Delta Sigma ADC for AVRs

    Version 1.2
    Uros Platise, uros.platise@ijs.si, MAPP 7. May 1999 (C) Copyright

    Key features:

    • 8 bit precision
    • 1/2 LSB INL, <1 LSB DNL
    • no missing codes
    • over 1 Mohm input impedance
    • acquisition time T_a = 10 ms @ AVR clock 8 MHz - 80000 AVR cycles
    • ultra low cost
    • simple circuit
    • very short software - 91 (S1200) or 110 (S*) words
    • maximum input slew rate 20 mV/T_a
    • no S&H circuit
    • ideal for measuring DC and low frequency voltages, temperatures, ...
    • supports complete Atmel AVR family MCUs

    The introduced Delta Sigma Analog to Digital Converter is built of only five elements: two resistors, capacitor, analog comparator which is a part of AVR and voltage reference of value half of the power supply voltage.

    1. Input

    High impedance input makes this ADC useful in many low cost applications. In its present form, the bandwidth is limited down to 0.3 Hz at AVR clock 8 MHz what makes it suitable for measuring DC or low frequency voltages. Frequency range may be extended using the sample and hold circuit (S&H). The result is proportional to the integral of the input voltage by time during conversion.

    The input range does not achieve full 8 bit scale but approximately 8 LSB less on the end of the scale. This is caused by the two types of error: constant offset and non-linear characteristics of the converter. Therefore the input voltage ranges from 0 to 247 what is 97% of the full scale. Inside that range there are no missing codes and standard deviation is 1/2 of the LSB. The scale constant equals:

    k = Ucc / 255 [V/d]

    and measured voltage Ui is then

    Ui = k x [V]

    where x is ADC result in the range from 0 to 247.

    Input noise can increase the non-linearity of the result over 1 LSB on the borders. That is why it is important to filter the input signal and limit the noise under 40 mV. This ADC is intended for measuring DC and very low frequency voltages so such a filter is definitely not problematic at all.

    2. Schematics

    The schematic design shows pin connections and the four elements with voltage reference. Resistors R1 and R2 are of 1 Mohm each and trimer P1 is 100 kohm. The optimal value of the capacitor C1 follows the equation:

    C1 = -(1/R1+1/R2)*4*dT / log((Urf-eh-Uh)/(Urf-el-Uh))

    Substituting the known variables and setting the R=R1=R2 we get:

    C1 = 9.5/(R f) [nF]

    where R is in Mohm, f in MHz and result C1 in nF. Some precalculated values for R=1 Mohm:

    f = 1 MHz, C1 = 9.5 nF
    f = 2 MHz, C1 = 4.7 nF
    f = 4 MHz, C1 = 2.3 nF
    f = 8 MHz, C1 = 1.2 nF
    f = 12 MHz, C1 = 0.8 nF
    f = 16 MHz, C1 = 0.56 nF

    The value of the capacitor should be rounded to the nearest available element.
    The trimmer resistor P1 is used to eliminate the linear error caused by the difference between resistors R1 and R2 and by the output voltages (Voh and Vol) of the digital output pin PB2. The same parts also produce constant offset which is compensated internally by the software constant during calibration process.

    For testing purposes, voltage reference was built of LM335-2.5 zener diode. But this is not always necessary. At certain conditions the cost of the voltage reference may be reduced down to simple resistor network.


    3. Calibration

    Supply two constant voltage sources to the input. Good choice is zero volts (0 V) and the voltage close to the maximum range (for example 4 V when Ucc = 5 V). Then rotate the trimmer resistor P1 and monitor the results until you get the constant offsets on both levels. Then check the other values such as: 2.5 V (128), 1.25 V (64), 3.75 V (192). Update the constant voltage offset error by means of macro DS_ADC_OFFSET_VOLTAGE, rebuild the code and programme the AVR MCU.
    Offset value compensates internal systematic error and voltage reference error. Typical values are 3 to 5 if voltage reference is inside of the +-1 LSB. By default is set to 4.

    4. Test Results

    The following two pictures represents test results using the AVR AT90S8515 running at 8 MHz. The first graph shows the U_adc(U_in) and the second absolute error U_adc - U_in. Number of samples per voltage for both graphs is 1. The absolute error is under 1 LSB (20 mV) on full input range from 0 to 4.86 V.





    Note: all voltages are in volts.

    Home