MSP430FR2311: Erratum ADC65 applies?

Hi Bruce,

Thanks for your post. I will report this to the right person who will investigate your findings. Could you share your test program and screen captures?

The program samples 4 channels with a timer trigger and MSC=0, so each timer pulse triggers one sample cyclically. There's a compile-time constant which says whether to use CONSEQ=0 or 3. With CONSEQ=0, the ISR manipulates INCH directly to provide the channel cycling.

The observation is that with CONSEQ=0, the idle time between samples is about 20uA (mostly REFO), as though the ADC has actually stopped. With CONSEQ=3, the idle time between channels 0 and 3 -- between bursts -- is about 20uA, but between the other channels -- mid-burst -- it's more like 100uA. The suggestion is that the ADC doesn't really shut down mid-burst.

I suppose this could have been happening even back in the F2 series, but before the FR2 series there was some solution (multiple MEMs, DMA, DTC) to cycle channels in a true burst, so there was no occasion to do triggering this way, so I never looked.

My notes (Nov 2018) say:

------------------------------------

Notes on the figures:

1. The ISR includes a “blip” using __delay_cycles. Each ISR spins for 1ms. The end-of-sequence ISR spins for an additional 5ms. Without these some of the ISRs were too short to be detected.

2. The Y scales are different, since the CONSEQ=3 spikes were higher. There's probably a Truth in there somewhere.

The measurements were obtained using an MSP430FR2311 (TSSOP) soldered to a breakout board, recorded using an LPM01A Energy Monitor board (V2.0), STM32CubeMonitor-Power software (v1.0.3).

------------------------------------

Here's the trace with USE_CONSEQ=0:

and this is with USE_CONSEQ=3:

Here is the program I used. The macro USE_CONSEQ is what makes the difference.

///
//  adc-2311a
//  See how much power sequence-of-channels takes.
//
#include <msp430.h> 
#include <stdint.h>
#define USE_REF         0       // Internal ref costs 60uA?
#define USE_CONSEQ      3       // CONSEQ=3
#define ADC_NCHAN       4       // See also ADCINCH_3 below
uint16_t ADC_Result[ADC_NCHAN];
volatile uint16_t i;
int
main(void)
{
    WDTCTL = WDTPW | WDTHOLD;   // stop watchdog timer
    // Configure ADC A0-3 pins
    P1SEL0 |= BIT0 | BIT1 | BIT2 | BIT3;
    P1SEL1 |= BIT0 | BIT1 | BIT2 | BIT3;

    //  LED1/2. I ran a jumper from the P2.1 pin to J10.
    P2OUT &= ~(BIT0|BIT1);             // LED2 off
    P2DIR |=  (BIT0|BIT1);             //  output

    // Everything else input-pulldown
    P1OUT &= ~(BIT4|BIT5|BIT6|BIT7);
    P1REN |=  (BIT4|BIT5|BIT6|BIT7);
    P2OUT &= ~(BIT2|BIT3|BIT4|BIT5|BIT6|BIT7);
    P2REN |=  (BIT2|BIT3|BIT4|BIT5|BIT6|BIT7);

    PM5CTL0 &= ~LOCKLPM5;       //  Engage GPIOs

    // Configure ADC
    ADCCTL0 = ADCSHT_2 | (0*ADCMSC) | ADCON;  // 16ADCclks, ADC ON
#if USE_CONSEQ == 0   // Check explicitly to avoid typos
    ADCCTL1 = ADCSHP | ADCSHS_2 | ADCCONSEQ_0 | ADCSSEL_0; // MODCLK, samp timer, TB1.1 trig, CONSEQ=0
#elif USE_CONSEQ == 1
    ADCCTL1 = ADCSHP | ADCSHS_2 | ADCCONSEQ_1 | ADCSSEL_0; // MODCLK, samp timer, TB1.1 trig, CONSEQ=1
#elif USE_CONSEQ == 3
    ADCCTL1 = ADCSHP | ADCSHS_2 | ADCCONSEQ_3 | ADCSSEL_0; // MODCLK, samp timer, TB1.1 trig, CONSEQ=3
#else
#error "Typo alert: USE_CONSEQ"
#endif
    ADCCTL2 |= ADCSR;               // 50kHz is more than enough
    ADCMCTL0 |= ADCINCH_3 | ADCSREF_0;  // A0-3(EoS); Vref=Vcc
#if USE_REF
    ADCMCTL0 |= ADCSREF_1;          // Vref=1.5V (we "know" ADCSREF_0==0)
#endif
    ADCIE |= ADCIE0;                // Enable ADC conv complete interrupt

    //  Trigger from TB1.1 (ignore typos in SLASE58C Fig 6-4)
#define PERIOD  (32768U/4/ADC_NCHAN) // Full group at 4Hz, pulse at 16Hz
    TB1CCTL1 = OUTMOD_3;             //  Set/Reset for short pulse
    TB1CCR1 = PERIOD-10;             //  short duty for no particular reason
    TB1CCR0 = PERIOD-1;              // Up Top
    TB1CTL = TBSSEL_1 | MC_0 | ID_0 | TBCLR; // ACLK, (Up), /1 (clear)

#if USE_REF         // extra 60uA
    // Configure reference
    PMMCTL0_H = PMMPW_H;            // Unlock the PMM registers
    PMMCTL2 |= INTREFEN;            // Enable internal reference
    __delay_cycles(400);            // Delay for reference settling
    __no_operation();
#endif

    i = ADC_NCHAN-1;                // ADCINCH_3
    ADCCTL0 |= ADCENC;              // Enable
    TB1CTL |= MC_1;                 // Start (Up)
    _EINT();
    while (1)
    {
        LPM3;                       // ADC ISR wakes us up every batch, about 4Hz
    }
    /*NOTREACHED*/
    return(0);
}
#pragma vector=ADC_VECTOR
__interrupt void
ADC_ISR(void)
{
    switch(__even_in_range(ADCIV,ADCIV_ADCIFG))
    {
        case ADCIV_ADCIFG:
            ADC_Result[i] = ADCMEM0;
            __delay_cycles(1000);       // Blip the CPU power for the chart
            if(--i >= ADC_NCHAN)        // unsigned underflow? Start over
            {
                __delay_cycles(5000);   // Bigger blip for the chart
                P2OUT ^= BIT0;          // Toggle (LP) LED2
                i = ADC_NCHAN-1;        // Always channels (N-1)..0
                __bic_SR_register_on_exit(LPM3_bits); // Exit LPM0
#if USE_CONSEQ == 1
                ADCCTL0 &= ~ADCENC;     // Disable
                ADCCTL0 |= ADCENC;      // Enable
#endif
            }
#if USE_CONSEQ == 0                 // Hand-type the INCH
            ADCCTL0 &= ~ADCENC;     // Disable
            ADCMCTL0 = (ADCMCTL0 & ~ADCINCH) | i;  // We "know" INCH is the LSb-s
            ADCCTL0 |= ADCENC;      // Enable
#endif
            break;
        case ADCIV_NONE:
        case ADCIV_ADCOVIFG:
        case ADCIV_ADCTOVIFG:
        case ADCIV_ADCHIIFG:        // above high
        case ADCIV_ADCLOIFG:        // below low
        case ADCIV_ADCINIFG:        // in between
        default:
             break;
    }
    return;
}

Hi Bruce,

Sorry for late reply.

I have measured the current using your code and settings, but removing the LED2 connection to exclude it's impact.

The power is measured by EnergyTrace in CCS. Since VCC supply is 3.3V, the current equals to power/3.3V.

Here's the trace with USE_CONSEQ=0:

Here's the trace with USE_CONSEQ=3:

It seems both the current in LPM3 and Active will increase with USE_CONSEQ=3. I will capture the TB1.1 and P2.0 toggle signal along with the current wave further.

I did similar experiments with the G2553 (ADC10) and F5529 (ADC12_A), and as near as I can tell neither of them exhibits this behavior.

Thanks for your findings. It's very helpful. We need put more resources to go through the other devices.

I tried the FR2476. It displayed similar behavior to the FR2311. The intra-burst current was even higher (190uA for FR2476 vs 100uA for the FR2311), in spite of the lowered between-bursts level (2uA for the FR2476 vs 20uA for the FR2311). [They fixed the REFO. Excellent!]

This is an "X" (not an "M") device, so I don't want to read too much into this.

Thanks for your deep investigation.