gw.sleep isn't very accurate on a Pro Mini 3.3V (8Mhz)

[TheoL]

Last week a ran into my first 8Mhz problem (see http://forum.mysensors.org/topic/1894/pro-mini-is-giving-a-frequent-time-out-on-dth11-read-out). But thanx to the kind people on this forum that problem is solved.

Now I have another question. I've noticed that my gw.sleep on that same Arduino (mini Pro 3.3V 8Mhz) is far from accurate. If I let the arduino sleep for 900000 milli seconds, I would expect 15 minutes, but it's usually around 16 or 17 minutes (I can monitor that on Domoticz). The longer I let my Arduino sleep the bigger the difference becomes. I have one Soil Moisture sensor that is battery powered (2 AA), that should sleep for 30 minutes. But it's more often 45 minutes or more.

It might be a hardware problem, I'm not sure. But could it possible that gw.sleep uses 16Mhz timing, which could explain the incorrect sleeping time? Otherwise I have no clue what could be wrong. Here's the sketch that is currently running on a battery powered Pro Mini 3.3V.

/**
 * The MySensors Arduino library handles the wireless radio link and protocol
 * between your home built sensors/actuators and HA controller of choice.
 * The sensors forms a self healing radio network with optional repeaters. Each
 * repeater and gateway builds a routing tables in EEPROM which keeps track of the
 * network topology allowing messages to be routed to nodes.
 *
 * Created by Henrik Ekblad <henrik.ekblad@mysensors.org>
 * Copyright (C) 2013-2015 Sensnology AB
 * Full contributor list: https://github.com/mysensors/Arduino/graphs/contributors
 *
 * Documentation: http://www.mysensors.org
 * Support Forum: http://forum.mysensors.org
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License
 * version 2 as published by the Free Software Foundation.
 *
 *******************************
 *
 * REVISION HISTORY
 * Version 1.0 - Henrik EKblad
 * 
 * DESCRIPTION
 * This sketch provides an example how to implement a humidity/temperature
 * sensor using DHT11/DHT-22 
 * http://www.mysensors.org/build/humidity
 */
 
/**
  * Include necessary libraries
  */
 
#include <SPI.h>
#include <MySensor.h>  
#include <dht8mf.h>  

/*
 * Define sketch constants
 */
#define CHILD_ID_HUM 0
#define CHILD_ID_TEMP 1
#define DHT11_PIN 3
#define BATTERY_MEASURE_CYCLE 48 // every 12 hours. Just for test purposes.
#define BATTERY_SENSE_PIN  A2 // determine battery percentage an A2
#define SENSOR_READING_DELAY  2000
unsigned long SLEEP_TIME = 900000;  // check every 15 minutes angd go back to sleep.

MySensor gw;
dht8mf dht;
double lastTemp;
double lastHum;
int oldBatteryPcnt = 0;
int batteryCycleCounter = BATTERY_MEASURE_CYCLE; // Send the status at the first cycle

MyMessage msgHum(CHILD_ID_HUM, V_HUM);
MyMessage msgTemp(CHILD_ID_TEMP, V_TEMP);

/**
 * Sketch initialization code
 */
void setup()  { 
   // use the 1.1 V internal reference
#if defined(__AVR_ATmega2560__)
   analogReference(INTERNAL1V1);
#else
   analogReference(INTERNAL);
#endif  
  gw.begin(); // Initialize MySensor for communication with gate way

  // Send the Sketch Version Information to the Gateway
  gw.sendSketchInfo("Basic roomsensor", "1.0");

  // Register all sensors to gw (they will be created as child devices)
  gw.present(CHILD_ID_HUM, S_HUM);
  gw.present(CHILD_ID_TEMP, S_TEMP);
}

/**
 * Read temperature and hummidity from sensor. Report values to gateway if the
 * differ from last reading.
 */
void readDHT() {
  delay( SENSOR_READING_DELAY ); // Wait the minimum ammout off time, so that DHT will give good reading
  int dhtStatus = dht.read11( DHT11_PIN ); // Change the method to your sensor see dht.h file
  if ( dhtStatus == DHTLIB_OK ) { // Check if Sensor readings are okay
      if (dht.temperature != lastTemp) {
        lastTemp = dht.temperature;
        gw.send(msgTemp.set(dht.temperature, 1));
      }
      if (dht.humidity != lastHum) {
        lastHum = dht.humidity;
        gw.send(msgHum.set(dht.humidity, 1));
      }
  }
}

void checkBatteryLevel() {
  if ( batteryCycleCounter == BATTERY_MEASURE_CYCLE ) {
    delay(100); // wait a while so that the analog pins can stabelize before reading.

    // get the battery Voltage
    int sensorValue = analogRead( BATTERY_SENSE_PIN );
    int batteryPcnt = sensorValue / 10;

    if (oldBatteryPcnt != batteryPcnt) {
      gw.sendBatteryLevel(batteryPcnt);
      oldBatteryPcnt = batteryPcnt;
    }
    batteryCycleCounter = 1;
  }
  batteryCycleCounter++;
}

void loop() {
  readDHT();
  checkBatteryLevel();
  gw.sleep(SLEEP_TIME); //sleep a bit
}

Just to be sure that it's not the battery power causing the problem, I hooked up the arduino through fdti to my computer. But that gives the same result. The arduino is sleeping much longer than instructed to :)

The <dht8mf.h> library is just the DHT library from the Arduino website. I refactored the name to DHT8Mhz Friendly.

[Dwalt]

I had a sensebender hooked up to a ftdi serial adapter and monitored the serial log for over a week (I was trying to capture rerouting issues). The sketch had the sensor gw.sleep for 60 seconds between readings and I noticed it actually took 62-63 seconds between readings. I assumed it was due to time to initialize the SI7021 and/or radio but that seems a little long for that.

[hek]

Sleeping is done in 8sec cycles. Not sure how much re-sleeping takes but during one minute it wakes up 7 times and goes back to sleep.

3 seconds sounds like a big readout time. @tbowmo, could usage of the internal oscillator affect sleep-time on the sensebender?

[tbowmo]

When sleeping, the internal watchdog timer is used to wakeup the MCU again. This runs off a 128Khz oscillator, which is highly dependent on both temperature and voltage.

See page 605 in the datasheet for the atmega (Figure 35-36).

Looking at the graph, 3V @25 degrees celcius, the oscillator runs at approximately 116Khz, which gives a sleep time of 9 seconds, instead of 8 seconds.

sleep time calculated as follows:

1/(116000/1048576) = 9,039 seconds. (1048576 = 1024 x 1024, which is the clock divider used when sleeping "8" seconds).

[hek]

Thanks for the explanation! Makes sense now.

[fleinze]

I noticed the same issue. My programmed sleep-time is 5 min, but the sensors actually sleeps for about 5:25. First I thought it has to do with the 8s-cycle, but the explaination of @tbowmo seems to be the answer.
Thank you!

[tbowmo]

btw. This sleep time, is independent of the oscillator used to run the MCU (8Mhz / 1Mhz). As the watchdog has it's own oscillator.