I have two sets (set A, set B) of 3 ADC channels that I'm sampling using continuous multi mode w/ DMA. For latency reasons, I'd prefer to sample set A for a while (let's say 1 msec), followed by sampling set B for a while, followed by sampling set A for a while, etc. I would prefer NOT to interleave every sample, e.g., by setting a sequence length 2.
As I have discovered, switching ADC configuration channels on-the-fly in DMA mode really can't be done. Ideally I would just like to call ADC_RegularChannelConfig(...) on-the-fly, but I get DMA errors. Makes sense. Referencing the RM and some online docs, I have a switch sequence similar to the following:
ADC_Cmd(ADC1, DISABLE);
ADC_Cmd(ADC2, DISABLE);
ADC_Cmd(ADC3, DISABLE);
DMA_Cmd(DMA2_Stream0, DISABLE);
while (DMA_GetCmdStatus(DMA2_Stream0) == ENABLE); // Wait for DMA shutdown.
// Switch ADC configuration.
ADC_RegularChannelConfig( ... );
ADC_RegularChannelConfig( ... );
ADC_RegularChannelConfig( ... );
// Re-enable ADCs and DMA.
ADC_Cmd(ADC1, ENABLE);
ADC_Cmd(ADC2, ENABLE);
ADC_Cmd(ADC3, ENABLE);
DMA_Cmd(DMA2_Stream0, ENABLE);
// Restart conversion.
ADC_SoftwareStartConv(ADC1);
This switch sequence works reliably. The major issue I have is that ADCs are somewhat inaccurate (I have plots taken over millions of samples to prove it) for hundreds of us immediately after calling ADC_Cmd(..., ENABLE). In my application, I can't wait around for ADCs to become accurate, I need the data right away. So you see my problem here.
Major questions are:
1) Is the switch sequence I have above optimal? Any alternative that doesn't make me disable ADCs to switch?
2) Is there anything I can do to make ADCs more accurate immediately after the switch sequence?
