mirror of
https://github.com/espressif/esp-idf.git
synced 2024-10-05 20:47:46 -04:00
353 lines
14 KiB
C
353 lines
14 KiB
C
/*
|
|
* SPDX-FileCopyrightText: 2015-2023 Espressif Systems (Shanghai) CO LTD
|
|
*
|
|
* SPDX-License-Identifier: Apache-2.0
|
|
*/
|
|
|
|
#include <stdint.h>
|
|
#include "esp_types.h"
|
|
#include "esp_err.h"
|
|
#include "esp_check.h"
|
|
#include "assert.h"
|
|
#include "hal/efuse_ll.h"
|
|
#include "hal/adc_types.h"
|
|
#include "driver/adc_types_legacy.h"
|
|
#include "esp_adc_cal_types_legacy.h"
|
|
|
|
/* ----------------------------- Configuration ------------------------------ */
|
|
#ifdef CONFIG_ADC_CAL_EFUSE_TP_ENABLE
|
|
#define EFUSE_TP_ENABLED 1
|
|
#else
|
|
#define EFUSE_TP_ENABLED 0
|
|
#endif
|
|
|
|
#ifdef CONFIG_ADC_CAL_EFUSE_VREF_ENABLE
|
|
#define EFUSE_VREF_ENABLED 1
|
|
#else
|
|
#define EFUSE_VREF_ENABLED 0
|
|
#endif
|
|
|
|
#ifdef CONFIG_ADC_CAL_LUT_ENABLE
|
|
#define LUT_ENABLED 1
|
|
#else
|
|
#define LUT_ENABLED 0
|
|
#endif
|
|
|
|
/* ESP32s with both Two Point Values and Vref burned into eFuse are required to
|
|
* also also burn the EFUSE_BLK3_PART_RESERVE flag. A limited set of ESP32s
|
|
* (not available through regular sales channel) DO NOT have the
|
|
* EFUSE_BLK3_PART_RESERVE burned. Moreover, this set of ESP32s represents Vref
|
|
* in Two's Complement format. If this is the case, modify the preprocessor
|
|
* definitions below as follows...
|
|
* #define CHECK_BLK3_FLAG 0 //Do not check BLK3 flag as it is not burned
|
|
* #define VREF_FORMAT 1 //eFuse Vref is in Two's Complement format
|
|
*/
|
|
#define CHECK_BLK3_FLAG 1
|
|
#define VREF_FORMAT 0
|
|
|
|
/* ------------------------------ eFuse Access ----------------------------- */
|
|
#define VREF_MASK 0x1F
|
|
#define VREF_STEP_SIZE 7
|
|
#define VREF_OFFSET 1100
|
|
|
|
#define TP_LOW1_OFFSET 278
|
|
#define TP_LOW2_OFFSET 421
|
|
#define TP_LOW_MASK 0x7F
|
|
#define TP_LOW_VOLTAGE 150
|
|
#define TP_HIGH1_OFFSET 3265
|
|
#define TP_HIGH2_OFFSET 3406
|
|
#define TP_HIGH_MASK 0x1FF
|
|
#define TP_HIGH_VOLTAGE 850
|
|
#define TP_STEP_SIZE 4
|
|
|
|
/* ----------------------- Raw to Voltage Constants ------------------------- */
|
|
#define LIN_COEFF_A_SCALE 65536
|
|
#define LIN_COEFF_A_ROUND (LIN_COEFF_A_SCALE/2)
|
|
|
|
#define LUT_VREF_LOW 1000
|
|
#define LUT_VREF_HIGH 1200
|
|
#define LUT_ADC_STEP_SIZE 64
|
|
#define LUT_POINTS 20
|
|
#define LUT_LOW_THRESH 2880
|
|
#define LUT_HIGH_THRESH (LUT_LOW_THRESH + LUT_ADC_STEP_SIZE)
|
|
#define ADC_12_BIT_RES 4096
|
|
|
|
/* ------------------------ Characterization Constants ---------------------- */
|
|
static const uint32_t adc1_tp_atten_scale[4] = {65504, 86975, 120389, 224310};
|
|
static const uint32_t adc2_tp_atten_scale[4] = {65467, 86861, 120416, 224708};
|
|
static const uint32_t adc1_tp_atten_offset[4] = {0, 1, 27, 54};
|
|
static const uint32_t adc2_tp_atten_offset[4] = {0, 9, 26, 66};
|
|
|
|
static const uint32_t adc1_vref_atten_scale[4] = {57431, 76236, 105481, 196602};
|
|
static const uint32_t adc2_vref_atten_scale[4] = {57236, 76175, 105678, 197170};
|
|
static const uint32_t adc1_vref_atten_offset[4] = {75, 78, 107, 142};
|
|
static const uint32_t adc2_vref_atten_offset[4] = {63, 66, 89, 128};
|
|
|
|
//20 Point lookup tables, covering ADC readings from 2880 to 4096, step size of 64
|
|
static const uint32_t lut_adc1_low[LUT_POINTS] = {2240, 2297, 2352, 2405, 2457, 2512, 2564, 2616, 2664, 2709,
|
|
2754, 2795, 2832, 2868, 2903, 2937, 2969, 3000, 3030, 3060
|
|
};
|
|
static const uint32_t lut_adc1_high[LUT_POINTS] = {2667, 2706, 2745, 2780, 2813, 2844, 2873, 2901, 2928, 2956,
|
|
2982, 3006, 3032, 3059, 3084, 3110, 3135, 3160, 3184, 3209
|
|
};
|
|
static const uint32_t lut_adc2_low[LUT_POINTS] = {2238, 2293, 2347, 2399, 2451, 2507, 2561, 2613, 2662, 2710,
|
|
2754, 2792, 2831, 2869, 2904, 2937, 2968, 2999, 3029, 3059
|
|
};
|
|
static const uint32_t lut_adc2_high[LUT_POINTS] = {2657, 2698, 2738, 2774, 2807, 2838, 2867, 2894, 2921, 2946,
|
|
2971, 2996, 3020, 3043, 3067, 3092, 3116, 3139, 3162, 3185
|
|
};
|
|
|
|
/* ----------------------- EFuse Access Functions --------------------------- */
|
|
static bool check_efuse_vref(void)
|
|
{
|
|
//Check if Vref is burned in eFuse
|
|
return (efuse_ll_get_adc_vref() != 0) ? true : false;
|
|
}
|
|
|
|
static bool check_efuse_tp(void)
|
|
{
|
|
//Check if Two Point values are burned in eFuse
|
|
if (CHECK_BLK3_FLAG && (efuse_ll_get_blk3_part_reserve() == 0)) {
|
|
return false;
|
|
}
|
|
//All TP cal values must be non zero
|
|
return efuse_ll_get_adc1_tp_low() &&
|
|
efuse_ll_get_adc2_tp_low() &&
|
|
efuse_ll_get_adc1_tp_high() &&
|
|
efuse_ll_get_adc2_tp_high();
|
|
}
|
|
|
|
static inline int decode_bits(uint32_t bits, uint32_t mask, bool is_twos_compl)
|
|
{
|
|
int ret;
|
|
if (bits & (~(mask >> 1) & mask)) { //Check sign bit (MSB of mask)
|
|
//Negative
|
|
if (is_twos_compl) {
|
|
ret = -(((~bits) + 1) & (mask >> 1)); //2's complement
|
|
} else {
|
|
ret = -(bits & (mask >> 1)); //Sign-magnitude
|
|
}
|
|
} else {
|
|
//Positive
|
|
ret = bits & (mask >> 1);
|
|
}
|
|
return ret;
|
|
}
|
|
|
|
static uint32_t read_efuse_vref(void)
|
|
{
|
|
//eFuse stores deviation from ideal reference voltage
|
|
uint32_t ret = VREF_OFFSET; //Ideal vref
|
|
uint32_t bits = efuse_ll_get_adc_vref();
|
|
ret += decode_bits(bits, VREF_MASK, VREF_FORMAT) * VREF_STEP_SIZE;
|
|
return ret; //ADC Vref in mV
|
|
}
|
|
|
|
static uint32_t read_efuse_tp_low(adc_unit_t adc_num)
|
|
{
|
|
//ADC reading at 150mV stored in two's complement format
|
|
uint32_t ret;
|
|
uint32_t bits;
|
|
|
|
if (adc_num == ADC_UNIT_1) {
|
|
ret = TP_LOW1_OFFSET;
|
|
bits = efuse_ll_get_adc1_tp_low();
|
|
} else {
|
|
ret = TP_LOW2_OFFSET;
|
|
bits = efuse_ll_get_adc2_tp_low();
|
|
}
|
|
ret += decode_bits(bits, TP_LOW_MASK, true) * TP_STEP_SIZE;
|
|
return ret; //Reading of ADC at 150mV
|
|
}
|
|
|
|
static uint32_t read_efuse_tp_high(adc_unit_t adc_num)
|
|
{
|
|
//ADC reading at 850mV stored in two's complement format
|
|
uint32_t ret;
|
|
uint32_t bits;
|
|
|
|
if (adc_num == ADC_UNIT_1) {
|
|
ret = TP_HIGH1_OFFSET;
|
|
bits = efuse_ll_get_adc1_tp_high();
|
|
} else {
|
|
ret = TP_HIGH2_OFFSET;
|
|
bits = efuse_ll_get_adc2_tp_high();
|
|
}
|
|
ret += decode_bits(bits, TP_HIGH_MASK, true) * TP_STEP_SIZE;
|
|
return ret; //Reading of ADC at 850mV
|
|
}
|
|
|
|
/* ----------------------- Characterization Functions ----------------------- */
|
|
static void characterize_using_two_point(adc_unit_t adc_num,
|
|
adc_atten_t atten,
|
|
uint32_t high,
|
|
uint32_t low,
|
|
uint32_t *coeff_a,
|
|
uint32_t *coeff_b)
|
|
{
|
|
const uint32_t *atten_scales;
|
|
const uint32_t *atten_offsets;
|
|
|
|
if (adc_num == ADC_UNIT_1) { //Using ADC 1
|
|
atten_scales = adc1_tp_atten_scale;
|
|
atten_offsets = adc1_tp_atten_offset;
|
|
} else { //Using ADC 2
|
|
atten_scales = adc2_tp_atten_scale;
|
|
atten_offsets = adc2_tp_atten_offset;
|
|
}
|
|
//Characterize ADC-Voltage curve as y = (coeff_a * x) + coeff_b
|
|
uint32_t delta_x = high - low;
|
|
uint32_t delta_v = TP_HIGH_VOLTAGE - TP_LOW_VOLTAGE;
|
|
//Where coeff_a = (delta_v/delta_x) * atten_scale
|
|
*coeff_a = (delta_v * atten_scales[atten] + (delta_x / 2)) / delta_x; //+(delta_x/2) for rounding
|
|
//Where coeff_b = high_v - ((delta_v/delta_x) * high_x) + atten_offset
|
|
*coeff_b = TP_HIGH_VOLTAGE - ((delta_v * high + (delta_x / 2)) / delta_x) + atten_offsets[atten];
|
|
}
|
|
|
|
static void characterize_using_vref(adc_unit_t adc_num,
|
|
adc_atten_t atten,
|
|
uint32_t vref,
|
|
uint32_t *coeff_a,
|
|
uint32_t *coeff_b)
|
|
{
|
|
const uint32_t *atten_scales;
|
|
const uint32_t *atten_offsets;
|
|
|
|
if (adc_num == ADC_UNIT_1) { //Using ADC 1
|
|
atten_scales = adc1_vref_atten_scale;
|
|
atten_offsets = adc1_vref_atten_offset;
|
|
} else { //Using ADC 2
|
|
atten_scales = adc2_vref_atten_scale;
|
|
atten_offsets = adc2_vref_atten_offset;
|
|
}
|
|
//Characterize ADC-Voltage curve as y = (coeff_a * x) + coeff_b
|
|
//Where coeff_a = (vref/4096) * atten_scale
|
|
*coeff_a = (vref * atten_scales[atten]) / (ADC_12_BIT_RES);
|
|
*coeff_b = atten_offsets[atten];
|
|
}
|
|
|
|
/* ------------------------ Conversion Functions --------------------------- */
|
|
static uint32_t calculate_voltage_linear(uint32_t adc_reading, uint32_t coeff_a, uint32_t coeff_b)
|
|
{
|
|
//Where voltage = coeff_a * adc_reading + coeff_b
|
|
return (((coeff_a * adc_reading) + LIN_COEFF_A_ROUND) / LIN_COEFF_A_SCALE) + coeff_b;
|
|
}
|
|
|
|
//Only call when ADC reading is above threshold
|
|
static uint32_t calculate_voltage_lut(uint32_t adc, uint32_t vref, const uint32_t *low_vref_curve, const uint32_t *high_vref_curve)
|
|
{
|
|
//Get index of lower bound points of LUT
|
|
uint32_t i = (adc - LUT_LOW_THRESH) / LUT_ADC_STEP_SIZE;
|
|
|
|
//Let the X Axis be Vref, Y axis be ADC reading, and Z be voltage
|
|
int x2dist = LUT_VREF_HIGH - vref; //(x2 - x)
|
|
int x1dist = vref - LUT_VREF_LOW; //(x - x1)
|
|
int y2dist = ((i + 1) * LUT_ADC_STEP_SIZE) + LUT_LOW_THRESH - adc; //(y2 - y)
|
|
int y1dist = adc - ((i * LUT_ADC_STEP_SIZE) + LUT_LOW_THRESH); //(y - y1)
|
|
|
|
//For points for bilinear interpolation
|
|
int q11 = low_vref_curve[i]; //Lower bound point of low_vref_curve
|
|
int q12 = low_vref_curve[i + 1]; //Upper bound point of low_vref_curve
|
|
int q21 = high_vref_curve[i]; //Lower bound point of high_vref_curve
|
|
int q22 = high_vref_curve[i + 1]; //Upper bound point of high_vref_curve
|
|
|
|
//Bilinear interpolation
|
|
//Where z = 1/((x2-x1)*(y2-y1)) * ( (q11*x2dist*y2dist) + (q21*x1dist*y2dist) + (q12*x2dist*y1dist) + (q22*x1dist*y1dist) )
|
|
int voltage = (q11 * x2dist * y2dist) + (q21 * x1dist * y2dist) + (q12 * x2dist * y1dist) + (q22 * x1dist * y1dist);
|
|
voltage += ((LUT_VREF_HIGH - LUT_VREF_LOW) * LUT_ADC_STEP_SIZE) / 2; //Integer division rounding
|
|
voltage /= ((LUT_VREF_HIGH - LUT_VREF_LOW) * LUT_ADC_STEP_SIZE); //Divide by ((x2-x1)*(y2-y1))
|
|
return (uint32_t)voltage;
|
|
}
|
|
|
|
static inline uint32_t interpolate_two_points(uint32_t y1, uint32_t y2, uint32_t x_step, uint32_t x)
|
|
{
|
|
//Interpolate between two points (x1,y1) (x2,y2) between 'lower' and 'upper' separated by 'step'
|
|
return ((y1 * x_step) + (y2 * x) - (y1 * x) + (x_step / 2)) / x_step;
|
|
}
|
|
|
|
/* ------------------------- Public API ------------------------------------- */
|
|
esp_err_t esp_adc_cal_check_efuse(esp_adc_cal_value_t source)
|
|
{
|
|
if (source == ESP_ADC_CAL_VAL_EFUSE_TP) {
|
|
return (check_efuse_tp()) ? ESP_OK : ESP_ERR_NOT_SUPPORTED;
|
|
} else if (source == ESP_ADC_CAL_VAL_EFUSE_VREF) {
|
|
return (check_efuse_vref()) ? ESP_OK : ESP_ERR_NOT_SUPPORTED;
|
|
} else {
|
|
return ESP_ERR_INVALID_ARG;
|
|
}
|
|
}
|
|
|
|
esp_adc_cal_value_t esp_adc_cal_characterize(adc_unit_t adc_num,
|
|
adc_atten_t atten,
|
|
adc_bits_width_t bit_width,
|
|
uint32_t default_vref,
|
|
esp_adc_cal_characteristics_t *chars)
|
|
{
|
|
//Check parameters
|
|
assert((adc_num == ADC_UNIT_1) || (adc_num == ADC_UNIT_2));
|
|
assert(chars != NULL);
|
|
assert(bit_width < ADC_WIDTH_MAX);
|
|
|
|
//Check eFuse if enabled to do so
|
|
bool efuse_tp_present = check_efuse_tp();
|
|
bool efuse_vref_present = check_efuse_vref();
|
|
esp_adc_cal_value_t ret;
|
|
|
|
if (efuse_tp_present && EFUSE_TP_ENABLED) {
|
|
//Characterize based on Two Point values
|
|
uint32_t high = read_efuse_tp_high(adc_num);
|
|
uint32_t low = read_efuse_tp_low(adc_num);
|
|
characterize_using_two_point(adc_num, atten, high, low, &chars->coeff_a, &chars->coeff_b);
|
|
ret = ESP_ADC_CAL_VAL_EFUSE_TP;
|
|
} else if (efuse_vref_present && EFUSE_VREF_ENABLED) {
|
|
//Characterize based on eFuse Vref
|
|
uint32_t vref = read_efuse_vref();
|
|
characterize_using_vref(adc_num, atten, vref, &chars->coeff_a, &chars->coeff_b);
|
|
ret = ESP_ADC_CAL_VAL_EFUSE_VREF;
|
|
} else {
|
|
//Characterized based on default Vref
|
|
characterize_using_vref(adc_num, atten, default_vref, &chars->coeff_a, &chars->coeff_b);
|
|
ret = ESP_ADC_CAL_VAL_DEFAULT_VREF;
|
|
}
|
|
|
|
//Initialized remaining fields
|
|
chars->adc_num = adc_num;
|
|
chars->atten = atten;
|
|
chars->bit_width = bit_width;
|
|
chars->vref = (EFUSE_VREF_ENABLED && efuse_vref_present) ? read_efuse_vref() : default_vref;
|
|
//Initialize fields for lookup table if necessary
|
|
if (LUT_ENABLED && atten == ADC_ATTEN_DB_11) {
|
|
chars->low_curve = (adc_num == ADC_UNIT_1) ? lut_adc1_low : lut_adc2_low;
|
|
chars->high_curve = (adc_num == ADC_UNIT_1) ? lut_adc1_high : lut_adc2_high;
|
|
} else {
|
|
chars->low_curve = NULL;
|
|
chars->high_curve = NULL;
|
|
}
|
|
return ret;
|
|
}
|
|
|
|
uint32_t esp_adc_cal_raw_to_voltage(uint32_t adc_reading, const esp_adc_cal_characteristics_t *chars)
|
|
{
|
|
assert(chars != NULL);
|
|
|
|
//Scale adc_rading if not 12 bits wide
|
|
adc_reading = (adc_reading << (ADC_WIDTH_BIT_12 - chars->bit_width));
|
|
if (adc_reading > ADC_12_BIT_RES - 1) {
|
|
adc_reading = ADC_12_BIT_RES - 1; //Set to 12bit res max
|
|
}
|
|
|
|
if (LUT_ENABLED && (chars->atten == ADC_ATTEN_DB_11) && (adc_reading >= LUT_LOW_THRESH)) { //Check if in non-linear region
|
|
//Use lookup table to get voltage in non linear portion of ADC_ATTEN_DB_11
|
|
uint32_t lut_voltage = calculate_voltage_lut(adc_reading, chars->vref, chars->low_curve, chars->high_curve);
|
|
if (adc_reading <= LUT_HIGH_THRESH) { //If ADC is transitioning from linear region to non-linear region
|
|
//Linearly interpolate between linear voltage and lut voltage
|
|
uint32_t linear_voltage = calculate_voltage_linear(adc_reading, chars->coeff_a, chars->coeff_b);
|
|
return interpolate_two_points(linear_voltage, lut_voltage, LUT_ADC_STEP_SIZE, (adc_reading - LUT_LOW_THRESH));
|
|
} else {
|
|
return lut_voltage;
|
|
}
|
|
} else {
|
|
return calculate_voltage_linear(adc_reading, chars->coeff_a, chars->coeff_b);
|
|
}
|
|
}
|