Question to people wanting CanBus in the next chip version

[PaulNi]

Hi guys,

Do you need CanBus for work on your own solution, or you want your device to work within an existing system (car electronics, can based industrial automation, etc?)

If you need it for your fully owned system, what makes you choose CanBus over rs485, i2c, i3c, semi-duplex daisy chain ethernet or other multidrop buses?

We have a an upcoming project in a rather big home automation/sensing deployment (one of biggest developers (like in real estate) in the country.) There is a hard pressure on BOM, one extra IC means like 25% unit cost increase in few extreme edge cases.

After a really long deliberation we settled on rs485 for the benefit of working with tons of 3rd party equipment, but we also noted that it is just the lowest common denominator and not the most optimal solution. Because of the later, I'm now researching what can we put as a second alternative bus in the coming future that will be used for higher speed devices. Rs485 will likely stay for use with really simple 3rd party devices, and the 2nd bus will be used for faster devices along it.

On one hand, we at least want to have electrical compatibility with some well supported standard with presence in automation space. On other hand, there is a need for us to not to be fully locked into the bus use case.

I do have a solution in mind, and it is not CanBus, but I want to listed what others are thinking.

[WiFive]

Rs485 goes up to 100mbps

[ESP_Angus]

Not sure I have any suggestions, but:

How many drops needed on a single bus, and what's the maximum cable length needed for a single bus?

[PaulNi]

Rs485 goes up to 100mbps

Yes, I researched the topic. Yes, you can do that if you can have a really fast transceiver and good wiring, but the question is where do you feed the I/O after the transceiver? Even most powerful MCUs around, way more powerful than even tensilica based ones are not that good at PIO at those speeds.

Another question, if you were to use a second rs485 bus just for high speed devices, you still have to take care of turning it into an actual bus. Just running Rs485 at 100mbps without some form of hardware MAC, will necessitate you to handle everything from link layer and up in software.

Third - Rs485 at 100mbps with your own protocol will be pretty much your own proprietary solution, without any kind of 3rd party support.

[PaulNi]

Not sure I have any suggestions, but:

How many drops needed on a single bus, and what's the maximum cable length needed for a single bus?

I think less than 30-40 devices per bus in an apartment. There will be 1 isolated bus for each apartment, and 1 for BMS (building management system.)

The maximum length should not exceed 25m in an apartment. On BMS side, there is not much of a concern about the length as we can break up the big bus into smaller segments as needed, and connect them to an ethernet network at will.

[jcsbanks]

Existing systems in cars.

I would not use CAN for a fully owned system as it is neither fast or cheap. Ethernet is so well supported generally with commodity hardware. Some new BMWs now have in the teens of ethernet nodes, but apart from between the gateway and OBD port it is 2 wire 100Mbps developed by Broadcom so non standard, but halves the wires.

I would give ethernet serious consideration against any other 2 or 4 wire solution.

[WiFive]

Ethernet is not multi drop and costs more to implement. Vehicles still use a combination of can, Lin, broadrreach, Ethernet, MOST ring. There are advantages to a lower speed multi drop bus like rs485 plus Ethernet where needed. Wireless is potentially cheaper if you can make it work with the device density.

[PaulNi]

Ethernet is not multi drop and costs more to implement. Vehicles still use a combination of can, Lin, broadrreach, Ethernet, MOST ring. There are advantages to a lower speed multi drop bus like rs485 plus Ethernet where needed. Wireless is potentially cheaper if you can make it work with the device density.

Well, you can coax ethernet into multidrop mode with few hacks, and quite a lot of people do that:

• CSMA/CD is still supported by some PHY/MAC combos in semi-duplex mode. The big problem is interoperability with devices which dropped the standard required CSMA function
• Option with a diode-opamp based 3 port hubs
• Option with 2 ethernet interfaces in bridging mode (STP nightmare)

The solution I actually settled on is 100Base-T1 (do not confuse with TX variety,) and 10Base-T1S. 10base-T1S is a modern multidrop PAM3 based ethernet standard that expects that some part of multidrop operation is done on board of a smart PHY. It is designed to reliably run on only one pair of wires, and uses some clever collision avoidance algorithms. On top of it, it is electrically engineered for work with POE by default.

I have not tried it with ESP32, but it looks to be very very very promising given that it has not only car people on board, but also building automation, networking, and industrial control. The showstopper is its electrical incompatibility with regular ethernet (PAM vs NRZ.)

I really hope it getting some attention from Espressif people, and, ideally, have it included as part of officially supported solutions.

Edit: there is a distinction in between the regular t1 variety and t1s. It is T1S that has hardware collision avoidance.

[WiFive]

Are there any cheap broadrreach/100base-t1 phys?

[PaulNi]

Are there any cheap broadrreach/100base-t1 phys?

No, and that the reason for me delaying committing to it.

In quantities financially feasible for out clients, the cheapest PHYs in stock we can buy in China is around $5 after duties and taxes (which is more expensive than the distributor price for ESP32.) This doesn't work for us when the cheapest devices we want to field will themselves cost around $5. This is how it goes for regular T1

For T1S, engineering samples are going around for $1k+ at the moment

TJA1100 from NXP is supposedly amazingly cheap from the vendor at $1.5, but god knows how one gets it directly from them. I keep calling NXP direct sales people for 2 months, with "keep waiting" being the usual reply or "fill in the qualification form" for the third time.

The current make up of the PHY makers is: TI, Broadcom, Marvell with minor NXP presence. Don't expect anything cheap until some company from outside the West makes a move on the market.

...But! I believe 100base-t1 and 10base-t1s both worthy of keeping RnD on them going. The traction from both car and industrial automation industry is immense. We ourselves got a client from construction industry who himself keeps track of the PHY and MCU markets with regards to base-t1s support.