RETURN

STUART BALL
f e a t u r e

Analog-to-Digital
Converters
The usual method of bringing analog inputs into a microprocessor is to use an
analog-to-digital converter (ADC). Here are some tips for selecting such a part and
calibrating it to fit your needs.

A
n analog-to-digital converter (ADC) accepts 5V/256 = 0.0195V or 19.5mV
an analog input—a voltage or a current—
and converts it to a digital value that can be Our 8-bit converter represents the analog input as a digital
read by a microprocessor. Figure 1 shows a word. The most significant bit of this word indicates
simple voltage-input ADC. This hypotheti- whether the input voltage is greater than half the reference
cal part has two inputs: a reference and the (2.5V, with a 5V reference). Each succeeding bit represents
signal to be measured. It has one output, an 8-bit digital half the range of the previous bit.
word that represents the input value. Table 1 illustrates this point. Adding the voltages corre-
The reference voltage is the maximum value that the sponding to each set bit in 0010 1100, we get:
ADC can convert. Our example 8-bit ADC can convert values
from 0V to the reference voltage. This voltage range is divid- .625 + .156 + .078 = .859 volts
ed into 256 values, or steps. The size of the step is given by:
The resolution of an ADC is determined by the refer-
Vref/256 ence input and by the word width. The resolution defines
the smallest voltage change that can be measured by the
where Vref is the reference voltage. The step size of the con- ADC. As mentioned earlier, the resolution is the same as the
verter defines the converter’s resolution. For a 5V refer- smallest step size, and can be calculated by dividing the ref-
ence, the step size is: erence voltage by the number of possible conversion values.

10 MAY 2001 Embedded Systems Programming

 ADCs come in various speeds, use different interfaces, and provide
differing degrees of accuracy. The most common types of ADCs
are flash, successive approximation, and sigma-delta.

For the example we’ve been using tors required, they tend to be power Successive approximation converter
so far, the resolution is 19.5mV. This hogs, drawing significant current. A A successive approximation convert-
means that any input voltage below 10-bit flash ADC may consume half an er uses a comparator and counting
19.5mV will result in an output of 0. amp. logic to perform a conversion. The
Input voltages between 19.5mV and A variation on the flash converter is first step in the conversion is to see if
39mV will result in an output of 1. the half-flash, which uses an internal the input is greater than half the ref-
Between 39mV and 58.6mV, the out- digital-to-analog converter (DAC) and erence voltage. If it is, the most sig-
put will be 2. subtraction to reduce the number of nificant bit (MSB) of the output is
Resolution can be improved by internal comparators. Half-flash con- set. This value is then subtracted
reducing the reference input. verters are slower than true flash con- from the input, and the result is
Changing that from 5V to 2.5V gives a verters but faster than other types of checked for one quarter of the refer-
resolution of 2.5/256, or 9.7mV. ADCs. We’ll lump them into the flash ence voltage. This process continues
However, the maximum voltage that converter category. until all the output bits have been set
can be measured is now 2.5V instead
of 5V.
The only way to increase resolution
FIGURE 1 Simple ADC
without reducing the range is to use
an ADC with more bits. A 10-bit ADC
has 210, or 1,024 possible output Input
codes. So the resolution is 5V/1,024,
or 4.88mV; a 12-bit ADC has a 1.22mV Output Bits (8)
resolution for this same reference.
Reference Voltage (Vref)
Types of ADCs
ADCs come in various speeds, use dif-
ferent interfaces, and provide differ- 1
ing degrees of accuracy. The most
common types of ADCs are flash, suc-
.8
Reference Voltage (Vref)

cessive approximation, and sigma-delta.

Flash ADC .6
The flash ADC is the fastest type avail-
able. A flash ADC uses comparators, .4
one per voltage step, and a string of
resistors. A 4-bit ADC will have 16 com- .2
parators, an 8-bit ADC will have 256
comparators. All of the comparator
outputs connect to a block of logic 32 64 96 128 160 192 224 255
that determines the output based on Output Value (Decimal)
which comparators are low and which
are high.
The conversion speed of the flash
TABLE 1 Example conversion, on an 8-bit ADC
ADC is the sum of the comparator
delays and the logic delay (the logic
Bit: Bit 7 Bit 6 Bit 5 Bit 4 Bit 3 Bit 2 Bit 1 Bit 0
RUPERT ADLEY

delay is usually negligible). Flash

Volts: 2.5 1.25 .625 .3125 .156 .078 .039 .0195
ADCs are very fast, but consume enor- Output
mous amounts of IC real estate. Also, Value: 0 0 1 0 1 1 0 0
because of the number of compara-

Embedded Systems Programming MAY 2001 11

ADCs

The primary disadvantage of the sigma-delta converter is speed.

Because the converter works by oversampling the input, the
and accuracy. For instance, while you
conversion takes many clock cycles. can get successive approximation
ADCs that range from 8 to 16 bits, you
won’t find the 16-bit version to be the
fastest in a given family of parts. The
FIGURE 2 ADC comparison fastest flash ADC won’t be the 12-bit
part, it will be a 6- or 8-bit part.
These charts are a snapshot of the
Sigma-Delta current state of the technology. As
Successive Approximation
CMOS processes have improved, suc-
cessive approximation conversion
Flash times have moved from tens of
microseconds to microseconds. Not all
8 16 24
Bits of Resolution
technology improvements affect all
types of converters; CMOS process
improvements speed up all families of
Sigma-Delta converters, but the ability to put
increasingly sophisticated DSP func-
Successive Approximation
tionality on the ADC chip doesn’t
Flash improve successive approximation
converters. DSP functionality does
10 100 1K 10K 100K 1M 10M 100M improve sigma-delta types because it
Maximum Conversion Speed, Hz (Log Scale)
enables better, faster, and more com-
plex filters to be added to the part.

or reset. A successive approximation sampling the input, the conversion Sample and hold
ADC takes as many clock cycles as takes many clock cycles. For a given ADC operation is straightforward
there are output bits to perform a clock rate, the sigma-delta converter is when a DC signal is being converted.
conversion. slower than other converter types. Or, But if the input signal varies by more
to put it another way, for a given con- than one least significant bit (LSB)
Sigma-delta version rate, the sigma-delta converter during the conversion time, the ADC
A sigma-delta ADC uses a 1-bit DAC, fil- requires a faster clock. will produce an incorrect (or at least
tering, and oversampling to achieve Another disadvantage of the sigma- inaccurate) result. One way to
very accurate conversions. The conver- delta converter is the complexity of reduce these errors is to place a low-
sion accuracy is controlled by the input the digital filter that converts the duty pass filter ahead of the ADC. The fil-
reference and the input clock rate. cycle information to a digital output ter parameters are selected to ensure
The primary advantage of a sigma- word. The sigma-delta converter has that the ADC input does not change
delta converter is high resolution. The become more commonly available by more than one LSB within a con-
flash and successive approximation with the ability to add a digital filter or version cycle.
ADCs use a resistor ladder or resistor DSP to the IC die. Another way to handle changing
string. The problem with these is that inputs is to add a sample-and-hold
the accuracy of the resistors directly ADC comparison (S/H) circuit ahead of the ADC.
affects the accuracy of the conversion Figure 2 shows the range of resolu- Figure 3 shows how a sample-and-hold
result. Although modern ADCs use tions available for sigma-delta, succes- circuit works. The S/H circuit has an
very precise, laser-trimmed resistor sive approximation, and flash convert- analog (solid state) switch with a con-
networks, some inaccuracies still per- ers. The maximum conversion speed trol input. When the switch is closed,
sist in the resistor ladders. The sigma- for each type is shown as well. As you the input signal is connected to the
delta converter does not have a resis- can see, the speed of available sigma- hold capacitor and the output of the
tor ladder but instead takes a number delta ADCs reaches into the range of buffer follows the input. When the
of samples to converge on a result. the successive approximation ADCs, switch is open, the input is disconnect-
The primary disadvantage of the but is not as fast as even the slowest ed from the capacitor.
sigma-delta converter is speed. flash ADCs. What the tables do not The figure shows the waveform for
Because the converter works by over- show is the tradeoff between speed S/H operation. A slowly rising signal is

12 MAY 2001 Embedded Systems Programming

ADCs

In a perfect world, the hold capacitor would have no leakage and the
buffer amplifier would have infinite input impedance, Since some impedance is in series
so the output would remain stable forever. with the hold capacitor when sam-
pling, the effect is the same as a low-
pass RC filter. This limits the maxi-
mum frequency that the S/H can
FIGURE 3 Sample and hold acquire. This is called the full power
bandwidth, and is specified in kilohertz
Buffer or megahertz.
As mentioned, the electronic
Input Signal To ADC switch is imperfect and some of the
input signal appears at the output,
Sample/Hold Control
Hold Capacitor even in hold mode. This is called
feedthrough, and is typically specified in
decibels.
The output offset is the voltage dif-
ference between the input and the
output. S/H circuit datasheets typical-
Input Signal ly show a hold mode offset and sample
mode offset in millivolts.

Software
Output Voltage An ADC system that uses a S/H may
have to accommodate the hardware
quirks. In some systems, the software
Sample/Hold Control directly controls the S/H control
(0 = Sample, 1 = Hold) input with a port or register output
bit. Typically, the S/H is placed into
sample mode, and the software must
ensure that the acquisition time
connected to the S/H input. While The ability of an S/H circuit to requirement is met. In some systems,
the control signal is low (sample), the maintain the output in hold mode is this can be accomplished simply by
output follows the input. When the dependent on the quality of the hold leaving the S/H in sample mode until
control signal goes high (hold), dis- capacitor, the characteristics of the a conversion is needed.
connecting the hold capacitor from buffer amplifier (primarily input After the S/H is placed into hold
the input, the output stays at the value impedance), and the quality of the sam- mode, another bit (or a write to an
the input had when the S/H switched ple/hold switch (real electronic switch- address or some other operation)
to hold mode. When the switch closes es have some leakage when open). The starts the ADC. After the conversion is
again, the capacitor charges quickly amount of drift exhibited by the output complete, the software reads the
and the output again follows the when in hold mode is called the droop result. However, a problem may occur
input. Typically, the S/H will be rate, and is specified in millivolts per sec- if any one interrupt (or a worst-case
switched to hold mode just before the ond, millivolts per microsecond, or stackup of interrupts) causes the out-
ADC conversion starts, and switched microvolts per microsecond. put of the S/H circuit to droop by
back to sample mode after the conver- A real S/H circuit also has finite more than one LSB. If this could hap-
sion is complete. input impedance, because the elec- pen, the software may need to disable
In a perfect world, the hold capaci- tronic switch isn’t perfect. This means interrupts before switching the S/H to
tor would have no leakage and the that in sample mode, the hold capaci- hold mode and re-enable them after
buffer amplifier would have infinite tor is charged through some resis- starting the conversion. This ensures
input impedance, so the output would tance. This limits the speed with which that the ADC will complete the con-
remain stable forever. In the real the S/H can acquire an input. The version before the S/H droop occurs.
world, though, the hold capacitor will time that the S/H must remain in sam- Software must also accommodate
leak and the buffer amplifier input ple mode in order to acquire a full- the charge time of the S/H. When the
impedance is finite, so the output level scale input is called the acquisition time, electronic switch closes and connects
will slowly drift down toward ground and is specified in nanoseconds or the input signal to the S/H capacitor,
as the capacitor discharges. microseconds. it takes a finite amount of time for the

16 MAY 2001 Embedded Systems Programming

ADCs

In the real world, though, the hold capacitor will leak and the buffer
amplifier input impedance is finite, so the output level will slowly However, the result is dependent
drift down toward ground as the capacitor discharges. on the value of the 5V supply. If the
supply voltage is high by 1%, it has a
value of 5.05V. Now the value of the
A/D conversion will be:
FIGURE 4 Reference voltage variation
(3.2V x 256)/5.05V = 16210 = A216

So a 1% change in the supply volt-

VREF (Nominal)
Nominal
age causes the conversion result to
change by one count. Typical power
supplies can vary by 2% or 3%, so
Maximum power supply variations can have a sig-
Input Voltage

Minimum nificant effect on the results. Power

supply outputs can frequently vary
with loading, temperature, AC input
variations, and from one supply to the
next.
This brings up an issue that affects
all ADC designs: the accuracy of the
reference. A typical ADC reference
might be nominally 2.5V, but can vary
512 1023
between 2.47V and 2.53V (these values
Output Code (10-bit ADC) are from the data sheet for a real
part). If this is a 10-bit ADC, convert-
ing a 2V input at the extremes of the
capacitor to charge because the tling time must elapse to allow the reference ranges gives the following
switch and whatever source is driving S/H capacitor to charge before the results:
the input both have nonzero imped- A/D conversion can start. The soft-
ances. If the sum of these impedances ware must ensure that this delay takes At Vref = 2.47V,
is large enough, the software may place. Result = (2V x 1,024)/2.47 = 82910
need to add a delay so the hold capac-
itor has time to charge to within one Conversion accuracy At Vref = 2.53V,
LSB of the final value before starting Some microcontrollers, such as the Result = (2V x 1,024)/2.53 = 80910
the conversion. Microchip family, allow you to use one
input pin as a reference voltage. This The variation in the reference volt-
Internal microcontroller is normally tied to some kind of preci- age from part to part can result in an
ADCs sion reference. The value read from output variation of 20 counts. Figure 4
Many microcontrollers contain on- the A/D converter after a conversion shows the effect a reference variation
chip ADCs. Typical devices include the is: has on the ADC result. Although the
Microchip PIC167C7xx family and the percentage of error stays the same
Atmel AT90S4434. Most microcon- (Vin/Vref) x 256 throughout the range, the numerical
troller ADCs are successive approxi- error is of course greater for larger
mation because this gives the best Some microcontrollers use the sup- ADC values.
tradeoff between speed and the cost of ply voltage as a reference. In a 5V sys-
real estate on the microcontroller die. tem, this means that Vref is always 5V. Software calibration
The PIC16C7xx microcontrollers So measuring a 3.2V signal with an 8- Sometimes you need an accurate ref-
contain an 8-bit successive approxima- bit ADC would produce the following erence, more accurate than the prod-
tion ADC with analog input multiplex- result: uct cost will support. When manual
ers. The microcontrollers in this fami- adjustment is out of the question, the
ly have from four to eight channels. (Vin x 256)/Vref software can compensate for refer-
Internal registers control which chan- = (3.2v x 256)/5V ence voltage variations. This is typical-
nel is selected, start of conversion, and = 16310 ly done by providing a known, precise
so on. Once an input is selected, a set- = A316 input, which is used to calibrate the

18 MAY 2001 Embedded Systems Programming

ADCs

This brings up an issue that affects all ADC designs: the accuracy of
the reference. A typical ADC reference might be nominally 2.5V, rected values were exactly right or off
but can vary between 2.47V and 2.53V. by just one. This degree of correction
is significantly better than the origi-
nal variation (10 counts) for a 2V
ADC. This reference can be very pre- division process consists of a series of input, and is all that many applica-
cise (and very expensive) because shift-and-add or shift-and-subtract tions require. If your application
only a few are needed for the produc- operations. can’t stand even this error, then you
tion line. The 2.47V example could be cor- may really need a better reference or
In the 2.47V example we’ve been rected by multiplying the ADC value you may have to resort to a manual
looking at, a precise voltage of 2V by .988. The same thing can be adjustment.
might be input to the ADC. When the achieved by subtracting 1/128 then This calibration technique can also
software reads the ADC, it knows the 1/256 then 1/512 of the initial value. be used to compensate for other sys-
correct value should be 819; the cali- Using the original 2V input example, tem inaccuracies, such as resistor tol-
bration constant is given by 829/819, and doing this with integer math, we erance stackup. If whatever you are
or 1.012. Similarly, the calibration con- get the following: measuring consists of a voltage input,
stant for the 2.53V reference is you can apply the precision voltage to
809/819, or 0.988. 829 – 829/128 – 829/156 – 829/512 that input and do one calibration to
This would seem to imply the need = 829 – 6 – 3 – 1 compensate for reference variations in
for floating-point math to correct the = 819 the ADC and resistor tolerance effects
ADC value. If you are using a proces- in the input conditioning.
sor capable of floating-point, this is an This result corrects the ADC read-
acceptable approach. On simpler ing to 819, which is the ideal value if Calculating and using the calibration
processors, though, you may not have the reference were the nominal 2.5V. value
the execution time or the code space Similarly, values read with the 2.53V The calibration value can be calcu-
available to implement floating-point reference can be corrected by adding lated by reading a known reference
calculations. 1/128 plus 1/256. and then finding which correction
One way to handle the ADC cor- Note that you do not need to factors (binary submultiples) to use.
rection is to use a lookup table. This apply the precise calibration voltage For the example given, the differ-
has the drawbacks of requiring suffi- to the input you are using. You can ence between the ideal and the
cient nonvolatile storage to maintain a use any spare ADC input, so long as worst-case ADC value will never be
lookup value for every possible ADC that ADC uses the reference you more than 1.2%, so there is no point
value—a 1,024-word table for a 10-bit want to calibrate. in starting with one-half or one-
ADC. You need to be sure that your ref- fourth of the original value. The
A voltage reference is fairly close to erence is sufficiently stable over your only values tested and used are
its nominal value—otherwise it would expected operating temperature 1/128, 1/256, and 1/512. You want
not be useful as a reference. Assuming range or the results will only be good to start with something close to the
that your reference is sufficiently sta- near the temperature during calibra- value you expect to see.
ble over your operating temperature, tion. If the temperature stability of Finding the correction factors is
the ADC error will be a constant per- the reference is not good enough, easy with a calculator, but if you have
centage of the value you read from the you will have to get a better reference to calculate it on the fixed-point
ADC. Since the ADC has a finite reso- or break your operating temperature processor you will be using in your
lution, there is no point in attempting range into multiple segments and application, you need an integer-
to correct the ADC error with any pre- use one calibration value for each based approach. Figure 5 shows in
cision greater than 1 LSB. segment. Of course, this implies that flowchart form the algorithm used in
Knowing this, you can simplify the you must have a thermistor or other this example to calculate and use the
ADC correction process. Instead of a means of measuring the tempera- calibration constant. In this method, a
lookup table, you store a value that ture, too. single byte (or word) is used to store
tells the software what (binary) per- This approach does result in the calibration constant. Bit 7 indi-
centage to add or subtract from the rounding errors caused by the trun- cates whether the reference voltage is
ADC reading to correct the error. You cation that occurs when you shift the low (calibration values need to be sub-
can add or subtract 1/8, 1/16, or result. I made a spreadsheet using tracted) or high (calibration values
1/24, all the way down to 1 LSB of the 2.47V example, and in all cases added). Bits 0, 1, and 2 indicate
accuracy. You only need to store a sin- the corrected value was within two whether the 1/128, 1/256, and 1/512
gle calibration constant, and your counts of the ideal value. Most cor- factors are used.

20 MAY 2001 Embedded Systems Programming

ADCs

If the temperature stability of the reference is not good enough, you

will have to get a better reference or break your operating nal computer can then calculate the
temperature range into multiple segments and use one calibration or table values and return
them to the microcontroller via the
calibration value for each segment.
same interface.
If the production equipment also
Of course, you could use a separate You will typically want to put the programs the microcontroller in-cir-
byte for each possible factor, with a processor into some kind of “calibra- cuit, the calibration data can be
fourth byte to indicate whether the tion mode,” possibly by grounding a embedded into the data pro-
reference is high or low. pin. The production test equipment grammed into the flash memory. If
can be programmed to apply a very the reference being calibrated is
Writing the calibration values precise voltage to an analog input and inside the microcontroller, the test
Whether you use a table or a calibra- ground the calibration pin. The micro- equipment may have to first load a
tion constant, how do you get the cal- controller can then enter calibration calibration program into the micro-
ibration values into the system? A key mode, where it reads the reference controller, perform the calibration,
component of any calibration value and calculates the compensation then load the actual application
scheme is the availability of non- value or creates a lookup table. code.
volatile storage. Many microcon- In some cases, you don’t have suffi- Finally, some very small microcon-
trollers have on-chip EEPROM. cient memory to add the calibration trollers simply don’t have enough pins
Calibration is typically performed code to the microcontroller. In this to give any up for calibration. In this
when the circuit board is tested. In a case, you can have the microcon- case, you can often make an output
high volume production environ- troller return the ADC value to an pin do double duty as a calibration
ment, this will probably be done by output pin (serially) or to a group of pin. You pull the pin up with an exter-
some kind of bed-of-nails automatic pins (parallel), where it is read by the nal resistor. The production equip-
test equipment. production test equipment. An exter- ment grounds the pin prior to power-
up to select calibration mode.
The way this works is the micro-
controller powers up with all the pins
in the input state. It reads the calibra-
tion pin before configuring the pin as
an output. If the pin is high, normal
operation is started. If the pin is low, it
must be externally grounded, so the
microcontroller enters calibration
mode. Of course, the output has to be
one that won’t damage anything when
the pin is externally grounded.
Finally, if you are calibrating the
reference applying the precise volt-
age to a spare ADC input, you can use
that input itself to put the system into
calibration mode. Use a resistor to
pull the spare input to the zero-scale
ADC voltage (ground, in the exam-
ples we’ve been using). Then have
the software enter calibration mode
when a voltage over some predeter-
mined threshold (say, two-thirds of
the full-scale voltage) is detected on
the pin.
When selecting a calibration volt-
age, you want to choose the largest
value that will not saturate the ADC
when the reference voltage is at its
lowest possible value. This ensures

24 MAY 2001 Embedded Systems Programming

 ADCs
Although it is sometimes difficult to know which ADC to use for
that you don’t lose accuracy in cal- your application, the wide array of parts available assures
culating the calibration constant (or that you will find one to suit your needs.
table) because of bit rounding
errors. This will typically put the cal-
ibration voltage above 90% of the down by 25%. Of course, the accuracy
full scale value, although you may of the sensor enters into this as well, Stuart Ball is an electrical engineer with
want to choose the nearest standard but that topic is beyond the scope of over 20 years of experience designing
reference voltage to make the this article. embedded systems. He is the author of three
design easy. Although it is sometimes difficult books about embedded systems. The materi-
In some applications, you can get to know which ADC to use for your al in this article is adapted from Stuart’s
around the reference issue by looking application, the wide array of parts latest book, Analog Interfacing to
for a change in the ADC input. You available assures that you will find one Embedded Microprocessors, which was
might be able to watch for an optical to suit your needs. Matching the soft- published earlier this year by Butterworth-
sensor to change by 10% instead of ware to the hardware ensures that you Heinemann. You can contact him at
comparing it to a fixed value, or you will get the accuracy and reliability [email protected] or stuart@stuart-
might watch for a temperature to go your product calls for. esp ball.com.

FIGURE 5 Calculating and using the calibration values

Calculate calibration constant.

Uses two variables: calib const Using the calibration constant
and temp Uses one variable, corrected ADC

No
(VREF High)
ADC Value > Ideal Value? Corrected ADC = ADC Value

Yes No
(VREF Low) (VREF High)
Bit 7 of Calib Const Set?
Set Temp = 0
Set Calib Const = 80H Set Temp = 0
Yes
(MSB indicates calibration Set Calib Const = 0H
(VREF Low)
values are to be subtracted (Calibration values are
from ADC values) to be added to ADC values)

ADC Value - ADC Value/128 No ADC Value + ADC Value/128 No No No

Bit 0 of Calib Const Set? Bit 0 of Calib Const Set?
> Ideal Value < Ideal Value?

Yes Yes Yes Yes

Temp = ADC Value/128 Temp = ADC Value/128 Corrected ADC Corrected ADC
Calib Const = Calib Const + 1 Calib Const = Calib Const + 1 = Corrected ADC - ADC Value/128 = Corrected ADC + ADC Value/128

ADC Value - Temp No ADC Value + Temp No No No

Bit 1 of Calib Const Set? Bit 1 of Calib Const Set?
- ADC Value/256 > Ideal Value? + ADC Value/256 < Ideal Value?

Yes Yes Yes Yes

Temp = Temp + ADC Value/256 Temp = Temp + ADC Value/256 Corrected ADC Corrected ADC
Calib Const = Calib Const + 2 Calib Const = Calib Const + 2 = Corrected ADC - ADC Value/256 = Corrected ADC + ADC Value/256

ADC Value - Temp No ADC Value + Temp No No No

Bit 2 of Calib Const Set? Bit 2 of Calib Const Set?
- ADC Value/512 > Ideal Value? + ADC Value/512 > Ideal Value?

Yes Yes Yes Yes

Temp = Temp + ADC Value/512 Temp = Temp + ADC Value/512 Corrected ADC Corrected ADC
Calib Const = Calib Const + 4 Calib Const = Calib Const + 4 = Corrected ADC - ADC Value/512 = Corrected ADC + ADC Value/512

On Exit, Calib Const Bits are.

Bit 0: 1 Means use 1/128 factor
Bit 1: 1 Means use 1/256 factor
Bit 2: 1 Means use 1/512 factor
Bit 7: 1 Means subtract, 0 means add

Embedded Systems Programming MAY 2001 25