Samsung PowerBot Virtual Wall.

[glnc222]

The top omni lens is also an on/off button.

[Bandi94]

Thx for the effort. In that case they might use some other frequency that is filtered out by the TSOP, that would be one reason why it is not picked up. For a more deeply analisis there would be useful a scope or logic analyzer to see what's going on directly on the output of the microcontroller used in the VW, there would be at least two frequency's used in there, a higher one to be able to filter out in the receiver the natural IR sources, and a slower one to encode the transmitted message, so i don't know how good a multimeter could analyze it. I will contact today a person who is selling a used VW in my country to ask if he still have it, if yes then i will buy it and next week scope it to see what magic is going on there. From images of a point to clean remote there are visible two lenses on the remote that i can't see in the VW. Anyway i got very interested in this "magic" so i am gonna study it even if my powerBot is not recognize the VW just to find out what they are using in there.

[glnc222]

I will have more data later today. AA cells did not power it adquately, but I do not see any significant AC noise in the adapter power supply. I also got an 18khz reading on the upward LED but on scope seems to be same as the on/off blinking indicator. Looking at the forward emitters.

To us this I have to point it at a 45 degree angle to the door opening and cover the omni emitter just to avoid false demarcation. It looks like an attempt to appeal to Roomba customers, while Neato Robotics had originally used mag strips on the first guided vacuums, because of the more refined behavior involved.

For Neato cliff sensor interfacing I found that an emitter mounted under the sensor could saturate the sensor in a way giving a cliff indication (stops the detector from distinguishing pulse on and off), avoiding opening the case. Those proximity sensors are also very sensitive to any extraneous voltage and easily damaged, so connect only to a high impedence input terminal on a comparator or op amp. They seem to have crude output circuits. There is a general caution IR sytems may not work in bright sunlight or certain incandescent lamps (large line power IR emission signal).

[glnc222]

Looking with both the data logger and the SimpleScope I do not see any high frequency pulses out the emitters on the Virtual Guard. Maybe better equipment would show otherwise. Emitters have current limiting resistors on the board.
The omni lens emitter has three leads, unusual.

[edit] No good -- 60 cycle power line noise maybe from A/C adapter. Need some "D" cell batteries.

data logger 1khz

Again, 180 cycle probably line noise, unless that is what they use. Suspicious voltage levels.
Possibly the duty cycle of high frequency bursts too low to capture without fancier instruments.

Again may not be in an infrequent coded burst not captured.

I don't see it worth tediously rebuilding my bumper mod opening the case to read the receivers inside the bot.

IN ADDITION, running the bot with heavy tape over the top front side receivers THE GUARD STILL DEFLECTED the bot.
They must be received by the proximity sensors which would not require high frequency coding like the remote.
There remains a question what wave length is used as was not detected on a fairly general 940nm phototransistor.
Our retail Radio Shack stores with components have closed in favor of mail order only, under the competition from Amazon. I would have to order a bunch of different ones from Digi-Key, cheap but takes a few days.
Advise if that would really help you.

[glnc222]

Maybe I will try to reproduce it with a 555 oscillator and IR emitters on hand. Difficult to match the frequency precisely, with an analog trimmer. Maybe use the scope, try to match pictures. Will see. Never programmed the Arduino which would be more precise.

[glnc222]

Even when the Gaurd is not recieved by the TSOP, its signal might still interfere with the remote per the manual. Maybe it does not, and they just were imprecise in the tech info. There is often problems in tech writing to get the info from engineers.

[glnc222]

[edit] Problem the graphs above show 60 cycle likely power line noise, A/C adapter. Getting "D" cells -- or a big capacitor and voltage regulator. But then there would be no signal at all left.
The omni lens led seemed to show the same low frequency blink as the power indicator led, 2.5s intervals 0.5 sec on.

I don't like the voltage range picked up by my instruments. It does not seem typical of a 2v LED drive signal switched fully on and off. On the dock, an indicator led for robot-connected is switched by a transistor bypassing the led, shorting it out to turn it off, a little unexpected. All depends on the logic circuits to interface, active low or high etc.

[edit] Next will try a comparator to exclude voltages below 1.5v to get rid of noise. And a digital latch to see if anything too brief to see occurs, though maybe noise free a usable scope trigger will exist. I may also just put a series of two switching diodes in series with the meter or scope to require a minimum voltage to pass, with a pull down resistor.

[Bandi94]

I contacted the seller and the VW is still available so i am going to buy it, next week i can run some test to. On that IC the logo look's like the microchip one, maybe it is a PIC micro, i can't tell from the photo. For getting some better results as you say'd there would be needed an Op-am or schmitt-trigger gate. The IR LED must be switched by a transistor so to get a good result, i would desolder the IR-LED, place a 10k pull-up resistor between the positive supply and the transistor collector. Then hook up the Saleae 24Mhz schmitt-trigger analyzer and capture couple of seconds of data, this way i will see exactly what data is sent to the transistors gate by the microcontroller. Also it may worth try to read the code of the iC and search a DS, and follow the transistor gate back to the Ic and check on the DS if that pin is indeed capable of PWM or they use some other linear output like a DAC. Then maybe it worth a try to place a know IR emitter instead the original VW and see if the powerBot still react's to it, if yes then the wavelength is not so critical. Anyway all this would mean to work on the VW, so if you want to keep yours functional as i think then i will make all this experiments next week.

[glnc222]

The 60 cycle signal above appears to come from the A/C adapter on the laptop, not the guard, reduced switching to all battery laptop power. I am getting nothing out of the emitters even though they work on the bot. A couple of switching diodes in series cuts 1v out of signals. Experience shows I would see any isolated spike over long intervals.
It is more likely the robots respond to the Guard with detection in the proximity sensors, maybe less software involved -- maybe just the existing obstacle detection does it. Those sensors are sensitive with tracking, locating a spot on the floor projected by the remote. I will try some emitter experiments easy with solderless breadboard, but without known frequency doubtful.

Maybe I can detect the Proximiy sensor frequency with phototransistor and try that. I think you can even see some visible red on those sensor emitters (which could suggest some shorter wavelength).

I doubt wave length is an issue looking at the Digi-Key catalog, most are 940nm, some go 850, but detectors are not extremely narrow band -- just emitters. My LiteOn 940 says for most IR uses. There has to be very little emission, very short duty cycle and low power, on the Guard emitters not to be seen with this. Maybe some spike too short for the sensor response time, but then what the heck does the robot use?
It is possible even a continuous signal would trigger the proximity sensors similar to an obstacle, with the pulsing just for power saving in the Guard. Proximiy sensors pulse to compare background illumination with the emitter beam. The bot might look for some pulsing input just to distinguish from background.

Label on the cpu is smudged so can't get spec exactly, might be F760. Could be a Samsung part or custom logic IC, programmable logic, whatever. The circuit board is multilayer so tricky, may need a special device to tap it, closer pins than the large pin DIP through hole logic chips with the test clips for 16 pins etc.

Very strange not to see anything on the emitters even though they do something and work on the bot. I measured both pins, and the resistor end for higher drive, relative to the 3v power supply ground. Low duty cycle is to be expected for battery power saving, as emitters -- several on this -- can take tens of milliamps to get any significant light.

[glnc222]

If not seen my thread here on Optical Cliff Sensor Limitations may be of interest; possibly how to solve the black floor problem using parallax similar to the short range lidars, but more simply.

[glnc222]

Proximity Sensors

A signal was obtained on phototransistor from the top proximity sensors at 1khz with about 20 per cent duty cycle, reasonable for such devices to compare intensity with emission and only background. For some reason the multimeter showed 11khz and 15 per cent duty cycle.
[edit] Proximity sensors operate only when the robot is running. To test remove the brush, cover the cliff sensors, tape in wheels to disable the suspension sensor, and mount on blocks with wheels free; then the remote manual driving can be used.

The Virtual Guard might send a higher frequency signal the sensor could distinguish or it might trigger an obstacle detection in the normal signaling. So far the Guard signal is not available. The proximity signal was detected on phototransistor but not the Guard, mysterious. More elaborate instruments seem needed.
Perhaps it is possible to tap the proximity sensor in the bot to see how it responds to the Guard, but a little difficult here because of a bumper mod (Samsung forum thread). This also depends on how the sensors are constructed, whether integrated off the shelf parts, or some custom design with separate components, all needing analysis. Integrated parts would just output an analog voltage corresponding to distance to targets, based on intensity as affected by the beam and reflection geometry.

Since the bot also detects a projected spot on the floor for "point" manual driving, some separate processing of detector output seems involved, well suited to capturing some high frequency Guard signal if present.

A lot of factors affect how the proximity detection operation would be affected by an external emitter signal, such as amplitude and interaction with the sensor pulses. Perhaps a 50 per cent duty cycle signal would at some time increase the sensor pulse return while only partly raising the signal in the trough, creating a phantom obstacle detection. It also depends on just how the sensor signal is processed and how its output is interpreted by the bot cpu, all unknowns. I will try a couple of emitter experiments but nothing certain. Couple days, have to review the 555 circuits.

Please post the Guard signal if you obtain one for testing. A Roomba VW is not likely the same, as those are passed by TV high frequency 38khz remote control receivers and the Guard was not.

[edit] This is how I power the Virtual Guard from wall plug. Package in heat shrink tubing in line, tuck inside the Guard, tape leads to battery terminals. Bypass resistor insures diodes conduct and A/C adapter loaded to stabilize, avoid any transients. I used a connector for the 5v line outside, for use on other things as well.

[Bandi94]

I will post the signal, as soon as i get the VW delivered. I have an adjustable bench power supply so it will be easy to power it, i will place a big 22000uF cap on the output to get rid of the noise and it should work. I also tried to 'fool' the front bumper proximity sensors with 38Khz 500us pulses with 8ms delay gap between them but the powerbot don't take it in consideration so those sensors must have some filtering to, hardware or software. I need to try to place the robot on something as you say'd to be able to drive it in manual mode and try to record the output of the sensors, the tricky part is if they are running all the same time or just one at a time to don't conflict with the other sensors.

[glnc222]

Point Cleaning Signal

"Point Cleaning" following spot projected on floor by the higher end remotes equipped; additional visible red leds show the IR spot. Detected only in reflection off the floor. The "down" nav remote button engages point projection, as manual driving omits reverse travel. Still the bot moves briefly backwards on occasion, but probably because no cliff sensors or bumper in the rear (wheel stalls from collisions still monitored).

Interesting low frequency pattern: two pulses at 500hz, then two at 1khz.

[edit] Additional 38khz encoded signal found from remote. This signal probably puts the proximity sensor system into following mode, stopping obstacle detection and following the projected spot instead, but who knows:

Ability to distinguish this signal -- and locate its direction on the floor -- suggests all sorts of simple modifications could be used for the Virtual Guard signal, like three short pulses instead of two etc. But then the signal would be easily detected while proving elusive with instruments used so far. Maybe it's a Startrek subspace communicator.

For securing the suspension when testing see "Powerbot Side Bumper Extension and Tilt Reduction Mod" thread in the Samsung forum.

[Bandi94]

I will check that thread to. Very good information about the remote, did you recorded it using the TSOP or phototransistor ? It would look odd if it is a recorded by the phototransistor and there is no modulation in there or maybe the scope / phototransistor is not fast enough to capture the "pulse train". It would be difficult to receive unmodulated signal as it can interfere with natural IR sources, that's the main reason why remotes/ IR gate barriers / etc ... uses modulated signal to be able to filter out natural IR sources.

Later edit: i just got the VW delivered by the courier company so this evening after i get home from work i will start making some test.

[Bandi94]

So first test was with the powerbot, the test was a fast one, my powerbot dose not see the VW . After that i opened up the VW to scope it. At the first look i was right, the IC is indeed a PIC16F716 microcontroller. After a bit of trace tracking i was able to find that pin RB3 ( capable of PWM from the Datasheet) is driving two transistors configured in darlington for high current capability, they drive 3 IR LED's. I soldered my probe to the pin and scoped it, the result was that there is no output, that was confirmed by my phone camera to, no IR light from the emiters. So i went further to check what is that fancy 3 pin emiter on the top and suprise, it's a TSOP recevier . The TSOP receriver's output is going directly back to pin RA4 on the microcontroller. I did not found any marking on the TSOP sensor to read out the decoding frequency.

So that's the reason you couldn't see any output on the scope, the VW is in stand by mode until the TSOP detect's some signal sent from the powerbot, when the signal is detected then the IR output must be driven until the powerbot turn's around, pretty clever method to save some battery power. Unfortunately my powerbot doesn't seems to output the necessary signal to turn the VW on so i am stuck. If you could scope either the IR output when the powerbot passes by the VW or the TSOP output on the VW to see what signal it is needed, then i would replicate the signal with an arduino to turn the VW on and scope the IR output, i am curious if the VW detection code is still there in the powerbot and they just don't output the enable signal to the VW or it is completely removed.

Just scoped the TSOP and it is responding to 38Khz Tv remote so it is a standard metal case 38Khz receiver.

[glnc222]

I will test further with the bot running near the Guard to activate the emitters. Good design to save power on the Guard, which can wear out batteries quickly; I had long observed they ought to work this way, but makers are often cheap (iRobot's Gilbert once mentioned, though, that Roomba had optical switches for bumper durability, compared to Neato cheap tact switches). This does explain how Samsung's Guard can be more expensive than Roomba parts. I did get the Guard working with newer AA's to reduce line power noise; I have an old two cell case from Radio Shack for experiments. The SimpleScope is proving most accurate. There are two input scopes more expensive useful for some logic things besides even more elaborate logic analyzers. I just do not do enough electronics to warrant a lot of equipment.

A possible complication can be the bot not responding to the Guard in manual navigation mode compared to automatic cleaning. Just a little dexterity on the PC should suffice.

Note I use the Guard with tape around the top omni lens because it seemed to interfere with proper aiming across a doorway with it close to the bot, no focus to it. Maybe this was unnecessary over kill, will check for a top emission. On Roomba the top just protects the unit itself from getting knocked around by the robot. So how does mine still work if it would seem the TSOP is blocked? I remove the tape when opening the Guard for measurements. There could be enough light leakage in from below for those sensitive TSOP's to get the signal. Using my TV receiver, maybe I can locate where on the robot it is emitting the signal.
But I observed the Proximity signal only on phototransistor at low frequency.

The "Point Cleaning" signal is from phototransistor. There is no sign of higher frequency components with the scope on high frequency scales, just a very tiny jitter no where near enough to activate an emitter(samples at 1mhz I think). But I can check again with the TV receiver. Were I seeing just a burst frequency with higher within, one would think there would be no gaps in the pattern.

The top TSOP's for the remote are not positioned to receive any reflection off the floor and this must be processed by the proximity detector equipment, several low and high on the front (those bulges in between the TSOP's with angled front faces), and down in the front and sides (see manual diagram for locations). Also with the bot emitting some TV-type signal for the Guard there could be confusion. The TSOP's on the top are positioned to get downward signals from holding the remote high above. The proximity equipment must be custom things with separate detector outputs for cpu analysis or some custom logic. I think I saw some central emitters with light pipes, never examined. The system does have two main microcontrollers, maybe one just for the video processing, whatever; plenty of capacity.

The bumper is interesting in using only two switches (lever snap action microswitches vs cheaper tact buttons), compared to Neato with four, distinguishing side from front impacts. Yet the behavior works well with side impacts, so they seem to have done their analysis. The 7000 has a longer side bumper I think because they moved the brush forward making more space for the bumper linkage, blocked on the 9000 with brush farther back.

[glnc222]

Signals Found

The Virtual Guard emits a low frequency signal (detectable by photosensor) when receiving a 38Khz TV remote format signal emitted by all the proximity sensors on the robot (on models equipped with this feature). The activation signal is emitted only when cleaning, not during manual driving (still possible when mounted on blocks for testing). This saves Guard battery power when the robot is not near. The top omni lens of the Guard contains only a TSOP TV remote type receiver, no emitter. The receiver still works if the lens is blocked, but should be kept open. First the signal emitted by the Guard when activated.

Instead of a top omni emitter like Roomba VW, the Samsung has an emitter in the rear as well.

The burst interval of the Guard Signal:

Virtual Guard emitter burst interval

Guard Signal detail:

Virtual Guard signal detail

Guard signal expanded

Vacuum Guard Activation Signal (from robot, 38khz carrier TV type receiver output, active Low; hard to capture with infrequent bursts, fast fingering on the scope button)
Activation burst interval:

Powerbot signal to Guard (38khz carrier) bursts

burst detail:

Guard Activation Burst Detail

Activation Signal Expanded (modulation of 38khz)

The IR sensor system appears to be quite complex with multiple functions. I have not traced its construction inside the robot. On models lacking the boundary marking feature it would be understandable the Guard activation signal from the robot could be absent, as a waste of battery power when not used. Those models also appear to lack the firmware components to detect and respond to the Guard signal. Newer models since 2016 in the U.S. have used mag strips similar to Neato Robotics instead of the expensive Virtual Guard system; mag sensors added to circuit boards. The lack of emitter focus in the optical Guards and reflections from walls all made that system a little dubious overall. The Powerbot Virtual Guard is the same as the one from Navibot.

[glnc222]

Adding Mag Strip Detection

To test a mag strip system just start the robot with the strip in detection position, and it will immediately pull back, same as if started with the optical sensors off the edge of a table.

In case of use some info on Neato Robotics mag strip sensor with component ID is at
http://www.robotreviews.com/chat/viewtopic.php?p=131834#p131834
Small stick boards with ends in wells in the bottom of the case. They need to be close to a mag strip to sense.
Output was found 2v quiescent and 4v when detecting.

Note the Neato mag sensor adds a ferrite antenna material bar to concentrate the field onto the sensor component.
Your own sensor construction might use different outputs than the Neato sensor, but they are available used on ebay in case of use.

The Neato cliff sensors (probably different than Samsung's) were interfaced to an optical signal detector at
http://www.robotreviews.com/chat/viewtopic.php?p=116780#p116780
The sensors analog output rises the closer to the sensor, so for a cliff the signal is grounded by replacing it with output of an op amp (I use one working good on 5v single power supply). Never manipulate directly the optical sensor output as sensitive to voltages and can be damaged. A comparator is set up to go active low (typically open collector outputs, with pull up resistors when needed) when a boundary is detected, to short out the op amp output in a resistor network, a trick for digital logic input manipulations without actual logic gate components, 7400 series etc.
The example above is for an active low TSOP detection, so the comparator would just be rearranged for the active high input from a Neato mag sensor. Again comparator useful on 5v supplies. There is a need for hysteresis feedback circuits sometimes with comparators, omitted in the above example by mistake, using slowly changing input signals, where the output is unstable very close to the crossing point of detection; you yank the input across the boundary with the output on the first switching action, using points on resistor networks. I can give examples if needed. So if you are detecting a rising input on the negative comparator terminal, and the crossing switches the comparator output low, a feedback resistor to the divider making a reference voltage on the positive terminal can lower it a bit to solidly get a switch. About a 10mv difference is sufficient with the TS912 I use. Comparators vary in their precision. Of course, various latching circuits can be constructed with such feedbacks.

Incidentally I crudely make compact circuit boards with 30 gauge wire wrap wires on the back of perfboard instead of making proper printed circuit boards expensive. Messy sometimes with lots of connections. Perfboard cuts easily with fine toothed hack saw.

Given that certain black carpet materials have proved to make phantom cliffs absorbing the proximity sensor IR beam, it might be possible to make boundary strips of some black material. Unfortunately most common black materials reflect well in IR. Only certain carpets are a problem, some from Ikea, I don't know which. I even tried some IR stealth paint which did not work (not the most expensive kind). A product based on this effect was once sold for Roomba but had mixed reviews. "Keep Off"pads and strips.
Of course, with some additional circuitry on the sensor output it might be possible to recalibrate the sensor to work with some available black material, but dubious as it must work on the normal cliff situations as well -- if you have any cliffs. Industrial robots and lawn mower robots detect current in buried signal wires.

Other Samsung Technical Anaysis
For other technical data on Samsung I did some analysis of the suction fan at
http://www.robotreviews.com/chat/viewtopic.php?p=136798#p136798
Elsewhere in that thread also experiments removing the cyclone extractor to increase suction -- not recommended in the end, the cyclone works good (and the max power mode is not needed, from pick up measurements -- nice it automatically goes on with the dirt detector; besides, the extra motor load may increase wear on the battery, not setup for that current level). This was on a 9250 higher power model. You see, member third_deg, Roomba designer pointed out earlier that the cyclone extractions systems take at least 20 per cent of the suction power from the motor. He doesn't like them, but I like washable filters -- a lot.

[glnc222]

Neato mag sensors robotshop Canada and France, ships intl.
http://www.robotshop.com/en/neato-xv-cliff-magnetic-sensors.html
If a U.S. delivery only vendor is cheaper, I might be able to forward U.S. Postal service $10-15 (they have a minimum).

[Bandi94]

Thank you for the captures. I am not very good at reading oscilloscope ( i am more a logic level guy) but i think i was able to get the required info. Correct me please if i am wrong.

Both use: 38khz carrier.
Activation signal: three 50ms pluses with 140ms gap followed by a 500ms delay before the next three pulses
Virtual Guard signal: 20ms pulses with 160ms gap

I will replicate today the activation signal to try to turn the VW on. For this kind of study i would recommend you a Saleae logic analyzer, you can find on ebay chines clones for only a few bucks and they are compatible with the original software that can be downloaded for free and it is very complex with a lot of tools.

I will try to tape some strong neodymium magnet's on the floor and run the powerBot over them to check if it is detecting the magnetic field or not. With the VW i tried both methods manual and auto clean but the activation signal is not sent, let's see if i will activate the VW with an arduino if the detection will be done or the necessary software is completely removed. Over all i can manage to use the powerBot without VW or mag strip i don't have any major places that it can't handle, when i want to run it only in one room i will place some obstacle in the doorway, after all it can handle all 3 room's by one charge, the other 2 rooms are not accessible because the door's threshold is to high to run over but i can leave it clean the first 3 rooms then let it recharge and next day move it by hand over the door's threshold and let it run the other 2 rooms. I don't have to big mess in the house so if i run it once every 3-4 day it's ok, it fill's a third of the bin, most of the dirt is dust from the carpets, from the 5 rooms 3 of them have carpets.

Also yesterday i opened up the battery bay and the 9040W model has a 21.6V 1900mAh one, not so big compared to the case. I wounder if they placed smaller capacity cell's or there is space left in the case after seeing that the higher end powerbot's have almost 3000mAh packs, in the same sized casing