mirror of
https://github.com/eclipse/upm.git
synced 2025-07-02 01:41:12 +03:00
docs: more header files edited
Signed-off-by: Mihai Tudor Panu <mihai.tudor.panu@intel.com>
This commit is contained in:

committed by
Mihai Tudor Panu

parent
55e8076988
commit
04edb9be04
@ -29,7 +29,7 @@
|
||||
|
||||
namespace upm {
|
||||
/**
|
||||
* @brief HX711 24bit ADC library
|
||||
* @brief HX711 24-bit ADC library
|
||||
* @defgroup hx711 libupm-hx711
|
||||
* @ingroup generic gpio electric
|
||||
*/
|
||||
@ -43,12 +43,12 @@ namespace upm {
|
||||
* @web http://www.dfrobot.com/image/data/SEN0160/hx711_english.pdf
|
||||
* @con gpio
|
||||
*
|
||||
* @brief API for HX711
|
||||
* @brief API for the HX711 Analog-to-Digital Converter
|
||||
*
|
||||
* The HX711 is a precision 24-bit analog-to-digital converter (ADC)
|
||||
* designed for weigh scales and industrial control applications to
|
||||
* HX711 is a precision 24-bit analog-to-digital converter (ADC)
|
||||
* designed for weight scales and industrial control applications to
|
||||
* interface directly with a bridge sensor. This module was tested on
|
||||
* the Intel Galileo Gen2.
|
||||
* the Intel(R) Galileo Gen 2 board.
|
||||
*
|
||||
* @image html hx711.jpeg
|
||||
* @snippet hx711.cxx Interesting
|
||||
@ -56,66 +56,66 @@ namespace upm {
|
||||
class HX711 {
|
||||
public:
|
||||
/**
|
||||
* HX711 module constructor
|
||||
* HX711 constructor
|
||||
*
|
||||
* @param data define the data pin
|
||||
* @param sck define the clock pin
|
||||
* @param gain define the gain factor
|
||||
* Valid values are 128 or 64 for channel A; channel B works with 32 gain factor only
|
||||
* @param data Defines the data pin
|
||||
* @param sck Defines the clock pin
|
||||
* @param gain Defines the gain factor
|
||||
* Valid values are 128 or 64 for channel A; channel B works with a 32-gain factor only
|
||||
*/
|
||||
HX711(uint8_t data, uint8_t sck, uint8_t gain = 128);
|
||||
|
||||
/**
|
||||
* HX711 module Destructor
|
||||
* HX711 destructor
|
||||
*/
|
||||
~HX711();
|
||||
|
||||
/**
|
||||
* Waits for the chip to be ready and returns a reading
|
||||
*
|
||||
* @return raw adc read
|
||||
* @return Raw ADC reading
|
||||
*/
|
||||
unsigned long read();
|
||||
|
||||
/**
|
||||
* Set the gain factor; takes effect only after a call to read()
|
||||
* channel A can be set for a 128 or 64 gain; channel B has a fixed 32 gain
|
||||
* depending on the parameter, the channel is also set to either A or B
|
||||
* @param gain define the gain factor
|
||||
* Sets the gain factor; takes effect only after a call to read()
|
||||
* channel A can be set for a 128 or 64 gain; channel B has a fixed 32-gain
|
||||
* factor depending on the parameter; the channel is also set to either A or B
|
||||
* @param gain Defines the gain factor
|
||||
*/
|
||||
void setGain(uint8_t gain = 128);
|
||||
|
||||
/**
|
||||
* Returns an average reading
|
||||
* @param times define how many times to read
|
||||
* @return the avarage reading
|
||||
* @param times Defines how many reading to do
|
||||
* @return Average reading
|
||||
*/
|
||||
unsigned long readAverage(uint8_t times = 10);
|
||||
|
||||
/**
|
||||
* Returns (readAverage() - OFFSET)
|
||||
* @param times define how many readings to do
|
||||
* @return the current value without the tare weight
|
||||
* @param times Defines how many readings to do
|
||||
* @return Current value without the tare weight
|
||||
*/
|
||||
double getValue(uint8_t times = 10);
|
||||
|
||||
/**
|
||||
* Returns getValue() divided by SCALE
|
||||
* @param times define how many readings to do
|
||||
* @return the raw value divided by a value obtained via calibration
|
||||
* @param times Defines how many readings to do
|
||||
* @return Raw value divided by a value obtained via calibration
|
||||
*/
|
||||
float getUnits(uint8_t times = 1);
|
||||
|
||||
/**
|
||||
* Set the OFFSET value for tare weight
|
||||
* @param times define how many times to read the tare value
|
||||
* Sets the OFFSET value for the tare weight
|
||||
* @param times Defines how many times to read the tare value
|
||||
*/
|
||||
void tare(uint8_t times = 10);
|
||||
|
||||
/**
|
||||
* Set the SCALE value
|
||||
* This value is used to convert the raw data to "human readable" data (measure units)
|
||||
* @param scale value obtained via calibration
|
||||
* Sets the SCALE value
|
||||
* This value is used to convert the raw data to human-readable data (measurement units)
|
||||
* @param scale Value obtained via calibration
|
||||
*/
|
||||
void setScale(float scale = 1.f);
|
||||
private:
|
||||
@ -128,9 +128,9 @@ namespace upm {
|
||||
|
||||
|
||||
/**
|
||||
* Set the OFFSET value
|
||||
* The value that's subtracted from the actual reading (tare weight)
|
||||
* @param scale value obtained via calibration
|
||||
* Sets the OFFSET value
|
||||
* This value is subtracted from the actual reading (tare weight)
|
||||
* @param scale Value obtained via calibration
|
||||
*/
|
||||
void setOffset(long offset = 0);
|
||||
};
|
||||
|
Reference in New Issue
Block a user