These dot displays were developed by Ferranti-Packard in Toronto in the 1970s. I developed several products with them. I visited the factory and met their self described magnetic expert. He was upset that I immediately understood he used square loop magnetic material and drove it to saturation just like core memory. He had not drawn the memory analogy and was entertained by it. They had arrays of 5x7 and larger that you could drive in a matrix. They sold a controller you could send RS-232 data to it. They also had a clever self decoded numeric digits. There were made with 11 separate wires sewn through the disk cores to drive the right dots for each of 10 numeric digits and also blank. It was actually very similar to a wire braded ROM. One pulse would flip disks on and off to erase any previous display and put up the new one. Even though there was a big nest of wires on the back it took quite a pulse to drive them. I drove them with SCRs and an 8048 MPU at the time. I was doing outdoor displays where their excellent sunlight contrast was important. LEDS at the time were just too dim.
You gave me an idea... It would be really cool to make a core memory with a large matrix of these dots. Would make for a very educational small computer.
LED screens in direct sunlight are a waste of energy - they require at least 5 times more energy than if they were in a shade, and even more compared to night conditions. It woud be much better to make some screens which use the sunlight as their source of light, by reflection. I know that Sony has had this on some of its camera displays, called it "transflective LCD" which passed sunlight through its cells and reflected it back through the same RGB cells. It also used a backlight which would pass through the LCD normally.
My dad used to work for FP a long time ago when this was state-of-the-art. He was an engineer that managed installs, repairs and some sales. We saw the background of places many rarely get to as kids like boards of trade, Indy 500 and pro bowling competitions. When they had large events, they had him on-site to quickly fix any issues.
I also have worked with these since the 90s at race tracks, with a company that used them from the early 80s. Tomorrow morning I’ll work on testing new software on one of 5 within 400km of me. We still use the FP541 controllers nearly 50y later. Used to love when track management would come to me and gripe about wasting power by leaving the infield board on all night…then I’d have to pull one of the indicators out to show them that they actually use no power to maintain a display.
good to see the overview of your findings you have mentionned! it has made me think of a possible removal idea ill fire you an email. the motor driver chips is certainly a stroke of genius.
We had two of these at the 2002 Commonwealth Games in the main athletics stadium. Lower resolution but larger flipdots which were loud when they flipped. The screens had multiple modules per screen and the design of the modules had a flaw that damp and humidity would make the dots stick. And with the weather conditions we had there, we ran some test patterns to keep them moving when they were not in use and the stadium was empty. One night we had some serious rain so we left the screens running the patterns all night so they didnt have to be manually sorted the next day due to early competition start. The following day we found out that the stadium had been visited by the police because locals had complained about the noise all through the night from the dots flipping, i think it was about once every 5 seconds.
Yes it seems fast enough to even do some 15 to 20 fps animation. I guess the limiting factor are those two plastic pivot pins that will eventually be ground down.
i was wondering that too. Anything mechanical is going to have a set lifespan rating, but being the main movement of the dot part is electro macgnetic, there's barely any touching. If the retaining posts of the dots and the dots 'arms' themselves are something like PTFE coated, i wouldn't be surprised if the units could reach or possibly outlive the chips/circuitry themselves. i would think ambient temps and environment would be the highest cause for failure perhaps?
I would bet it’s pretty darn high, considering that they were designed to live in bus marquees that could operate maybe once per second for the better part of a 24-hour day. Of course pixels did occasionally get stuck-I wonder whether that tended to be electronic (including things like solder joints) or mechanical failures?
Pretty much the only wear part is the pivot, and that is a small diameter, so not travelling much distance. There is very little friction, so I'd expect life to be well into the millions
From what I recall I once saw a number that was either 1-5 or 15-20 million updates. I have a really old flipdot panel that's been heavily used, and the only issue I've noticed is that minor dust buildup causes slower flip time(significant, enough to notice).
Only simpler I can think of is up to 1 khz AC and triacs. Like you only need a single pulse to flip them, in sync with the AC drive. The selected half-wave decides flip direction.
The flip-LED displays that Siemens Desiro trains use are also quite interesting, the daytime readability of a flip-dot display with the LED's for night time, surprised they didn't just backlight the flip-dots but Siemens never do things by halves. Used to be (and maybe still is) that the guard had a control panel and could put in any text they wanted to with sometimes comedic effect!
I used to design controllers for these things in a previous life. Typically, flip-dots are driven in a bipolar matrix built out of octal sink and source drivers (e.g. TBD62783 & TBD62083), because that's the cheapest way to do it. The disadvantage of that approach is, you can only really energise one dot at a time, which is why you usually see flip-dot displays scan from left-to-right when they update. If you're doing some sort of art installation requiring animation, then yeah, individual h-bridges are the way to go.
I love these kinds of displays so much, I wish they were more commonplace so that you could buy them without having to go through ebay to find old used ones
A few years back I made an installation for a car manufacturer where they wanted a wall of irises that opened and closed to mirror people in front of them. We couldn't get anything fast enough for a nice effect, so had a nice 3D mask made with round holes, in front of 6 huge video wall units. The crowd was far enough away that it was hard to tell it was using video screens. We also used speakers around the edges to position sound where irises were switching, so there was a nice chattering noise 😁 The way it worked was with IR distance sensors in between each iris hole, all going through comparitors to create simple bit-streams that were being scanned at 100fps using a little 8 bit MCU to a PC running a WebGL app - basically a pixel shader so it was fast even at a huge resolution. The wiring loom was huge! The event was at a venue in London that I recognised when I saw it in a scene from the film "The Krays" a year or so later! Damn, I just checked, it was 8 years ago! How time flies...
You can flip them with capacitor discharged directly into the dot’s coil, is enough energy to make the dot flipping correctly so can be done with CMOS. It's nice because you get the flip back without any energy and you can make them silent.
The nice thing about these flipdots is that there is no precision mechanical mecahnism that might get clogged with dust over time, the axial component can be pretty loose and still work well even years later.
You can also flip these without an H-bridge by using a split rail power supply. Tie one side of the coil to common and then pull the other positive or negative. It might not be any simpler though.
We did try a round version if I remember correctly, instead of flat disc's they had half coloured balls. Went on to use 10mm leds in a large out door sign in a local shopping centre. Fun time in R&D back the day.
Funny how we independantly came to similar solutions, I also ended up using a little motor driver chip (iirc the 6110) to control the seven segment flip segment clock I built. It really was the cheapest/simplest solution to the problem.
Wow. These are indeed much more subtle than I realized! Interesting approach with the "parasite" PCB approach, just attaching it to existing flipdot displays is a great idea. I've only just started on a desoldering project, and I've realized as awesome as desoldering vacuum guns are, they're not magic and you definitely need to think before you start desoldering stuff.
I work for a company that manufacturers Bus display signs and most of our signs consisted of flip dot technology up until about 18 years ago. That board looks like it could have been one made by us, Hanover Displays?
I just started to unsolder a Brose board. I use a strip of flat copper that fits just between the legs (75x6,5x1mm). With a big tip in my weller its very easy to heat the strip in a few places and so melt the solder on all legs in one go and then take the 7 dot unit out without any damage. The only shitty thing is that the board is covered with conformal coating so i have to tin each pin before starting to clean it up. That takes some time so a whole board will keep you busy for a while. What i learned is that the strip should be a bit wider with notches so it gets more around the pin. Now i have to check that each pin has enough solder and is in good contact with the strip.
Some similarities with ferrite core memory makes me wonder if you could read as well as write the current state of these... Could you use the display as memory? There is no sense wire, but perhaps you could detect the difference between successfully flipping vs. trying to flip to the state it's already in.
For sure you can see the current will be different if you try to flip it one way or the other from current position. Possibly you could check without upsetting it which would save time.
Interesting video 2x👍 The company I worked for build and serviced many different types of signs. I got thee job of flying to Dublin and fixing the flip-dot sign at The Curragh Racecourse, the thing was full of water so I ended up bringing much for the internal driver boards back. It was just after September 11th so I ended up getting more than a scan at the airport, with all my tools and loads of PCBs and wires. 🤦♂ But I guess at the time people were a little more sensitive.
Fascinating details and design process. Thanks. FYI about using magnets to reduce coil current and make bi-stable: Years ago, I needed high current DPDT relays which required no power after activation, little power to activate and the same current capability and contact resistance in both states. Note that most relays (without the help of the permanent magnets) have higher on resistance to to NC contacts when not energized, often double ~ quadruple of the NO contacts. This is made worse by vibration, where the NC contacts can bounce open with shock or vibration. The Panasonic 25A relays I found had NC and NO contact resistance within 1~2 milliohms in all modes and about 10x the resistance to shock and vibration. The absolute best way to make these flip faster is to remove the main resistance: air. Put the whole thing in a vacuum and the flip time will be limited by the mass of the disc indicator. I don't think ball bearings on the little pivot pins would help much. 😂
I was thinking maybe a double gate driver would have worked, but a motor driver is a better choice. I've actually used cheap DC motor driver IC's as low side gate driver IC's with really good rise times. In my case i was driving a bank of a dozen MOSFET's from a single IC for a spot welder. I think this application would be an ideal spot to use the 3 cent padauk micro to perhaps make them work exactly like Neopixels do because that would make it very easy to use whatever existing neopixel code there is to drive the display. Basically piggybacking off the existing neopixel architecture.
Desoldering these are nightmare. I abandoned my project because I didn't found a way to remove them. Event 7 segment ones are made from single use plastic.
@mikeselectricstuff this is awesome. Thanks so much for this, I've been using Servos to make Wood panel pixel displays (192 pixels/servos...). These flip dots are way more compact. Really appreciate the deep insight into their inner workings.
I played around with some flipdots awhile back, but mine were a much different construction. Instead of having little coils and armatures like on these, mine had the coils as pcb tracks on the board! there were two diodes on each pcb coil, set up as a positive matrix and a negative matrix. I ended up driving it with 8V and anything higher seemed to be diminishing returns. The pcb track coils only had 60ish turns maybe, 15 on the top layer and 15 on the bottom, and then two coils- one wound backwards from the other, on the two corners of each dot. I added a one shot to mine so that a coil could not get stuck on, which would cook the pcb in pretty short order. On mine, the length of the pulse mattered; if it was too short, the dot wouldn't complete its travel and wouldn't change state. This one seems to be better in that regard; it will accept the much thinner pulses and then the magnetic field completes its travel.
Wow, I thought these we a thing of the past. I worked for a company back in the mid 70's that used these as an event recorder on equipment that was used in the power generation industry. They'd used them to record over current, under frequency, etc. It's neat to see something on these devices.
It was nice meeting you. Will be intresting to see hwat you disign with these. I have my own panel to play with, meany other projects still in a half done state that need to be finnished first. One is even time sensertive.
Thanks for sharing your insights on the accelerator magnets on the bottom of the flip-dots. I wonder how flip speed could be accelerated if those bottom magnets where electromagnets instead being able to be timed to accelerate the dots even faster as it approched the E-magnet. Like others had commented it would curoius to see them operate in vaccum. Other than flip speed the vaccum could also enhance the lifespan I would think. Looking forward to seeing the project move forward!
That took me on quite a little journey across youtube and wikipedia ... the hackaday article about the 74000 pixel flip dot display was quite fascinating .. you always have such interesting topics... In the german wiki it says that the axis of the dot is designed to act a little bit like spring to support the action and help the dot retain its position . also I read about diodes to control the dots (in the early days) like a keyboard, but the other way around.. anyhow, good luck with the project ! maybe shear them off ? with a really sturdy blade, like a Microtome ?
I have subscribed to your channel because you know what you are doing and talking about, and you go into details much like I do, which sometimes comes close to OCD. 😁
A while back I ordered some engineering samples of flip dots from Alfa-zeta. They are fascinating devices, but I didn't have enough of them to make any useful prototypes. There are many ways to drive them, and there nonvolatile mechanism is a bonus. I'm looking forward to your journey with these and the final result which will no doubt be awesome.
Another idea to drive the devices with the half bridge idea, is using MOSFET gate driver chips. Those can supply amps for short amounts of time, and usually have inbuilt logic to prevent simultaneous turn on of the upper and lower FET, and sometimes even current limiting built in. There's of course also dual gate driver chips which could be combined into a full bridge.
If you revisit the row/column control, one idea that came to mind would be to have the two rows or columns or what have you have the pulse PWM'd inversely to each other, so the driver always delivers the same current but alternates which flip dots it goes to. The idea being that it might be more efficient to deliver a longer but less powerful pulse with regards to the flip dot overcoming its inertia or whatever.
If the two coils are centre tapped (didn't say if they were or not) you could hold the centre at v+ and just pull one side down.. would save the motor drivers!
Idea I pondered in my head is that T12 soldering iron cartridge tips are so cheap that you could maybe build custom desoldering jigs with many of them.
I have always wandered how it works (sorry for my cripple EN). Now I understand! Such a simple and yet effective solution for large boards! It's almost doable at home.
Ferranti-Packard made similar flip dots back in the early 1970's that were assembled into groups each to display one alpha-numeric character. Used in airport arrivals boards iirc.
Transistors (BJT and FET) have a slower turn-off time than turn-on time. If you don't use a delay circuit (or a proper gate driver) and connect the upper and lower gates together, you will eventually destroy them.
Super interesting stuff as always, Mike. This made a lightbulb go off in my head. At least in principle, if you have individual pixel control, nothing is really stopping you from getting a decent frame rate with these under certain conditions. You'd just need to sequence the transition time to where you want it. Doesn't matter if a single transition takes 65 ms, if you can sequence two transitions to happen say 16 ms apart for a 60 Hz update rate. BUT for this to work... - The transition time would need to be consistent between flips. - You mustn't flip any individual dot too often, which you'd probably have to figure out in software, and maybe creatively adapt the material shown for this restriction. - You'd need to be able to handle the pulse current from a lot of simultaneous transitions. If you beef up the capacitors on the driver boards, you'd need to prevent communication glitches. - You might now have issues with the magnetic field from a certain dot affecting other dots, which might ruin the consistency of the transition time. Excited to see what you'll come up with in the end.
He already has 16 fps just driving them regularily, and if one was to send a pulse to flip the dots to "frame 2" while half of the display time of "frame 1" had already passed you could get it up to at least 24fps.
Your analysis of the flip-dot element is quite impressive, much more detailed than my investigation. About 7 years ago, I've got some flip-dot panels which have identical pixels like yours, and I hand built the circuit to drive the panel quick enough to show some really interesting animations. Here is the video. (The video was not uploaded by me but someone else, since I haven't got my channel back then) ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-ko0z3SfXpm8.html In my experience with my 32x14 panel, I used multiplexed driving method, with 8 L293N 4-channel half bridges driving the columns and discrete MOSFET bridges driving the rows. Actually I wanted to use L293N to drive the rows as well, but since the rows are connected in series with two diodes, there're actually 2 individual lines for each row, one pull up and one pull down, which made integrated half bridge unusable. I used 74138 3-8 decoder for row selection, so only one of the row lines can be selected each time. I've also added an R-C delay circuit with a diode on each MOSFET gate, to create a hard wired dead-time, so as to prevent shorting the rails. To increase the refresh speed, you actually just need to speed up the magnetization speed, since the mechanical rotation time is much longer than the magnetization time. You could skip to the next group of pixels after the magnetization time is enough, and the dots will flip by themselves, kinda like a pipelined work. I increased the magnetization speed simply by increasing the driving voltage. Also, flipping a dot takes quite some current, so you can't flip too much dots at the same time. My setup flips 4 horizontal dots each time, limiting the current. Also, my code stores two adjacent frames, compare them and skip the unchanged pixels, which increases refresh rate a lot. Hope my little information will give you some help, also wish you good luck with your project!
Bet they could be PWMed continuously to speed up the charlieplexing. Power all the rows in a single period with a 6% duty cycle to give the illusion of all of them being simultaneously driven.
just out of curiosity , could u make a video about lm 3914n and lm 3915 ics , or perhaps stuff with cd4000 series ic's ? (cd4060 would make a neat binary led counter for example)
Our trains up to London and all the original Siemens Desiro's have the even more interesting flip-LED display that has both flip-dot and LED's so you get the superior daylight readability of A flip-dot whilst the screen is still visible at night, could have just used a backlight of some description but not. Interestingly these displays were invented by one of the Plessey divisions in Canada.
Hot air gun on a CNC rig, I suspect you'd have trouble making a desoldering bar that would make good contact across the entire width of the board and dread to think what it would take to heat it and what it's thermal inertia would be. But if the pins are a tight fit in the holes then any desoldering method is going to be a nightmare. On balance your best bet likely will be to use the boards as is and interface to them.
what about something like the old ic desoldering iron tips, the ones with a U shaped piece of copper, drilled with a dimple to locate on the ends of the terminals, it should then push the pins out when melted, with an additional pull on the module once it moves.
Once you've overcome the initial interia, i.e. after it's started moving, I think you might be able to push energy in at a greater rate - maybe a short inversion and then a bigger pulse of the original polarity.
Interesting investigation. When you were measuring the time until the dot had flipped, it looked like it was still bouncing quite a way off the right edge of the oscilloscope display, not really becoming stable within the time period you were viewing - did you look into what was going on there? At least in the slow motion footage it looked much more stable than the reading from the optical sensor.
No - anything after the initial position isn't really of interest as there's little you can do to control it. The amount of overshoot is small, so doesn't really affect the visual appearance - the only real noticeable difference driving it harder is the sound, which gives the illusion of it being faster
May you could try force the field on the way the dot is already flipped (charging/saturating the coils) and then flip the polarity, this should boost the bounce (a little) a may reduce the delay
The magnetic field takes time to change and all that mass is a problem too. Byt the time you bumped it, then reversed it, waited for it to bounce and reversed it again, it would already have completed a full flip.
I'd think that it's a special ceramic poles that the wires are wrapped onto that stores the magnetic values, basically a reprogrammable permanent magnet - it's like a mechanical version of core memory of the ancient computers, only you can see it. And I have seen the flip-dot sign maintained by a college around here, long before they finally upgraded to LED, probably due to the fact that the hinges have finally failed or just the electronics tired of flipping the flip-dots all the time for decades.
I stand corrected. Still, it's interesting that the flip-dots basically demonstrate the hysteresis effects of flipping whenever it's magnetized, regardless of how it's driven.
solder fountain for desoldering? i think the right approach is probably using the boards as is minus the edges and maybe cutting the matrix a bit. i think i know where these are going and i'm glad you're the one doing the work, sam's a great bloke but i'd hate to watch him get bogged down in some kind of mechanical sequencer madness trying to drive this
Are 4000 series logic chips as cheap as the 7400 series? If your logic runs at 12v you can skip the level conversion. Some paralleled outputs might be able to drive the dots directly, but a pair of emitter followers certainly could.
No you can't as 4000 input thresholds are still 1/3 and 2/3 of the supply voltage. Unfortunately they never made the 4000 series equivalent of 74HCT parts
@@mikeselectricstuff I was thinking you could have just a few level converters (Clock, Data, Latch) on the controller. Then the shift registers on the driver boards could use 12v-only logic.
@@mikeselectricstuffould the CMOS be driven by an open drain circuit with pull up. Anyways, using open drain high and low drivers, the diode matrix can be driven using only 2 pulses per column, with heavy current in column drivers. So first pull column high and pull select rows low, then column low, select rows high. All others high Z. Low side row drivers can be traditional 7-channel relay drivers like ULN2003, high side would be loose p-channel or PNP driven by second set of relay driver outputs, column drivers would be beefier discrete transistors controlled by duplicates of row circuits. Thus a 90x160 display would need 250 driver pairs for each polarity, 90 of which are high current and driven by a 90 step scanning counter, the other 160 pairs are normal current driven by data. As each polarity is used only in odd or even cycles, the LSB of the counter can control master gating of all outputs for that polarity perhaps via an extremely strong outer transistor feeding the individual drivers. Possible circuit becomes: Row driver: Pwr------+ +---p | +------- High line | | +------ Low line Bit--n---n | | odd-N N-even | | gnd-+---+ Where N being big transistors, p small transistors and n are channels in a darlington driver chip. Column driver is same but with N and P transistors as a second output stage. The odd/even shared control line ensures only one polarity is active at a time, while the inversion in the second stage ensures active column is opposite polarity to rows. Each data bit/counter output drives both polarities, 1 for change/select, 0 for no change/deselect The 160 row bits would be loaded via high speed SPI (20x 74HC595), while the 90 column bits would come from CD4514 4-to-16 decoders with enable pins driven by cascaded CD4515 decoders of higher counter bits, counter clocked by falling edge of even signal. max row count is limited only by column driver transistor strength divided by single dot current. MCU needs to output odd and even master pulse bits and an SPI stream pumped into row registers with a hold line to freeze outputs while next data set is streamed during previous pulse, and an input bit sensing column 0 active, thus 6 MCU pins total. Use ¾ of a CD4001 to generate odd and even from counter LSB, and SPI hold pulse doubling as counter clock and driving pulse , to need only 4 MCU pins and prevent accidental shorting via code bugs. Actual counter would be straight 8 bit reset by decoding to column 91. The design duplication in the output would allow one pair of 7 channel drivers to be split between row and column bits, needing only 72 uln2003-like driver chips for this hypothetical 14400 dot display.