Frame Errors (Character Stretching) of CP2102N

Question

When I send a data stream from my PC to a CP2102 USB-to-Serial Bridge, the data coming out of the RS232 side is sometimes corrupted. The receiving micro (an ATmega2560) reports frame errors when this happens.

I send a command of 30 characters every 180ms. for every 10000 commands there are around 15 to 30 failures.

I use 9600.8N1, so a start bit plus 8 data bits should take 0.9375ms. Flow control is set to None.

It is always the first byte and sometimes (1 in 20) the second byte that is corrupted. I've captured the data on an oscilloscope and measured the byte times (beginning of start bit to beginning of first stop bit).

If I have corrupted data, the time is around 1.000ms. Good data takes between 940us and 960us.

It is always the last few data bits (i.e. the highest bits) and the stop bit that is the problem.

See image below for an example of the second byte being corrupted:

The character between the vertical cursor lines should be a 'P', but it takes 1.000ms and is received as 0xd0 instead of 0x50. The byte before it is a '*' (0x2a) and is fine, it takes 0.940us.

What could cause these frame errors?

Answer 1

I eventually found the problem: It is caused by setting the serial parameters while using the port at the same time.

I read the serial port in a separate thread that calls the following method in a loop:

def read(self):
    serstring = ""
    try:
        if self.is_reading_lines:
            self.serial.timeout = READING_LINES_TIMEOUT
            serstring = self.sio.readline().rstrip()
        else:
            self.serial.timeout = READING_CHARS_TIMEOUT
            serstring = self.serial.read(100)
    except:
        pass
    else:
        if len(serstring):
            self.process_read(serstring)
    return serstring  # if called directly

I use different timeouts for reading lines (responses to commands) and for reading binary data.

I update the timeout attribute before every read, not realising that this results in a call to method _reconfigure_port in file serialwin32.py in the pyserial module.

One of the calls in method _reconfigure_port is to win32.SetCommState. This is the ultimate cause: This affects the hardware of the CP2102 chip and causes it to interfere with outputting serial data on the RS232 port, which causes stretched data bits.

It makes perfect sense that updating the hardware settings interferes with the operation of the chip, even though the settings remain the same.

I replaced updating the timeout attribute in the read method with a separate method to set it only once and this fixed my problem.