STM32F401RE - ADC SMPRx 'sample time setting'

[JohnBirchmeier]

Documentation Error? RM0368 sections 11.12.4 and 11.12.5 list sample A2D clock times for 8 different sample time settings. For all but setting 0, the data seem to disagree with my testing and performance measurements. My measurements indicate:

// SampleTime setting vs Conversion time in A2D Clocks (tested on board NUCLEO-F401RE):

// - CPU Clock = 84 MHz

// - A2D Clock Pre-scaler = 2

// - Three channels per scan (ADC IN0, IN1 & IN4) with DMA transfer ADC to memory.

//

// 0 => 12.5 + 2.5 = 15 A2D Clocks / convert ( 3M conversions in 1.0714 s) << ONLY real match to general document a 15 clocks = (3 + 12)

// 1 => 12.5 + 10.5 = 23 A2D Clocks / convert ( 3M conversions in 1.6427 s)

// 2 => 12.5 + 19.2 = 31.7 ... ??? very odd (30M conversions in 22.6189s)

// 3 => 12.5 + 28.5 = 41 A2D Clocks / convert (30M conversions in 29.2856s)

// 4 => 12.5 + 56.5 = 69 A2D Clocks / convert (30M conversions in 49.2855s)

// 5 => 12.5 + 75.1 = 87.6 ... ??? very odd (30M conversions in 62.6187s)

// 6 => 12.5 + 96.5 = 109 A2D Clocks / convert (30M conversions in 77.8568s)

// 7 => 12.5 + 320.5 = 333 A2D Clocks / convert (30M conversions in 237.8558s)

//

[TDK]

Interesting that you're measuring less than the expected time, and not more. How are you measuring these durations exactly? I've done the same on STM32F429 (or maybe F405/F411) and they matched the RM exactly. I suspect it's the same ADC module on all of them.

(Note that conversion time is 12 cycles exactly, not 12.5.)

[JohnBirchmeier]

I have a 10KHz timer. I grab the value at the start of the DMA and grab the time again from within the DMA callback and I hit 10 million scans (i.e. 30 million A2D conversions complete).

I also verified the 10KHz timer operation by dumping its value to the console every 10s and used a stop watch to verify that timer is accurate.

[JohnBirchmeier]

NOTE: I just noticed that I have set the ADC clock pre-scaler set to 2, whereas the Device Configuration Took will only allow a setting of [4, 6 or 8]. Perhaps I'm violating some rule here ... I'll update my code and retest.

[TDK]

So you have a DMA conversion complete every 3 samples? So every 1us at the lowest sample rate? Doesn't seem like the code would be able to keep up with that.

[JohnBirchmeier]

The DMA completes after 300 conversions (i.e. 100 scans of 3 channels). The DMA is configurated for continuous requests with callbacks for 1/2 complete and complete.

Do you want me to strip down my test code and send you a copy? Do you have access to the NUCLEO-F401RE board on which to test (or something similar)?

[waclawek.jan]

You've changed sampling time for only 2 of the 3 ADC channels, the 3rd remained at 0 i.e. 3 cycles for sampling (or 2.5, depending on which table are you looking at).

JW

[JohnBirchmeier]

I think I have all three channels set the same sampling time setting. For example, for setting 1, this is the ADC1->SMPR2 (offset 0x10) register value before the DMA starts and the value is unaltered when it stop. See image:

ADC1 offset 0x10 set to 0x49 (i.e. all three channels are configured for '1')

[TDK · Accepted solution]

> (ADC IN0, IN1 & IN4)

You have channel 4 sample rate (SMP4) unchanged. Setting up SMP2 affects channel 2 (IN2).

[JohnBirchmeier]

You were right, I just could not see the issue.