diff --git a/lib/HLW8012_1.1.1/src/HLW8012.cpp b/lib/HLW8012_1.1.1/src/HLW8012.cpp
index 086e6dd38c..53397e0957 100644
--- a/lib/HLW8012_1.1.1/src/HLW8012.cpp
+++ b/lib/HLW8012_1.1.1/src/HLW8012.cpp
@@ -22,6 +22,8 @@ along with this program.  If not, see <http://www.gnu.org/licenses/>.
 #include <Arduino.h>
 #include "HLW8012.h"
 
+#include <GPIO_Direct_Access.h>
+
   #ifndef CORE_POST_3_0_0
     #ifdef ESP8266
       #define IRAM_ATTR ICACHE_RAM_ATTR
@@ -74,7 +76,7 @@ hlw8012_mode_t HLW8012::toggleMode() {
     return new_mode;
 }
 
-double HLW8012::getCurrent() {
+float HLW8012::getCurrent() {
 
     // Power measurements are more sensitive to switch offs,
     // so we first check if power is 0 to set _current to 0 too
@@ -89,40 +91,40 @@ double HLW8012::getCurrent() {
     }
 
     const unsigned int current_pulse_width = _current_pulse_width;
-    _current = (current_pulse_width > 0) ? _current_multiplier / current_pulse_width / 2 : 0;
+    _current = (current_pulse_width > 0) ? _current_multiplier / static_cast<float>(current_pulse_width) / 2.0f : 0.0f;
     return _current;
 
 }
 
-unsigned int HLW8012::getVoltage() {
+float HLW8012::getVoltage() {
     if (_use_interrupts) {
         _checkCF1Signal();
     } else if (_mode != _current_mode) {
         _voltage_pulse_width = pulseIn(_cf1_pin, HIGH, _pulse_timeout);
     }
     const unsigned int voltage_pulse_width = _voltage_pulse_width;
-    _voltage = (voltage_pulse_width > 0) ? _voltage_multiplier / voltage_pulse_width / 2 : 0;
+    _voltage = (voltage_pulse_width > 0) ? _voltage_multiplier / static_cast<float>(voltage_pulse_width) / 2.0f : 0.0f;
     return _voltage;
 }
 
-unsigned int HLW8012::getActivePower() {
+float HLW8012::getActivePower() {
     if (_use_interrupts) {
         _checkCFSignal();
     } else {
         _power_pulse_width = pulseIn(_cf_pin, HIGH, _pulse_timeout);
     }
     const unsigned int power_pulse_width = _power_pulse_width;
-    _power = (power_pulse_width > 0) ? _power_multiplier / power_pulse_width / 2 : 0;
+    _power = (power_pulse_width > 0) ? _power_multiplier / static_cast<float>(power_pulse_width) / 2.0f : 0.0f;
     return _power;
 }
 
-unsigned int HLW8012::getApparentPower() {
-    double current = getCurrent();
+float HLW8012::getApparentPower() {
+    float current = getCurrent();
     unsigned int voltage = getVoltage();
     return voltage * current;
 }
 
-unsigned int HLW8012::getReactivePower() {
+float HLW8012::getReactivePower() {
     unsigned int active = getActivePower();
     unsigned int apparent = getApparentPower();
     if (apparent > active) {
@@ -132,15 +134,15 @@ unsigned int HLW8012::getReactivePower() {
     }
 }
 
-double HLW8012::getPowerFactor() {
-    unsigned int active = getActivePower();
-    unsigned int apparent = getApparentPower();
-    if (active > apparent) return 1;
-    if (apparent == 0) return 0;
-    return (double) active / apparent;
+float HLW8012::getPowerFactor() {
+    const float active = getActivePower();
+    const float apparent = getApparentPower();
+    if (active > apparent) return 1.0f;
+    if (apparent == 0) return 0.0f;
+    return active / apparent;
 }
 
-unsigned long HLW8012::getEnergy() {
+float HLW8012::getEnergy() {
 
     // Counting pulses only works in IRQ mode
     if (!_use_interrupts) return 0;
@@ -151,34 +153,35 @@ unsigned long HLW8012::getEnergy() {
     f = N/t (N=pulse count, t = time)
     E = P*t = m*N  (E=energy)
     */
-    return _pulse_count * _power_multiplier / 1000000. / 2;
+    const float pulse_count = _cf_pulse_count;
+    return pulse_count * _power_multiplier / 1000000.0f / 2.0f;
 
 }
 
 void HLW8012::resetEnergy() {
-    _pulse_count = 0;
+    _cf_pulse_count = 0;
 }
 
-void HLW8012::expectedCurrent(double value) {
-    if (_current == 0) getCurrent();
-    if (_current > 0) _current_multiplier *= (value / _current);
+void HLW8012::expectedCurrent(float value) {
+    if (static_cast<int>(_current) == 0) getCurrent();
+    if (static_cast<int>(_current) > 0) _current_multiplier *= (value / _current);
 }
 
-void HLW8012::expectedVoltage(unsigned int value) {
-    if (_voltage == 0) getVoltage();
-    if (_voltage > 0) _voltage_multiplier *= ((double) value / _voltage);
+void HLW8012::expectedVoltage(float value) {
+    if (static_cast<int>(_voltage) == 0) getVoltage();
+    if (static_cast<int>(_voltage) > 0) _voltage_multiplier *= (value / _voltage);
 }
 
-void HLW8012::expectedActivePower(unsigned int value) {
-    if (_power == 0) getActivePower();
-    if (_power > 0) _power_multiplier *= ((double) value / _power);
+void HLW8012::expectedActivePower(float value) {
+    if (static_cast<int>(_power) == 0) getActivePower();
+    if (static_cast<int>(_power) > 0) _power_multiplier *= (value / _power);
 }
 
 void HLW8012::resetMultipliers() {
     _calculateDefaultMultipliers();
 }
 
-void HLW8012::setResistors(double current, double voltage_upstream, double voltage_downstream) {
+void HLW8012::setResistors(float current, float voltage_upstream, float voltage_downstream) {
     if (voltage_downstream > 0) {
         _current_resistor = current;
         _voltage_resistor = (voltage_upstream + voltage_downstream) / voltage_downstream;
@@ -190,33 +193,44 @@ void IRAM_ATTR HLW8012::cf_interrupt() {
     const unsigned long now = micros();
     _power_pulse_width = now - _last_cf_interrupt;
     _last_cf_interrupt = now;
-    _pulse_count++;
+    _cf_pulse_count++;
 }
 
 void IRAM_ATTR HLW8012::cf1_interrupt() {
 
     const unsigned long now = micros();
-
-    if ((now - _first_cf1_interrupt) > _pulse_timeout) {
-
-        unsigned long pulse_width;
+    const unsigned long time_since_first = now - _first_cf1_interrupt;
+
+    // The first few pulses after switching will be unstable
+    // Collect pulses in this mode for some time
+    // On very few pulses, use the last one collected in this period.
+    // On many pulses, compute the average over a longer period to get a more stable reading.
+    // This may also increase resolution on higher frequencies.
+    if (time_since_first > _pulse_timeout) {
+        // Copy value first as it is volatile
+        const unsigned long last_cf1_interrupt = _last_cf1_interrupt;
+        const unsigned long pulse_width = 
+          (last_cf1_interrupt == _first_cf1_interrupt) 
+            ? 0 
+            : (_cf1_pulse_count < 10) 
+              ? (now - last_cf1_interrupt) // long pulses, use the last one as it is probably the most stable one
+              : (time_since_first / _cf1_pulse_count);
         
-        if (_last_cf1_interrupt == _first_cf1_interrupt) {
-            pulse_width = 0;
-        } else {
-            pulse_width = now - _last_cf1_interrupt;
-        }
-
         if (_mode == _current_mode) {
             _current_pulse_width = pulse_width;
         } else {
             _voltage_pulse_width = pulse_width;
         }
-
-        _mode = 1 - _mode;
-        digitalWrite(_sel_pin, _mode);
+        
+        // Copy value first as it is volatile
+        const unsigned char mode = 1 - _mode;
+        DIRECT_pinWrite_ISR(_sel_pin, mode);
+        _mode = mode;
+        // Keep track of when the SEL pin was switched.
         _first_cf1_interrupt = now;
-
+        _cf1_pulse_count = 0;
+    } else {
+        ++_cf1_pulse_count;
     }
 
     _last_cf1_interrupt = now;
@@ -228,13 +242,20 @@ void HLW8012::_checkCFSignal() {
 }
 
 void HLW8012::_checkCF1Signal() {
-    if ((micros() - _last_cf1_interrupt) > _pulse_timeout) {
+    const unsigned long now = micros();
+    if ((now - _last_cf1_interrupt) > _pulse_timeout) {
         if (_mode == _current_mode) {
             _current_pulse_width = 0;
         } else {
             _voltage_pulse_width = 0;
         }
-        toggleMode();
+        // Copy value first as it is volatile
+        const unsigned char mode = 1 - _mode;
+        DIRECT_pinWrite(_sel_pin, mode);
+        _mode = mode;
+        if (_use_interrupts) {
+            _last_cf1_interrupt = _first_cf1_interrupt = now;
+        }
     }
 }
 
diff --git a/lib/HLW8012_1.1.1/src/HLW8012.h b/lib/HLW8012_1.1.1/src/HLW8012.h
index 99ea860537..8454d7f753 100644
--- a/lib/HLW8012_1.1.1/src/HLW8012.h
+++ b/lib/HLW8012_1.1.1/src/HLW8012.h
@@ -79,31 +79,32 @@ class HLW8012 {
             unsigned long pulse_timeout = PULSE_TIMEOUT);
 
         void setMode(hlw8012_mode_t mode);
+
         hlw8012_mode_t getMode();
         hlw8012_mode_t toggleMode();
 
-        double getCurrent();
-        unsigned int getVoltage();
-        unsigned int getActivePower();
-        unsigned int getApparentPower();
-        double getPowerFactor();
-        unsigned int getReactivePower();
-        unsigned long getEnergy(); //in Ws
+        float getCurrent();
+        float getVoltage();
+        float getActivePower();
+        float getApparentPower();
+        float getPowerFactor();
+        float getReactivePower();
+        float getEnergy(); //in Ws
         void resetEnergy();
 
-        void setResistors(double current, double voltage_upstream, double voltage_downstream);
+        void setResistors(float current, float voltage_upstream, float voltage_downstream);
 
-        void expectedCurrent(double current);
-        void expectedVoltage(unsigned int current);
-        void expectedActivePower(unsigned int power);
+        void expectedCurrent(float current);
+        void expectedVoltage(float current);
+        void expectedActivePower(float power);
 
-        double getCurrentMultiplier() { return _current_multiplier; };
-        double getVoltageMultiplier() { return _voltage_multiplier; };
-        double getPowerMultiplier() { return _power_multiplier; };
+        float getCurrentMultiplier() { return _current_multiplier; };
+        float getVoltageMultiplier() { return _voltage_multiplier; };
+        float getPowerMultiplier() { return _power_multiplier; };
 
-        void setCurrentMultiplier(double current_multiplier) { _current_multiplier = current_multiplier; };
-        void setVoltageMultiplier(double voltage_multiplier) { _voltage_multiplier = voltage_multiplier; };
-        void setPowerMultiplier(double power_multiplier) { _power_multiplier = power_multiplier; };
+        void setCurrentMultiplier(float current_multiplier) { _current_multiplier = current_multiplier; };
+        void setVoltageMultiplier(float voltage_multiplier) { _voltage_multiplier = voltage_multiplier; };
+        void setPowerMultiplier(float power_multiplier) { _power_multiplier = power_multiplier; };
         void resetMultipliers();
 
     private:
@@ -112,22 +113,21 @@ class HLW8012 {
         unsigned char _cf1_pin = 0;
         unsigned char _sel_pin = 0;
 
-        double _current_resistor = R_CURRENT;
-        double _voltage_resistor = R_VOLTAGE;
+        float _current_resistor = R_CURRENT;
+        float _voltage_resistor = R_VOLTAGE;
 
-        double _current_multiplier = 0.0; // Unit: us/A
-        double _voltage_multiplier = 0.0; // Unit: us/V
-        double _power_multiplier = 0.0;   // Unit: us/W
+        float _current_multiplier = 0.0; // Unit: us/A
+        float _voltage_multiplier = 0.0; // Unit: us/V
+        float _power_multiplier = 0.0;   // Unit: us/W
 
         unsigned long _pulse_timeout = PULSE_TIMEOUT;    //Unit: us
         volatile unsigned long _voltage_pulse_width = 0; //Unit: us
         volatile unsigned long _current_pulse_width = 0; //Unit: us
         volatile unsigned long _power_pulse_width = 0;   //Unit: us
-        volatile unsigned long _pulse_count = 0;
 
-        double _current = 0;
-        unsigned int _voltage = 0;
-        unsigned int _power = 0;
+        float _current = 0;
+        float _voltage = 0;
+        float _power = 0;
 
         unsigned char _current_mode = HIGH;
         volatile unsigned char _mode = 0;
@@ -136,6 +136,8 @@ class HLW8012 {
         volatile unsigned long _last_cf_interrupt = 0;
         volatile unsigned long _last_cf1_interrupt = 0;
         volatile unsigned long _first_cf1_interrupt = 0;
+        volatile unsigned long _cf_pulse_count = 0;
+        volatile unsigned long _cf1_pulse_count = 0;
 
         void _checkCFSignal();
         void _checkCF1Signal();
diff --git a/src/_P076_HLW8012.ino b/src/_P076_HLW8012.ino
index cd7c5aa349..106dc45f2a 100644
--- a/src/_P076_HLW8012.ino
+++ b/src/_P076_HLW8012.ino
@@ -41,10 +41,10 @@ int StoredTaskIndex = -1;
 uint8_t p076_read_stage = 0;
 unsigned long p076_timer = 0;
 
-double p076_hcurrent = 0.0f;
-unsigned int p076_hvoltage = 0;
-unsigned int p076_hpower = 0;
-unsigned int p076_hpowfact = 0;
+float p076_hcurrent = 0.0f;
+float p076_hvoltage = 0;
+float p076_hpower = 0;
+float p076_hpowfact = 0;
 
 #define P076_Custom       0
 
@@ -450,12 +450,12 @@ boolean Plugin_076(uint8_t function, struct EventStruct *event, String &string)
       }
 
       if (command.equalsIgnoreCase(F("hlwcalibrate"))) {
-        unsigned int CalibVolt = 0;
-        double CalibCurr = 0;
-        unsigned int CalibAcPwr = 0;
-        if (validUIntFromString(parseString(string, 2), CalibVolt)) {
-          if (validDoubleFromString(parseString(string, 3), CalibCurr)) {
-            validUIntFromString(parseString(string, 4), CalibAcPwr);
+        float CalibVolt = 0;
+        float CalibCurr = 0;
+        float CalibAcPwr = 0;
+        if (validFloatFromString(parseString(string, 2), CalibVolt)) {
+          if (validFloatFromString(parseString(string, 3), CalibCurr)) {
+            validFloatFromString(parseString(string, 4), CalibAcPwr);
           }
         }
 #ifndef BUILD_NO_DEBUG