
(and reliable/scalable) to have each "slave" act as initiator
when *it* wants to talk to the "master" (let's call it a "controller"
and them, "motes")

Depending on the power requirements of your motes, you may be able
to source power centrally (from the switch) and thus eliminate the
need for locally sourced power at each mote.

[It is incredibly liberating to be able to place a mote anywhere
that is convenient -- instead of having to make considerations
for how you'll access power.  E.g., I have devices up at ceiling level,
inside walls, *above* the ceiling, etc.  I have neither the
unsightliness of "power supplies" scattered all around -- nor the
regulatory issues (e.g., hidden boxes).  Likewise, one less
aspect of an attack surface]

You may end up rethinking the application's feature set.  E.g.,
perhaps you incorporate provisions for two-way audio (intercom)
between each mote and ????.  Or, a fingerprint sensor -- with
the authentication performed by transferring the image to the
controller at the enhanced throughput available.

With (likely) more smarts at a mote, there are other options
available to enhance reliability.  I.e., what happens when your
controller is not accessible (perhaps the network has been subverted
or the controller has crashed or the system is in overload)?

> Actually RS485 can be managed by a very cheap MCU with a simple embedded UART 
> and a very cheap external half-duplex transceiver.

If your focus is exclusively on the (recurring) cost of comms, you
will end up chasing an /ad hoc/ bastardization of *some* protocol
to "make it work".  You should, instead (IMO) think about what you can
do with a different capability suite.

> What will I need to have 10BASE-T1L in a node? I think at the minimum an 
> Ethernet MAC and an external T1L transceiver.