K20 (ADC or DMA issue) ADC results are half of the correct value

[nicholasf]

I am using the ADC on a K20F120, and getting results that are exactly half what they should be (according to an oscilloscope which I am measuring the input with).  My configuration is as follows:

core clock = 120 MHz
bus clock = 60 MHz

ADC clock = 7.5 MHz
ADC mode = 16 bit differential

ADC trigger = software triggered, free running

ADC triggers DMA transfers on conversion complete

ADC is calibrated before being used

ADC is running at correct speed

No PGA configured

ADC range is 0V-3V (approximately the same as Vdd)

I can post the configuration if that would help (I am not using processor expert), but I am wondering if anyone has encountered this issue before.  It looks as if the results are being right-shifted.  The problem could be something with the ADC or DMA.

Any input would be greatly appreciated. Thanks for reading my post!

[AleGuzman]

For the Kinetis 16-bit SAR differential reads means that you have a pair of external pins.

The ADC digital result is the subtraction between the differential inputs

ADCRn = DPx -  DMx

     DPx -- Differential input plus

     DMx -- Differential input minus

This doesn't means that the ADC accepts negative input voltages, the ADC input voltage range is between VREFL and 31/32 VREFH

The ADC output format is in 2's complement 16-bit signed will be

[nicholasf]

Yes, this is exactly what I thought; my VREFL = 0V, and VREFH = 3V DC.  I am wondering if there is anything in my setup causing the results to be divided by two.

My ADC configuration is as follows:

{

ADC_CFG1_REG(pBase) = ADC_CFG1_ADIV(3) | ADC_CFG1_MODE(3) | ADC_CFG1_ADICLK(0); // divide clock by 8, 16-bit mode, select bus clock as input

//ADC_CFG2_REG(pBase) = ADC_CFG2_ADHSC_MASK; // 2 ADCK cycles added to the conversion time to allow for higher speed clocks

ADC_SC2_REG(pBase) = ADC_SC2_DMAEN_MASK; // enable DMA trigger

ADC_SC3_REG(pBase) = ADC_SC3_ADCO_MASK; // continuous mode

ADC_SC1_REG(pBase,0) = ADC_SC1_DIFF_MASK| ADC_SC1_ADCH(signal);// select differential mode, and correct signal channel

}

The DMA is configured as follows:

disable_int(INT);                    // not ready for this interrupt yet. Plug vector first.

DMA_CR = DMA_CR_GRP0PRI_MASK; //writes to the CR register must be performed when the DMA channels are inactive, that is when TCDn_CSR[ACTIVE] bits are cleared

DMA_SOFF_REG(DMA_BASE_PTR,number) = 0;

DMA_ATTR_REG(DMA_BASE_PTR,number) = DMA_ATTR_SMOD(0)             // Source address module disabled

                  | DMA_ATTR_SSIZE(1)           // 16-bit source data transfer size

                  | DMA_ATTR_DMOD(0)            // Destination address module disabled

                  | DMA_ATTR_DSIZE(1);          // 16-bit destination data transfer size

DMA_NBYTES_MLNO_REG(DMA_BASE_PTR,number) = 2;                    // Minor byte transfer count (on minor loop mapping is disabled)

DMA_SLAST_REG(DMA_BASE_PTR,number) = 0;                          // Last source address adjustment

DMA_DOFF_REG(DMA_BASE_PTR,number) = 2;                           // Destination address signed offset

DMA_DLAST_SGA_REG(DMA_BASE_PTR,number) = 0;                       // Adjustment value added to the destination address at the completion of the major iteration count 

if(number <16){

  DMAMUX0_CHCFG(number) = DMAMUX_CHCFG_ENBL_MASK | DMAMUX_CHCFG_SOURCE(40+number);

}else{

   DMAMUX1_CHCFG(number-16) = DMAMUX_CHCFG_ENBL_MASK | DMAMUX_CHCFG_SOURCE(40+(number-16));

   }

enable_int(INT);                     // Enable DMA1 interrupt

[egoodii]

I guess we need to be more explicit.  The setting of 'differential mode' means that IN+ can be 3V, IN- 0V, and the reading will (ideally) be 0x7FFF (positive max).  Conversely, IN+ can be 0V, and IN- at 3V (for a 'net' -3V signal) and the 'ideal' reading will be 0x8000 (max negative).  This is all 'not counting' the errata that limits positive voltages to 31/32 of VrefH (2.9V on a 3V reference like REF3030) or full-scale 0x8400 thru 0x7BFF.

[nicholasf]

The maximum reading I am getting is 0x3EA8 (16040) when the input is near Vdd, which is about half of what it should be; the readings are always half of what they should be.  I get similar results for the 'negative' readings.  I am not sure whether this error is being caused by the ADC or DMA configuration.

[egoodii]

You say 'the input is near Vdd' but 'differential' requires two inputs, so what is the voltage in this test on the OTHER pin?

[nicholasf]

The other input (my Vref) is at 1.5V, halfway between Vdd and Vss.  It appears Vref is unlikely to be the source of the problem, as I am getting half the readings I should throughout the full range of the ADC.

[egoodii]

So if the 'other input' is 1.5V, then the 'differential voltage' measured relative to the 'primary' input is thus at most 1.5V above, or 1.5V below, or half of full scale of +/-3V (a full 16-bit span of 6V), exactly as you report.  To get a 'full scale' differential signal will require one of the inputs at ground, the other at 2.9V.  If you are 'simply biasing' the '-' input at 1.5V, and are NOT 'routing' it along with the '+' input to the ultimate source, in a manner to reject some AC/DC interference, then I strongly suggest you just sample the '+' input in single-ended mode (and invert the MSB to turn it into a sign-bit, so that 'half scale' 0x8000 becomes 0x0000).  I am a little confused by your use of the term Vref -- to the ADC, VrefH is the 'source of scaling reference' against which the analog input voltage (single-ended, or the difference of two inputs in differential mode) is defined.

[nicholasf]

Freescale specifies "VREFH" as the "ADC reference high"; in my case this is 3V, which is also my Vdd.  The "VREFH" is "ADC reference low", and is 0V, and my Vss.  My inputs are both coming from an analog circuit on the PCB (the ultimate source), which has dual-supply amplifiers referred to a 'Vref' level of 1.5V (this is my 'in-') and output a signal called 'signal' which is my in+.  This part of the configuration is working perfectly well.

My issue is that when the voltage difference between 'in-' and 'in+' goes from 0 mV to +1500 mV, the ADC should go from ~0 to ~32000, but the results I am getting are going from ~0 to ~16000.  This is why I say that my problem is that the readings are half what they should be.  There is no problem with voltage offset, or the rest of the circuit, it is simply that the ADC readings are exactly half as large as they should be, through the whole range from 0V-3V (which should be -32000 to 32000 on the ADC, but which I am reading as -16000 to 16000).

[egoodii]

The 'whole range', considered differentially, is 6V, from -3V to +3V.

[AleGuzman · Accepted solution]

Those are actually the expected ADC result. Please be aware that the ADC result is a 16-bit register, therefore the sign will be on D15. For your configuration the ADC result will be around

-32768 when DP = 0 V and DM = 3 V

32767 when DP = 3 V and DM = 0 V

To calculate the LSB you need to use 15-bit resolution because the most significant bit is used like a sign

Using 91.55 uV with 1.5V differential input the ADC result will be ~16384

[egoodii]

"ADC mode = 16 bit differential" means the ADC range is -3V to +3V.