NAT7210 to host data transmission problem

Question

Hello!
Years ago, our company developed a GPIB interface board, using a NAT7210 GPIB controller which is hosted by an Atmel 8 bit MCU. For a test, the board is connected to a NI GPIB PCI interface, using MAX and NI-488 Communicator for communication.
Everything works fine so far, except one thing: a customer complained about occasionally "distorted" responses to the *IDN? query. We verified this to be true.
The response sometimes contains one additional character. For example, if the response string to the *IDN? query would be 70 bytes (including end token 0xA), sometimes it is 71 bytes and the additional character, which is mostly a ">" (0x3E), is inserted in random position.
Sometimes it's even worse. Then the NAT7210 sends the string twice, making it 139 bytes. This error occurs every few times when querying *IDN?.
The MCU that controls the NAT7210 does not send this additional character. I can verify in the debug mode, that the buffer content that is sent to the GPIB controller is OK and the code is OK, so that sending data is not interfered. We don't have any clue how to stop this. The NAT7210 has no command to flush its output buffer, nor we know if it does automatically after sending the message via GPIB.
Does anybody have any idea what's causing this and how to stop it? TIA

Answer 1

About NAT7210:
The problem with those GPIB controllers is that there are too many "ORs" in the manuals. "Do this OR do that". Too many options when setting up the config of a controller making it almost impossible to find the right one. We strictly adhered the initialisation sequence as told in the manual. But NI or other GPIB controller manufacturers don't mention a word about timing of the control data between host MCU and NAT7210. I also couldn't find example code for best configuration. 
However, the chip and the GPIB device are working smooth on the bus, except that one single problem. So I doubt if the controller is set up incorrectly.
I can not ensure that there is no noise on the bus, but I only have that one controller chip as talker/listener on the bus.
The number of queries after the problem occurs again is arbitrary. Sometimes 1, sometimes 5, sometimes 10 queries. The frequency of the double string is not so high as that of the additional character.
About rich text editor:
I'm using Chrome. The behaviour with the missing button strip and mode switching only happens when starting a new topic after searching the forum and pressing the blue button "Post" above the search result window. When replying to a thread, everything is fine, even switching between Rich Text and Preview modes. In the other case, the windows contents were deleted when switching view mode.
Best regards,
Maik