////////////////////////////////////////////////////////////////////////////
// **** eDog **** //
// //
// Electronic Dog Home Security System //
// on the //
// STM32F4-Discovery //
// //
// Copyright (c) 2014 David Bryant //
// All Rights Reserved //
// Distributed under the GNU Software License (see COPYING) //
////////////////////////////////////////////////////////////////////////////
The idea is simple. If someone thinks you have a big mean dog in your house,
they are far less likely to try to break in and steal your stuff (or hurt
you), and until now there were only two ways to achieve this. You could buy
a big mean dog or you could buy a "beware of dog" sign and put it on your
door. The former is a lot of trouble and the latter is probably not very
effective. Well, now there's a third way: the eDog!!
Implemented on the incredibly cheap (about $20) STM32F4-Discovery evaluation
board from ST Microelectronics, the eDog continuously monitors your home with
its on-board MEMS microphone. When it detects someone knocking at your door or
ringing your doorbell, it renders the menacing sound of a big dog barking
through your audio system (which it is connected to). At the correct volume
level, this is very convincing. In fact, not a single person who has come by
thought anything other than that there was a dog in the house.
The STM32F407VG CPU on the Discovery board is great for this project because
it has the Cortex-M4 core with the single-precision FPU that makes the audio
processing algorithms so much easier to implement. I was often surprised how
little my code was taxing the CPU, and I was amazed when I realized that the
ST demo code sampling the digital microphone was executing 64,000 interrupts
per second without a hiccup!
***** Building and Installation *****
ST provides a complete development tree for the Discovery board that works
with several commercial development environments including Keil's uVision4,
which is the one I chose. The free evaluation version of this toolset has a
32K target size limit, but that is plenty for this task. I started with their
Audio_playback_and_record project as a reference and added and modified code
as needed, and then set it up as a new project. You will also need ST's STM32
ST-LINK Utility to flash the raw PCM audio data into the internal flash,
above the code (see Projects/eDog/bin/readme.txt).
The knocking detection algorithm works without any configuration because it
simply finds appropriately spaced transients. However, the doorbell detection
uses a narrow bandpass filter tuned to the fundamental frequency of the actual
bell. I have an old standard electro-mechanical doorbell from about 1980, and
I measured the fundamental frequency (of the higher "ding", not the lower
"dong") to be 770 Hz. This may or may not be the most common frequency around.
I also measured another, newer, wireless doorbell (that had no "dong") to be
785 Hz, and that filter is in the source too (commented out).
To generate the custom coefficients for the biquad filter, you can use this
online tool and parameters:
Site: http://www.earlevel.com/main/2013/10/13/biquad-calculator-v2/
Type: bandpass
Sample rate: 16000
Fc (Hz): 770 (the measured fundamental frequency of your bell)
Q: 100
Enter the five generated coefficients into the call to biquad_init() in the
scan.c module. Be sure to add the "F" at the end of the constants.
Of course, if you want to just program the Discovery board without actually
building the program, you can use the prebuilt binary (in Projects/eDog/bin),
but this will not allow you to customize the filter for your doorbell (which
might not be exactly 770 Hz).
It might very well be possible to build and flash the whole thing using free
tools on Linux, but I have not tried that yet.
***** Using the eDog *****
Simply attach the headphone output of the Discovery board to an unused audio
input on your AV receiver. You might need to use a cable with a 1/8" stereo
jack on one end and dual RCA connectors on the other end.
The Discovery board will also need power. The easiest way is to simply plug a
USB power adapter into a wall outlet and connect that to the ST/LINK USB
connector (CN1) on the Discovery board with a standard USB cable. It might be
possible to simply attach that USB connector to a USB jack on your AV receiver
(if it has one) but I found that my AV receiver would only power the board for
a second or so before shutting it down (presumably because it did not like
what enumerated there).
When powered, the board will immediately start running the eDog application,
indicated by the flashing green LED. There are two settings available from the
blue "user" button. First, there is the "one-bark" mode which causes the eDog
to produce a single, short bark at each detection instead of rotating through
the regular long (about 5 second) sequences of barks. This is handy for
testing the sensitivity because you don't have to wait for the barking to stop
after each trigger. This mode is indicated with the blue LED.
There is also a high sensitivity mode. This mode, indicated by the rapidly
flashing green LED, loosens several detection criteria and so will catch even
the faintest "knocks" in any part of your house. However, this mode will also
more often falsely trigger on normal activity around the house (like walking
and talking), and can even trigger without any apparent noise at all (although
this is rare).
You should experiment with the two sensitivity modes to determine which mode
is the most appropriate for a given situation. For example, if a neighbor
could be disturbed by sudden bursts of barking in the middle of the night,
then that might be a good time to NOT use the high sensitivity mode. If your
neighbor has trouble sleeping, or has no sense of humor, or is an avid gun
collector, you might want to avoid using the eDog at night altogether!
In summary, here are the 4 operating states that are accessed with the blue
user button and how they are identified:
1) Normal sensitivity mode (green LED flashing slowly, blue LED off)
2) Normal sensitivity one-bark mode (green LED flashing slowly, blue LED on)
3) High sensitivity mode (green LED flashing quickly, blue LED off)
4) High sensitivity one-bark mode (green LED flashing quickly, blue LED on)
When the eDog is triggered, the orange LED flashes while the barking audio is
playing and, when it's finished, the green LED resumes. This can be used to
verify operation without connecting the audio.
The red LED indicates the that input microphone is overloaded. This should
normally only happen with very loud nearby sounds, or when handling the
Discovery board. It does not indicate a problem and is mostly just an easy
way to see that the board is running properly.
The eDog automatically adjust its sensitivity to the environment so that it
can react to fairly quiet knocking (say, on the other side of the house)
without falsely triggering all the time when there's more activity. However,
if you are active near the eDog when it's on (even in the normal sensitivity
mode), it WILL falsely trigger occasionally, so it's not really recommended
that you do that (unless being startled every few minutes does not bother
you).
***** The Code *****
The code is based on the Audio_playback_and_record demonstration project
provided in the Discovery firmware tree, and so many of the files in that
project are carried over, but are heavily modified and cleaned up. There
are also several completely new modules:
scan.c -- this is the code that scans PCM audio for knocks and rings
scantest.c -- this is a harness for non-embedded testing of scan.c
serial.c -- provides buffered debug logging output on USART2
The main functionality is implemented in waveplayer.c, and contains, in
addition to the eDog function, the ability to generate sine waves into the
headphones and also to echo the microphone directly to the headphones with a
delay determined by the defined buffer size. These other two options are
enabled with macro definitions.
***** Liability Limitations *****
In addition to the liability limitations outlined in the GNU license and
the ST license, I want to make other liability limitations clear. Neither
I nor ST are in any way responsible if
* the eDog fails to trigger at a crucial moment and someone burgles your
home
* the eDog startles someone and they have a heart attack or soil their
clothing
* the eDog crashes and generates noise that blows out your speakers
* the eDog starts barking all night and your neighbor comes over and,
well, you get the idea