It's already been a while since the idea of an Ethernet implementation in automotive has been launched. During this time, the recipe improved, the "cookie" baked and it is now ready to be served. Therefore, to have a corect impression of the "taste", some helpful details are in order.
Nowadays, a high-end car can contain easily over a hundred ECU's and you can imagine that the flashed software went beyond the GBytes threshold. Therefore, having a fully functional and a highly reliable network in cars is not only a desirable thing, but also a mandatory one. Basically, the automotive industry developed different solutions to acomplish needs, with technologies like: LIN (Local Interconnection Network), CAN (Control Area Network), FlexRay … MOST (Media-Oriented System Transport). All these were and still are very good in doing their job, but none is able to offer a common connection point since each is based on specific communication protocols. More than this, none offers a layer model to allow for a network with switching capabilities. Well, Ethernet (ENET) can do this.
[1] - www.renesas.com - In-Vehicle Networking Solutions
In contrast to the other existing solutions, ENET comes with a layering flexibility which goes from the application-layer multiple choices, reuse and exchangeability on different protocols levels, and different layer speed, up to the fact that adding a new participant will not disturb all the others connected elements and will not limit their bandwidth. They will influence only the one they are directly connected to. You will also have an environment capable of passing the emission and immunity tests, reducing cost and enabling new functionalities.
However, it should not be understood that all these advantages brought by the ENET implementation in the car architecture can be used by the simple use of the ENET host, without an Ethernet PHY (ENET PHY), which is one of the key components of the physical layer. The ENET module (integrated as a peripheral element in a microcontroller architecture) is not enough for NODE_A to send data to NODE_B in an Ethernet network. As part of the physical layer, each node (ENET module) needs an ENET PHY that takes care of the physical interconnection. This means fulfilling all the analog signal modulation requirements. As an interesting note, more than one PHY (up to 32) can be connected to an ENET host.
There are several available solutions for ENET PHY's, but all the vendors use a standard interface to allow the interoperability between the PHY's and the hosts. The standard interface is called Media Independent Interface (MII) and is used to interface the Ethernet Host's MAC (Media Access Control) layer with the ENET PHY.
The signals of the MII are shown in the picture bellow:
MII has 18 pins organized in two different groups:
On one side we have the configuration of the PHY, using SMI (Serial Management Interface) commands over two pins, MDC (Management Data Clock) and MDIO (Management Data Input/Output) as defined by IEEE802.3. The clock is provided by the host's MAC over the MDC pin and has to respect the maximum frequency of the PHY depending on the vendor. A typical frequency is 2.5MHz. MDIO is a bi-directional pin on which configuration commands are sent and status registers are read out.
With the goal of making the layout simpler for complicated circuits like switches (where there are a lot of signals), some derivatives of the MII were created. The most common are:
MII-lite: has only 14 pins (12 for data) and can achieve the same data payload. Signals CRS (Receive Carrier Sense), TX_ERR, RX_ERR and COL (Collision signal) are missing. Other than this, the way MII-lite interface works is similar.
RMII: Uses about half of the signals compared to MII. TX_CLK and RX_CLK are combined in a single clock signal which is provided by an external circuitry (or by the host in some cases). Since the data pins are also half (TXD [1:0] and RXD [1:0]) to achieve 100MBS the clock is 50MHz.
In a "normal" switch, router or a PC, the used PHY is the 10/100BASE-TX between the host and the RJ45 connector. On one side, you can see the connection with the host via the MII interface, and, on the other side, there is a 4-pin connection to the RJ45 connector. Two differential pair data pins transmit the data over the UTP cable. It transmits 100Mbits/s per twisted pair at a 125MBaud unidirectional using Block Coding (4B5B) and MLT-3 (Multi Level Transmit on three levels).
[2] - NXP internal documentation
UPT Cat-x cable has 4 pairs of twisted wire. In total, there are 8 wires, out of which 4 are used to send data by the 10/100BASE-TX ENET PHY's. Looking from the automotive perspective, this means a big cost, high EMI, and most importantly … a point of improvement. This is why BroadR-Reach PHY was designed. The biggest innovation is that the BroadR-Reach PHY only needs one twister pair cable, while still being full-duplex. It is a cheaper solution, which allows the reduction of wire harness cost, since the full Cat-x cable is not needed. The signal modulation used allows it to exceed the automotive standards of noise cancellation and jitter, making it the perfect automotive replacement for 100Base-TX.
It is developed and promoted by the OPEN Alliance SIG (One-Pair Ether-Net Alliance Special Interest Group) under the IEEE 802.3 standard for 100BASE-T1 [3].
The already-stated data is transmitted over a single copper pair, using Physical Coding Sublayer (PCS) and 3 bits per symbol - PAM3 (Pulse-amplitude Modulation on three levels.). It supports only full-duplex, transmitting in both directions simultaneously. The twisted-pair cable is required to support 100Mbps, with a maximum length of 25 m. The data is transmitted at 66.6Mbaud bi-directional. No specific connector is defined.
One important note that must be added is that in a BroadR-Reach connection between two PHY's, one has to be configured as master and one as slave. When we say node, this means a microcontroller, a PC or the port of a router or switch. If the PHY connected to the microcontroller (over MII interface) is configured as Master, the switch/router port to which it is connected must be configured as slave. This is a big difference compared to 100BASE-TX PHY's where you don't have the master/slave concepts.
Basically, the physical transceiver consist of an integrated single CMOS chip found at Broadcom as BCM89810 for a single port communication or at NXP as TJA1100.
As a use-case example, in the development phase, we needed the ENET host (the microcontroller) connected in a test network using a router. Several nodes were configured to communicate over TCP/IPv4. Only one node used BroadR-Reach PHY, which can't be directly connected to the 10/100BASE-TX PHY. Remember, the 100BASE-TX transceiver is using two twisted pair cable uni-directionally, while BroadR-Reach is using one twisted pair cable bi-directionally. The problem is solved by using a media converter (Easy CON Converter in our case which can be configured either master or slave), which links a BroadR-Reach PHY to a 10/100BASE-TX PHY (can be another node, a switch or a router).
Thus, from all the presented facts and figures, the clear advantages of the SW compatibility, the lighter and shorter wires, and cost effectiveness recommend it as the one best solution on the way right now.
[1] - www.renesas.com - In-Vehicle Networking Solutions
[2] - NXP internal documentation
[3] - www.opensig.org
[4] - https://www.broadcom.com/collateral/pb/89810-PB00-R.pdf