Posted on May 17, 2011 at 12:40After a few million scans my set of scan data shifts (circularly) one channel. It stays shifted until I reset the processor.
This is an unlivable bug, and I can't reliably detect if the shift has occurred just be looking at the data.
I am using DMA to store the conversions. Since the data shifts circularly (DMA only writes to the correct addresses) it seems that the adc sequencer must be at fault.
I trigger the scan via the software start bit, called from a 1kHz periodic timer routine (the system timer tick). I use bitband access for the start scan bit (cr2 bit 22).
I set the settling time to the maximum (to minimize one channel's value from affecting another's).
I process the scanned values in the DMA ISR.
When I stop the processor after the fault all of the ADC configuration values are unchanged from my initialization.
Since a single software conversion is actually a scan of 1 channel I have no confidence that asking for conversions one at a time is going to solve the problem.
Am I alone in my using the adc in this way?
Am I alone in needing the hardware to work correctly more than 30 million times in a row?
----
The only truly weird thing I do is to use structs for access to the adc controller registers. Using a struct requires that the controller register support single byte accesses, and the manual doesn't state which registers require 16 or 32 bit access. Some peripherals don't work with 8 bit accesses but I can't imagine that being intermittent, and after configuration is completed I use bit-band access so the use of a struct should be moot.
