Building Ultra Low Power with MicroPython

[volkerjaenisch]

How to achieve the lowest possible power consumption with MicroPython?

Probably there is such a topic already out there - then please give me a pointer.

The above question seems a bit naive since it depends on ... a lot of things.

Motivation:
The project that made me open this topic is building a solar powered low cost digital moisture sensor with WiFi (or LoRa) out of widely available components.

The components are IMHO straight forward:

• ESP32 Dev-Board
• Digital-Moisture sensor
• LiPo Accumulator
• Solar panel
• LiPo solar loader

Problems:

• Hardware:
  • There are ESP boards out there with voltage regulators that draw 10-100 times more current that the ESP in deep sleep
  • Sensors are drawing a lot of power if kept unchecked
• Software:
  • All ESP Controllers are featuring a machine.deepsleep functionality, but each bit of the ESP32 hardware has to be shut down manually by the programmer before entering deepsleep. This fact is not documented.
  • And it is not really clear how to invoke a hardware component and how to set it to sleep. For most hardware components micropython features a "init" method, which is complemented by a "deinit" method. But there is a lack of documentation if e.g. the "deinit" method really shuts the hardware down, or not.
    • And there is e.g. no method to "deinit" for the I2C bus.

Status:
With my current code I am able to quench the current down to 0.4mA on a DFRobot ESP8266 Firebeetle in deep-sleep. But is this the limit?

Goals:
Let us document what hardware in combination with what code achieves the minimal current.

Cheers,
Volker

[davefes]

I have been moving all my solar-powered remote sensors to LoRa.
Also, I replace the AMS1117 with HZ73333 (4-8uA Quiescent Current)

All ESP Controllers are featuring a machine.deepsleep functionality, but each bit of the ESP32 hardware has to be shut down manually by the programmer before entering deepsleep. This fact is not documented.

This is news to me ... which bits of hardware stay powered-up?

Have you read this:
https://github.com/glenn20/upy-esp32-experiments/blob/main/ESPNowvsWifiEnergyUsage/README.md

[karfas]

each bit of the ESP32 hardware has to be shut down manually by the programmer before entering deepsleep. This fact is not documented.
And it is not really clear how to invoke a hardware component and how to set it to sleep.

This is not true. The micropython docs are terse, yes, but other sources of information are

• the ESP-IDF documentation
• data sheets
• source code.

I really have no idea why so many people seem to expect that uPy docs should cover every possible hardware configuration under the sun.
Programming bare metal (or freeRTOS) does require at least minimal knowledge about hardware, protocols, etc.

[peterhinch]

@volkerjaenisch I suggest looking at the FeatherS3. The ESP32-S3 has extremely low deepsleep current (unlike ESP32) and the board is well designed. I haven't managed to get a measurement of deepsleep current for the entire board powered from LiPo, but it's below 1μA. See also this doc from @glenn20 who has done a lot of work on minimising energy use on boot.

[davefes]

3V3 at 20mA, which is the default drive setting. E drop = 0.02A * 30 Ohms = 0.6V drop. I'd check the voltage actually on the sensor.

[volkerjaenisch]

Dear @davefes

Please be gentle with me. My electronics knowledge is sparse and English not my mother language, so many terms you are familiar with are a riddle for me.

1. What do you mean by "drop"?

2. The sensor needs a min. of 3.0 V for the resonant circuit (according to the producer).

3. You are completely right that I should measure the sensor characteristics at the sensor :-). But my aim was to first understand the overall power consumption.

4. I have to clear my other workbench to get a working environment. Current measurements are a bit improvised :-)

5. Let us discuss the sensor and its powering via GPIO in another thread. I will open it when I get a more robust working environment. I may also invite the developer of the sensor so we can use his first hand knowledge. He has lots of customers struggling with ESP32 and low power.

6. @ALL: Many thanks for the support

7. Disclaimer: This is a spare time project. I am the maintainer of the SoMoSe-Micropython open source driver since I own one and needed a driver.

Cheers,
Volker

[davefes]

1. at the default drive current setting they say the source impedance (resistance in this case) is 30 Ohms. So, when you try to draw 20ma you end up with a decreased GPIO drive voltage. 0.6V "lost" across this internal source resistance takes you below the 3.0V min voltage for the sensor. Even at drive level 3 you are going to "lose" or drop 0.3V, which is at least still in spec. Measure the actual voltage on the sensor and verify it is about 2.7V, indicating a 20mA current consumption. Then try drive level 3 and then I can suggest another approach in your new thread.

[volkerjaenisch]

Dear @davefes
Thank you so much for your explanation. Now I got it.

I will open up a new thread concerning the power supply of the SoMoSe Sensor.

To the gods of this forum: Is it OK to discuss such a specific electronic topic here?

Cheers,
Volker

[jimmo]

Is it OK to discuss such a specific electronic topic here?

Yes, of course. Try and keep each top-level discussion MicroPython-relevant, but absolutely it is common for the discussion to go into the details of the electronics.

[volkerjaenisch]

This is news to me ... which bits of hardware stay powered-up?
I do not know. But the current my board is drawing in deep sleep (4mA) is higher than expected.
Therefore I came up with the idea to talk about this topic.

For instance I2C can be initiated by Micropython but there is no corresponding deinit() function as there is for other components like e.g. WLAN or UART. This asymmetry I do not understand.

Cheers,
Volker

[jimmo]

For instance I2C can be initiated by Micropython but there is no corresponding deinit() function as there is for other components like e.g. WLAN or UART. This asymmetry I do not understand.

I forget exactly the conclusion from last time we talked about it, but I think this is something we need to add. It's just a matter of someone putting the time into implementing. For example it's also a good way to let the pins go back to their default state.

[davefes]

With a ESP32-WROOM DevkitC board, 3v3 regulator removed and the supply cut to the CP2102 I am under 1mA in deepsleep.
Provide a link to your actual hardware.

[davefes]

On Aliexpress try searching for:
ESP-32S Breakout Board Converter Module
If you use these I would suggest a dab of heatsink paste on the little square in the middle before mounting or better still reflow the module on to the carrier.

[volkerjaenisch]

Yeah! Got it! Ok, this I can solder.
Sorry for being such a NOOB, and thank you for the support.

[davefes]

I have been building hardware for 60 years and just discovered these particular break-out boards, so I would be surprised that everybody knew about them :)

[volkerjaenisch]

You are welcome.

[glenn20]

Or, there is this monstrosity I built a few years ago when prototyping my own low power esp32 devices... 😉. It's actually not hard to solder onto the bare modules.

I even soldered smd LDOs and caps directly onto the ribbon cable. (BTW. - not a recommended solution, but very compact).

[volkerjaenisch]

each bit of the ESP32 hardware has to be shut down manually by the programmer before entering deepsleep. This fact is not documented.
And it is not really clear how to invoke a hardware component and how to set it to sleep.

but other sources of information are

* the ESP-IDF documentation

* data sheets

* source code.

But what costs the sentence "If you set your board to deep-sleep, please make sure that all components you have invoked are shut down first." in the documentation.

I really have no idea why so many people seem to expect that uPy docs should cover every possible hardware configuration under the sun. Programming bare metal (or freeRTOS) does require at least minimal knowledge about hardware, protocols, etc.

1. I do not understand why you are reacting so aggressively to my friendly proposal.
2. Your response does not contain any usable information beside the rant : "Use the force read the source".

And yes, you are right. I have only a physics diploma but I have no deep electronics knowledge. And therefore I come from a different context. And we all know how important context is. I am a Programmer for 40+ year starting with machine language on a SHARP PC1500 and I am now developing applications. And my context knowledge is that when I enter a deep-sleep on e.g. my linux-system the OS will shut down the hardware. And I wrongly assumed that MicroPython would do the same thing on a ESP.
The above mentioned additional hint in the deep-sleep documentation would have given me the correct context and spared me and others a lot of red herring chasing.

3. I gave an concrete example that e.g. for I2C there is a deinit() only for certain boards. And your answer is simply:

This is not true.

For that bold and absolute statement you gave zero reason.

Can you please be in the future a bit more constructive.

Cheers,
Volker

[volkerjaenisch]

Dear @glenn20 !

I just ran a lightsleep and deepsleep test on my FeatherS3 with the PPK2:

* Normal idle operation (wifi off): **46mA**

* Light sleep:  **1.3mA**

* Deep sleep: **0.043mA**

That is quite impressive. Many thanks for your demonstration.

Now I only need a distributor that sells/ships a FeatherS3 in/to Germany. Any hints anybody?

Pimoroni where I bought Adrafruit products in the past is "out of stock" and also their product pictures for the FeatherS3 do not show the second Stemma QT-Port. Are there different variants/versions of the FeatherS3?

Cheers,
Volker

[peterhinch]

There are two variants, one with an onboard antenna, the other with a Ufl connector - see this page. Both have two Stemma ports.

[volkerjaenisch]

So the product images are indeed wrong on the pimoroni website. Thank you for clarification.

[volkerjaenisch]

Dear @glenn20

I know that you are using an esp8266 device, but I think the best opportunities for achieving very low power operations are probably with the ESP32 devices (partly due to the availability of modern well-optimised boards like the feathers3 mentioned by Peter) and much more comprehensiive documentation (though often ambiguous).

The esp8266 was a bad buy (never trust amazon - I ordere a DFrobot ESP32 but got a 8266), but since it also utilized the RT9080-33GJ5 I gave it a try.

But as @davefes found out the voltage regulator is not the problem on this board but probably the CP2102.

Just ordered a FeatherS3 for eye watering 34EUR via Amazon.uk. Will be a fortnight to get it delivered.
I will come back to share my experiences.

Cheers,
Volker

[glenn20]

If they are clever, the cp2102 may not be powered when running off lipo.

[glenn20]

Just ordered a FeatherS3 for eye watering 34EUR via Amazon.uk. Will be a fortnight to get it delivered.

I have the good fortune that these devices are built only a few suburbs from where I live ;).

[UnexpectedMaker]

@volkerjaenisch I ship directly to DE from Australia - Either from my website, or from Lectronz (that handle the VAT/Fees for you).
https://lectronz.com/products/feathers3

Issue shipping from Australia is the insane postage prices since Covid that unfortunately will never go back down :(

BTW, Pimoroni have the correct pictures for the FeatherS3 on their site - maybe you were looking at the FeatherS2 instead?
https://shop.pimoroni.com/products/feathers3-esp32-s3?variant=39762596298835

Cheers :)

[volkerjaenisch]

@UnexpectedMaker Thank you so much for your support, I am overwhelmed!
Will look how the Amazon-UK delivery works. Then lectronz I will try as the next option.

The EU CE certification is the new fourth rider of the apocalipse. It brings fear, desease and decay onto us makers.

I developed a device to fotograph and classify rain filter probes for the German Weather Service. All of its components (RPi, industrial camera, power supply, servos, calibrated color card) were CE certified. My added work was some 3D-printed gears, mechanics and housing (and a lot of Software). But the whole thing had to be anew CE certified. Luckily I found a hole - instruments build for scientific research do not need a CE sign. But only for 3 years duration.

(Edited by @jimmo to remove a paragraph that was not appropriate)

Cheers,
Volker

[UnexpectedMaker]

The FeatherS3 should be able to get down to < 30uA deep sleep - but it depends on what you have connected to it, and how you are waking it. I've had it as low as 20uA @ 3.3V, and 26uA at 4.2V, both with nothing connected to it.

But honestly, anything under 50uA it perfect for deep sleep current - The biggest cost is always going to be when the board is awake and the radios are either calibrating, or you are transmitting data. A super low deep sleep current is only helpful if you plan on having the board sleep for days at a time without waking, and then you have the issue of RTC drift with the internal ESP32 RTC :(

[volkerjaenisch]

Dear @UnexpectedMaker !

How are the hardware UARTs located in Pin-Space on the Feather S3?

I am using UART 1 with RX=17 and Tx=18 for my custom logging. Can this be the problem?

Are there guidelines how to debug such a situation? If not please give me hints.

And to state it again: My code runs fine and stable for hours powered by USB. But it fails instantly while battery powered.

The Rx/Tx log from the board is not really readable. Here I need some pointer to improve.

QTTI�PQ�IDLD�DDI��1323
�EZ��
Ma 21 201
214& (BRCQFAUTCRST),bcc4 (SPIABASTCFLASHCBAAT)
SPIQP
0ee
-cdeDIA, clc!k $i1
laad01fae0,leff34
d!!$
01!100,de&0c4
,cad403cc300,lef02a
ef42 403c&4
�QTTI�PQ�Y&DD�DDI��123
�EZ��
Ma 1 2
214& (BRCSFAUTARST),bcc (SPIAFASTCFLAQHCBCCT)
SPIQP
0ee
-adeDIA, clcck di61
lcad3fce30$lef0f14
lcad0403c100,lef0c44
lcad0403cc300,le&02c3
%&42 403c&4
QTTI��Q�IDLD�DDK��103
�EJ��
Ea 21 21
214& (BRCSFCUTCRST),b#a4
0 (QPICFASTCFLASHCBAAT)
SPIQPe%
mcdeDIA, clcck di61
,cad03fce30,lef0f34
dca$
03c100,lefc
,!!d401cc0$leb02c
%&42 401af4
�Q*TI�PS�Y&DD�DDK��1032
�EZ��
Ma 21 201
214
0f (BRCSFCUTCRST),bc#40 (SPICBASTCFLASHCBACT)
SPISP%%
mcdeDIC, clcck di61
lcad03fce0,leb0f34
lca$
0403a30,lefc
dc!d001cc0$$%b2!1
ef42 001f4
QTTK�PQ�IDDD�D�I&D�DDD�DC��EZ��
Ma 23 2021
2140f (BRCSFCUTCRST),bcc40 (SPIAFASTCFLASHCBCCT)
SPISP0ee
-cdeDIA, cdcck di1
lcad03fce30,lef0&34
lcad003c10,lef0!44
d#!$
040!a10,le"02a3
%f42 401cf4
�QTTI�PQ�IDLD�DDI��1023
�EZ��
Ma 21 2021
2340f (BRCQFCUTCRST),bcc40 (SPIAFASTCFLASHCBACT)
SPISPe%
mcdeDIA, cdcck di61
laad01fce30,lef0f34
l!!d
043c10,def0c4
,cad040cc300,lef02c3
e&42 0403cf4
QTTI�PQ�IDLD�DDI��10323
�EZ��
Ma 23 201
2340f (BRCSFCUTCRST),bcc0 (SPICFASTCFDASHCBCCT)
SPISP0ee
m!deDIC, clcck $i61
d!!d03fc%30$$%f
f34
lca$4c100,lef0!44
lcad003cc100,lef02c3
ef42 0cf4
QTTI�PS�YDDD�DDI��1023
�EZ��
Ma2 21 21
240f (BRCQFAUTCRST),bcc (SPIAFASTCFLASHCBCAT)
SPIQP
e%
-adeDIA, clcck di61
,aad03&ce30,lef0f34
lcad003!300,lefc44
,aad0403cc300,lef02c
%&42 0403cf4
QTTI�PQ�YDLD�DDI��123
�EZ��
Ma2 21 2

Cheers,
Volker

[UnexpectedMaker]

That looks like a baud rate issue, there should be no issues using any IO as HW UART pins on the ESP32 - you could just use UART0 that are TX/RX on the headers. They should already be dumping boot info as the ESP32 starts up.

[volkerjaenisch]

The connection diagram:

Sensor -> Feather:
GND -> "-"
VCC -> 3.3V LDO2
SDA -> SDA
SCL -> SCL

Serial Code:
UArt1: RX=17, TX=18

Cheers,
Volker

[UnexpectedMaker]

How do you know it's not working on battery? Are you using the RGB LED or something to indicate the wifi status? Is it just the wifi connection that is failing?

[UnexpectedMaker]

Are you on discord? Maybe at this stage, it might be easier to troubleshoot this via my Discord server?
https://discord.com/invite/DZPA5rgyGt

[robert-hh]

Now I only need a distributor that sells/ships a FeatherS3 in/to Germany. Any hints anybody?

exp-tech.de have them on their web site. You have to register as company first, and then they ship it. They stopped shipping to non-commercial customers. Maybe they got tired of people shipping back their purchase.
Eckstein-shop.de as well, but with 2 weeks delivery time, indicating that they buy it for you from US on request. In that case it uses to fail.

Note: I had bad experience with shops from the US promising to handle VAT/customs for you. That failed, and I had to pay VAT/Customs twice + handling fee. You can reclaim that, but it is a formal process which takes weeks. That whole customs stuff only worked for me with Aliexpress, Mouser & Digikey.

[UnexpectedMaker]

Neither of those companies are buying wholesale from me - They are just buying via Adafruit when they stock other Adafruit boards, so their price is much higher than it should be. That's ok, if the lower shipping rates even it out :)

I can't vouch for VAT pre paid from US vendors - but I use Lectronz for EU orders for those that want the VAT taken care of as part of the purchase, and Lectronz are an EU based company, and they do indeed manage the VAT etc.

I've not had any customers that bought via the Lectronz store have any VAT/Duty/Fee issues on delivery.

Seon

[UnexpectedMaker]

The reason they only do B2B (and don't sell to consumers) is it's illegal to sell them in the EU to consumers as they are not CE certified. B2B is ok though, so that is why they require business sign-up/registration.

It's why I don't have any EU distributors - Silly EU :(

[robert-hh]

Yes, Mouser, Exp-Tech and sell them as Adafruit products. And Exp-Tech does B2B for all of their products, even those manufactured in DE. They just stopped selling to end customers. Maybe end customers returned too many products.

Lectronz are an EU based company, and they do indeed manage the VAT etc

If that works, that's fine. I just looked it up at Lektronz: €44.26 (incl. €17.21 shipping and €7.06 VAT/taxes). This whole customs stuff is so medieval! I stopped purchasing from UK companies since Brexit, just because of that additional trouble.

[UnexpectedMaker]

Yeah, the shipping costs from Australia are really bad :(
A wonderful country to live in, but a terrible country to export goods from !

[volkerjaenisch]

Looks like a brown out? From the behaviour, or because you actually see a brownout confirmation in the boot logs via TX/TX ?

It's very unlikely a brownout if it doesn't brown out from 5V power source, as it's the same path in, so it if is actually a power problem, it's likely noise being caused by something externally connected to the FS3 and powered from the FS3.

If you do find time to update us, I'd love to see a simple diagram of how everything is connected to the FS3.

That looks like a baud rate issue, there should be no issues using any IO as HW UART pins on the ESP32 - you could just use UART0 that are TX/RX on the headers. They should already be dumping boot info as the ESP32 starts up.

The log output was from the TX/RX on the header.

I will look into this - but its 4:00 local time - time to sleep.

Cheers,
Volker

[xyzzy42]

I've also got a UART issue related to power management on the ESP32-S3 that looks very similar to this. I've tracked it down to the UART changing baud rates when the frequency is set to 40 MHz.

If pm config is set to max freq 160 MHz, min freq 80 MHz, then the UART works just fine. Set it to max 160 and min 40, then it stops working. It's not entirely broken, some data comes through, especially when code is running, like at boot up. But at an idle repl prompt when BLE or wifi is not enabled, it doesn't work at all. Except it does work, the baud rate is just half what it should be!

I haven't checked to see if I can replicate this in a pure ESP-IDF based app or if it only happens in micropython.