Thanks for the input. I had worked with RS-485 quite a bit in the past, but this was my first time setting up a new link in a few years. I never had a problem like this before. Most of the gear I used in the past had chassis ground referenced ports and with just a simple termination resistor at each end, we never had data issues on the buss. I am sure the issue is with the power meters and not the Nano10. Maybe they suffer from some offset, noise, or other issue, I can't be sure without digging into the circuit, but I can be sure that having the data lines fail safe biased cured my problem. Now that I am getting every data request, I am not going to dig deeper into them. They are cheap, and if this project works out well, we may upgrade to better components.
This time, I spent about 3 days messing around with the two power meters and the Nano10 trying to get reliable data. The part that made this hard to diagnose, was that when I would try to monitor the buss, it would start to work. The Nano10 port never seemed to have a problem, but the power meters just were either not seeing the requests, or the request was invalid due to a data reception error. When I first connected the buss tot he PC to monitor it, I was using the cheap RS-485 adapter that came with the power meters. This USB to RS-485 device supplied a mild bias, with just a pull down on the D- lead. It only has two terminals for the buss, and no signal ground lead. It worked ok when connected to just one power meter, but was not completely reliable with the 2 connected. The U-485 adapter that I purchased with my Nano10 made the buss completely reliable as soon as it was connected to the UBS port. Looking at my scope, I it appears to be providing bias to both data lines. Even with only the D+ and D- connected, it was instantly rock solid.
The power meters do have isolated ports. There is a tiny switching power supply to power the 485 interface chip and a pair of optocouplers to get the data from the logic to and from the interface chip. I did not measure any leakage current from the line measuring side to the interface side. I agree, the isolated port was part of the problem, but it was not the only problem. I did try tying the negative power supply of the Nano to the signal ground of the isolated port, but that did not cure the situation. Termination only across the two signal wires also did not help, but once the buss was biased, the data became rock solid.
The power meters even have the interface 0v signal ground and 5 volt available on the terminal block with the data D+ and D- data pins. That made it very easy to just add the resistors there on the unit at the end of the data buss. In normal operation, my data buss is only 6 feet. At 9,600 bps, I did not expect any issues with data. The two power meters are in a common box, with just a few inches of wire between them. The CT's to measure the current are remoted to the breaker panel. I did not want to place the electronics in the breaker panel and run the data in there. I only have the signal ground connected at one of the power meters, the other one only has the 2 data wires connected, at this time, and it is responding to every query and no data errors. I thought I might need to tie the signal grounds, but it has not shown to be an issue at all.