Friday, May 25, 2018

Fun with Home Automation Part 1: The most convoluted doorbell

This is the first of a two part series on using a music keyboard for home automation. Here we describe using a motion detection system to trigger two audio events:

1.) MIDI notes on a keyboard (because why not)
2.) Your Google Home device (this might actually be useful to some people).

The result is that when someone approaches our front door, two notes are sounded on a music keyboard, and the Google Home speaker talks to us as well. The future is now!

In this example, I receive motion events from my security camera software (iSpy) via a http GET request. The actual camera feed is from a streaming Raspberry Pi server (running RPi-Web-Cam Interface). iSpy allows me to view multiple streams, and manage motion detection for each stream. I also (as a lazy option) store stills during motion events into a dropbox folder, so I can see it easily on other devices without any custom application. This particular setup probably warrants a separate post at some future date...

Back to the main feature in question: the doorbell itself!

Here I'm using node-red again, running on a Raspberry Pi. Here's the overall system block diagram:

And here's what the flow looks like: [node-red flow code here]

On top of a stock node-red install, you will need the following two addons:

node-red-contrib-midi: for talking to the MIDI port
node-red-contrib-google-home-notify: for sending text to speech snippets to your Google Home device.

Relatively simple: on the top left you see the incoming motion /get request. I actually assemble a simple html reply and shoot it back for testing purposes, so you can test emitting this event with your browser and it will return an html page with "OK" on it.

1. Triggering MIDI notes

The Raspberry Pi running the node-red server also has a MIDI keyboard (in this case, a cheap Casio CTK-2300) connected via USB. The midi out object will automatically find any class compliant ports, and list them in a drop menu.

To emit the MIDI events, we have two triggers in the middle of the flow that basically turn on, and then off a particular MIDI note (you need to emit both ON and OFF messages otherwise the key will be stuck forever, even after it fades out and becomes inaudible). I put a delay on the second note so they are played one after the other. The message format is simply an array containing the raw MIDI bytes. You can take a look at the trigger objects (or midi out object info) to see the exact messages (notes) I'm sending.

2. Triggering Google Home

When we first got our Google Home device, one thing I really wanted to do is to be able to emit custom events in the form of audio notifications on the speaker. Turns out one of the easiest ways is to use google-home-notifier. The gist of how it works is that on a local network, you simply need to know the IP address of your Google Home speaker, and then audio can be directed straight to it! So the notifier application does a bit of text to audio first with your input, and simply transmits it to the Google Home, and thats it! Much simpler than I imagined. Obviously if you want more complex two way interactions you'll probably have to dig into the actual Google Assistant API...

Monday, August 21, 2017

ESP32-based wireless BBQ probe

At work I've been using the ESP32 lately, and its proving to be a much more capable platform compared to the cheaper (but more mature) ESP8266. The great thing about the ESP32 is that it has a dual core processor, and a really adaptable I/O matrix that allows you to route hardware ports to many different pins. The flip side is that it costs a bit more (dev boards are around $10, while the chip itself is about $5. This makes the ESP8266 about half the price). Additionally, the software support is nowhere as developed and there are many bugs and undocumented stuff throughout, so for that the ESP8266 may be a more stable platform.

Over the past weekend, I played around with adapting a teensy-based thermocouple 

Part List

ESP32 dev board (this is the cheapest one with a lot of pins at $7USD currently. You can get a fancier one with built in OLED, or 18650 battery holder and charging circuitry for about $10. There's also one that has a standard Arduino Uno shape if thats what you're into...)
- Mobile phone power bank (I use a 5000mAh one that lasted comfortably overnight. a 10000mAh one like this one would probably power it for 2 days straight ;)
- Computer for programming, and monitoring temperature over http (via a browser)

Hardware Setup

The MAX6675 thermocouple driver uses a simple SPI connection. However, unlike many common setups, the chip doesn't take in any commands from the master so there is no MOSI line and only CS, MISO, and CLK are required. Once the chip select pin is pulled low, the resultant data gets sent out on the MISO line bit by bit for every CLK cycle. 

In this case, I used pins 18, 19, and 21 on the ESP32 for CLK, MOSI, and CS respectively.

The dev board has a micro-USB input for power (and programming), so to power it portably I simply use a power bank.

Software Setup

I'm using the Adafruit driver which can be found and downloaded automatically using the Arduino library manager. Note that like the ESP8266, you may need to add this change to explicitly implement the delay function as util/delay.h does not exist for the Arduino ESP32 library. The library allows you to hook up the MAX6675 to any pin that supports digital I/O, and using additional digital pins for the CS pin supports more than one probe driver if necessary.

To actually serve temperature values, I built upon the SimpleWifiServer example so that the user can access the value through the browser. As a lazy way to make the display update, I added a refresh tag to the served html so the browser will reload the page every 5 seconds. Hacky, but works!

The code is available here.

Test and Use

Here's a short test comparing it with a Thermopro wireless system: off by less than 2F at room temperature. Not bad!

I did a longer test while practicing the "snake" method which is a way to have a long and low temperature cooking technique for smoking on a Weber-style bowl BBQ. Here are some not-so-well-lit shots of the starting of the snake:

With a relatively short snake that spanned less than a third of the grill I managed to keep the temperature around 250~270F for about 3 hours. This bodes well as typical cooking times for pork ribs is around 3-5 hours (depend on a number of factors; here's a good recipe). And the snake could always be lengthened for really long cooks for brisket and larger roasts.

Next Steps

Here are some things I'd like to add to the system:

- a better web user interface!
- multiple probes: at least one for the grill and one for the food. almost essential to do any legitimate smoking.
- explore possibility of powering it directly off a thermoelectric module, making use of temperature difference between the BBQ enclosure and surrounding air.

*note: the MAX6675 is actually discontinued, but it is still widely available on most online retailers. Its replacement, the MAX31855, is also double the cost. What you get as extra are an extended temperature range, as well as some additional sensor info (e.g. probe short circuit detection, etc). Might play around with the newer chip at some point, but for now the 6675 will do... ;)

Monday, August 14, 2017

iPhone 7's slo-mo mode

The newer iPhones have this "slo-mo" mode, which is basically very fast video (240fps) that you can then tag a certain section to have it played back slowly.

On recent trips I managed to capture some interesting footage, for example: some seagulls that flock around the ferry in Helsinki harbour, on the way to Suomenlinna Island:

Or leaping dolphins at the National Aquarium in Baltimore: (note that we had been going here everyday for almost a week, which is why I got relatively good at anticipating when they do neat tricks. What I enjoyed about this aquarium is that they don't just make the dolphins do crowd-pleaser tricks, but take time to explain the kind of interactions and trainings they do to keep them healthy. Keeping marine mammals in captivity is a controversial subject...

Kinda cheesy, but pretty neat for the first while at least...

Sunday, July 23, 2017

Cheap 3" Drone for taking HD video

Ever since the intern Canadian UAV regulations were released, I've been wanting to put together a relatively cheap, but HD Camera carrying drone that is light enough to be exempt from any regulation. At the threshold of 250 grams, it was actually quite doable at a relatively low price given the extremely rapid developments in the DIY quadcopter industry over the past few years.

This is the second, but much more used iteration of the particular frame. The first had a 4 in 1 ESC and F4 controller which I blew up on a hard crash (did not disarm when it dove into the ground in acro mode), and due to parts I had available it was rebuilt using slightly less optimal (weight-wise) parts but ended up flying very well, so I kept it even after replacement parts arrived.

Before delving into the details, here's a google photos-created video from some footage I took while travelling in Finland for a conference. The small size of the quad means I can carry it easily in a backpack containing my work laptop (13" MBP) without adding too much extra weight.

Here's the finished build "in action" next to the transmitter, out on the field on Aalto University campus:

Parts list and build details

Frame: Realacc RX130 that came with a Matek PDB. (Used to be able to find it for about $13-15 on Aliexpress)

Motors: ReadyToSky 1306, 3100kV (available only on AliExpress from what I can find; example link. They are super cheap clones of the DYS 1306 at half the price, but the latter are much better in quality. Supposedly 4s capable but I've only ran them on 3s so far. Works fine!

ESCs: EMax 12A "BlHeli" which actually had to be manually programmed with "real" BlHeli. They're not super light but can be found for cheaper from eBay and AliExpress merchants. However, I would actually rebuild it with a BS412 (25x25mm) or maybe a RacerStar 4in1 (30.5x30.5mm). The latter is 20A and would work with 1407 motors which could support a larger craft (160mm frame swinging 4" props). A 4in1 (even larger 20A one) would also be considerably lighter than the 12A EMax's I'm using. For discrete ESCs the EMax Bullet series may be better as they're much smaller and lighter.

FC: I'm using a strange Piko BLX clone board that has a curious 23mm!? hole spacing. It fits in between the standoffs of the 30.5mm spaced stack, but is not ideal. I would probably just go with a full size 30.5mm FC for this frame next time... or else go 20x20.

Receiver: FlySky A8S. Tiny, supports failsafe fine despite product description stating otherwise, and no reliability issues that seem to be a problem with older versions that are no longer sold on BG...

Batteries: ZOP power 3s 500mAh 45C. They're not rated for consistency, and of the two I have one seem to work better than the other. Having said that both allow me to zoom around for about 3 minutes or so lugging the HD camera, or do more aggressive manoeuvres without the camera.

Props: DYS 3045. The only 3" props I've used so far. Seems to be a good match and they're pretty, so not too much to write about at this point... ;)

Camera: the trusty RunCam 2 HD. Weighing in at 50 grams with battery and SD card, its not the lightest option but its a nice action camera to boot. I could have shaved about 15 grams off by not using the battery and powering it off the 5V rail of the PDB instead. Another interesting option would be the new Split that weighs about 20 grams or so and has a faster analog signal for realtime FPV feed (requiring an analog transmitter of course), but the Split is pretty much a dedicated flight camera and can't be used elsewhere by itself.

Transmitter: The trusty FlySky i6 which I modded to 10ch, and the i6ab receiver is still working well on the 250 class quad.

The build weighs in at exactly 250 grams with the camera, which is about 15 grams more than when I had the BS412 4in1 ESC and F4 controller with built in 5V regulator which eliminated the need for a PDB. This means I don't have room for a lipo voltage checker/alarm (such as this one), and I have to go by gut feeling and go by experience from more open locations when I didn't need to conform to regulation.

Previously, with the lighter 4in1 ESC:

Here's some raw, unedited zooming-around-the-park footage:

All in all, a very fun quad that gets a lot more action than the 250 (5") one due to the size and legal status. The Canadian regulations have recently been updated to be a lot more reasonable, but anything that weighs over 250 grams (and less than 1kg) is still relatively restrictive in terms of space.

Tuesday, May 30, 2017

Sonifying Plant Moisture using a simple Algorithmic Arpeggiator

Putting together a number of projects related to the Raspberry pi, sensors, and node-red, this post describes the sonification of sensor values in a relatively simple but practical context.

After putting together the automated plant watering system and appending to a physical actuator to it, here we look at other ways of communicating the status of the system to the user: using sound and music!

Mapping sensor values to music

The concept of mapping sensor values to music in the context of new digital musical instruments is a deep and fascinating area of research (ok, I might be a bit biased here since this is the subject of my PhD work... :P). In a nutshell, the mapping problem can be laid out by the following components:

Input -> Mapping -> Synthesis

The input is the sensor signals that correspond to some physical phenomena of interest. Here we have a value that gives us a sense of the humidity levels (to be precise: it is difficult to get a sense of exactly how much water is in the soil, and I'm not even sure what would be the appropriate units to measure it... but what we do get is a relative number that moves up and down as the soil dries out, and that's what we're interested in here!) In our current situation, it is simply a single value coming from the moisture sensor, perhaps scaled to a more convenient range (lets say 0-100...)

The mapping is the part of the system where the sensors values are translated in a meaningful way into control parameters to drive the sound producing module. . While a rather abstract concept that is usually completely implemented virtually for most digital systems, this is an important part that ultimately defines how the system behaves (i.e. produces what kind of output, for a given input).

The synthesis component refers to the part that actually generates sound - typically attached to some kind of physical sound producing device. Here we build a small algorithmic arpeggiator that takes in parameters and generates notes that are being emitted in real time through a MIDI port which can be used to control any hardware synthesizer. (It probably makes sense to look into using a software synth on the Pi itself, for a more standalone solution in the future...)

Design of Mapping

The general behaviour was inspired by the system presented in the Music Room project. Here, two values (speed and proximity between two people) are used to drive an algorithmic composer that can respond in real time to generate music in a "classical" style. The sensor values here are mapped to concepts like valence and arousal, which in turn affect the tempo and key of the music being produced. (For more details check out the paper/video!). In our case we take a simplified but similar version of the concept.

Algorithmic Arpeggiator/Sequencer

The sequencer is implemented as a Python script, building on one of the examples in the mido library. It simply randomly selects from an array of notes and emits an note-on followed by a note-off message after a certain duration. I expose the relevant parameters via OSC to control the tempo (i.e. note duration), and the key (each scale name changes the contents of the note array). The Python code is available here and the OSC address/message handling is quite straight forward.  Of course, the possibilities for incorporating more complex and interesting musical concepts is endless, but we leave it here for now... :)

Software Architecture

Since most of our existing system is built in node-red, we simply add the necessary nodes to talk to the sequencer. The OSC node takes care of formatting the message, and then we simply pipe it out a UDP object to the correct local port where the Python script is listening.

node-red Configuration

Here's what the node-red configuration looks like. The top function node divides the moisture range into "pentatonic", "major",  and "minor" scales with decreasing moisture values. The tempo map function below provides an exponential scaling of the note durations, which cause a rather sharp change as you reach a "critically low" moisture value (to be fine-tuned).

The blue OSC nodes ("key" and "tempo") take care of the formatting of messages, and here they're sent to port 7002 on the same host where the Python sequencer is running. The entire flow is available here.

This is what the dashboard looks like, showing the changes in "key" and tempo (in Beats Per Minute) as the moisture decreases:

Audio/Video recording to come...

Soft Synths

It is possible to do all the sound generation on the RPi itself. I have experimented with amsynth running on the Pi, and after battling a bit with alsa configurations, managed to get the onboard soundcard to generate the actual audio as well. The universality of MIDI means you can have it any way you like!

Tuesday, May 02, 2017

Playing a MIDI keyboard in node-red

While likely not originally designed for typical IoT platforms, it is possible to play a MIDI keyboard in node-red via the GUI dashboard using a relatively simple setup.

For this somewhat unusual exercise, you will need:

- A Rasbperry Pi (can do this on a desktop platform, but whats the fun in that?)
- A USB-MIDI capable Keyboard. (You can use a USB-MIDI adapter with an older MIDI keyboard without USB of course)
- Install of the most recent node-red, with the dashboard. The only additional node you'll need is node-red-midi.

For further details check out the detailed writeup here.

Here's what it looks like in action. The computer monitor shows the flow as well as dashboard UI. Excuse my rats nest under the desk.

Some neat features:
- Works on any platform in the browser
- Allows concurrent connections, so more than one person can play with it at the same time on different devices 

Some obvious limits I can think of:
- limited UI from node-red-dashboard for this purpose; a row of buttons is not a great interface for an instrument
- multi-touch
- not tested for performance (latency, etc).

Next step: hooking up some music to the plant moisture sensor!

Tuesday, April 18, 2017

Raspberry Pi Plant Watering System

Here's what I've been working on at Infusion lately: documenting the Raspberry Pi based plant watering system that measures the moisture content of the soil and triggers a relay-driven valve to water the plant:

Control software was built using node-red, with a FRED endpoint to talk to the cloud-hosted node-red instance that allows a public dashboard to be displayed (public dashboard available here).

One of the awesome things about node-red is that you can easily view the dashboard from a browser, and with FRED you can access it anywhere:

Detailed hardware and software write-ups on the PiShield site.