Having tried battery life with some Poundland NiMH AA cells I decided to give a LiPo a go.
I've a big pile of 18650 cells recovered from laptop batteries and in anticipation of trying this I bought a few of those ubiquitous TP4056 charger & protection boards as these cells don't have protection built in. Running down a LiPo below about 2.5v is a death sentence for it.
Running the ESP8266 straight off the cell wasn't an option this time either. The 4.2v from a fully charged LiPo is in principle enough to kill it. Having managed to run microcontrollers completely out of spec with no trouble before I expect 90% will survive just fine, at least for a while but if I'm ever to turn this into a completed project I wouldn't want to rely on that across a large number of devices. So I also stuck a little 3.3v regulator in line.
This made for a slightly untidy test rig, but it's functional.
As expected the LiPo lasted much better. I lost track of uptime as I reprogrammed it a few times while working on my code, but I got over 30 hours of constant running.
So I reckon 18650s are a great option for the wearable parts of the project I have in mind, even if they do add complication.
Update: I did this test again and got twenty seven hours runtime.
I have been impressed with the projects out there to make a webcam out of an ESP32 and generic camera module.
Thinking of the faff in putting this together tidily I've not bought the bits, which is good as M5stack released a complete module for a very sensible £10.
It really is very small. More when I can fiddle with it.
As power consumption and battery life are on my mind I did an unscientific brownout/rundown experiment with an ESP8266.
I took an ESP-01S module and a couple of part used AA batteries and left it to run sat on my mesh network until it stopped responding.
This won't be the first time somebody has tried this kind of thing but I was impressed it worked down until almost 1.8v. It ran for ten hours on a pair of batteries that weren't great to begin with, coming in at 2.7v when I started.
It would have been sending packets every few seconds all this time, so it's not like it was sat there doing nothing. I know my code currently causes quite heavy power use, averaging at about 80mA.
This is telling me that my aspiration to run a wearable mesh node 'all day' on normal alkaline batteries is almost certainly achievable. A twin AA battery box is not egregiously large, I can probably desolder the onboard LEDs in a final version and maybe improvements in power management in the code will have some effect.
Of course I'm currently ignoring that I want to connect a GPS module to the wearable and this will eat a consequential amount of power, but I'll worry about that later.
A LiPo battery is what most people would go with but I prefer the wearables to have field replaceable batteries. This is because they will literally be used in a field/forest, the sort of place where you worry about being able to charge your phone as the evening draws in. Having a few AAs to hand is very easy to manage if something goes flat.
For my next experiment along these lines I'll try the same with a 18650 cell and 3.3v regulator. If AAs won't cut it 18650 cells are the sane removable LiPo option in my opinion. There are also 14500 cells which are AA sized so very convenient but these have a much lower capacity.
Want to know exactly which ESP8266 board your sketch is compiled for in the Arduino IDE and act on that in your sketch?
Why do I need this? I'm building for a couple of different flavours of ESP8266 boards and wanted to know which board is the target so I can change a value to match the board automatically.
I'm using the ESP8266 core as an example but this should also apply to other boards supported in the Arduino IDE with some tinkering.
Find the file 'boards.txt' in ESP8266 Arduino core. On Windows this will be somewhere like "C:\Users\YOUR USERNAME\AppData\Local\Arduino15\packages\esp8266\hardware\esp8266\2.4.2\boards.txt"
Search for the name of your board as shown in the Arduino IDE, for example "LOLIN(WEMOS) D1 mini Pro". You should find a line that looks like "d1_mini_pro.name=LOLIN(WEMOS) D1 mini Pro"
Immediately below this there should be line similar to "build.board=ESP8266_WEMOS_D1MINIPRO".
The compiler passes the build.board value on as a #define but it prepends "ARDUINO_".
So if "ARDUINO_ESP8266_WEMOS_D1MINIPRO" is defined in your code you know it has been compiled for a Wemos D1 mini pro.
I am using this so I can do a readVcc() and get an accurate value for Vcc across a couple of different flavours of module. Here's a usable code fragment...
float vcc()
{
uint16_t v = ESP.getVcc();
#ifdef ARDUINO_ESP8266_WEMOS_D1MINI
return((float)v/918.0f);
#else
#ifdef ARDUINO_ESP8266_WEMOS_D1MINIPRO
return((float)v/918.0f);
#else
#ifdef ARDUINO_ESP8266_GENERIC
return((float)v/1024.0f);
#endif
#endif
#endif
}
The slightly different values are because the boards have a slightly different set of resistors on the Vcc monitoring connection and these are the values I've found generate realistic readings verified by a meter.
The IDE also sets the environment variable ARDUINO_BOARD, which you can use, for example...
Serial.print(ARDUINO_BOARD);
You could use this to avoid wading through boards.txt, or code the above example differently.
Unusually, I posted one of my videos publicly and I got a good swathe of comments so I think I'll do this more often in the future.
One of the questions was how big does the mesh scale?
Frankly I don't know.
My aim is to support about 40 nodes actively sending data every few seconds because location tracking is my primary goal. This feels achievable.
However to prove this I'll either have to build it or use some network modelling software. I like building stuff.
I have over time bought quite a few ESP8266 modules for various projects, a lot of them speculatively or for things that are over and done with.
12x Wemos D1 Mini
12x Wemos D1 Mini pro
10x ESP-01S 1MB flash
10x ESP-01S 512KB flash
So when I allow for a few I've lost, given away or killed that's about forty.
Plugging all this in at the same time would be a pain in the neck. So I've built a little test rig that allows me to get fourteen ESP-01S running off my bench power supply.
This circuit is simply a bare minimum of suplying power and a pullup resistor to CH_PD so it boots. In principle you need pullups on GPIO0 and GPIO2 as well but in my experience they boot fine with these pins left floating, at least reliably enough for a little testing.
This took a few leisurely hours to build as there was quite a chunk of soldering, especially as I added individual on/off switches.
Running off my bench PSU it draws 1.1-1.2A which is quite a lot. I need to work on my power efficiency, but given my desired runtime is 'all day' I reckon I can get there. The outdoor nodes lasted about eight hours, which is about in line with this power draw given the rubbish batteries I used.
Compared to a sensor that sleeps almost all the time and draws microAmps when doing so this power usage is awful, but my kit just can't sleep as it has other jobs to do, one of which is always being there to relay traffic for other nodes.
If the nodes were connecting to APs, the various radio sleep modes would reduce consumption massively as the DTIM table means they know when to wake up. Without an AP to manage this I'd have to create my own scheduling algorithm equivalent to DTIM. This is a job for later if I can't make improvements in other ways. I've already got a time sync protocol it might just need to be more accurate.
One of the things I really liked about PainlessMesh was it had a built in time protocol that synced across all the nodes. If you're building a mesh of interacting things then having them share a common clock is useful.
The PainlessMesh developers have gone the whole hog and implemented an NTP inspired protocol taking into account latency and jitter of communication between nodes.
While I don't have the enthusiasm to do this, I have come up with a simple clock syncing option that seems to work fine at low mesh sizes.
I was already sharing uptime information across the mesh so my very simple scheme is as follows...
The node with the highest uptime is considered the time server.
All other nodes work out their offset from this figure as NHS packets come in.
There's then a function that returns this calculated 'mesh time'.
If the clocks drift then every NHS packet from the time server tweaks it back into line.
If the time server goes away, the node with the next highest uptime takes over, faking its own uptime to be what it understands 'mesh time' to be.
If the previous time server comes back the current time server stops.
This simplistic approach seems to be working just fine so far and the sum of all clock drift corrections over several hours is in the tens of milliseconds. Which means it really doesn't need to sync very often.
The nodes aren't perfectly in sync down to the millisecond (mostly because ESP-Now iterates through sending packets to its peers) but nobody is going to notice in real use. This is entirely about making events happen in sequence on human timeframes. It doesn't need to be more accurate to achieve this.
It was our end of year LARP social event this weekend which includes a little bit of shooting at each other in the woods with Lasertag guns, so I took advantage of this to do a field test of my code and hardware.
With the current sketch loaded onto the six nodes I originally built for testing PainlessMesh I headed off to Banbury. It's been a real success, with a couple of provisos.
First the good news. The range of five of the nodes was very solid, I could see direct ELP neighbour discovery packets from them using an ESP-01 plugged into my laptop inside a large wooden hut. This was just with the trace antenna. I think the sixth node has a fault where I've moved the link to connect the external antenna. This node had crappy connectivity despite being no further way from its neighbours than the others.
Secondly propagation of OGM packets works perfectly, I was getting packets forwarded from every node and the ESP-01 connected to my laptop looked to be making good routing decisions.
Finally I got almost eight hours runtime out of at least one of the nodes with nasty cheap pound shop NiMH AA batteries in the freezing cold and pouring rain. This bodes well for better batteries and my target is only to have 'all day' life from a set, fitting fresh at the start of each day. When I got home and opened the nodes they were all bone dry, despite my vague worry they'd leak at the antenna.
The big proviso is this was only a small site (42 acres) so I simply couldn't expect much in the way of a range test. I think I could have achieved coverage of this site with PainlessMesh.
The other proviso is I've now got eight hours of logs to wade through before I'm sure what I've just said is true. Mostly I was there to socialise so peering at a screen for hours wasn't going to happen.
I now have until March for the next field testing opportunity, unless I make a special trip somewhere.
In the meantime I need to make a device that actively uses the mesh so this isn't all just peering at MAC addresses and routing tables. Given my end goal is GPS tracking and status reporting I think it's time to build a minimal GPS tracker that reports back to a specific node.
I'm still wrestling with making what I hope will be a fully featured mesh protocol for battery powered ESP8266 devices.
Having put in the main building blocks of ELP and OGM from BATMAN IV I wanted to add something of my own to monitor the 'health' of the devices. I've coined the term Node Health Status (NHS) for these packets. Because British.
This is relatively simple just now, it periodically sends uptime, free heap memory, starting heap memory, supply voltage, number of peers and radio transmission power to all its neighbours. Nodes store this and make it persistently available.
My reasoning for doing this is it will allow me to 'visualise' the network more accurately without having direct access to each node and certain features need an election process to decide which nodes to run on.
The first use I'll have for this is that nodes with fewer peers will run as a SoftAp to help discovery of new nodes. I'm hoping having fewer peers is a passable heuristic for being somewhere 'sparse' on the network.
Monitoring of supply voltage and uptime give simple measures of battery health and if a node has rebooted. Heap monitoring is something that may feed into feature elections.
I also spent a fair bit of time playing with management of transmission power. So long as it is not experiencing consequential packet loss each node lowers power until packet loss starts. Then it raises power and sets a 'floor' it will not go below. Periodically it lowers the floor slightly to see if power can go down further, but will immediately raise it again if packet loss starts. If we had an easily accessible RSSI measure for ESP-Now peers this wouldn't be necessary but the only RSSI information exposed to us in the standard libraries is the RSSI for an AP association.
It might be possible to infer from transmission power, number of peers and so on whether a node actually has to act as a router or not, switching that off/on to conserve resource. However that's probably out of scope for now and is making a big change to the BATMAN IV routing algorithm that would have big unintended consequences.
