Very DeepSleep and energy saving on ESP8266 – Part 1: Basics

Intro: For the seasoned ESP user I might be over explaining things here, but apparently some people struggle with basic things around deepsleep, hence my perhaps somewhat overly long explanation.

For battery fed projects, low energy consumption is a must.
The ESP8266 can help with that, using DeepSleep.

Earlier I published a rather simple DeepSleep program, combined with MQTT, but from an energy conserving point of view  there are more things in setting up DeepSleep that can make a difference. There are also couple of other things to take into consideration.

The ideas here are not new. I have gotten some inspiration from Erik Bakke’s OppoverBakke site, even though there was a mistake in his code snippets.

I will just describe what I did to bring my ESP8266 to minimal energy consumption.

1. Pick the proper ESP8266 board.
You want your ESP board as lean as possible. No other power hungry other components such as FTDI chip or LEDs. This basically excludes most modules such as Wemos D1 mini or NodeMCU and only leaves the bare chips such as ESP-03, ESP-04 and ESP12 on a carrier board. The ESP12 is the easiest though as the ESP-03 and ESP-04 for instance do not have the GPIO16 or RST pin broken out to a side connector, but to a pad on the PCB. The ESP M2 and ESP M3 can be used too as they have gpio16 broken out
The ESP8266-01 would be perfect after you remove the LED, but it does not have the DO/GPIO16 pin broken out. If you have young eyes and a steady hand you can try to solder a wire on the GPIO16 pin.

2. Pick the right voltage regulator.
If you are using 2×1.5V AA cells (unlikely) or LiFePO4 battery (3.6V) you could get by without a regulator, but if for instance you are using 3×1.2V NiCd-s or a 3.7V Lipo (that can go up as high as 4.2Volt), you are going to need a regulator. You dont want a regulator that needs a big input voltage in order to maintain 3.3Volt, so you will need a Low drop regulator, a so called ‘LDO’
Possible candidates are:

Type	Quiescent current	Drop Voltage	Max Input Voltage	Max Output	Package
Torex XC6206	1μA ‎	160mV@100mA	6Volt	200mA	SOT23-3, SOT89,USP-6B
Torex XC6204	70μA* ‎	200mV@100mA	10Volt	300mA**	SOT23-3, SOT89,USP-6B
Analog Devices LTC3525-3.3 ‎	7μA ‎	Step up converter ‎	6Volt ‎	60mA at 3.3V from a 1V Input, or 140mA at 3.3V from a 1.8V Input‎
Holtek HT7333	4μA	90mV@40mA	12V	250mA	SOT89,TO92
Holtek HT7833	4μA	360mV@500mA	8.5V	500mA	SOT89,TO92
ME6211	40μA*	100mV@100mA	6.5V	300-500mA	SOT23, SOT89
SPX3819M5	90μA*	340mV@full load	16V	800mA	SOT23-5, SOIC8, DFN8
Microchip MCP1700	1.6μA	178 mV@ 250 mA	6V	250mA	SOT23, SOT89, TO92
Microchip MCP1825	120μA	210mV@ 500 mA	6V	500mA	SOT23, SOT89, TO92
Microchip MCP1826	0.1-220μA	250mV@ 1000 mA	6.5V	1000mA	SOT23, SOT89, TO92
LT-1763	30μA	300mV	20V	500mA	S8
TPS783xx	500nA	130-175mV	6V	150mA	SOT5
TLF80511TFV33	38μA	100mV@100mA	45V	400mA	PG-TO252-3
MCP1825S33	0.1μA	210mV@500mA	6V	500mA	SOT223-5, DDPAK-5. TO220-5
TC1262-3v3AB	70-130μA	650mV@500mA	6V	500mA	TO220-5
LM2937IMPX-3.3/NOPB	2mA	600mV	26(60 peak)	500ma	To220,DPAK,SOT223

*) The SPX3819 also has an enable pin that needs to be pulled HIGH for it to work. In standby mode  its current draw is practically zero. Ditto for the ME6211 A &H types, these have a standby current of 0.1μA. Ditto for the XC6204, that has a standby current of 0.1μA. This will be used in a yet to appear chapter 7 in this series.
**) only the E-H types. Otherwise 150mA

The problem here is that many of the LDO’s that have a very low quiescent current, also have an output current that is on the edge of what the ESP8266 needs at startup.

The MCP1700 as well as the HT7333 are often advised,but they are really at the verge of what an ESP8266 needs in current using WiFi. If the 90mV drop of the HT7333 is important for your project then consider adding a beefy capacitor.

In all honesty I would advise AGAINST using any regulator giving less than 300mA.

Apparently it is possible to use two HT7333 chips in parallel. The HT7833 seems a pretty good choice. It typically needs a 1uF elco on the input and a 2.2uF elco on the output to be stable. A higher value is ofcourse a good idea.

The ME6211 is often used on the Wemos D1 mini board (which does not mean it is the best choice for low-drop/low Iq). Same goes for the SPX3819M5 on the Lolin NodeMCU.

The LM2937IMPX is a bit of a questionable advice as it has  a rather high quiescent current and relatively large voltage drop, but it  can handle fairly high input  voltage….for the rare case that would be needed in a Low Power setup, but for that very rare situation, i am going to leave it there.

3. Pick the right internet connection
Often we pick a DHCP connection by default. A DHCP connection usually  takes longer to establish than a static IP connection. It is therefore best to chose for the latter. With the ESP it is simply a matter of providing the following details (mind you, the IPnrs are just an example)

//We will use static ip
IPAddress ip(192, 168, 1, 35);// 
IPAddress dns(192.168.1.1);
IPAddress gateway(192, 168, 1, 1); //
IPAddress subnet(255, 255, 255, 0); // this one is pretty standard

and then make the connection with

WiFi.config( ip, dns, gateway, subnet );
WiFi.begin( WLAN_SSID, WLAN_PASSWD );

That is not the only thing we can do making the connection. We can try circumvent some oddities that the ESP8266 shows in connecting to the WiFi network:

the ESP8266 persists the network connection information to flash, and then reads it back when it next starts the WiFi function. It does this every time, and it takes at least 1.2 seconds (source Erik Bakke).

The chip does this even when you pass connection information to WiFi.begin(), like:

WiFi.begin( WLAN_SSID, WLAN_PASSWD );

This will actually load the connection information from flash, ignore it and use the values you specify instead, connect to the WiFi and then finally write your values back to flash.

We can disable that with the command WiFi.persistent( false )

There is more we can do to save time though. The Wifi.begin(SSID,PASSWORD) command does a scan for the proper network amongst the ones advertised. It is possible to supply a WiFi channel number and a BSSID to the WiFi.begin() command that takes away the need for a scan.
The ESP does not know the channel number or BSSID, so we need to read that info from the first connection, store it, and submit it at the next connection.
For that we need to replace the Wifi.begin(); statement with a piece of code that retrieves and manages the necessary information.
Fortunately there is such code available, and as I saw it in various places, I would not know  who to credit for it.
We begin with defining a structure:

A structure aka. struct is a way to group variables together, possibly of different types. One can have several instances of a declared structure. The variables within a structure are called members.

We define the following structure:

struct {
  uint32_t crc32;   // 4 bytes
  uint8_t channel;  // 1 byte,   5 in total
  uint8_t ap_mac[6];// 6 bytes, 11 in total
  uint8_t padding;  // 1 byte,  12 in total
} rtcData;

We use a CRC checksum to see if the data we use perhaps got corrupted. We use 1 bte for the channel and 6 bytes for the Accespoint information.
We add one superfluous byte to come to a multiple of 4. (here is more about storing data in rtc memory).

Then on connecting we make a few checks:

1. 1. 1. 1. is the CRC data correct? If not, make a normal connection.
         2. if after 100 tries it is not working, we reset the WiFi and go make a normal connection
         3. If  after 600 tries still no connection, we put the ESP8266 to sleep and will try later again.

We do need a routine to do the CRC-check, but that’s easily done.
Once we made the WiFi connection, we can connect to where we need to go: an MQTT broker or make an HTTP request for instance.

4. Make sure you do not connect to WiFi before you have to.
Making the connection at the right moment is important though. You do not want to waste energy on your 2.4 GHz signal when you are still reading sensors. As it is, the ESP8266   already has its radio on by default.
So the first thing we need to do upon wake up, is to switch off the radio.
We do that like this:

void setup() {   // disable WiFi, coming from DeepSleep, as we do not need it right away
  WiFi.mode( WIFI_OFF );
  WiFi.forceSleepBegin();
  delay( 1 );

Only once we have done that, we let the program do what it must, like reading sensors, and only then do we start thinking about making the  WiFi connection. Ofcourse we first need to switch the modem/radio on and we do that with:

//Switch Radio back On
WiFi.forceSleepWake();
delay( 1 );

That brings us almost to the end of what we need to do, software wise:
We Switch off the radio, read the sensors, switch on the radio, try make a quick connection and if necessary a regular connection, we then connected to where we want to go (Thingspeak,  MQTTbroker, whatever)and then it is finally time to put the ESP8266 into DeepSleep.

5.Putting the ESP8266 into deepsleep
The simplest form of a DeepSleep program is to do the stuff you need to do  (read yr sensors,make connections) and then call:
ESP.deepSleep(SLEEPTIME)
There is however more that we can do. remember that I said that on wake up the radio is on by default? Well we can give the ESP8266 a wake up instruction so it knows what to do when waking up. We can already tell the ESP8266 that it should NOT switch on the radio by default. We do this with:

ESP.deepSleep(SLEEPTIME, WAKE_RF_DISABLED);

I know that in an earlier step we started with switching the radio off and strictly speaking that step would no longer be necessary once we use the wake up instruction, but I left it in anyway, just to make sure/for illustration. You could opt to leave it out though.
Truthfully there is a bit more superfluous code. e.g. it is not really necessary to switch off the WiFi immediately before DeepSleep.

Only one thing remains now: How long will we put the ESP8266 to sleep?

That ofcourse is a very personal decision, but lets see how long we CAN put the ESP to sleep.

The max deepsleep USED to be max 71 minutes. That was simply because the system_deep_sleep parameter was defined as an unsigned 32 bit integer. The max value of a 32bit unsigned integer is 0xFFFFFFFF, that is 71 min in usecs.
Since the 2.4.1 core, it is a 64 byte unsigned integer. That comes to 18446744073709551615 us or 5,124,095,576 hrs or about half a million year. Nobody wants to wait that long, so expressif seem to have capped that to about 3.5 hrs

This capped number is defined in the ESP.deepSleep(Max) parameter. So when I defined my deepsleep call as ESP.deepSleep(ESP.deepSleepMax(),WAKE_RF_DISABLED );, my ESP went to sleep for 3hrs and 9 min. I noticed though that this cycle tended to get shorter over time. Also if you define a specific period, e.g. 1 hour, with 3.6e9 (=3.6×10⁹) then you will find that that might not be exactly an hour. in my case it was about 58 min.

Testing on a Wemos D1 mini. Connect RST pin with a jumper tp pin D0, as shown here

You will find a DeepSleep program as described above, here. The only thing you need to add other than your web credentials is some code to read your sensors and some code to send your data to where you want to store it (e.g. Thingspeak or yr own webserver, or a broker). I will be adding such code in several follow up articles.
Upload the code and AFTER your upload, connect D0/gpio16 to the RST pin.

6. Think about your sensors
Your sensors use power as well, even when your ESP8266 is sleeping. Perhaps there is a sensor that does the same as another sensor, but uses less power. Consider feeding yr sensors from a gpio pin that you switch LOW when the sensor is not used. Keep in mind that some sensors need a warm up or settling time.  If you do that, ALWAYS make sure your sensor does not draw more than an ESP8266 pin can deliver (12mA), but chances are that if it does, it isn’t one that you should feed from a battery anyway. If you really have to or want to, you can opt for a FET to switch on the sensor. A very popular ‘sensor’ in battery fed projects is measuring the battery voltage. As the ADC of the ESP8266 only goes up to 1 Volt, some form of resistor voltage divider is necessary. This means a constant drain on the battery, so it is best to chose high values for the divider. If you use a 420k divider, you would be constantly draining 10uA on a full battery. That may not seem much, but it is also what your entire ESP8266 uses in deepsleep. If you are using a pFET for HighSide switching of a voltage divider measuring your battery you will encounter the problem that a full LiPo battery may carry 4.2Volt. It will be nearly impossible to close the pFET with only 3.3V on its Gate. Johan Westlund found a solution for that.

If you are not using a power regulator, for instance when you are using 2×1.5Volt batteries, then it is better to use the internal measurement of the Vcc. Do that with:

ADC_MODE(ADC_VCC);
float vccVolt = ESP.getVcc()/1024.0F;

Another possibility is to switch off the voltage divider with a p-FET.The idea is to only put current to the voltage divider when it is needed. We can do that with a circuit like this:

Through a “HIGH” on the input of the circuit, the BC547 will conduct and pull the gate of the p-FET down. The P-FET will open and allow  voltage to be on the voltage divider. After the measurement, the input should be pulled “LOW” which will block any current to go through the voltage divider. As it is, the  (3.3-0.6)/1000=2.4mA, ONLY when the pin is pulled HIGH. It is possible to increase the  base resistor: the current going through the collector will be roughly 3.3/10.000=33uA. with an hFE of roughly 100, the current through the base would only need to be 0.33uA, so most likely a 68k base resistor would still be OK. The FET needs to be a type with a low RDSon and a Vgs of around -2-3 Volt (remember the Vgs(th) is when the Fet starts to conduct, still not up to full power). The values chosen for the resistors are pretty ‘safe’. Yet they also will consume power. It is  possible to use a higher value. Make the 10k a 100k and you could make the 1k even a 1Mohm, provided the transistor has a decent hFE. The downside of using a 100k gate resistor is that the switch off time takes a bit longer (think ms). A suitable P-FET might be the NDX2301, that is a surface mount though. It has a Vgs  of 1.8 Volt. It will not even need the transistor driver in front of it.

Another possibility is to give the sensors their own voltage regulator, that can be switched off by the ESP8266. The SPX3819 is a suitable one. Connect a GPIO pin to the ENABLE pin of the regulator.  A HIGH (<= 0.25V) will switch it on, a LOW will switch it off. Make sure that the voltage regulator has a lower quiescent current than the peripherals you are switching with it (the SPX3819  has a quiescent current of 90uA, max 16V input, a voltage drop of 340mV and a max current of 800mA). Something similar can be done with the ME6211 (40uA quiescent current).

High-side switching can also be obtained by a PNP transistor, like so:

According to some older expressif documentation the status of the gpio pins is maintained in DeepSleep but can only drive 2uA. It is a bit unclear though in the updated version, saying the state of the pins on DeepSleep is no longer defined (table 1-1), though it seems to then contradict itself (page 5/9).
When I measured myself. I always found the same status of my pins during deepsleep, regardless whether they were HIGH or LOW before deepsleep. Status was as follows:

• D0 0V
• D1 0V
• D2 0V
• D3 1V68
• D4 3v3
• D5 3v3
• D6 3v3
• D7 3v3
• D8 3v3

Ofcourse from some pins this is to be expected (D0,D3,D4) but I had expected D8 (GPIO15) to be LOW.

The results teach us though that if you want to feed a sensor with HighSide switching through a a p-FET or PNP transistor, you’d better use pins D4-D7 to close it during deepsleep.

Some sensors –like the BME280– can be programmed to go to sleep in between measurements. For the BME280 this is the ‘Forced Mode’ if you then have a 1/min reading (Bosch calls this the ‘weather monitoring state’) the power consumption is 0.16uA. Using forced mode is also a good way to prevent/limit internal warming up of the chip, which would give skewed readings.

An ADC like the PCF8591 has a relatively low current use in ‘standbye’ as Vin equal to Vss or Vdd, which is not well possible in situ. (Chapter 14.1, Table 8).

The ubiquitous DHT11 has an operating current: 0.3mA (measuring) and 60uA (standby), as 60uA is still rather high, cutting off the power would be better. In that case however the DHT11 needs a 1 sec warming up period. The HTU21D is a better choice with a 0.14uA sleep current.

7. Do you need to send your values?
Rather than sending every measurement when made, you could opt to store results in RTC/SPIFFS memory and only send them say once a day,  or you could decide that if a new measurement result is say within 5% of the previous reading, there is no need to send it. If you are weary of RAM wear, you could add some FRAM, e.g. the MB85RS64V (8k, Operating current 1.5 mA. Standby current 10 μA) or MB85RC256V (32K, Operating current 200μA. Standby current 27uA) (library). (The 8k uses SPI, the 32k uses I2C).

8. Choose your batteries wisely
Once you get to uA levels, self discharge of the batteries becomes an important competing factor.
Nick Gammon states the following self discharge for a number of batteries:

Type	Capacity mAH	Discharge %/month	Self discharge (uA)
CR1212 (3V)	18	1	0.250
CR1620 (3V)	68	1	0.950
CR2032 (3V)	210	1	3
NiCD AAA (1.2V)	350	20	98
NiMH AAA (1.2V)	900	30	375
NiCd AA (1.2V)	1000	20	270
Alkaline AAA (1.5V)	1250	2	35
NiMH AA (1.2V)	2400	30	1000
Alkaline AA (1.5V)	2890	2	80
Li-Ion (3.7V)	4400	10	600

The LiFePO₄ (Lithium Iron Phosphate) is worth considering as it’s max voltage is 3.6Volt and it’s nominal voltage 3.2Volt with a very flat discharge curve. One could opt to use these without a voltage regulator.

TIt is also possible to use a Lithium Thionyl Chloride battery without a voltage regulator. These batteries give 3.6 Volt…which is really at the top spec for  the ESP8266. These non-rechargeable batteries generally have a very long life, with about 1% self-discharge per year. Keep in mind that in cold temperatures their internal resistance will increase  significantly, so they need to be buffered when used in cold temperatures or your ESP8266 may crash when it needs power to connect to WiFi. You cannot do this with say 3×1.2 NiCd cells in series, as they may  be well over 1.2 volt when fully charged and you could end up feeding 4,35 Volts to your ESP8266, which it really does not like. You could use a Lithium Thionyl Chloride battery with a voltage regulator, just to be safe, as long as you pick one with a voltage drop of less than 300mV

Further reduction of Power
The Expressif documentation mentions 8 further ways to reduce power (chapter 4.5) that I have not tried myself and to be honest I have seen some back and forth on the various fora about it not working, about it being fixed and about it still not working.

Connecting D0/GPIO16 to RST
In order for the wake up pulse to work, D0 needs to be connected to RST. The drawback of that is that you need to remove that link when you want to reflash the ESP8266. some people advise connecting the pins via a resistor. That has mixed results though. A better way is to use a Skottky diode -e.g a 1N5819– with the cathode to pin gpio16/D0 and the anode to the RST pin.

In a follow up article, I will publish some code that uses this deepsleep framework to read a sensor and send it to a database.

Part 1 -DeepSleep General
Part 2 -DeepSleep HTTP publishing
Part 3 -DeepSleep MQTT Publishing
Part 4 -DeepSleep MQTT Subscribing
Part 5 -Deepsleep ESP-NOW
Part 6 – Power Down