CheapWeather: Sensor node design

The design

The ESP8266 changed the game for making this project happen.  Even more, the Arduino support for it had made building it easier through the standard libraries that I already knew how to use.  Let’s go back to the list I had initially made of what I wanted sensors to tell me:the

  • When my furnace or AC was on
  • What the temperature at the thermostat was when it went on
  • What the temperature in each room was, throughout the day
  • What the temperature outside was, to know the delta between inside and outside temps.
  • A temp sensor for every room, or at least the rooms that count
  • A humidity sensor, only really need one of these within the house
  • An outside station that captures
    • light, to know when the clouds have overcast
    • temp
    • humidity
    • barometric pressure (maybe?  nice to have?)
  • The battery level at each station.

While I’m at it, I’ll throw in some nice-to-have features:

  • solar power for the outside node
  • multi sensor per node so that I can have one sensor out in the garage that senses ambient temperature but also freezer internal temps so I can know if my freezer is dying.  Also multi sensor so I don’t have to have two nodes in one area.

The long laundry list of features you want it to have is how most projects start.  I’ve learned that you have sort that list and come up with the “minimum viable product”.  This means that you prioritize the features you really need at the cost of putting off the others until later. This isn’t all bad.  At work we use Agile as our way of organizing our priorities.  When we have good ideas, they go into our backlog.  We don’t take on good ideas right away, we stick to what we’ve said we’d accomplish in our current sprint (a sprint is a “work unit” of time, so for us, it’s every 2 weeks).  One good side-effect of it is that sometimes good ideas are not necessary.  We find that over time when we’ve solved the necessary issues, sometimes those good ideas that got stuck in the backlog have become irrelevant in light of the new knowledge about the problems we’ve been solving.  Sometimes those good ideas ripen with age, and you add more information to them as you think about them while you work on other items.  That way, when you finally get around to implementing that good idea, if it’s still useful to have, it’s matured and is ready to be put into action.

So looking at the laundry list of items, the minimum viable product that should come first would need to have at least these features:

  • What the temperature at the thermostat was when it went on all the time (if a regular node was put on top of the thermostat)
  • What the temperature outside was, to know the delta between inside and outside temps.
  • A temp sensor for every room, or at least the rooms that count that measured throughout the day
  • An outside station that captures
    • temp
    • light (for knowing if the sun is shining or not)
  • The battery level at each station.

This should satisfy the immediate problem of knowing what temperatures my rooms are at throughout the day.  The battery level should help me keep an eye on when the batteries die so that I can change them quickly and keep getting data.  A side effect of having the battery knowledge will also be getting data on how much life one sensor can get out of a pair of AA cells.  Should I keep this system going long term, I will need to write a script that pulls from the database and notify me if any of the nodes stop reporting.  But that’s another nice to have feature that can wait.

My initial design also included the need to configure each without having to reprogram it.  This meant some kind of system of being able to change it’s config through the serial port.  For a while I’ve been doing shell-like interfaces to all of my projects, where you can connect up using a serial cable, and talk to the system.  This enables me to dump registers through functions I’ve coded up, or do any number of things needed to either be useful, or debug.  Mostly these are primitive serial based shells.  And I do mean primitive.  Hardly any processing of the command line, very little flexibility to be used between projects.  I decided to factor out the shell functionality and start building it into a library that I can use across projects without having to munge it for every one.  I was inspired by some of the structure of ChibiOS’s shell, in that it used function pointers to tie a string for the command to the function responsible for it.  Similarly, in my SimpleShell library, I require a table of strings and function pointers to be handed in.  This table allows you to create commands that you can then call from the serial shell.  SimpleShell is as platform independent as I could get, and will not read or write to any input/output streams directly, but instead will only take a string (like the unprocessed command line slurped in via a Serial.readUntilByte() call) and execute a command, or return NOTFOUND.  The command functions return a ExecStatus type, and it’s up to the person using the library to make something of the return codes.

I was contemplating how to put the node into a “command mode”.  I needed to figure this out because the ESP8266 wakes up from a sleep mode via reset and not where it left off, so i was not guaranteed (at least that I could find) that any characters that I typed on the serial terminal while it was asleep would trigger the Serial object to show that it had data in it’s FIFOs when it woke.  I decided that I’d use a weak pull up on one of the pins and read it on startup.  If the pin was pulled low, then I would go into a “Command Path”, which would put the sensor into a loop that read the serial object and process it through the SimpleShell object.  The other code path, the “Sensor Path” would merely start up, read the sensors, send the data and go back to sleep.

Of course, a command mode needs commands.  The sensor node is coded with commands that fit into two categories: config commands, and interactive commands.  The config commands do what you’d expect, they set the persistent EEPROM data that the system uses when it powers up and connects to send it’s data.  The interactive commands allow a user to test the node.  Things like sleep, sending data, getting sensor data are all commands that invoke the code paths that would be invoked by the sensor path.   The interactive command set also included things that would dump registers, or show config data.  This is how SimpleShell can help debug problems in a system where there isn’t a in-system-debugger readily available.

As for the dev process, I had a solderless breadboard setup that allowed me to plugin modules and program them pretty quickly.  If I do more with these modules, I’ll most likely create a protoboard version of this circuit, so that I will not wear out the area on my solderless breadboards.

20161225_093643I like this particular version of protoboard from RadioShack.  I haven’t found any that are close.  I’ve always thought that I’d have to scrape off the lacquer layer that is on top of the copper, but it turns out it’s not lacquer, it’s a type of flux.  It keeps the copper from oxidizing while not in use, and you can solder straight through it no-problemo.  These boards are definitely nice for prototyping.  I was able to get two sensors out of each one, with a bit left over.

The temp sensor is an Atmel AT30TS74 which is an LM75 clone that is a broad voltage range part (1.8v to 5.5v).  I don’t know why I hadn’t initially thought about it, but with two AA batteries I wasn’t going to be operating at 3v all of the time.  I would need to go down to the 1.8V that they say the ESP8266 can go to.  This sensor can do that. In my experience now that I’ve had enough time for most of my modules to expire a set of batteries, they tend to last down to 2.3v before giving up completely.  With a sleep interval of 5 minutes, it took about three months for my modules to get to 2.3v.

The assembled module was really simple.  Just a bit of wire, the temp sensor, the ESP module and the battery holder.  20160612_001023 20160612_164839

All of this amounts to a sensor node that can be configured easily, report temperatures, and sleep until the next time interval.  The system sleeping every 5 mintues means I can get about 3 months on a pair of AA batteries.  So far, it’s been a success.


Ah, EEPROM.  What a pain in the neck.  EEPROM is normally easy in the AVR Arduino environment but doesn’t work as advertised in the ESP8266 environment.  I found that in the ESP8266 Arduino environment (and maybe in the SDK as well) writing to any other location than the first location in EEPROM would not save the data.  It took me coding an EEPROM shadow object to save data in locations other than 0x00.  Normally this is a bad idea because you have to use up the same amount of RAM for the shadow copy of data as the EEPROM size.  This means that for my 1kB sized EEPROM, I’m eating into my runtime RAM space by the same amount, solely for the purpose to keep config data around.  This isn’t too much of a problem on the ESP8266 though, because it has about 80kB of RAM, with about 40kB usable by user programs.  Most of what I’m doing here won’t go over the 40kB limit.  In fact, most projects I can see applying this microcontroller to won’t go over that limit.  Since the EEPROM is really an emulation of EEPROM and stores all of your data in the flash memory chip, if memory does get tight the library for the ESP8266 allows you to use just what you need.  Only need 32 bytes?  The shadow will only take 32 bytes in RAM.  Any time you set data on part of the shadow, it writes out the entire shadow area to the EEPROM.  While it’s ugly, and I hate writing out the whole shadow each time, I’ve found that it works and I’m not writing config data too often.

[UPDATE 03/01/17: the problem mentioned in the next paragraph about programming these was largely to do with the breadboard circuit I had been using.  After creating a more permanent programming solution, all of the suspect modules programmed and booted fine.  The comment about the voltage regulators and the bum LM75s still stands… ]

Speaking of flash, that reminds me about programming them.  Ugh.  Flash.  Re-flash.  Flash again.  Register dump.  Reset.  Re-flash.  Sometimes it takes, sometimes it doesn’t.  Sometimes it takes after four times.  Sometimes never.  Sometimes, it doesn’t boot, it just keeps register dumping to the screen.  Beware of getting parts off eBay cheap.  My friend Dan always chuckles and shakes his head at me when I mention any of these troubles.  “When will you ever learn?  Stop wasting your time and spend a little extra for known working parts…”.  They’re cheap, so if you’re going to get cheap, you’re going to have to deal with the potential 50% garbage parts that you get.  Marginal, or swept off the floor.  Or marginal in construction.  Any way you cut it, there’s a percentage of your cheap parts that will be trash.  About 25% of my modules have issues.  About 100% of a set of LM75 sensors don’t work, period.  They tie the clock line low, and sayonara, no talkey.  A few of a set of 3.3v regulators are trash.  You can input voltage all day long and they don’t output 3.3v, they regulate it to 0v.

If you’re using a serial cable without the RTS line being tied to reset, this means you have to hold the part in reset until the proper time and then release it so the bootloader comes alive just when the esptool utility is needing it to.

The SDK isn’t as straightforward as I’d like.  I had accidentally moved a line of code for turning the ADC pin toward the VCC rail.  This made it stop working, but it wasn’t as obvious why.  Eventually I figured it out, and it was back to reading the battery voltage.

Also, this is one of my first major lead-free ventures.  Beware of your solder joints.  While you can’t easily tell anymore if they’re cold joints or not, you can with an ohm meter.  I don’t know if the pins weren’t taking to the solder, or what, but the joint i had on the VCC was 16kohms.  No wonder it was rebooting, it wasn’t getting the current it needed and would brown out, which meant once it hit a certain voltage on the battery, it would reset loop until it burnt out a set.

Next up, the design of the outside sensor node.

You can find the code for this at GitHub:

CheapWeather: First Steps

If data was what I needed, the question was “how was I going to get it?”.  I already knew that I wasn’t going to get the data if I had to walk to each room, on a schedule, and measure it.  So automation was the way to go.  But just because a sensor would be read on a schedule still didn’t solve the part of collecting the data in one place.  That’s where it made logical sense to make each sensor wireless, with one station collecting the data for the entire set.

The design revolved around two major factors:

  • It had to be cheap, as I was going to be putting more than one of these together.  The minimum would be seven sensor nodes, but potentially more.
  • I wanted it to be easy.  I didn’t have all the time in the world to devote to this project, and like any good geek, I have a million and one other projects I’m working on.  Not to mention the normal obligations of a day job and family.

Other factors that bubbled up while contemplating the design of the system:

  • The sensor node had to be battery efficient, or would have to be able to be close to a wall socket.  Battery efficient meant more than 3 months on a set of batteries.
  • It needed to have a central repository where the data was collected and available on the network for me to get to from my laptop.  This could be a cloud service, or a database on a raspberry pi, or whatever.  Being available also meant I could work on it on my lunch hour, or when monitoring other activities.

With this small set of requirements, I started looking into my options for wireless communications.  There were a couple of options available to me at the time:

  • cc3100 wifi chipset
  • 433mhz wireless link
  • ZigBee module (like xbee), or other 802.15.4 chips and stacks like Microchip’s MiWi stack on top of ZigBee hardware
  • Nordic Semiconductor’s nRF24L01+ chipset

This one was costly at first, somewhere around $30/module (it’s predecessor was, it’s now down around $10).  It was out because it was too expensive for the small single purpose sensors.

433mhz wireless links

These are the cheap $2-$3 modules on eBay.  They’re easy to interface with, but they’re also not smart, so collision avoidance and noise are things you have to put into your software stack to deal with the physical layer that this is.  So they meet the cheap requirement, but they don’t meet the easy.

ZigBee chips or modules

I’ve loved the idea of ZigBee from the first time I heard of it.  A mesh network for embedded communications is an awesome idea.  Unfortunately, I’ve found that it’s a heavy lift to get into the ZigBee stacks. The benefit of ZigBee’s design is that the software stack is where most of the work is done for mesh networking.  The downside though is that you need a stack, and if you’re not coding it up yourself, you’re going to need to get one prewritten.  That can mean extra license fees.  Microchip has a chipset and stack called MiWi that attempts to solve the problem that ZigBee was made to solve, with a lighter weight stack.  MiWi might be better, but again, it’d take some dedication to get into it.  They’re also not too cheap, or weren’t.  The later chips and popularity have brought these down to affordable prices for small modules.  The modules from Microchip are about $10.  The XBee modules are still in the $30 to $40 range, so that’s too expensive for this project.

Nordic Semiconductor’s nRF24L01+

These chips (and the modules I bought) met the requirements of cheap, but they weren’t as easy as I would have liked.  However, there were good examples and there was good code to use to begin understanding the details of using them.  You can get them for a good price on eBay, but beware, the really cheap modules that you can get for sub $2, are probably counterfeit and your failure rate might be high.  Out of the modules I bought, half of them had a problem with ACK’ing packets sent to them.  They could read, they could write, but while the other side saw the packet and had no problems with the exchange, the sending side said that it didn’t get an ACK.  Dan, a good friend of mine, has used them in his product for quite a few years.  His are the genuine chips, and don’t have the problems mine did.  Mine were either counterfeit or marginal chips that got made into modules and sold for surplus prices.

Since the nRF24L01+ met the cheap, and mostly easy requirements, it was the wireless link I started moving forward with.  The design started with an Arduino mini pro driving a nRF24L01+ module for each sensor node, with a Raspberry Pi on the other end listening to a nRF24L01+ module on it’s SPI port.  The Pi was going to be the gateway of the sensor net.  I got to the point of getting the example code pinging back and forth between two arduinos, and between an arduino and the pi.  I was able to do strength tests throughout my house, and found a sweet spot for where the Pi could sit, and hear the entire house.

This is where the project got derailed.  Any geek will tell you there’s far too many projects that beg to be completed, and usually if a project isn’t in high demand, it gets shelved in favor of spending that time on the new and shiny.  During the winters, since the demand for AC wasn’t there, there was always a “I’ll get back to it before summer” excuse, and off it went to the mental shelf to collect dust.  It’s not that I couldn’t use the sensor net throughout the year, my house has issues with the cold just as it does with the heat, but it’s easier solved with space heaters, so the demand for this project just wasn’t there in the winter.

The game changer: ESP8266

It wasn’t until the advent of the ESP8266 that this idea became a reality.  The ESP8266 solved many problems, not just the two main criteria of easy and cheap.  The ability to program this module with code meant that I also could consolidate the microcontroller functionality into it.  Now we’re less one component in the design.  It’s built-in support for WiFi removed the need for an additional computer, like a Raspberry Pi, for use as a gateway.  The sensors could communicate directly with the data store.  Given the Arduino environment and boot code, this module also could use standard libraries for many things.

Now it looked as if my sensor core design could be simplified quite a bit.  The microcontroller and the wifi link could be one module/chip and the data collector could be anything, anywhere I could get to on the network.

Next article I’ll detail the server design and choices.  I started with the server because I could get that done quickly, and test it without needing to have the sensor core done


Easy is quite subjective here.  If you’ve never heard the phrase, “penny wise but pound foolish”, it perfectly describes this term.  There are many times we pick something to be “easy” (cutting some corners) but end up paying more elsewhere (for example the time spent learning a new platform just to make use of a library).  Luckily, that wasn’t the case here.  The new platform I learned was an good choice for future projects.

The CheapWeather project

tl;dr – I’m finding a way to gather data so I can figure out how to better utilize energy in heating/cooling my home.

Air conditioning is a marvel of our modern age. Using the phase change of a gas to a liquid and back to a gas, between two pressure domains, heat is absorbed in one and discharged in another.  The system is driven by a compressor and an expansion valve.  The gas is compressed, and sent through a coil where it dumps it’s heat to one domain (usually outside), sent through a valve where it can expand (into the lower pressure side, like the evaporator coil in the furnace) and by expansion evaporate from a liquid into a gas and take heat with it.  From the evaporator coil, it’s back to the compressor to repeat the cycle.  This is amazing feat.  But it takes work.  The compressor and fans involved are necessary to move this gas through it’s phase change stages.  This work can be expensive.  And if you throw in any inefficiencies, like equipment that has aged and is no longer operating at peak efficiency, you end up with a very expensive air conditioning bill.

This is my current problem.  I’m always feeling like I’m never getting the cooling I’m paying for.

Why don’t I replace it?  It’s just not an option at this time.  While the system lives and breathes, it won’t get replaced.  Until it’s demise, I need to investigate alternatives that could help.  Turning on the blower fan all the time tends to even out the temperatures throughout the house,  or so I was told by an AC guy.  Attic fans could also help reduce the enormous thermal battery that is my attic (testified to by my friend who put them in his attic and immediately felt a difference).  Keeping the house at a warmer temp and using fans.  These are all strategies that might work.

The key to knowing, is data.

Data is hard to pin down if it’s a subjective feeling.  Feeling like it’s working can differ from person to person and day to day.  In addition, remembering it is even harder and sticking to a schedule to write down each data point is near impossible.  That’s where having a system that can report the temperature throughout the day and record them is critical to making decisions.  I needed a way to record multiple points of data throughout my house.  I needed to know data points over time such as:

  • When my furnace or AC was on
  • What the temperature at the thermostat was when it went on
  • What the temperature in each room was, throughout the day
  • What the temperature outside was, to know the delta between inside and outside temps.
  • A temp sensor for every room, or at least the rooms that count
  • A humidity sensor, only really need one of these within the house
  • An outside station that captures
    • light, to know when the clouds have overcast
    • temp
    • humidity
    • barometric pressure (maybe?  nice to have?)
  • The battery level at each station.

The central server piece of software needed to satisfy these requirements:

  • Would be a simple REST based web application, hosted on my in-house server.
  • Needs to run on free packages available.
  • Needs to be something that could run multiplatform.  Linux on a PC at a minimum, Raspberry Pi being a secondary target, with MacOS and Windows last.
  • Data needed to be stored in a concurrent safe manner, that is if two sensors sent data at the same exact time, they wouldn’t collide, they’d be queued properly and inserted.

Stay tuned, in future posts I’ll get into the sensor design itself including the firmware running it, the design of the server software, the design of the outside sensor node, and the processing of the data that it’s produced so far.

Redneck fume extractor

As I was getting ready to build the motor driver controller board, I realized that it would be a good idea to use a fume extractor.  This was a major soldering build, more than I’ve done in one sitting in quite a while.  I’ve seen these used, and my local hackerspace has one:  Hakko Fume Extractor


I didn’t have an extractor, but I did have the parts for one.  A spare wall wart power supply, a surplus squirrel cage fan, an activated carbon filter, and a bit of plastic to mount it on.   The first item to be selected was the fan.  As any good hacker, I have a junk box and happened to have a squirrel cage type in the pile.  Squirrel cage fans are good for moving air, which I need if I’m going to be sucking it through a carbon filter.  I already had the formed plastic on hand, it just needed a hole cut for the fan.  A 3″ holesaw did the trick to cut a hole for the intake.  A 12v wall wart provides enough power to spin the fan at it’s highest speed, which is too noisy and moves too much air.  Downgrading that to a 9v wall wart seemed to give enough air movement without being too noisy.


It sits nicely over the PCB holder in my PanaVise.  It worked well during the soldering of the motor controller board, and it’s continued to serve me well.