Learning How To Learn

Mon, 20 Feb 2023 00:00:00 +0000

If you’re a little bit like me, some subjects could seem too complicated. Other times you’d find it very hard to get started with something and keep procrastinating it. Or you might find yourself very tired to learn something new. And frankly, if we overcome all these we might find it was all in vain and we forgot most of it. Let’s fix this!

It’s mostly a thing of having a well-rested brain, a proper mindset and a proper technique to improve your learning a lot.

As part of my “Learning How to Learn: Powerful mental tools to help you master tough subjects” Coursera class I am writing this post. It’s supposed to be a gift that keeps on giving. It helps me master what I learned by repeating the concepts and it will teach someone else concepts that are supposed to be life-changing.

The big picture

Many of us enjoy exploring new interests and learning new things. I think I always did, except maybe for half my life when I was in school 🙂. The school system in my country favored going for the grades and not for the education so pretty much everybody that learned, did it just before getting tested, we got our grades and then we would forget most of it. I used to think that even though in 19 years the school didn’t teach me much, at least it taught me how to learn. But it seems that is not the case either.

Anyways, this is not about school systems it’s about our personal systems. I often need (for making money) or want (for health, hobbies, the sake of others, etc.) to learn more, faster, better and I find myself accomplishing way less than I think I could accomplish. It’s time to get better at learning before we can get better at other things.

The Mindset

Like many of the other things which follow, it’s essential to have the proper mindset. Knowledge is made by people FOR people. We all start as a blank slate. We learn a HUGE amount of things by the time we grasp reading and basic arithmetic. Let’s not stop learaning, ever!

If you read or listen to interesting things which you’d like to learn but you don’t apply yourself in the direction of learning them, likely that will not happen.

Looking at scientific research tells us that most of us think we’re above average. Let’s use this fallacy to our advantage, believe we can learn everything and increase our chances to become above average.

Sleep

When you’re tired you don’t remember. As the day progresses, the brain accumulates substances that interfere with strengthening neural pathways.

During sleep, the brain clears all the substances which inhibit learning.

I highly recommend reading the Why We Sleep book (ISBN 9781501144318). It presents many other advantages to having proper sleep which will also be good for your learning.

Thinking modes

Our brain has two modes of thinking:

Focused mode thinking
Diffused mode thinking

In focused mode thinking we can point our mind to e narrow subject, just like a searchlight. It helps us understand that specific thing so we can then store it in memory, thus forming a chunk of information. Focused mode thinking uses a lot of mental energy and it’s not sustainable for prolonged periods. Also, when we’re in focused mode our vision is very narrow and often we can’t clearly see the bigger picture.

Diffused mode thinking is a more relaxed way of thinking when we just let our minds wander. We let the thoughts pop into our minds as they please and it’s often a very good source of general problem-solving and creativity. We can stimulate diffuse mode thinking by walking or doing other relaxing activities that don’t grab your focus. During the diffused mode thinking, the chunks we formed in focus mode interact in a relaxed fashion.

Both modes are very important. Thinking works best by alternating between these two.

Memory

Although our memory is very complex, for the sake of grasping the main ideas as quickly as possible we are going to touch a little on the main two branches, short-term memory and long-term memory.

Our short-term memory, often described as working-memory, is handled by our prefrontal cortex and it can only store 4±1 items for just 15-30 seconds, according to Cowan (2001) and Atkinson and Shiffrin (1968). We’re using this memory pretty much for any type of thinking and it’s important to understand its limitations.

This memory is small, easily distracted, and very short-lived. If I am focused on my work and notifications fire on the computer and the phone, they bump items out of my short-term memory. If I actually check out those notifications, the short-term memory gets wiped out and I need to recreate it. This process requires energy and our brain gets tired faster in the process.

When we’re working with chunks of information (pieces that we already have in the long-term memory), they use less space in our working memory. To solve complex problems we first need to develop the necessary chunks.

Long-term memory is VAST. It is comparable to a warehouse where pieces of information stay until they’re needed. Like a warehouse, different areas seem to be used for storing different types of information. Walking, biking and writing are part of our unconscious memory and we call this implicit memory.

Knowledge about factual information, like how you use some software on your computer, is called semantic memory and it’s part of our explicit memory. Semantic memory is subject to fading and we can protect from that through deliberate recall.

Things get into the long-term memory when after the information is manipulated in the short-term memory and then it’s encoded by the hippocampus. Repeating the information activates and strengthens the neural pathways related to that piece of information. If the hippocampus gets damaged you can still retrieve memories but you can’t form new ones. The protagonist of the movie Memento had this problem.

Chunking

As we learn a concept, it becomes a chunk of information and it gets stored in the long-term memory. When we grasp it clearly, this chunk takes only one slot in our working memory.

Our diffused mode thinking interacts with different chunks, sometimes even from different areas of knowledge, to come to a solution for a complex problem.

Spaced repetition

We mentioned before that repeating a piece of information makes it stick. Instead of doing all the work of memorizing in one big session, it’s way more effective to split it across multiple days into much shorter sessions. If you learn the subject at once you’re way more likely to forget it and have a bad day.

This technique of learning was studied multiple times since 1932 and has been proven to increase the rate of learning.

Spaced repetition allows us to be more chill about learning and learn better.

For example, Duolingo, the popular language app employs spaced repetition as the main teaching mechanism. Headway tries to apply this to books. A general-purpose tool for doing spaced repetition is Anki. Let me know if you find something better.

Procrastination

When we think about starting to work on a complex task our brain actually feels pain. Not like skin pain, but brain pain. And it wants to protect itself by changing our attention to something way easier, like surfing on the Internet or talking to other people.

If we get started with our task, the pain goes away very quickly and we can then focus on our task. A notification would break that focus and we’d have to go through that initial pain again. And another problem is that going through this pain uses the limited willpower we have during a day.

There are a few key actions that we can do here:

Start early with the task that would require the most willpower
Minimize notifications so we don’t waste energy on overcoming the pain over and over again
Take deliberate breaks to rest and do diffuse mode thinking
Treat yourself so that you train your brain that at the end of a work session something good is coming

The most well-known technique for fighting procrastination is the Pomodoro method. Its basic principle is that you work for 25 minutes and then take a break for 5 minutes. After two hours you’ve done 4 pomodoros and you can take a 30-minute break.

The Pomodoro method doesn’t really work for me. I need longer than 25 minutes sessions so I will invent the Rădulescu method. 50 minutes of work and 10-20 minutes of break. It’s not yet clear how many of these sessions I would do in a day but I’ll try different things.

Illusion of competence

If you know how to drive a car and you look at F1 races, you might think you could drive an F1 car. This is a well-known cognitive bias called the Dunning-Kruger effect.

When I started to write this blog post about learning how to learn I thought it was going to be quick and easy. I got 100% on my test so to me that looked like I mastered the subject matter. In reality, I am 5 hours into this and still learning.

You can only know for sure you are competent in a subject when you do it yourself. Test yourself!

What next?!

If you feel it’s worth keeping anything out of this I will give you an assignment. Drop me a message with 3 key points which you think will improve your learning the most. This exercise WILL strengthen neural pathways especially if you go over those key points on different days during a week.

Looking forward to your message!

Taking my first photo of the Milky Way

Sat, 03 Aug 2019 00:00:00 +0000

While searching for a hobby for my wife I got interested in photography. It seems that I am especially drawn towards low light photography which I understand is what “normal” photographers don’t usually do. I played a little with my phone in low light situations and even though it’s amazing where things are in 2019, it’s still not good enough for what I am looking for.

The gear

Of course, part of the fun is getting the gear so after doing a little bit of research I decided to get the Sony A6000 because “they” say it’s a very capable compact mirrorless camera. I was also considering RX100 and A7 because the first one is way more compact and the second one more capable but bulkier, I feel this one I chose is a good middle ground.

About lenses

My understanding right now is that the glass is the most important thing if you have a decent camera. Also, there is no point in having a very good camera if you don’t have a decent lens because it will probably not make a difference.

There are 4 main properties to consider for lenses and some other less important things:

Focal length
Aperture
Manual or not
Image quality

1. Focal length

Generally speaking, is a measure of that fits into the frame. A small focal length makes wide pictures which fit a lot into the frame and a high focal length number “zooms” a lot bringing the subject closer. The image below should make it the basic concept clear.

The focal length also changes how compressed things appear in the background. In this example, the zoom is done with the legs so the subject is kept in the frame. As you can see, the background changes drastically.

The first lens I got is a versatile 18-200mm which means that I can take both wide and significantly zoomed-in photos.

2. Aperture

The aperture is the opening through which the light enters the camera. Lenses have a variable aperture but we care about the smaller number which represents how much the aperture opens thus allowing more light in.

A lens with a smaller aperture is called a fast lens because more light comes in thus we can expose for a shorter time and catch faster-moving subjects.

Changing the aperture also influences what is kept in focus.

The second lens I got for photographing the stars is an f/2.0, 12mm which should allow for capturing lots of lite and a decent portion of the sky.

3. Manual or not

Manual lenses are just what the name says. They don’t have motors or electronics and can easily be adapted to different camera systems. For setting the focus and aperture they have some ring which you turn. The zoom lenses also have a ring for that.

Normally I would prefer a lens with autofocus but for some applications such as night photography, it doesn’t matter because you set it once and live it like that.

The 18-55mm kit lens of my camera has a lot of “auto” in it. Autofocus, auto-aperture, optical stabilization and even motors for zoom. Of course, all these auto things can be set to manual and configured with buttons from the camera.

I wouldn’t say one is better than the other, it just depends on what you do with it. I like that my 12mm, f/2 lens is manual because to me it means that most of the cost comes from having better glass. And I also like that my “walk-around” lens has a lof of auto because that optical stabilization is a very nice feature when taking pictures zoomed in and handheld.

4. Image quality

Glass is not perfect and physics not on our side when light is bent in complex ways. The science of producing a good lens deals with issues such as distortion, chromatic aberration, vignetting, lens flare and others. To fight these issues lenses use multiple glass elements combined into groups. The issue with this is that the more elements are used the more light is lost due to reflections. The glass can have antireflection coating but I guess you can see why lenses can get expensive.

Distortions are more of an issue with wide-angle lens and chromatic aberrations with faster lens used wide open. Combine that with a high contrast image of stars against the dark void of space and you probably have the least favorable scenario.

The 12mm, f/2 which I got for night photography is supposed to be pretty good quality.

About the camera body

I wanted to get a camera which is affordable, compact and way better than my phone.

It’s a lot to consider when you get the first camera and don’t know what you get into:

Body type: DSLR, mirrorless, point-and-shoot, stick to the phone :)
Sensor size: 1 inch, APS-C, Full Frame
Brands: Canon, Nikon, Sony, Fuji, etc

1. Body type

The main thing I considered here it’s that it’s supposed to be compact and capable. DSLR is too bulky for me and I think it’s just older tech without any real advantage. The point and shoot cameras “feel” too limiting since you can’t change lenses on them, even though RX100 really seems way better than a phone.

2. Sensor size

For the same number of megapixels, a larger sensor has a larger pixel and thus can collect more photons of light. Then the signal to noise ratio is higher and thus the image can be better. This is especially important in low light photography when we need to amplify the signal a lot (set a high ISO) to see the very little available light.

The most recent phones which produce usable images in low light with a very small sensor do a lot of post-processing and what you get is not what the sensor sees. As a fun fact, a sensor in a phone is more than 10x smaller than a sensor in a cheap mirrorless camera.

The most common sensor sizes are full-frame, APS-C, 4/3 and 1 inch. Full-frame cameras are too large for my taste and the lenses are more expensive and they also need to be bigger. I don’t see myself caring these with me. The next best thing is APS-C which looks like something that I’d drag with me on vacations or to the park.

3. Brands

There is a lot of variation here and once you choose a brand you kinda get locked in the ecosystem. Manufacturers usually have their own mount for lenses and while adapters exist it’s not ideal especially for lenses with electronics#### 3. Brands

Sony seems to dominate the mirrorless game so far so I choose the A6000. It looks like their lenses are more expensive and less available than the Nikon lenses but so far I don’t regret my choice. Having almost no experience with other cameras I have nothing to complain :)

Preparing to photograph the stars

In this case, I need to find a very dark place on a very dark day with clear skies.

The very dark day was 1st August 2019 when it was a new moon which means the moon was completely in the shadow. We have one of these every month. I was quite lucky the first one had clear skies and that I didn’t have something else to do.

The very dark place can be found looking at a light pollution map like lightpollutionmap.info. I live in Bucharest so I can find some very dark places ~100km away.

I’ve ended up ~70km away because I thought it was dark enough and very convenient. The next day I found out that I was in a class 4 zone on the Bortle scale. If I’ve driven for an hour more I certainly could have reached a darker place (class 3).

Once you have the day, the place and the weather, all you have to do is put the camera on a tripod and make these settings on the camera:

exposure: for my 12mm lens, the exposure is supposed to be 27 seconds to avoid star trails. I’ve set it to 30” because the other option on my camera was 20” and that was too little.
aperture: opened it all the way to f/2
ISO: I experimented between 1600-3200. I feel like 3200 was a bit too much but the picture I’m going to share was shot with ISO 3200 and then the exposure was taken down in lightroom because I liked it more from the bunch.

The result

What next?!

I realize it can be better in various ways, but to be honest I’m satisfied with this first attempt. If there is going to be the next time, I want to go to a place with even less light pollution, try to make a timelapse and learn how to actually edit these. Also, I could experiment with stacking and if I find something interesting to put in the frame also experiment with light painting.

How to give access on AWS to a consultant

Wed, 03 Apr 2019 00:00:00 +0000

I am a DevOps consultant so I have access to a bunch of AWS accounts, in addition to my own. If I was to use individual credentials for each account, it would be very inconvenient especially since I have to switch between accounts multiple times a day. For a while now, AWS has a nice feature which allows me to access multiple accounts from my own AWS account. This feature is called Roles.

The official documentation for this is easily available but maybe not so easy to understand by everybody.

This guide is intended to be the most straightforward way of giving access to a person who uses more than one AWS account.

Setting up a role for access from another account

Go to IAM -> Roles and click Create role. Ask for the account ID.
Go through each step one by one and select the appropriate permissions.
Cases matter. I use Bogdan every time so it’s easy to remember.
Click Create role and if you’re only in charge of giving access your job is done.

Using this new role in your main account

Click your account name from the top right corner and find the Switch Role button.
Fill in all the info.
Enjoy this faster way of switching AWS accounts.

There is a limitation though to a maximum of 5 accounts but if you just learned about this I am sure you can live with this limitation :)

Conclusion

This is a nice time-saving feature which facilitates using multiple AWS accounts. AWS was somewhat late in the game with this feature. Others from my toolbox that support something similar are CloudFlare, GoDaddy and obviously Google.

Internationalized domain names ...in Linux

Wed, 01 Aug 2018 00:00:00 +0000

The Internet was made for Latin script, more specifically a-z, 0-9 and a hyphen. Of course, I’m talking about Internet addresses, which is exactly how you reach content online. The problem is that around 2 billion people actually use Chinese, Arabic, Devanagari, Cyrillic and other writing systems. Even the French, Germans and Romanians have non-latin characters so let’s see how those are handled online.

I got the domain ă.cc. How can I use it?

Some background info

The DNS, which is the system that helps us get around online using names instead of IP addresses is restricted to only ASCII characters. It makes sense to me because:

The Americans who invented DNS can’t really be blamed that they didn’t think of characters they didn’t understand.
Coordinating the update of many servers on the Internet to support some major new feature is not reasonable.
Adding UTF-8 could be a huge liability. It has some strange characters such as blank and go back one character. This would open interesting possibilities.

Nevertheless, in 1996 a guy from Zurich felt that we need domains with all types of characters so he wrote a draft. People implemented, debated and more than a decade later, in 2009 ICANN brought the languages of the world to the global Internet.

Internationalized domain names (IDNs for short) are domain names which use non-ASCII characters and could be helpful to more than 30% of the world population.

But all this time DNS didn’t change, so how does this work?!

At some point, some guy proposed a standard for converting any character (Unicode) into ASCII. This is called Punycode. For my domain, ă.cc, Punycode looks like xn--0da.cc. The character ă is actually 0da. xn-- tells applications this is Punycode.

So, IDNs are implemented at the application level. Internet Explorer started to support this late 2006 and others a little bit earlier but it seems that 12 years later, support is poor in pretty much any other basic tool.

The Linux situation

SSH, traceroute and many others don’t support IDNs.
The browsers, nslookup and dig are OK.

The thing that handles resolving domain names in Linux is glibc, which has some resolver code copied mostly from BIND. BIND is that DNS server which we talked about in the beginning, that only supports ASCII.
Pretty much all programs link against glibc and when they need to resolve some address, glibc handles it for them.
There is a library in Linux, called libidn, which handles the conversion to Punycode.

nimblex:~# ldd /usr/bin/nslookup 
 linux-vdso.so.1 (0x00007ffc3dde8000)
 libedit.so.0 => /usr/lib64/libedit.so.0 (0x00007fed5f979000)
 libdns.so.1100 => /usr/lib64/libdns.so.1100 (0x00007fed5f551000)
 liblwres.so.160 => /usr/lib64/liblwres.so.160 (0x00007fed5f33e000)
 libbind9.so.160 => /usr/lib64/libbind9.so.160 (0x00007fed5f12d000)
 libisccfg.so.160 => /usr/lib64/libisccfg.so.160 (0x00007fed5ef01000)
 libisc.so.169 => /usr/lib64/libisc.so.169 (0x00007fed5ec89000)
 libcrypto.so.1.1 => /lib64/libcrypto.so.1.1 (0x00007fed5e802000)
 libcap.so.2 => /lib64/libcap.so.2 (0x00007fed5e5fd000)
 libjson-c.so.4 => /usr/lib64/libjson-c.so.4 (0x00007fed5e3ee000)
 libpthread.so.0 => /lib64/libpthread.so.0 (0x00007fed5e1cf000)
 libxml2.so.2 => /usr/lib64/libxml2.so.2 (0x00007fed5de6a000)
 libz.so.1 => /lib64/libz.so.1 (0x00007fed5dc53000)
 liblzma.so.5 => /lib64/liblzma.so.5 (0x00007fed5da2d000)
 libm.so.6 => /lib64/libm.so.6 (0x00007fed5d692000)
 libdl.so.2 => /lib64/libdl.so.2 (0x00007fed5d48e000)
 libidn.so.12 => /usr/lib64/libidn.so.12 (0x00007fed5d25a000)
 libc.so.6 => /lib64/libc.so.6 (0x00007fed5ce70000)
 libncurses.so.6 => /lib64/libncurses.so.6 (0x00007fed5cc46000)
 libtinfo.so.6 => /lib64/libtinfo.so.6 (0x00007fed5ca1a000)
 /lib64/ld-linux-x86-64.so.2 (0x00007fed5fbb1000)

nslookup links against libidn and that’s why it can resolve my domain; ă.cc

Now a few things come to mind:

link glibc against libidn to support resolving internationalized domains names. Well, this is stupid because glibc is primordial. libidn links to it, not the other way around.

nimblex:~# ldd /usr/lib64/libidn.so.12
  linux-vdso.so.1 (0x00007ffe1cbdc000)
  libc.so.6 => /lib64/libc.so.6 (0x00007fc7f9785000)
  /lib64/ld-linux-x86-64.so.2 (0x00007fc7f9da3000)

merge glibc and libidn. It seems this was done before ICAN made their grand announcement, almost a decade ago, but then abandoned.
link tools like ping, ssh and others against libidn.
- it looks like for ping, libidn was encouraged since 2015 but many distros don’t support it yet. Still, I’m sure ping will support IDNs in most distros soon.
- for ssh the territory is virgin.

Conclusion

In 2018 we have AI which recognizes my face better than my family but most tools don’t work with domains such as рнидс.срб. It seems seriously limiting and I don’t like that. I guess I will start bothering some people about this.

Pedal car electric conversion - preparation

Tue, 26 Jun 2018 00:00:00 +0000

I did an impulse buy and got a Baghera Legend Red for my son and I decided to convert it from clean and boring pedal power to dirty and fun electric power. In this first part, I will do calculations and planning so I don’t order too many useless parts.

The motor

Of course, it all starts with the motor. It needs to be:

cheap (it’s a pedal car conversion)
fairly small and light
powerful enough to move 20Kg

At ~24V there is a good selection of motors which would do it for this application. Electric scooter motors start at 250W so I guess I want something that handles 30-150W.

With motors I don’t think it’s proper to rate them only by voltage, current and power because you’re putting electricity through a wire/coil which creates a magnetic field that makes the motor turn. If the motor draws more current then it heats up and the wire insulation could be damaged. Usually, motors draw more current when they are loaded so I consider the rating loosely and I’d keep an eye on the temperature.

I have no problem getting a motor rated for 30W nominal and run it at 100W for a minute or so. It would only shorten the life of the motor slightly and it would provide the bursts of acceleration which are all the fun.

I got a good deal on this motor which is going to work just fine at 50-100W. This one has a nice diagram showing the specs. Most of the cheaper ones don’t.

The battery pack

Once we have a general idea about our motor size we can think about the battery pack. We don’t really have real-life numbers with it but again, we make some assumptions.

The battery pack needs to satisfy 3 things:

voltage ~24V. This will allow us to reach higher speeds but also to extract power from the motor.
current handling >5A. To run our motor at 100W. We need this to accelerate faster or to go uphill.
capacity >50Wh. To get a decent play time.

It could be as easy as 2x12V 7Ah UPS batteries. Both are less than 30$ new around here. The main disadvantage is that they weigh 4KG. Other than that these would be good because they have excellent current handling capability. One disadvantage though is that SLA batteries have this capacity if they are discharged at a very low rate, in 10-20 hours. If we discharge the SLA battery in 1 hour it’s capacity would be closer to 2/3 of what it says on the tin. Smaller capacity UPS batteries cost a little more because they’re not that common and weight just a little less.

Li-Ion has a better energy density (smaller & lighter) but the pack would be more expensive. One cell is 3.7V nominal so 24/3.7 => 6 batteries and a half. I’d go for 7 batteries in series because I want to have enough voltage even when the cells are discharged. When the batteries are fully charged the voltage is closer to 30V but that’s not bad for us. One cell is usually 2.2-3Ah and it’s not recommended to pull more than 4A of one of those cells. There are some high discharge cells on the market but those are more expensive. We can improve both current handling and capacity by adding more batteries in parallel. It looks like the pack that we need is 7S2P (7 series 2 parallel). This would provide a similar runtime for 1/4 of the weight and 2x the cost.

The speed controller

To control the speed of the motor we need to control the voltage that gets to it. Usually, this is done with PWM (pulse width modulation) on a MOSFET. Something similar was done for a flashlight in the past.

To control direction and braking, we need 4 of those MOSFETs in an H-bridge configuration. Also, those MOSFETs are coupled with a flyback diode to dissipate the inductive energy that is characteristic to motors.

The cheaper speed controllers are just 10$ so I’m going to buy one because this project is not about the speed controller. Knowing how it works should help in case it breaks.

The drivetrain

To transfer power from the motor to the wheels, I think there are 3 practical options:

gears (hard to mount)
chain drive (more efficient)
belt drive (less noisy and doesn’t require lubrication)

I am choosing belt drive because I care about noise and not having to grease it.

There is a large variety for belt drive and I am going for timing pulleys of the HTD3M variety because they are fairly common, with decent performance and not expensive. When transferring power from the motor we need to reduce the RPM and increase the torque. The driving pulley is the smaller one and the math is like this: If the driving pulley has 15T (T = teeth) and the driven one has 120T then RPM is decreased 8 times. Our motor is rated for 3300 RPM so, at the axle we’d have 3300/8 ~ 412RPM. Because the diameter of our wheels is 10” then the theoretical top speed is 20KM/h. That’s very nasty for a pedal car which will be driven by a toddler.

After doing a little bit more reading, it seems that because of the higher RPM I can’t really use a 15T pulley, I have to use a 20T one. This means lower torque and high speed up to 26KM/h. I see these options:

limit top speed electronically (torque stays the same but it’s simpler for me)
add another set of pulleys (gain more torque with lower top speed)

Browsing pages 67 to the end in this document indicate that given our 108 Ncm peak torque a 6mm belt wouldn’t do it and a 15mm one would be more than enough. We are using a very short belt so our length correction factor is ~0.7 and to account for beginner errors I am just going to go for the beefier option.

Once we have our choice of pulleys we need to calculate the belt length. Since the math is a bit more involved for that I provide a calculator below which does the heavy lifting.

D=Pitch Diameter Large Pulley

d=Pitch Diameter Small Pulley

C=Center Distance

L=Belt Pitch Length

Center Distance Known
Large Pulley D:
Small Pulley d:
Center Distance C:
Belt Length:

Belt Length Known
Large Pulley D:
Small Pulley d:
Belt Length L:
Center Distance:

In my case, 405mm is close enough to a standard belt size but I’m getting a 408mm belt to make sure it can be mounted.

Servo steering

I am thinking to implement remote control so I need to be able to steer the car. I am just going to do it like they do it on RC cars, with a servo motor.

Basically a servo motor it’s special because it can rotate or push to a specific angle or distance and then keep that position steady. To do that it has a positive feedback loop which compares the input signal which specifies the position, with the actual position and tries to keep them in sync.

Generally, servos are controlled with PWM at 50Hz and rotate 180 degrees.

They tend to have high torque because they have a gear reduction mechanism but I’m getting one which is way bigger than the ones used for RC cars because the car is much bigger. I’ll use a SUPER200 servo which is rated for 200Kg/cm which means it can lift 200Kg at 1cm distance from the shaft. If the lever which moves the wheels is mounted 20cm away from the shaft I still get 10Kg of force to move the wheels so that’s good.

Quick CAD modeling

I believe I have pretty good imagination and I can see in my mind how things fit together. That was a problem in the past because there are always differences from reality so I decided to do a little bit of modeling to confirm all would fit together.

Errors were visible right away so look at the image below.

The large timing pulley cuts in the dark blue metal plate which supports the motor near one of the edges which supports it. I will try to mount the motor 180 degrees rotated so the metal plate would have better structural integrity.
The two timing pulleys collide. I would have to make the metal brackets which support the motor narrower so I can place the motor 10mm closer to the back edge.
The small timing pulley rubs on the motor. I will mount it 2mm offset on the shaft and hope one of the sides will not fall off. I don’t want to use a thinner pulley because we determined earlier 15mm would do it.

After putting a bit more thought into it, this assembly makes more sense like this:

To view the actual 3D assembly click HERE.

BOM for this conversion

pedal car
GR 63X25 DC motor
24V battery pack
motor speed controller
10A fuse
acceleration pedal
threaded rod as a drive train
20T & 120T HTD3M timing pulleys
408mm timing belt (15mm wide)
SUPER200 servo

We’re getting close to 300$ in parts, without considering the cost of the car.

… I’ll think about it.

Operating Systems hidden on your PC

Thu, 16 Nov 2017 00:00:00 +0000

Whether you use a Mac or PC with Windows or Linux, your Intel computer runs some other operating systems in the background without bothering you with this piece of information. These OSs run all the time, have amazing capabilities and are impossible to remove.

Some time ago, computer scientists devised a mechanism called protection rings which has the purpose of allowing operating systems limit access to resources. It goes from ring 0 (most privileges) to ring 3 (least privileges). For example, the kernel of your operating system runs at ring 0 and your web browser at ring 3. The kernel should have the capability to access all the RAM to allocate it to different programs, but your web browser should not be able to read the memory of your bitcoin wallet software.

Then it came hardware virtualization and that hypervisor runs one level below the kernel so that the guest OS can run just as before. We call this ring -1.

Bellow ring -1 there are many other things lurking in the shadows.

System Management Mode (SMM)

This is a special operating mode of your CPU, which runs code from the firmware completely independent of the Operating System. Some people call this ring -2.

It does things such as:

control fans or shut down the computer when the CPU overheats
resume the OS from standby when the laptop lid is open
emulate a USB keyboard as PS/2 so it can be used in a super legacy OS like DOS
control voltage regulators to manage how much power your CPU gets
vendors use it to implement features specific to their devices

SMM was designed so it can’t be overwritten by the main OS when it’s enabled and it reserves 8MB of RAM which is not visible anymore.

SMM runs the vendor code all the time and you can’t do much about it. You don’t see what and when it runs and it has excellent control over USB or other peripherals.

The code that runs in SMM takes time away from applications and in current implementations, it switches all the cores even to add 1+1.

But of course, you should trust your vendor will do a good job here. It’s not like NSA exploited SMM against the most reputable vendors like Dell, HP and Juniper.

Unified Extensible Firmware Interface (UEFI)

The word “extensible” sounds good in general, but not so good for things you want to be specific, simple and secure.

This is a piece of software that replaces the BIOS which has some limitations such as it can only run in 16bit mode, address 1MB of memory and use up to 2TB drives. The specification was originally developed by Intel but in 2005 this responsibility was transferred to the UEFI Forum.

The main features of UEFI I would like to debate are:

CPU independent device drivers
The possibility to run EFI applications
Graphical features which allow having a GUI before loading the OS
Network booting (PXE)
Secure Boot

This list of advantages sounds like having an Operating System?!

Let’s take this great features one at a time:

Universal drivers should be a great feature. I speculate that many of these drivers could be used by the main OS directly through the UEFI API. The truth is this most often they are implemented twice; once for firmware and once for the OS.
The application I willingly run is the GRUB boot loader which gives me flexibility when I start my operating system. I can’t think of anything more that I would want to run at this stage so I can’t appreciate this capability too much.
I don’t really need anything before I load the OS. I only need my OS to load fast so I wouldn’t mind a couple of seconds of black screen before the kernel of my OS initializes the display.
Network booting is a neat feature. I have also used it in the days of the BIOS when network cards had a dedicated chip which included the software implementation for doing that. This feature allowed me to either run stateless machines without HDDs or to automatically install the OS without handling CDs. My concern now is that in the context of UEFI, this means an obscure application can communicate over the network and applications now are not limited like they used to be.
Secure boot is complete bullshit! It was mostly used by Microsoft to restrict people to only use the Windows which comes preinstalled. It restricts the firmware to only load a kernel signed with a private key which corresponds to a public key which is loaded in the firmware.

Now let me tell you another concept. In UEFI there are boot services and runtime services. The boot services run until your operating starts loading and runtime services run also when your OS is running.

Most of UEFI is closed source and only the vendor knows what’s there. Some parts of it are OpenSource but basically that doesn’t make a difference for the end user.

In a nutshell, UEFI is an extremely complex proprietary kernel which runs on the main CPU even while your OS is running. Exploits where demonstrated this year which work even when updates and security measures are in place.

Intel Management Engine (Intel ME)

This is the masterpiece. RING -3

It is a separate processor core embedded in the CPU package, with internal ROM, RAM and DMA (direct memory access) to the main system memory. In addition to that, it has network access with its own MAC address through the ethernet controller.

When you apply power, the first thing that starts is ME, which loads its firmware from the flash chip where you usually also have the UEFI firmware and SMM stuff. Parts of this firmware are signed by Intel and if it doesn’t find what it expects it either doesn’t start the main CPU or it reboots the computer every 30 minutes.

One nice component of ME is Active Management Technology (AMT) which can be used remotely even when the PC is turned off.

Another nice one is Protected Audio Video Path (PAVP) which allows ME to communicate with the GPU so it can read and write to the screen. I wonder about the audio capabilities which are hinted in the name.

In a nutshell, ME is the nicest control tool which runs a proprietary MINIX based OS that can do everything on your computer:

power ON/OFF
access all memory
view/paint your screen
read your keystrokes
… ?!

Having so nice features it was surely bound to be vulnerable for 9 years. If you have an i3, i5 or i7 computer bought before the end of 2017 then you are likely vulnerable.

If you want to know more, in the PCH datasheet at chapter 5.25 you can see a short overview. The PCH is the die on the right side of the image below.

Can we do something about it?

Technically speaking, the options are very limited. All these beautiful things are burned on a flash chip like this one you can see on the right side of the CPU in my PC.

In theory, we could do the following:

Disable parts of ME with this Linux tool or the Windows one.
Buy a SPI flash programmer to dump all the firmware.
Use me_cleaner to remove a good part of the ME firmware.
Flash the minimized firmware back to that chip.
Hope for the best
To improve the UEFI and SMM issue, it’s going to be much harder. You’ll have to learn about heads, coreboot, the UEFI specifications. Then you’ll have to build a custom UEFI of your own which will start a Linux kernel very early on (at BDS stage). This kernel would take a good chunk of the EFI responsibility and it could also run your Linux OS. For me, it will be a little more difficult because I have 32bit UEFI and it would only start a 32bit kernel which I could then use to start the main OS kernel.

… or maybe I will just buy a Purism computer next time.

If you are a Windows or MAC user, I think your only option is to complain to the government. Spreading awareness is something that I am sure you can do.

We only talked about Intel here but AMD does something similar. Chromebooks are a little less affected by this but they only run Google stuff.

Conclusion

Most of us are always running a version of MINIX on their computers, which makes Tanenbaum very happy because MINIX is used more than Windows, Linux or MacOS.

To access AMT you can use the user “admin” with an empty password and you have complete control. The right janitor in the wrong network would make a very interesting story.

We also run UEFI and SMM stuff all the time which was proven to be exploited very effectively time and time again.

Basic function generators

Mon, 06 Nov 2017 00:00:00 +0000

If you have electronics as a hobby it’s unlikely you are going to get away without using a function generator. Its role is to make electrical waveforms over a range of frequencies while allowing to control the amplitude of the waveform. Some basic use cases include testing speakers or amplifiers, calculating inductance for coils or experimenting with resonant circuits such as for wireless energy transfer, induction heating or tesla coils. In some cases, I got away without spending a cent but at some point I got a 10$ kit and a while later I spent 50$ for a finished product. A “normal” generator starts from 300$ so what exactly is inside the 50$ one?

Specs to consider

Waveform types should be at least sine, square, triangle. Sawtooth and reverse sawtooth can also be useful but an ARB generator will do everything.
Frequency range: I would be satisfied with 0.1Hz-1MHz but if you only play with audio applications pretty much any generator would cover the range.
DC Offset is not really a must-have feature but I would be happy if I could add +5V DC offset to the signal. For example, it would help to fake digital signals.
Duty cycle control is often used to control the power supplied through MOSFETs. For example, with PWM I am adjusting the light level in a flashlight.
Output level represents the maximum and minimum level of the signal. If it does 10Vpp then you can have +/-5V signals and it would cover a lot.
Output impedance is 50 Ohm standard but with crappy generators is larger. If the impedance is larger it would have less driving capability for a load.
Modulation would come handy if you plan to do more advanced experiments. Amplitude, frequency and phase modulation are the common types.

Analog or digital

If we want to do it in less than a day, analog is the best option. We should be able to do it with a few components and for very little money. The first chip that comes to my mind is XR-2206 and in its databook from more than 30 years ago, you can see some good notes starting with page 42. The other one I am thinking about is ICL8038 and this one has the advantage of being readily available in my parts bin.

They are both designed for generating signals and I guess with a dual opamp, three potentiometers, some caps and resistors will be all done. Half of the dual opamp I could use as a summing amplifier for offset and the other half as a buffer. The potentiometers would be frequency, amplitude and offset.

This approach with analog would be fun, easy, quick and cheap but lacking in some areas. It wouldn’t be easy to precisely set the frequency, not high frequency and having two channels in sync with controllable phase shift is not easy with these. Not to mention they are not being manufactured anymore.

Buying a decent analog generator is not really attractive to me because it’s bulky and I live in an apartment.

With a digital generator, we dial in exactly the frequency that we want. Coupled with some simple software it can sweep between two frequencies in a predefined amount of time or even generate some arbitrary waveforms with the fancier ones. These unlock a lot of other possibilities and are by far the most common commercial function generators these days.

Out of the digital type, the most common ones use direct digital synthesis, DDS. They are very versatile but of course more complicated and more expensive.

DDS uses a phase accumulator, a look-up table containing a digital representation of the waveform, and a DAC. The phase accumulator advances position each time it receives a clock pulse. This position in the look-up table contains a digital value that is feed in a DAC.

As a DIY approach, I would probably consider using AD9833 which is not expensive and would allow me to generate signals up to a few MHz. I would add a second one for having two channels, some Arduino compatible board, a nice LCD, rotary encoder and plenty buttons to make it easier to jump to the functionality we care about. I don’t like instruments where you spend time navigating through menus. Of course, just like before we need the opamps for controlling the offset or amplifying the signal.

The free generator

Yeah, it’s just a jack cable which I hook to my phone where I run the Function Generator Android App.

Sacrifice a pair of old headphones and you got yourself a dual channel function generator for free. Its sample rate is bad, the output level is horrible and there is no DC offset but you can’t beat this price point. This thing is just good enough to test audio stuff and it can be set up in no time.

The 10$ kit

I got this mostly because I felt like building a kit and I imagine it would come in handy at some point. I like its design for its simplicity but I don’t enjoy using it because it’s awkward to operate. It didn’t come with a power supply so I adapted the picoPSU I’ve had around. The power supply was way more expensive than the kit!

The design of this seems to be copied from ScienceProg.

The Atmega16 micro generates signals which go in an 8bit R-2R DAC that then goes in an OpAmp for setting the amplitude and offset.

Its frequency range (0-65KHz) and output levels (10Vpp) are a few times better than the Android generator but I ended up using the Android generator way more than this. Without having it in a proper case with buttons it’s just not fun to use.

The 50$ unit

It comes fully assembled as a product, packed in a cardboard box, with charger and some cables for 50$ including shipping from China! Its specs are impressive to me:

0.01Hz-25MHz for sine and up to 6MHz for the other waves
5mVpp~20Vpp amplitude
12 bits amplitude resolution
Arbitrary waveforms with memory for my own
PC control over USB
… and the list goes on and on

How is this possible?! What’s inside?

It’s actually got quite a few things which are not necessarily the cheapest if we got 1000 of them from a reliable supplier like DigiKey or Mouser.

From left to right:

There is the switching power supply which converts 5V to +13V and to -13V. ~ 1$
Next to the power supply it’s the ST 8bit, 16MHz microcontroller which most likely handles the LCD, menu and all that user interaction. < 1$
Bellow the power supply it’s the USB-SERIAL converter which allows us to talk to the micro and an EEPROM for storing ARB waveforms and settings. < 0.3$
Then it’s the Lattice MachX02 FPGA which is one of the cheaper ones but still one of the most expensive single chips from this board. < 6$
The R-2R resistor ladder is a good way to make two DACs without spending money. When a 12bit DAC is implemented, 0.1% resistors should be used but here cost is the driving force so normal resistors were used. < 0.1$
Then the signal flows in the two AD8017 variable gain amplifiers. < 6$
Just before these but not in the signal path there are two LM OpAmps which control the DC Offset. These are super cheap. ~ 0.1$
Last, under the way too small heatsink there are the most expensive chips. Those are the output buffers which actually drive the load that we hook up to our generator. My board had THS4022 which are pretty high-speed chips so at 25MHz the sine was still looking good. > 12$

Quicky adding these up, then making some very optimistic estimations for PCB, connectors, plastic case, LCD, power adaptor, cables, cardboard box and labor it probably goes to a minimum of 35$ for the cost of goods sold.

This also shows that both resellers and manufacturers operate on very thin profit margins. I am sure they pay less than my estimations and I would really learn how they do that.

Improvements I did

Replace THS4022 with THS3092 which is slightly cheaper and can handle more current. This means we maintain the amplitude better at higher frequencies.
Replace the heatsink with a much larger one. At high frequency or with low impedance loads this thing gets hot.
Add some decoupling capacitors on the input of the power supply to remove some noise. On my board I added 1600uF and it was enough for medium loads.
Add a small heatsink to the XL6007 switching converter. This works at the very limit of its output current when the generator is driven hard. In it’s SEPIC topology it should do 0.6A but the OpAmps want way more.
Blocked the light from the LCD to bleed to the LEDs.

Improvements I wish the manufacturer did

Much better SEPIC converter. This implementation can’t handle enough current.
OpenSource software. I am sure nobody can make it cheaper so if the software was OpenSource people would be happy to improve it. For example, I don’t see why this unit wouldn’t support modulation.
Better load handling.
Lower noise.

A professional generator board

Sigilent SDG 5000 series are entry level arbitrary waveform generators which might actually be used in a professional environment. This one is more than $500 and as you can see below, the main board is way more complex. The obvious things are a fancier FPGA, SRAM, a nice DAC and many OpAmps in parallel on the output.

Another board does only the software and control stuff and this one is also quite impressive to me.

Conclusions

Function generators can be fun :)
As usual, you get what you pay for so don’t get the 10$ Kit.
The 50$ MHS-5200A unit is good bang for the buck and with some small improvements and OpenSource firmware, it can be even better.

Flashlight with software

Thu, 12 Oct 2017 00:00:00 +0000

So, you know when you are tired or bored and you go to strange places online? I somehow found this flashlight online and from the picture, you can see many high power LEDs + many high power batteries + adequate cooling => awesome light output. The price at that time was ~50$ so I ordered because it seemed I get a lot for my money.

As a technical guy, I see 10x CREE XM-L T6 and I go straight to the datasheet to see what that is about. It looks like one LED is able to output 692 lumens at 2A and 3.2V. This is the rated power but the datasheet also suggests it can be overdriven too, so I could speculate one of these LEDs can reach 1000 lumens. 10 of these would output 10000 lumens and they would consume either MANY amps (in parallel) or volts (in series).

Anyways, from the theoretical (not the advertised) specs it looks like a 100W LED flashlight which can output 10.000 lumens. This should be great.

I ordered, it arrived, and it wasn’t great. The light output was not impressive so I disassembled it to see what’s up.

Principle of operation

All those LEDs are in parallel so to reach max output it should draw more than 20A and that’s insane. That’s also at the very limit of the 4 parallel 18650 cells but I guess in theory it should be OK to go to maximum for a few seconds. A quick measurement revealed it draws under 1A, so something is clearly not right.

Looking at the electronics we can quickly deduce it works like this:

Pressing the button is registered by the microcontroller.
The microcontroller switches the MOSFET on/off. By changing the amount of time the FET is ON vs OFF the light output can be controlled.
Some resistors limit the current flow between batteries, MOSFET and LEDs.

More is less

The first thing to try is shunting those resistors so more current will flow to the LEDs. I can immediately see the light is way brighter and now it’s drawing around 3A. It’s not even close to the expected maximum not now at least is nice.

I played with it for a little more and suddenly it was dead. No smoke or something like that, it just didn’t work anymore.

Probing the circuit a little I was able to determine the microcontroller was faulty. If it was the MOSFET then I could easily replace that and get it working. The microcontroller had no markings and I had no software to flash to it anyway. Before I abandon this I decided to hook the LEDs directly to the bench power supply to see how much light can these actually deliver and it seems not more than 4A. It seems the LEDs are also fake. :)

I clearly didn’t get what I paid for.

Less is more

One year has passed, and that weekend when I made time to do something different arrived.

The footprint of an ATtiny13 matches what used to be on that board so I incorrectly assumed it should fit right in. Reality is that it doesn’t even match the VCC and GND pins but it’s OK; we cut a few traces, bypass with a few wires and we make it work.

Basically, we need to power the ATtiny13, connect the MOSFET to a pin that can be used for PWM (pin 5 or 6) and the button to a digital IO pin which is pretty much any other pin.

I made this simple programming rig where I used an Arduino nano as an ISP and connected a button and an LED to the ATtiny13.

Code

All I want from this flashlight is to be able to control the light intensity.

I keep the button pressed from Off and the light starts ramping up.
I keep the button pressed from On and the light reverses until it turns off.

That’s easy to use for everyone and easy to implement.

I have an array of 16 intensity values which should be somehow logarithmic since perception of light is not linear. We go through those every 100ms and when you lift the finger of the button light stays at that level.

Initially, the array had 64 values for much smoother ramping but ATtiny13 doesn’t have enough RAM for that. :)

The signal that is sent to the MOSFT when the button is pressed looks like this:

With this code, the microcontroller works all the time it’s powered. Fortunately this micro is not power hungry.

Just as a quick and easy power saving method I disable the ADC and the Analog comparator to save 25% energy and if I configure it to work at 1.2MHz instead of the default 9.6MHz. This cuts the energy consumption in half once more. It now uses under 1mA so it’s good enough for now.

End result

After fitting our new microcontroller on our board, it looks like this:

… and the end result looks like this: I actually press the button twice. Once to go to max and again to go from max to zero.

YEAH! The flashlight is alive again!

Possible improvements

Read the battery voltage with the ADC and reduce the output or produce a warning strobe when the voltage is too low.
When the flashlight is off power down the microcontroller and wake it up with an interrupt when the button is pressed. This would improve the standby time a little.
Automatically reduce the output from 100% to 50% after a number of minutes. This should help a lot with heat and battery life.
Go hardcore and make another board which can accommodate more MOSFETs for better power dissipation, temperate sensor and ATtiny85 which has more memory.

Since the flashlight is usable now, I am going to stop here but if I was to develop a product, I would definitely do those improvements.

Power meter for the entire apartment

Wed, 19 Apr 2017 00:00:00 +0000

In the weekend I wanted to do something fun for improving the visibility of energy consumption. I ended up with a nice LED display that shows usage in realtime and some cool charts on my phone.

Principle of operation

We learned in middle school that current passing through a wire generates a magnetic field and that a magnetic field passing through a wire generates electricity. Later I learned that this is the basic principle of the transformer. Our current sensor is essentially a transformer which generates a voltage across a burden resistor proportional to the current passing through the wire we are clamping the sensor to.

Now we need to read the voltage from the sensor and do some calculations to determine the power usage.

Since our voltage is AC, it has negative values and we would have to add a DC offset to it so the alternation would only happen above zero.

Our microcontroller has an ADC which would convert the voltage value to a numerical value depending on it’s resolution. A 10bit resolution could have a value between 0-1023.

Finally, the circuit going to our ADC looks something like this:

Implementation

Our BOM (Bill of materials) is short and cheap (~$20):

Microcontroller: Wemos D1 mini
CT Sensor: YHDC SCT-013-000
Display: Dot matrix with MAX7219 driver
Power supply: HLK-PM03

Yeah, I am a fan of the ESP8266 microcontroller even though it doesn’t have the best ADC. Still, it has WiFi and works well with the Arduino SDK.

To determine the burden resistor for the CT sensor we have two options:

Do some math

Assuming we want to measure up to 9kW then the peak current would be: 9000/230=39.13 ~ 40A
Since we have AC, we have to multiply with √2 to get peak values: 40A × 1.414 = 55.33A
Our sensor is rated 50mA on the secondary for 100A on the primary so we have 100 x 1000 / 50 = 2000 turns.
The peak current we can expect on the secondary will be: 55.33A / 2000 = 0.0276A
Now the burden resistor should be chosen so that voltage doesn’t go over half of 3.3V: 3.3 V / 2 / 0.0276 A = 59.7 Ω This is close to 56Ω which is a common value so we can choose that.

Do some measurements

Took out the old scope and measured Vpp with various burden resistors and with all the major appliances running. ;) WTF is this?! Where is my sine wave?! Yeah, with all those reactive and non-liner loads (Fridge, TV, CFL and LED bulb, computers, etc) it will not look like a sine wave. If I plug everything off and only run the stove or some other resistive load it looks like a very pretty sine wave but with normal devices, never.

It seems that the sweetspot is 2 x 47Ω and adding all the running appliances sums up to ~6kW. I added a calibration feature in the Android app anyways and this way we could get slightly better results.

After a bit of soldering and coding, our prototype looks like this:

Here my meter is measuring 1525Wh and the commercial thing I have is measuring 1550Wh. The error margin is less than 2% so I’m happy with this result. We don’t know how well the commercial meter is calibrated anyways.

Code

The SDK we are going to use is Arduino and a bunch of libs that make things simple.

Blynk will help us with the Andorid app and communications through Internet.
ArduinoOTA will allow us to do wireless firmware updates.
MAX7219_Dot_Matrix is obviously the lib which paints the display.
EmonLib does all the math for energy monitoring.

After the prototype testing I installed it in the fuse box and it looks like this:

… and the Android app looks like this:

Possible improvements

Add more channels.
Improve the resolution.
Predict the cost.

The ESP8266 has a single 1V, 10bit ADC built in. This isn’t great if you want precision. Using an external ADC such as ADS1115 would allow us to go to 4 channels, 5V and 16 bit. It would be a huge step up and just a small increase in cost.

Hacking the washing machine

Tue, 11 Apr 2017 00:00:00 +0000

I got the perfect washing machine in my early 20s when I was single. It has some predefined programs, no display or any other complicated features. I was using mostly “power + quick wash”. Now, in my early 30s, my wife uses programs that take hours and staring them after work means the cycle would be done after midnight. So, let’s make that hunk of metal controllable from the Internet.

Step 1 - Understand how it works

At first impression all is simple. It’s a relatively unknown micro with many GPIOs, some used as inputs for buttons and some to switch transistors that drive LEDs.

I got the control board out from the machine and hooked it to a bench power supply but nothing happens. Obviously there is also a power module with relays and stuff for controlling valves and the motor. It turns out that on this power module there is an EPROM which identifies the model and which allows the control board to actually run. That is not going to be an issue because we can test with the complete circuit.

Probing around with the multimeter made it clear quickly that the buttons pull GPIOs to GND. LEDs on the other hand, are NOT connected to GND or to a voltage rail directly through a resistor. The LEDs are connected to GND through individual transistors and power is applied to all of them through another transistor.

Doing a bit more investigation, seems that there are a lot of failsafe mechanisms so the machine wouldn’t run with a jammed engine or failed valves. Still, we don’t care about all that so we move on.

Step 2 - Interface a wireless microcontroller

Our microcontroller of choice is going to be the notorious ESP8266 for obvious reasons; it has builtin WiFi, it’s cheap, easily available and works well with easy to use SDKs such as Arduino. It has some minor shortcomings, such as it’s logic is 3.3V while the micro in the washing machine is 5V. That it’s easy to overcome with some simple voltage dividers or a level shifter circuit so we move on.

We just want to do two simple things:

Push buttons to actually start wash cycles.
Read LEDs so that we know remotely the state of the machine.

If we only did the button pushing thing then we would have no guaranty things are actually happening.

For pushing the buttons we will connect some transistors to ESP8266 and use them as a second set of switches. This will use 3 GPIOs and handle Power, the 40°C program and the 60°C program.
For reading the LEDs I decided to use a level shifter I had around. We will use all the 4 channels to detect the power, lock, 40° and 60° LEDs.

It would be a good idea to improve the 5V power supply included in the power module because our microcontroller draws extra 100mA and the added consumption takes us very close to the limit. Still, I will do that some other day.

Code

The SDK we are going to use is Arduino and with just a library or two this is going to be really simple.

Blink will help us with the Andorid app and communications with the microcontroller through Internet.

WiFiManager will help us not hardcode the AP credentials in the code so that when we change the wireless router, we don’t have to reflash the washing machine.

After some testing with the ESP8266 disconnected I procedded to solder it in and it now looks like this:

… and the Android app looks like this:

Step 3 - Enjoy

To me, this is still the best washing machine because I don’t have to buy a new one :)