I have an application where i need to operate a gigE x-ray sensor over an existing 120 ohm twinax cable about 12 meters long. Our particular application has two parts or cable runs. Run 1 is from the sensor to the base of the mobile unit. We have a speblox at the sensor and at the base. In the base we have an ethernet switch. So, sensor >> spedblox>> twinax cable 10meters>> spedblox >> switch. The second leg is from that switch to the pc, so switch >> spedblox >> twinax >> spedblox >> PC. We find this works, however, we do get about 1% to 3% packet loss. What is odd, the packet loss is always in one direction. Sensor to PC sees no packet loss, but PC to sensor sees the loss. We added a 100 ohm terminating resistor at the sensor and at the pc and the packet loss dropped to less than 0.6%. Why does the terminating resistor help? Is there a better way to match the spedblox to the twinax cable? We use the two center conductors connected directly to the spedblox. The shield is connected to the chassis ground. Any suggestions would be appreciated.
Well done on solving the issue. This is the kind of question we love to answer as it takes us back to our pure electrical engineering roots! Let’s break this down…
SPEBlox 1G uses 1GBASE-T1, which uses a symbol rate of 750MHz with PAM3 encoding. What does that mean?
Well PAM3 means there are 3 states for the signal. This is unlike normal NRZ signals where there are only 2 states. PAM3 looks like this.
What is symbol rate? Well it’s the underlying “frequency” of the signal; it’s how fast the signal changes. That means the period of the signal is 1/750Mhz = 1.33 nanoseconds. In terms of the “length” of that signal in real terms, we usually say that electrical signals travel down a wire at 2 nanoseconds per foot. So 1.33 nanoseconds is around 0.7 feet = 0.2 meters.
Now we need to talk about reflections. When the length of a cable is close the length of the signal, you have to deal with the signal reflections. This happens when there is a significant change in the impedance of a conductor. For example, an unterminated pair of conductors will go from having a 120Ohm impedance to essentially having infinite impedance. That causes a reflection, that reflection bounces around between both ends of the cable and interferes with the signal. Putting a 120Ohm resistor on the end of your line terminates your line with 120Ohm, which kills the reflection.
Note that you don’t need to consider this when your signal/symbol period is much shorter than your cable length. As you’re using a 12 meter cable, and the symbol period is 0.2 meters, then we have to consider reflections.
That all being said, I am a bit surprised this works because the 1000BASE-T chip we use has an integrated 100Ohm termination resistance. That means that, in theory, both sides should already be terminated. There is a mismatch between your 120Ohm twinax and the 100Ohm termination in the chip, but that’s still not enough to make me think you would need an additional termination resistance.
The only thing I can think is that the additional components on SPEBlox between the chip’s internal termination, and the length of your cable, are causing a small reflection between the connector on SPEBlox and your cable.
Super interesting though, but if it works, it works!