Thursday, February 4, 2021

Dead Instrument Cluster, part: The Second

 See my previous article and then know that 4 days later it died again. 

I was lucky in that I never cancelled the appointment to get it fixed so when that day rolled around I drive it in without any instrument display. The only danger was that I didn’t know how fast I was going which was easily solved by running Wayz on my iPhone and that I didn’t know how much gas it had in it but my wife assured me it was nearly full when it went dark again ;) 

They took over 2 hours to decide that it was actually broken again and that they would need to order a new instrument cluster. I told them that in moment one, but OK whatever. Since the last one failed only 4 months ago they are going to replace it under warranty. 

They really didn’t want me to take it back home without a speedometer, so I had to sign a hand written phrase on the receipt to the effect that I had been counseled about the dangers of driving it without a speedometer. I signed it, I have one. At least approximate on my phone. I tried to explain that too.

Sadly the part is back ordered so they don’t know how long it will take to get a replacement.

Luckily I decided to take it apart yet again and do some lower level debugging of the connections to the board. This time I was going to take it apart, actually de-solder a component, and then re-solder it with leaded solder and put it together enough to do a quick test. That way I can be sure what I touched and didn’t unlike the first time when I just randomly reflowed anything that was large enough for me to safely handle without a lot of prep work.

Step one was just the connector to the board itself. I used a combination of a solder sucker and solder wick to remove maybe 80% of the solder holding the connector in place from the back. And then generously fluxed it and resoldered it with standard leaded solder. Put the device together enough to test it with both displays connected again. and the screws holding them in installed so that the very weak little connectors for them wouldn’t get broken or torn and it lit right up.

So the problem is definitely in the connection of the connector and not in the other power supply components that I resoldered in the first entry of this saga.

If you think you can do this I would encourage you not to. See my first entry that has more info on the very delicate connectors for the screens and other parts. But since it’s dead anyway and they were going to replace it anyway I decided it couldn’t get any more dead than it was so I took a chance.

If it is still working when I finally get the call from them in the future that the real part is back in stock and to bring it in for replacement it’s going to be a tough decision what to do. Do I keep the one I’ve been fixing and hope it keeps working? Do I let them replace it with another one that is definitely going to have the same problem but now I know how to fix it? Do I let them even notice that I fixed it? Or just claim that it spontaneously started working again and has been on and off for all that time? I like that last one but I don’t want to lie to them. I’d really like to keep this one as a spare and let them replace it with a new one. Thats the offer I’ll make them I think. I’ll be sure to update with more information if the backordered replacement ever arrives ;) When we got this first replacement we had to tell them what the odometer reading was so that they could get it burned into the new one. They didn’t ask me this time so either they will call back when it is ready to go and then spend more time getting that burned in or it will just come with it set to 0. Since the car only has 40k miles on it it would be hilarious to reset it to 0. I’m pretty sure that they will call to ask me. The temptation is to tell them less than the last time just for fun ;) We keep our cars until they are ready to be junked so I won’t be using that to sell it to someone else fraudulently, but just curious what they would say...

Friday, January 29, 2021

Instrument Cluster Board Level Repairs And The Evils of Lead Free Solder

 My wife drives a Ford C-Max “energie” plugin hybrid. It is an awesome little car that Ford no longer bothers to sell in the US because they failed to market it well enough and it became unprofitable for them to keep them around. Or it is possible that Americans really are only interested in double sized trucks and SUV’s and simply don’t want smaller cars at all. I don’t know but I suspect a combination of both problems exist. In any case I was in Chicago 4 months ago caring for my mom and I got a call from home saying that the dashboard on the car was completely blank. Had to manage remotely getting her a rental and getting the car fixed at the dealership. A couple of thousand dollars later we had a new instrument cluster as the old one was dead. It took weeks to get fixed as a new one had to be ordered with the proper odometer reading burned into it. 

This morning it did it again! Not even a full 4 months later! Car starts but the instrument cluster stays dark and eventually the touch screen/ radio comes up with a message that says “Vehicle Network Communication Error” and will not do anything else. Other interesting side effects are that the climate controls defaulted back to Celsius without being able to talk to the instrument cluster. 

Since there is no way to control most of the car and no way to know easily how fast you’re going the car is simply not safe to drive that way beyond just not having to have it towed to the dealership. 

The local Ford dealership said that since this was an electrical problem it required special attention from special people and they wouldn’t be able to even begin to work on it for 5 days! Don’t bother bringing it in before this coming Wednesday! And no willingness to tell me if this would be a warranted repair since they just did it. 

As I sat here fuming about how long we’d have to rent a car for as they ordered another new one for us I decided that since it was dead anyway I’d take it out and have a look at it. I’ve rarely regretted doing that and have often found that I could make a repair to something that would have cost a lot of money to replace unnecessarily. 

It turns out that it’s very easy to remove the instrument cluster from most Ford cars. There are lots of youtube videos about it but basically you pull the cowling that connects the rubbery sheet that covers the top of the adjustable steering column to the plastic under the cluster straight towards you and then that reveals the 2 7mm bolts that hold in the cluster. Make sure to adjust the steering wheel all the way down and forward or you can’t get it out. On this C-Max there are 2 clips that you need to force to let go directly behind it, but I have no idea if just yanking it straight out is the right thing to do for your model or not. It was the right thing for mine. Grab the cluster and pull it straight towards you until they let go. If you do it wrong or yours is held on differently though you might break something expensive so find a take apart video for your specific model before you start applying excessive force to anything.

There is a single connector on the back that you can reach around and find the pinch spot to easily remove it. You may want to put a towel over the cowling part or any other exposed plastics to keep from scratching them as you figure out how to get the cluster out from between the steering wheel and the dashboard. It is possible, but only just.


It actually look me longer to clear enough room on the desk to work on it than to get it out of the car ;) After carefully removing all the other plastic coverings you get to the point of having to remove the speedometer meter pointer. There is no D shaped or otherwise locking connection point for this. It is strictly a friction fit piece of plastic pushed down over the pin that comes up from the stepper motor that drives it. At this point it is recommended that you take a piece of tape, I used electrical that didn't seem to damage the plastics or paint or printed parts underneath it but go easy on it, and make that line up with exactly where the pointer is at it's sopping point. You'll notice it is a little below zero but if you get this wrong when you put it back together you'll be driving at a speed other than what you think you are. "I took the dash apart Officer and evidently did not get the pointer back on right" is not going to get you out of a speeding ticket. Put some tape on the underside lined up with it's stop point so that you get it back together right.

I used a "spudger" and very gently pried it up on both sides. The plastic these parts are made of feels like the lowest quality styrofoam and I believe they would break if you so much as looked at them wrong. If you don't have a proper spudger put some cloth or something to protect the black plastic background as you pry upwards or you will damage it and have to look at the light soaking through your scratches forever.

The next thing you have to remove to get to the board are the connections to the 2 screens on either side. I didn't capture a picture of this but I HATE these kind of board connectors. I've ruined many similar ones on Raspberry Pi camera connectors or others. In this specific one the connectors had a black plastic bar across the ribbon connector that you VERY GENTLY pried upwards to release the ribbon cable and then could slide the ribbon cable out. For me they worked very well but if you have any qualms about this sort of thing don't do it. You must remove them to safely work on the board though because once you take off the front plastics nothing else is holding them in place and any force or movement will be transferred to the connector which will either rip the cables or tear the connectors right off the board. It does not take much force to do that at all so really be seriously careful at this point. I would also make sure you don't confuse the left and right screens. I couldn't see any difference in looking at them but it seemed like a good idea not to put it back together backwards. They would certainly fit just fine backwards but who knows if they have different firmware or something.

And finally you'll have the nice board you can have a good look at!

It is, of course, almost entirely surface mount and there just isn't much you can do about that unless you're much more of a professional soldering god than I am. I put it under the inspection microscope and can see a lot of connections that looked a bit iffy to me. The first thing I did was to put some flux over the through hole pins that go to the connector on the opposite side of the board. You can see them as two rows of parallel pins in the picture aboe right in the middle of the board towards the top. I then carefully touched each one up with some nice antique leaded solder making sure not to over do it. I then searched the board for other such connections and found lots that looked like they were not particularly well done. I also touched up the larger connections in the several different power regulators on both sides of the board though I don't know if that or just the connector itself was what finally brought it back to life.

Many of the CPU connections looked quite badly done to me as well but I did not try to reflow any of them in the initial attempt to get this working as I was almost certain that if I touched them I'd make things worse and not better.

I re-attached the screens and the plastic case parts necessary to hold them and provide for the cable attachment point to the car and took it back to the garage and low and behold it lit up!

Took it back upstairs and re-affixed the speedometer indicator and the rest of the plastics and it kept working! 

This is almost certainly a failure to make proper connections while using lead free solder. The connections seem to become brittle and just break. I have assembled only 1 device with lead free solder, a wonderful clock kit and every 2 years I have to take it apart and resolder one of the LED displays with good old fashioned lead free solder as the original connections fail one after another. 

It should be possible to do such things properly with lead free solder, but it is my opinion that it requires slightly different design and probably different or better flux and other such things. The solder joints themselves look so horrible with lead free solder that I think it would be impossible to validate them with visual inspection the way you can with leaded solder. A cold leaded solder joint is obvious. The ones that failed with lead free solder look identically awful to me to the ones that are still working. 

In any case, if you have the skills you should take things apart and look at them before you pay someone many thousands of dollars to replace it. We may be putting less lead into the land fills but we're putting a lot more entire assemblies into them because of it. It seems there should be a solution that properly recycles leaded boards while still letting you use what is so obviously a superior way of making things that should last for more than a few years.

I'm off to cancel the appointment at the Ford dealership for next week and to complain to someone who has no idea what the parts are made of and who won't know who to contact to complain at corporate that their instrument cluster boards are patently defective because they have not carried lead free solder construction techniques over into their new lead free board designs. I know it's pointless but I'm still going to complain to them...

Monday, August 3, 2020

Power Supplies for ESP32 Arduino Based Projects

You’ll be surprised to learn that not all power supplies are created equal ;) Or if you’ve bought more than 1 or 2 online you’ll not be surprised by that at all. There are some that are so wonderfully cheap and seductive that you purchase them only to have them fail on day 3 or do something even worse. I have fallen prey to that in the past, but not anymore. This post is nothing about any of that, but just the well documented differences between name brand (not rip off!) imported power supplies. There are still issues you need to be aware of when selecting a power supply for your Arduino or IOT based project and evidently, specifically for your ESP32 based projects. 

Here you’ll see 2 name brand, Mean Well Din rail mounted power supplies of similar ratings.

The one on the left is a MeanWell model: HDR-15-12 with a slightly blurry sticker on the top which I apologize for. It accepts line voltage input and outputs 12v at 1.25 amps. It is also very nicely tiny and uses only a very small amount of your din rail space. It’s also cheaper than any other Mean Well supply I’ve seen of similar output. It also did not work reliably with an ESP32 based IOT device. The one on the right is also a Mean Well supply model DR-15-12. It also outputs 12v at 1.25 amps but it has the “DC OK” indicator on it.

The cheaper one on the left caused my ESP32 based device to reliably startup in firmware upload mode if you waited long enough between re-powering for the internal capacitors to be discharged. If the power was disconnected for more than a minute or so the device would not properly restart when power was applied again! This is obviously not acceptable as you need to be able to plug in a device and have it actually start up. The DR-15 model on the right starts up the device every single time without any care for how long it’s been disconnected and is only slightly more expensive and only slightly wider than the cheaper one.

So whats the difference? The only thing I can find is that the “Rise Time” on the spec sheet for the 2 is significantly different. It is listed in the “output” section of the sheet and along with the “setup” time. The setup time is the amount of time between when power is applied before the output actually turns on. If you’ve used many switching power supplies you’ll have noticed that there is a delay between applying power and any voltage coming out of them. This is not generally an issue, just the housekeeping and charging up of everything that the power supply must do before it is ready to supply regulated output. It’s the “Rise Time” that seems to be important. 

In the case of the cheaper HDR-15 model the rise time is 80ms. Which doesn’t sound like a lot. The slightly more expensive one is only 30ms. I have measured these on the scope and it does indeed seem like they are the problem. In my case they are running a secondary 3.3v switching supply on the board, so your exact requirements and the point at which is causes you trouble will be different based on that.  

It does seem that a longer Rise Time can cause the ESP32 (and possibly ESP8266) based designs to startup in firmware upload mode, as if you were holding gpio0 to ground even when you are most definitely not, rather than starting up normally.  There is some allusion to this in the ESP32 documentation about making sure that if you’re running from batteries that you have a power controller that will not apply power to the ESP32 until it is above a certain level. Or as I’ve learned if it is going to within 30 or 40ms but not as long as 80ms. Keep in mind while testing that if I unplugged and replugged the supply within say 10 or 20 seconds it would come up fine. It was only with power outages that lasted long enough for it to discharge all it’s internal caps that it would fail.

While the exact timing that causes your design problems will be different based on the 3.3v regulator design if you have this specific issue then I would first recommend testing with a power supply with a faster rise time. In my case the chip just did nothing when powering up. Re-applying the programming cable revealed the “waiting for download” message from the ESP32 making it clear that it thought I was grounding the gpio0 pin even though I wasn’t. With a faster rise time it starts up reliably every single time. 

Friday, July 3, 2020

1-wire Temperature Sensors and Lightning (and other sources of interference)

I use a lot of 1-wire temp sensors around the house for various purposes. One of which was a sort of hack to know when the hot water heater was in use. I placed one on the inlet side and one on the output side of the tank and when the 2 temperatures diverge by enough I know that the hot water is actually flowing. Within a few minutes of someone turning the tap off the temperature has equalized enough that it can tell the hot water is no longer in use. 

The main reason I set this up initially is that I wanted to have my bathroom lights go off after there was no motion in the bathroom for a while, but I didn’t want them to go off while I was in the shower nor did I want to set the timeout to be so long that they would stay on unnecessarily at other times. So now if the hot water is in use the logic of the master bathroom will not turn off the lights on you, but will keep checking back to see if the water is now off or if there has been more motion.

My day job is home automation software so all the logic for that is running in XTension.

Here is the graph of the output from last night:

The blue and red lines are the inlet and outlet temps of the water heater. Physically they are just a standard 1-wire temp sensor taped to the pipes a few inches up from the water heater. The yellow highlight is when the hot water is considered to be in use. The light green line is another 1-wire sensor just hanging in the attic showing the ambient temp in the attic. You can see 2 spikes to 185 degrees between 5:30 and 6pm last evening and one identical spike on the inlet sensor. 

Those were all caused by very sharp close by lightning hits. These particular sensors are connected to a Barix Barionet device which is a fantastic remote host for 1-wire sensors as well as digital inputs and relay outputs. Very handy for home automation and fully supported by my plugin for it in XTension. I have another device that is just basically an arduino that has a similar sensor that is glued to the side of a power supply so that I can make sure it is not overheating as it is in an electrical box with no ventilation. When those lights are on that sensor reads the same 185f or 85c repeatedly until the load decreases. These 1-wire sensors are just prone to interference and when that happens they return a reading of 85c. 

The reading of 185c is within the range of temperatures the device can actually read so it isn’t good to just filter that out in the driver or plugin or hardware. I am considering for a future firmware build of the 1-wire to wifi kit that if I get an 85c reading I will attempt another reading a few seconds later and only report it if the reading is the same twice. For the moment though the solution to filter them is easy enough inside the XTension ON script for the Unit. In the case of the above graph I am not going to filter it as it is interesting and happens rarely and does not throw off any other calculations or anything. For the sensor on the power supply it happens so frequently that it is frustrating to look at the graph. In that case I use the Change Future Value verb to change the value back to the previous reading if it is exactly equal to 85c. That does potentially ignore some valid measurements, but it does let anything through that is 84.9 or below or 85.1 and above and so I’m not terribly worried about that. The script is very simple and just looks like this:

if (future value) is equal to 185 then

change future value to (value of (thisUnit))

write log "ignored bogus 185 value for porch temp sensors"

end if


futureValue tells you the value that the unit will be set to when the script is finished running. The (value of (thisUnit)) portion returns the current value of the unit. This is sometimes initially confusing to people new to working with XTension but it allows you to do things like this or anything where you need to compare the current value with the new value and potentially override or change it. It also lets you do very specific smoothing of readings or throwing out ones that are out of range such as this. The call to ThisUnit just returns the name of the unit that you would normally have to pass to the value of verb so that if you renamed the unit you wouldn’t have to remember to edit the script and if you had many of the devices you want to do this to you could either cut and paste the code, or create a single script that you could call for all of them.

This 185/85 reading is actually a known issue with some 1-wire temp sensors. I don’t see a difference between powered or parasitically powered networks. The sensors in the above graph are parasitically powered but the one that is on the power supply is powered. The parasitically powered ones are on an absolutely horrible network of wires flying through the attic and to say that it is not a single bus or a standard star configuration would be a dramatic understatement. There is no shielding though they are on a single twisted pair of bell wire. Anyplace I wanted to add more sensors I just spliced into it and extended it. That attic network is my test of the absolute worst case I can imagine for testing 1-wire reading devices. The barix does exceptionally well under almost all circumstances except for these lighting storms.

That network now has 10 sensors on it all over the attic and it does sometimes drop a reading for one of them entirely. It also reads the intake and output temperatures from 3 different AC units so that I can verify the temp differential and know if the units are leaking gas and will need a recharge (or in one of the older ones a replacement is coming up I’m afraid) before they actually stop working and we have to make an emergency call to the HVAC folks. That reading also lets me run the fan after an AC or Heat cycle the best length of time to get all the cold or hot out of the unit. I turn the fan to On when a cycle completes and then start watching the temp sensors. As soon as the 2 temperatures are within a few degrees of each other I turn the fan back to Auto. 

I am also going to update the hardware in the next revision of the wifi temp sensor to use the better interface that is outlined in the 1-wire best practices documentation to see if that will help with bad values or interference.  In any case, overall 1-wire sensors work great, are inexpensive and there are a lot of options for getting them into your home automation system. 

Friday, January 3, 2020

Python Motion Detection on a Raspberry Pi Zero

I have been working non-stop the last few months to make my main project software compatible with Apple’s latest OS version. I finally realized how badly this was depressing me and how much I didn’t want to be doing this and how much it was making me hate Apple. I was spending literally man months of time to rebuild from scratch things that weren’t broken before Catalina and it was making me miserable. My rants about the absolute stupidity of everything that Apple chose to do to this alleged OS update will have to wait as I can’t afford the dental bills to fix the damage that gritting my teeth that hard will cause.

So in the interest of sanity and preserving the small quantity of it with which I have been granted  I took a short vacation from everything MacOS. I have been fighting with the horrible firmware of even the best available IP security cams for some time as I struggle to write software that can connect to them and insert their video streams into the web and other interfaces of XTension. I believed that there should be software for the Raspberry Pi that could leverage that camera for a much more reasonable price that would be terrific. Sadly even the best software out there for it just wasn’t doing the things that I wanted or at least with the quality of image and recording that I wanted.

The idea of controlling everything about a security camera is compelling! Not having to setup a firewall to keep it from contacting it’s own “cloud” and making sure it isn’t setting up a UPnP passthrough even when you told it not to have revealed to me several times that even commercial products don’t really do what you think they are. I wanted one based on software that I have written myself and so knew exactly what it’s doing.

It should also run well on a Pi Zero W so that you can make something small, though if it runs better on the latest multi-core Pi4’s thats OK too. Those little machines are worth every penny and very very nice!

The built in “motion” system is really interesting, but it’s remarkably difficult to get it to run at a high framerate and a larger frame size. Who wants a non-HD security cam now days anyway? The separate OS versions available may be spectacular but I want a software package for the stock Pi. After fighting with trying to get openCV to compile on a Pi Zero for the last week I wanted something that didn’t need that either. (Which I suspect is why they went with a totally separate image for their security cam OS as getting that stuff to install properly is almost hilarious)

The python camera support for the Pi is absolutely awesome. I can stream 1080p video and I can record it to disk and then post process it to standard MP4 files in almost real time. Those were not insignificant accomplishments ;) But the real sticking point was going to be the video motion detection. Using the pycamera code as a starting point and the hardware H264 encoders ability to output it’s changes from one frame to another makes it absolutely possible to do this at least as well as any commercial product. I’ll eventually open source this project but for the moment I’m working on it just myself. Here’s the 720p stream, part of the web interface and also a stream of just the motion sensing. This works as high as 1080p but at lower framerates though I’m sure I can optimize the calculations a bit more for that.

The image on the right is a live stream of the motion content and I’m just really enjoying playing with this at the moment.

I am currently building the web interface to the ability to mask out portions of the video for motion detection. The math for that is working great and adds very little if any overhead. Lots of work yet to do of course before something that is download/install ready but it will come before long. This is making me happy! Apple is making me mad.

Lest anyone think I’ll abandon developing software for Apple products over how much they drive me crazy sometimes really needs to spend some time using any of the other truly commercial desktop OS’s available out there. Don’t worry. XTension is undergoing a lot of changes because of this, but it’s not going away and unless I sell enough of it to hire a support team in another time zone I wont even consider doing a windows version. Sorry but if Apple frustrates me then Microsoft would kill me outright.

Thursday, October 17, 2019

RGB Fairy Lights with only 2 wires (part the first)

I am a huge fan of all things blinky and light up. I’m less interested in things I cannot control remotely by writing a plugin for XTension or by building a new controller for it. While looking for RGB “fairy” lights on Amazon (“fairy” lights here being defined as some strands of what looks like magnet wire with surface mount LED’s soldered between then and then sealed into a little blob of epoxy) I came across these RGB ones and was intrigued since they appeared to have only 2 wires. There was no control wire or separate PWM lines for the RGB channels. (Amazon Link) These are by “Yihong” but there may be others that are the same.

They came with an RF remote which is nicer than the IR ones that some similar things come with. Since you don’t have to point it at the controller box. Interestingly enough the remote also has an IR led and so could be used by either type of system. I have not tested to see if it is running on the standard 433mhz band or does something special. Given that it appears the same as the thousand other such remotes I’ve gotten over the years for other LED displays or just for remote arduino projects I would guess that it does.
In spite of the picture on the box showing different colors on the strand at the same time, you cannot control the LED’s individually, only as a whole strand.

This is the very tiny control box. The strand is powered from a USB plug and this box does the work of receiving the remote signals and adding whatever data is necessary to the power lines going down the strand. I have not broken the case open yet to see whats in there.

While there is tons of information out there about the “neopixel” type LED’s and the newer “dot star” that Adafruit sells and are amazingly wonderful there is very little information about anything similar to this. If anybody knows the type of chip or LED that they are made from I’d be interested to hear about it.

This you tube episode from Julian Ilett shows him reverse engineering a similar protocol. The system used by my lights looks to be very different however. So the first step to seeing what it’s doing is to get it on the scope. Since the wires are just enameled I scraped some of the insulation off them at the end of the strip. It occurs to me that I might have gotten a cleaner signal by tapping it at the front of the string instead. I’ll try that in step 2 and see if it makes it easier for the scope to trigger on it.

These LED’s are different than Julians as they don’t pulse all the way to off. The data pulses drop the voltage to the lights down to 1.5 volts, but not all the way to off. This potentially makes sense as it would let the little processor in the LED chip continue to operate as it’s being updated.

Here you can see the amazingly noisy USB power supply, either from the hub I was using to power this or from the controller device itself I don’t know, but it didn’t seem to bother them. You can also see the tiny blips of the control pulses coming in every 100ms or so. 

Zoomed in on those a bit you can start to see the data.

There are 2 lengths of pulses. Since the on portion of the pulse is always the same length the entire duration of the data packet changes based on how many 1’s there are in it.

The short pulses, which I will call a binary 0, are approximately 5us with the bus at 1.5v and then 10us back at 5v. The actual measurements I got from the scope was 6us and 12us but they varied enough that I assume they meant to use 5 and 10us. That may have to be adjusted later if the device is very picky about the timing. The 1 binary pulse is 10us at 1.5v and 10us at 5v.

Every data packet starts with 8 “0” or short timing bits. These do not appear to change no matter what the string is displaying so I will assume they are a preamble that helps the receiver lock into the data rate. In the screen shot above you can also see the end pulse which is twice the length of the “long” or binary 1 pulse, 20us at 1.5v. 

There are 32 bits in total, one byte for the preamble and 3 more bytes which obviously encode the RGB values that the LED should display. I did several captures from the various solid colors the three different intensity levels, leaving out the preamble byte: I’ve added the decimal value of the binary next to them:

RED:   FULL:  0010 0110(38)  0000 0000     0000 0000
       MED:   0100 1100(76)  0000 0000     0000 0000
       LOW:   1001 1000(152) 0000 0000     0000 0000

GREEN: FULL:  0000 0000      0010 0110(38)  0000 0000
       MED:   0000 0000      0100 1100(76)  0000 0000
       LOW:   0000 0000      1001 1000(152) 0000 0000

BLUE:  FULL:  0000 0000      0000 0000      0010 0110(38)
       MED:   0000 0000      0000 0000      0100 1100(76)
       LOW:   0000 0000      0000 0000      1001 1000(152)

Its quite obvious already that the 3 bytes are RGB values. You can also see that even when on at full they dont send just an FF. Perhaps the LEDs would overheat at full on. There is some visible PWM even when a solid color is set to full on. Or there is a limit in the controller device itself that is not capable of 255 individual levels. They are rotating that bit pattern across for dimming the LED. At this point Im thinking that the value is actually how long to wait into the PWM cycle before turning on that LED. That would make sense, it would let a larger number actually be a dimmer display. Just for fun sampled some of the other colors as well:

ORANGE: FULL:  0011 0001(49)  0101 0000(80)  0000 0000
        MED:   0110 0010(98)  1010 0000(160) 0000 0000
        LOW:   1100 0100(196) 0100 0100(68)  0000 0000

YELLOW: FULL:  0010 0110(38)  0100 1100(76)  0000 0000
        MED:   0100 1100(76)  1001 1000(152) 0000 0000
        LOW:   1001 1000(152) 0011 0000(48)  0000 0000

CYAN:   FULL:  0000 0000      1101 0010(210) 1101 0010(210)
        MED:   0000 0000      1010 0100(164) 1010 0100(164)
        LOW:   0000 0000      0100 1000(72)  0100 1000(72)

and now Im wishing I hadnt sampled those extra colors as they dont seem to follow the same start point of PWM. Or perhaps they are just compensating for irregularities in how much the LEDs output at any given level. They wont be the same certainly, generally red needs a much higher value than green and blue to look the same. 

In any case there is enough information here to get started trying to build an Arduino controller for the thing. I’m not entirely sure how to provide the 1.5v pulse level in hardware. Will it be enough to just run the stream power supply through a resistor? Will have to calculate the power draw of the LED’s and do the math to drop the voltage to that level. Or are the LED’s in sync with this pulse and so will be off when the data starts and therefore I’ll need a different value since the load would be less. Or will it be necessary to provide a 1.5v regulator and use a transistor or mosfet to bypass it when the strip should be given the full voltage. Some experimentation will be necessary. When I’ve had a chance to play with that I’ll post again.

Tuesday, May 7, 2019

WiFi Signal Strength as presence indicator, part 2

As a followup to yesterdays post about trying to filter the wifi RSSI of an ESP8266 board in order to sense movement I added a standard variance calculation on the last 2 minutes or samples or so. I added a threshold of 1 just for testing and have let it sit here for a while. The graph of that output looks like this:

When the variance goes above 1, or whatever preset threshold, I turn on the presence register which shows in green. You can see that it definitely does work to know when I’m coming and going, but at this sample rate, buffer size and threshold it also comes on randomly with non-people moving through the room wifi interference.

In order to register anything reasonable at all you have to have the device very close to the person you want to know if they are there. This may be of use for this particular application as I plan to have the device sitting on my desk to provide some display of info for me while sitting here. It might even be able to blank and wake up it’s own display based on that. That would be useful and fun. Unfortunately if I sit very still it eventually thinks I’ve gone away. It’s decided that while I’ve been typing this paragraph actually. Lost of tweaking of settings would be necessary to make it at all useful.

I’m not sure that it will ever be useful as a generic motion sensor. You just have to be too close to the device to get it to see enough variance in the signal. Still an interesting experiment and I will keep fiddling with it to see if something useful can be done with it. It’s also possible that with the newer ESP32’s or other devices you could get better resolution or even access to things like the path being used to send the info or other similar data that would let you do a much better job. But for something that wakes up when you walk up to it and then goes to sleep shortly after nobody is standing next to it this could potentially be used.

Calculating a variance is a standard statistical technique so that you can gauge if you’ve got enough data points to realistically represent what you’re looking for among other thing. The formula is:

S2 = ∑(xi - x̄)2 / (n-1)

Which translates to the Variance squared being equal to the sum of each value in the dataset minus the average of the dataset squared and then divided by the number of points in the dataset minus one. So the first thing you need is an average of the dataset.  It’s slightly complicated by the fact that my buffer array of read RSSI values may contain 0’s or values that are not a valid RSSI. As I place the values into the buffer I am making them all positive as I am not worrying about the actual db value, only the variance of the data. So the first thing you need to do is to call the scanRSSI function every second. To do this you need to compare millis() with the calculated next time you want to do something inside the loop. You can’t just compare simple unsigned longs as are returned from millis() as they will overflow back to 0 at some point which will break everything. So you can do the clever trick of casting it to a long and then subtracting it later which will yield the proper result of timing even during an overflow condition.

#define MAX_RSSI_ARRAY 128
bool currentPresence = false;
byte rssiArray[MAX_RSSI_ARRAY];
byte rssiIndex = 0; // the index into the array where we are now
unsigned long nextPresenceScan = ((long)(millis() + 5000)); // 5 second start

the code above sets up the buffer array as bytes, since the numbers will never be more than 255 there is no reason to waste memory by using 2 byte ints or anything else. We also need an index into the array so we know where to save the next value and the nextPresenceScan value which is where the trick comes in. Casting the value of millis() + 5000 as a long integer and then putting it into an unsigned long variable we get around the overflow problem as far as the comparisons later are concerned. Very clever and described in great detail on the Arduino website if you’re interested.

Next we need some code to run inside the loop() to check the time and call us when it’s time to run. Something like:

void loop( void){
 if (((long)(millis() - nextPresenceScan)) >= 0){
  nextPresenceScan = ((long)(millis() + 1000));

Same trick as before, sets up the nextPresenceScan and then calls the function to do the work.

void scanRSSI(){

 float currentRSSI = WiFi.RSSI();
 if (currentRSSI > 0){
  // sometimes the WiFi.RSSI() call can return a number greater than 0 to indicate
  // an error or that it simply couldn't get the info at this time. If that happens
  // just ignore it.

 // add the new value into our buffer at the current rssi buffer index
 // multiply by -1 to turn it into a positive number
 rssiArray[ rssiIndex++] = currentRSSI * -1;
 // if the index is now beyond the size of our buffer reset it back to
 // the beginning of the buffer for the next read
 if (rssiIndex >= MAX_RSSI_ARRAY){ rssiIndex = 0; }
 // first value we need for the variance calculation is the average
 float accumulator = 0;
 // since the buffer may not be filled with valid data yet 
 // we need to count how many non-zero values there are for the
 // average calculation rather than just divide it by the 
 int averageCounter = 0;
 float average = 0;
 for (int i=0; i<MAX_RSSI_ARRAY; i++){
  // ignore values of 0
  if (rssiArray[ i] > 0){
   if (rssiArray[ i] < mn){ mn = rssiArray[i];}
   accumulator += rssiArray[ i];
 // if we don't have at least 10 valid values in the array then just drop out
 // there is no point in calculating the rest unless we have enough values to 
 // look at
 if (averageCounter < 10){ return; }
 average = accumulator / averageCounter;

 // now loop again to calculate the SUM portion of the variance
 // reuse the accumulator 
 accumulator = 0;
 // loop through the buffer array again, this time subtracting the average and then
 // squaring the result adding it to the accumulator
 for (int i=0; i<MAX_RSSI_ARRAY; i++){
  if (rssiArray[ i] > 0){
   // dont want to waste the memory to include the floating point or C math
   // libraries, so just put the calculated value into a local variable and
   // multiply that by itself
   float work = (rssiArray[i] - average);
   accumulator += work * work;

 // here we divide by n-1 
 float variance = (accumulator / (averageCounter - 1));
 // the variance may be less than 0? if so take absolute value
 if (variance < 0.0){ variance *= -1; }
 // now do something clever with the variance data.


That is actually pretty horrible code. I also do realize that I’m never taking the square root of the variance as calculated. That doesn’t seem necessary at this point and the poor arduino is already working hard with all those floats. If I make the value even smaller it will be harder to watch whats happening. It’s also not memory compact as it needs a full buffer. I’m sure some other kind of rolling average could be used. To get anything useful you’ll have to adjust the threshold and the scanning frequency and probably a lot of other stuff too. An interesting starting point to something potentially useful.

The device in question here is sending it’s values back to XTension for graphing and potential use. The graph above was generated by that program. XTension: Home Automation for the Mac, is my day job.