Powerbot Mag Strip Addition Mod

[glnc222]

[edit]See last post for failure of the cliff sensor after some usage with mod.

A sensor is added to Samsung Powerbot models not equipped with the mag strip boundary sensor present in some models in the U.S. since 2016. Models with optical Virtual Guards are still sold in Europe, but as detailed later, the Virtual Guard has many deficiencies. While models lacking the feature have board connectors and mounting brackets for Samsung's sensor, the software to use it might be absent and the Samsung parts expensive or unavailable.
[edit] All robot brands with mag strips use the same magentization of strips, for the Hall Effect field sensors used. These have the same pole all along the top and bottom, compared to common magnetic stick-on material with alternating poles, not useful. Neato Robotics makes a heavy plastic strip which I prefer to the stick-on ones from Samsung; also more widely available. Works with the mod here.

[edit] Installation shown on a 9000 series model; 7000 series disassembly later post http://www.robotreviews.com/chat/viewto ... 01#p145801

A DIY sensor is shown here and works by combining the mag sensor output with the existing front cliff sensor, producing the same avoidance behavior. Samsung uses a single mag sensor in front of the center cliff sensor.
Mag strips from Neato Robotics and others as well as Samsung's own strips can be used.

The Powerbot cliff sensor is a four wire type with unprocessed emitter and detector lines processed by the cpu, compared to integrated types with internal processing and a single output voltage varied with distance to objects. Neato Robotics uses such a three wire Sharp sensor, and the Neato mag sensor component is used here, the Allegro A1324LUA-T, the most sensitive in their series. As in the Neato assembly, a ferrite antenna core rod is used to double sensitivity of the Hall Effect field detector.

Robot mag boundary strips are magnetized with the same pole all across the surface for the Hall Effect sensors. Commong stick-on magnet material is "multi-pole", alternating and cannot be used. A YouTube video test of several robot strips shows they are all the same, and Samsung can use those made for other robots. https://www.youtube.com/watch?v=WnX86fztht4 The Neato strips are the most widely available, since marketed for years, and are likely cheapest. The Powerbot strips from Samsungparts.com are $31. Samsung strips have self-stick tape on them. Neato strips are more substantial and made to be portable.

The Interface Circuit

Samsung's cliff sensor outputs a ground level voltage when sensing a cliff (nothing detected), so grounding the input to the cpu simulates a cliff detection when a mag strip is detected. A Texas Instrument TCL393 low power, low voltage comparator is used to connect the two sensors with response on the smallest field detection. The output is active low, and open collector with no effect in absence of a mag strip.

[edit] An alternative circuit employs a dual Op Amp TS912A; this circuit avoids inserting any resistance between the Cliff Sensor and the Cpu input. Tested with the mag sensor working but am awaiting a replacement Cliff Sensor as it seems not working even without the mag sensor interface. The robot cleans but the front cliff sensor no longer responds to distance, even though having a 496hz signal out. [update] New circuit working after correcting polarity of the sensor cable connection.
[edit]TS912 OpAmp is obsolete in 2020, may not be found at distributors but some from China on ebay. A close equivalent is the newer Texas Instrument LM358, LM358B etc.

The Op Amp works to adjust output up or down until the inverting or minus input, connected with feedback from the output, equals the input signal non-inverting input. With no feedback output goes to either rail as needed. Comparators are made of Op Amps.
If you insert this IC backwards it will burn up with smoke -- fails the "smoke test" of assembly. Maybe insert wires into the female cable connector to put it on breadboard as assembled for testing before installing.

Resistors 1/8 watt for compactness. Alternative values possible to divide 5v power supply to 2.6v reference, and R1 is set to R2 just to minimize sizes needed. e.g. 36/39K, 33/36K, 47/51K.

No hysteresis feedback sometimes used with comparators to avoid oscillations near the crossing point is used here, for simplicity and because it is unlikely in the circumstances to affect the cpu interpreting cliff sensor signals. It is not like the robot pauses for long periods on a marginal boundary, and the Allegro sensor might already have a fairly quantized response (not tested).



Perfboard Construction

[edit] This circuit is so simple perfboard mounting is not strictly required. You could bend out sideways the IC socket pins and solder to them, and wrap the whole thing in duck tape, if such crude methods are satisfying. I just happened to have all the stuff on hand and left over from other things. The perfboard provides strain relief on the cable connections to help prevent wires breaking in handling a lot. Just putting the socket with attached wires onto a bit of carboard to stiffen when wrapped in tape might also help.

TLC393.jpg (12.81 KiB) Viewed 3457 times

An 8 pin DIP socket is used for the TCL393 to avoid soldering heat damage.
The 4x20mm ferrite rod is covered with heat shrink tubing or tape for insulation over bare sensor wires.
[edit] Correction -- the ferrite rod is non-conducting, does not need covering.
Sensor mounted with bare wire section for applying soldering heat sink clip for protection.
26 guage stranded hook up wire was used; about 12 in. leads from the interface to the mag sensor.
Drill out perfboard holes 1/16 in. for passage of insulation, forming strain reliefs.

A special problem exists with the Samsung cable connector. 4 pin 2mm pitch header. Samsung's plug size is slightly smaller than the commonly available JST PH 2mm connectors and the Samsung unavailable. It is necessary to construct a plug to fit the Samsung socket for the cliff sensor cable on the system board by removing the pins from a JST PH 2mm connector (not wider pitch ones with larger pins) by pressing the exposed locking tab and pulling out the back. The pins are then covered with heat shrink tubing and wrapped, oriented the same, with a single layer of thin masking tape or scotch tape. For the Samsung plug, the a JST header can be attached to a JST plug, or a substitute socket made from leads from a 1/4 watt resistor as a similar pin size (as done here).

[edit] Another possibility was received late, standard 2mm pitch female board headers, without shell enclosing plugs, usable as plugs instead of JST plugs supplied with attached leads. These are definitely thinner than the JST plugs but the same width, though with these it would be possible to grind the sides thinner if needed. You would still use heat shrink tubing attaching to the pins and have to supply wire. e.g.
4-Pin 2.0mm Pitch Female Header

female header 2mm pitch

Header2mmPitch.jpg (12.15 KiB) Viewed 8960 times

ebay https://www.ebay.com/itm/4-Pin-2-0mm-Pi ... 2749.l2649

[edit] It is probably easier to skip the board for the sensor and just mount on the ferrite rod, wrapping with the capacitor onto leads spliced to the three wire cabling; put in heat shrink tube. This may then pass through the closed wiring hoop eliminating splices on wires to the interface.



Installation

Assembly is shown here for the 9000 series Powerbot. Additional 7000 series use will be posted later when an expected service manual arrives from Samsungparts.com. 7000 series electronics will be similar to 9000.

The mag sensor is mounted in front of the center cliff sensor and the wiring joins the cliff sensor wires in clips, around to the side of the sensor assembly board and over the top side. Because the wires pass through a closed hoop fitting at the side, an unsoldered splice is used on the mag sensor wires at the interface instead of an additional connector (though the JST connectors come in bunches so available if wanted).

The cliff sensor is accessed in the bumper switch assembly frame under the sensor system board in the front section of the powerbot (see photo above); two screws, sides of the frame, into the top section above.
Note the white bumper switch actuators are free on pivot pins so can fall out and need to be saved.



The robot front section is opened with four screws on the bottom, two under the brush. For screws in deep wells a paper clip wire bent into a hook can help pulling them out. Various tabs around the cover are worked free to lift.
The sensor assembly is lifted by removing two screws to the sides, and unplugging the wide sensor system plug cable (release locking tab while prying up at the sides; note this cable passes through a hoop to reach the socket, in case gets pulled out, for reattaching). Work the back end on one side over the brush motor to free movement. Then the frame underneath is freed with screws to the sides.

Mag Sensor Mounting Bracket

See photo above under Perfboard Construction.
The bracket for Samsung's own mag sensor provides too little clearance for the ferrite rod using the slots present, so the mag sensor here is merely inserted between the cliff sensor and the bracket front, with a very tight fit holding it in place against the heat shrink tube over the rod. There was no room for a second rod retaining loop near the top but was not needed. Insulated cable wires on the back of the perfboard increase thickness, and maybe routing on top of the perfboard, exiting to the back, might allow different fits, not tried. With looser fits the mag sensor perfboard can be taped to the cliff sensor front.
[edit] The ferrite rod later proved non-conducting so will be thinner without insulating cover, fit will be looser.

The bottom of the Samsung mag sensor board bracket slots fit against the case bottom, so this is where the level at which mag sensor here is positioned.

Interface Mounting

The interface is plugged between the socket and plug for the front cliff sensor located in the middle atop the sensor assembly. The pill box sized module mounted vertically to the right is the WiFi radio. The interface is tucked behind the radio module (with integral antenna; WiFi does not appear degraded so far).

Operation

The sensor has worked well and has good sensitivity which may even reach under carpets, depending on thickness. It may be better than Neato Robotics own sensor on which it is based, due to refined small signal detection. The sensor has a range of about 3/4 in. below the sensor face, which might translate to 1/2 inch below the floor level, but needs checking for the details of mounting.

Motivation -- Problems With Virtual Guard

Before introducing mag strips with the 9350 model in the U.S. 2016, some models with Virtual Guards from the old Navibot line where sold, and are still sold for the moment, in Europe etc.

These guards do not produce a focussed beam and have omni direction reflectors above each of three IR emitters. While resembling Roomba virtual walls, they work very differently. The clear top lens on the Roomba wall is an emitter to protect the wall unit itself, while a partly focussed beam emits from the front. The top Samsung lens is for a receiver which gets a coded signal, on the TV remote standard at 38khz, from the proximity sensors around the robot. The emitters only power when receiving an activation signal from the robot. So the Guard protects an area all around it, with emitters on both sides and the rear, governed by the robot being anywhere near it. There is no sharply defined boundary line such as wanted say, across a doorway; the boundary of this omni direction signal area is not at all clear.
This design appears suited to a random nav robot -- if at all -- and should never have been used for a systematic guidance system like Powerbot, which needs to map any boundaries encountered. It was probably just offered because available from Navibot. The units sold with Virtual Guards have the connectors and brackets for the mag strip sensors introduced later, so there seems to have been just a product development plan of some kind in place.

[edit] The only way I was able to use the Virtual Guard across a doorway was placing it inside the door to the side against the wall, so the wall provided some direction to the beam when encountered from outside the door. This arrangement worked in one case, but there is an issue of blocking the activation signal from the Powerbot.

For more technical details of the Virtual Gaurd see other hacking thread
http://www.robotreviews.com/chat/viewto ... =4&t=20151

and http://www.robotreviews.com/chat/viewto ... =4&t=20196


[edit] For other mods to the Powerbot 9000 bumper see Samsung forum thread
http://www.robotreviews.com/chat/viewto ... 22&t=19388

Powerbot 9000 disassembly http://www.robotreviews.com/chat/viewto ... 22&t=20207
Powerbot 7000 disassembly http://www.robotreviews.com/chat/viewto ... 22&t=20188

[edit]Repair: after a couple years the TLC393 comparator chip, handily mounted with a socket, needed replacement. For unknown reason it stopped fully grounding the input, only down to 0.1v, and was not triggering the cpu (while the cliff sensors continued to work). Replacement got the mag sensor working again.
However the problem later got worse, and seems to involve bad coupling with the crude connector made; awaiting 2mm header to make a better one.

[glnc222]

Advantages

Physical instead of electronic boundaries are fairly easy to make, with sticks, molding, and right angle plastic wall corner guard material. The mag strip is neater, rolls up for storage, and is flat for not tripping over. The Neato strips are portable, temporary if wanted, compared to Samsung's strips with self-stick tape.
The Powerbot installation works better than the old Neato one where at a height transition from carpet to hard floor, a wood strip had to be used to raise the mag strip up to the carpet level in order to work with Neato's sensor. I do not know how Samsung's own detector works, but the one here lets the mag strip work as expected for a change.

The Virtual Guard worked in this case only by covering it partially and angling it in just a certain way hard to determine. Not attractive. Besides, they use batteries unless an extra constructed wall plug power supply is adapted to substitute, which is not available ready made. At least Samsung's system saves the batteries by activating only when the robot is nearby.

Samsung's own mag sensor does not appear to have a ferrite rod snsitivity enhancement though hard to tell in this photo; perhaps one could be taped on their board to increase its sensitivity. I do not have a model equipped to test their unit's performance.

[glnc222]

Additional Connector News

Another possibility for the problem Samsung cable socket was received late, standard 2mm pitch female board headers, without shell enclosing plugs, usable as plugs instead of JST plugs supplied with attached leads. These are definitely thinner than the JST plugs but the same width, though with these it would be possible to grind the sides thinner if needed. You would still use heat shrink tubing attaching to the pins and have to supply wire. e.g.
4-Pin 2.0mm Pitch Female Header
ebay https://www.ebay.com/itm/4-Pin-2-0mm-Pi ... 2749.l2649

[glnc222]

Boardless Sensor Mount

It is not even necessary to mount the Allegro sensor on a board. Slide insulation over the leads, bend at right angle at the sensor, solder the tiny capacitor onto the pair at the top along with cable spliced on, and tape or shrink wrap the ferrite rod onto the leads. Tape in place onto the cliff sensor.
It just saves work to put the capacitor with the sensor than on the interface board.
[edit] This way might even pass through the wiring hoop without needing splices on the sensor cable at the interface.

[glnc222]

Optical Boundary Marking Alternative to Virtual Guard

The same interface can be used for a proper optical boundary system compared to Samsung's dubious unfocused Virtual Guard. Some differences would exist from the Neato forum project "Optical Boundary Marking" http://www.robotreviews.com/chat/viewto ... ry#p122722
The comparator here is a dual package, and both outputs can be wire-or'd, connected, to combine two sensors, either triggering the cliff sensor input.

Because TV remotes used for Neato above are on the same 38khz frequency as Powerbot's remote and dock signals, a different frequency is needed to keep the remote and dock from triggering a boundary; the bot itself also emits 38khz signals to activate Virtual Guards neaby. A boundary emitter would need a 555 IC based oscillator as a number of projects for Roomba Virtual Walls and DIY remote controls show; one was at 36khz instead of TV's 38khz. The TSOP receiver components come for a number of frequencies. A crystal controlled oscillator would be preferred but might require a microcontroller; maybe an inexpensive small 8 pin PIC might do, unknown.

A single receiver over the front is required for behavior keyed to the cliff sensor; very small. Thin wire could be run over the back or the front section, down to the joint in the top cover, right where the added interface board is -- through a little wear on a corner...
Optical boundaries are interesting for very long barriers (never supported by Roomba) invisible with only the emitter unit tucked away somewhere (with a wall plug for power instead of batteries), and may be cheaper than large lots of mag strip.

[edit] receivers Vishay TSOPyyyxx where xx = frequency. Lots of 30khz ones on ebay cheap vs 38's for TV's, robot.
TSOP34830 30khz

[glnc222]

For wiring an IR receiver a ventilation grill is located on the back of the front section case behind the camera. A receiver can be placed in front of the camera and a cable passed around behind the camera. I will see if such an addition works as more parts arrive.

[glnc222]

7000 Series Assembly

No separate front and rear covers. Open with six screws on the bottom, four in the front and two in the rear. Do not lift off the cover fully until unplugging first the display cable in the rear (fastened into the top...) then a bumper cable in the front.
The mag sensor bracket on the cliff sensor similar to 9000. Observe wiring layout.
The cliff sensor signals should be checked for similarity to the 9000. These models introduce automatic detection of hard floor and carpet difference, and not yet known how they do that (Roomba was described as using laser mouse pointer tech to measure texture of the floor, but other ways might be possible such as effects on drive wheel drag and motor loads).

[glnc222]

Optical Boundary Impractical (or at least difficult)

When using the front cliff sensor input to signal a boundary, the high rear end of the Powerbot 9000 with the bin and suction fan creates a geometry problem for any boundary beam detection. Were the robot to approach the boundary at a sharp angle as shown below, the only way it could work would be with a sensor in the front up high on a post, to see over the back. This would interfere with contact with over head obstacles, maybe interfere with the camera, and be just generally unattractive.

The Samsung system, for what it is worth, handles the above case with the proximity sensor receivers on the sides, and has behaviors programmed consistent with those positions on the robot. The maneuvering behavior programmed for the front cliff sensor might not work with any sensors added on the side, triggering the front cliff sensor. There are also side cliff sensors which could be used, but the installation would become very elaborate, with two interface circuits etc. (though without mag strip, the dual comparator in the one IC could be used). It might be possible to mount receivers inside detecting the boundary beam through the IR transparent dark plastic.

Multiple receivers can easily be connected to one interface comparator using diode isolation and a pull up resistor, but leaves these behavior issues unresolved.

Vishay TSOPxxxx remote control receiver(s) can be added to the mag strip interface comparator with minimal adjustments to the perfboard layout. The reference voltage is shared between the two halves of the comparator, and an RC filter is in line with the receiver output to remove oscillations. The receiver is active low compared to the positive going mag sensor.

When experimenting with these receivers, a low power 2ma LED with a 2.2K resistor can be connected to the output to indicate when a signal is received (for 5v power).

Several problems were found with the Vishay receivers: first, the 30khz version still responds to 38khz remote control signals, so the Powerbot remote could potentially trigger boundaries. Only digital encoding, requiring a microcontroller, would distinguish the signals. Possibly geometry, with the boundary sensors facing horizontal would prevent remote signals from high up from being detected. Unfortunately, the dock shows some 38khz signal as well which would could only be distinguished from a boundary emitter with digital code processing. Eliminating the Powerbot remote alone would not suffice (can't have the dock becoming a boundary!).

A second problem with the receivers is the angle of detection, which is only about 90 degrees, where maybe a full 180 degrees horizontally would be needed. Two detectors would then have to be mounted together at an angle, on each side (though they are small -- just more expense and work to build). In the diagram above, one has to actually point towards the back. It is possible to construct a conical omni directional reflector for a top facing receiver, with paper and aluminum foil, but a lot of tedious detail -- with unknown results.
[edit] Actually this problem was limited to using the 555 based emitter; a TV remote gave 360 coverage with the receiver facing up, from the side. See later post
http://www.robotreviews.com/chat/viewto ... 20#p145820

Finally there is the problem of collimating or focusing the boundary emitter beam for a reasonably precise boundary definition. A large cylindrical lens made from a strip reading magnifier worked in experiments (see Neato forum Optical Boundary thread), but a smaller and cheaper version made by cleaving an acrylic rod (like a diamond cutter), did not seem to work so far, maybe with more work. [edit] This turned out to be only a focal length measurement issue, correct distance to the lens. Maybe still useful.
Actual lens products are very expensive. Emitter LED's do come with the case molded lens focus down to around 15 degrees, but not as tight as wanted for a long beam reach (presumably the advantage of optical systems compared to mag strips).

In initial experiments, a 555 oscillator driven emitter could be detected by the Vishay receiver only four feet away, compared to older results with TV remote emitters -- even over a range of output power. Since the 30khz version did not exclude the 38khz signals, the cheap keychain TV remote would seem the preferred emitter (these remotes can over drive emitters on short duty cycles).

Maybe with a small 8 pin PIC microcontroller (inexpensive, and compact enough for additions to the robot, compared to the larger ones), something could be done, but too much work for my needs, not being equipped for microcontroller programming at the moment.

So for the purposes here, adding optical boundaries improving on Samsung's dubious Virtual Guard is found impractical even if possible. Others may pursue it further if more interested.

[glnc222]

Remote Control Encoder/Decoder

Instead of a microcontroller dedicated IC's encode and decode signals for Vishay type IR receivers.
e.g. Motorola MC145026, MC145027. Data sheet
https://media.digikey.com/pdf/Data%20Sh ... ,27,28.pdf

Not available at Digi-Key but listed on Ebay, in Germany, and at Jameco. Perhaps other equivalents are made but Digi-Key lists a lot of obsolete, discontinued ones.

16 pin DIP package practical for inclusion in robot mods.

A project using these for robot controls http://www.electronics-base.com/general ... bile-robot

So at least with these it could be practical to make a boundary emitter and receiver which is distinguished from all the IR coded signals used in the Powerbot.

It would still take receivers on the sides as the Powerbot has its little towers on the front corners for similar.

Another possibility is tapping into the receivers in the Powerbot to extract a unique boundary signal distinguished from the Powerbot dock and remote signals.
The side cliff sensors could be interfaced with a single dual TLC393 comparator, if not the decoder alone (have to check) dedicated to the boundary system, but it looks like two decoders are needed, to separately trigger each side cliff sensor, for a possibly useful maneuvering response -- still a question.

[edit] One decoder might suffice using shared input from two sides, using signal presence to enable which cliff sensor to trigger. This assumes the boundary emitter will be received only on one side at a time.

Using a keychain TV remote for emitter, which previously showed some range, the code from that would be what is to decode. Examine with a scope or logic analyzer.

[edit] The Motorolla parts handle a 5 bit device ID address with a four bit function code returned. This might not work with TV codes, will have to see. Sparkfun has a similar remote with 32 bit codes, 16 bit addresses and 16 bit function codes. Arduino code for remotes [url]ps://learn.sparkfun.com/tutorials/ir-communication[/url]
Sparkfun remote https://learn.sparkfun.com/tutorials/ir ... okup-guide
Arduino code for various brand TV's mentioned.
A PIC microcontroller based project uses a 14 bit code referenced as one particular standard RC5 for 36khz carrier
http://www.picprojects.net/rc5_decoder/
Another PIC specifically for TV's 38khz http://www.pyroelectro.com/tutorials/in ... _receiver/
The standard coding methods are described in Texas Instruments document, RC5 and SIRC standards
http://www.ti.com/lit/an/slaa134/slaa134.pdf

I do not consider it very practical to add a large microcontroller to the Powerbot just to make a boundary detector, what with expense, space, power drain etc.
Were it possible to get an emitter and receiver made from simple photodiodes/transistors and emitter LED's to work with sufficient range, some simple methods could be used to exclude 38khz modulated signals from the Powerbot's systems, like requiring very long pulses, without decoding anything, just an RC filter and comparator -- already available in the mag strip interface.

[glnc222]

Better TV Remote Results

The angle coverage of the TSOPxxxx does not seem to be a problem with TV remotes. Facing upwards, the TSOP receives the TV remote from all sides at the level of the sensor, at a very large angle from vertical. Reception was also good out at least 15 feet, like before with the Neato experiments. I do not know what makes the difference from the 555 with emitter tried, the power, emitters, whatever. The 555 emitter specs 200ma peak power momentary, and I only went to about 100. Some TV remotes are mentioned going to 500ma on short duty cycle, tuned not to over heat the LED, which governs the maximum. Add water cooling...

So two of these receivers, on each front corner of the Powerbot, or maybe inside, or on top of the little side towers for Samsung's receivers, would work, facing up with effectively 180 degree angle capacity from the side.

However, the keychain remote on hand showed a short time out on repeated output holding down the buttons, just a few seconds.
[edit] time out vanished when the switch jumpered, an external power supplied, instead of button push. Also needs a full 3v supply, not 2.9v worn battery.
The TV's regular remote has some longer time out too long to quickly measure. The keychain has a coin battery so understandable to preserve power. The keychain also repeats slower than the regular remote, more power saving. So a 555 circuit with transistor can be used to activate the button contacts with a duty cycle to interrupt, re-press, say once every second or two.

To use this the remote needs to be decoded with the Motorolla IC to distinguish from the Powerbot's elaborate IR signaling system. Frequency change to 30khz did not work, the filters are not tight enough.

[glnc222]

Conclusion

An improved optical virtual boundary could be made with a TV remote as emitter; a high frequency system is needed to filter ambient light interference. To filter out all the Powerbot's high frequency signals the remote must be decoded by a microcontroller (the Motorolla parts use different formats and meant to be used in transmitter/receiver pairs). Two TSOP3848's can be mounted on the side and combined into a single signal with diodes for a decoder, triggering the front cliff sensor, or separately the side cliff sensors -- experiments needed.

There is extensive ready to use code libraries for IR receiver decoding with Arduino, but only the UNO etc. not the tiny versions. I am not familiar with all the possible Arduino's.
https://learn.adafruit.com/using-an-inf ... are-needed

There is lots of code online for the 18 pin PIC controllers shown by Google, but not for the more attractive 8-pin 12F683, which could compactly fit with the mag strip comparator on an interface board and maybe replace the comparator as well, for both mag strip and optical on one board. Yet the features needed are likely there, just different than used on the others, some interrupt thing, so a new coding project would be needed.
The keychain TV remote in default code setting -- they can match any TV -- for the volume-up key, looks a lot like an RC5 standard code for remote controls, at which the PIC projects are directed, but just a guess.

So a separate optical boundary mod might be posted in future if I happen to get into microcontrollers. Anyone interested doing that can pursue it further at the moment.

[edit] It is still possible to use an unencoded, 555 based 38khz transmitter -- by decoding the dock and remote Powerbot signals to exclude them, by difference. It is just those TV remote transmitters seem to have optimized power for range, besides dirt cheap, and compact in the keychain ones.


[edit] If you add mag strips to your Powerbot as shown here, your experience posted here would be of interest.

[glnc222]

Addendum:
Error messages for the Powerbot include C07 for debris blocking the cliff sensor, like dirt on the window. [This means the sensor is detecting a surface closer to the detector than the normal distance to the floor. As the sensor output goes to ground when no floor is present, a cliff, the sensor output must be higher voltage than normal for the floor, for the cpu to see this difference.]
[edit] Incorrrect -- covering the sensor window lowers voltage oddly, while covering the cover opening below (window recessed) gives a standard floor reading. See following post. The covered window outputs around 30mv compared 10mv for cliffs, while the normal floor is 0.567v. The error detection might be based on some inconsistency between the three sensor, hard to tell. Samsung has its own proximity sensors instead of the typical off-the-shelf parts used in other brands. Some of Samsung's are even discrete emitters and receivers, doing double duty as transceivers for encoded IR signals with docks, remotes, Virtual Guards, and detecting a spot projected on the floor which the bot follows around in "Point Cleaning" (vs Spot Cleaning small area).

No C07 errors were encountered when using the mag sensor adapter above, but in case of defective components or other oddities, it may be advisable to note baseline sensor output voltages with and without the adapter installed, and with and without cliffs, and debris on the sensor window -- for diagnosing any hardware problems.

Note the cpu analog input will pull up the voltage on the sensor line unless loaded by some sensor input to regulate it, in the nature of these input cpu port circuits on the chip. A disconnected sensor input might produce the C07 high voltage but have not measured the case. Software might detect the disconnection and issue some other message or otherwise not function.

[edit] Error C07 can also occur for unusual reasons. When the drive wheel extension sensors are disabled as I did to prevent certain traps, lifting off the floor causes C07. Cliff sensors detect a cliff and the drive wheels turn to pull the robot back, but the cliff remains sensed, which it would not in actual motion Software then interprets this as a cliff sensor failure, indicated by C07. Were the wheel extension sensors tripped normally the cpu could tell the cliff sensors were acting as expected and no error would be shown.

[glnc222]

Circuit Breakdown Encountered Ater 2 Years

After being used a couple years and dozens of cleaning runs, two problems were encountered with the addition, after a long distance move to a new house.

1. The TLC393 failed to fully ground the cliff sensor signal, going down only to 90mv, too high to signal a cliff, and had to be replaced (socketed for easy replacement). The new chip worked well for a couple dozen runs, but another, fatal problem emerged.

[edit] The problem has later been traced to possible poor connection with the crude plug assembly for the cpu input socket. At the same time, a new interface using an OpAmp IC instead of the Comparator has been made, added to the top post in thread. This eliminates any resistance inserted between the Cliff Sensor and the cpu input.

2. The vacuum started showing C07 cliff sensor errors intermittently. At first restarting the robot would work, but the error repeated after a few minutes run. Then the problem worsened, showing the error as soon as the robot started driving away from the dock (though it would back out of the dock without error).
Unplugging the mag sensor adapter stopped the errors, but it has not been possible to diagnose the failure, as the signals all seem correct.

Examining the sensor signal voltage, the normal floor output is 0.567v, found with and without the mag sensor adapter inserted (RMS of 496hz oscillation to separate background illumination). The cpu input must be high impedance because with the separation resistor between the cliff sensor and the comparator, the voltage across the resistor is only 2mv. Both the cliff sensor and the mag sensor bring the signal down to around 10mv.

These measurements are made with the robot stationary, while the errors all occur after the suction fan and drive wheels are running. This could affect the power supply causing some problem with the 5v logic supply used, and the TLC393 comparator and mag sensor add only slightly to the load on this (theTLC is especially low power vs other comparators). But this was not a problem previously, so the change would seem to be in the Samsung circuits, which cannot be fixed, only replaced. This 9250 is four years old, had one battery replacement, and might be due for entire replacement.

[edit] One possibility not checked is some noise in the 5v power emerging from system board wear, from the motors running, affecting only the mag sensor circuit. Perhaps a filter capacitor on the mag circuit power line might help, but has not been tried (such extra local filtering is common with a lot of digital components). Besides, the wear is in the Samsung components which might need replacement anyway. It is tedious to set up the oscilloscope to try and see any such situation.

It is also possible the software just ignores sensors until the robot is traveling, and doesn't process this type of error until then. Power is on as soon as unit is turned on.

It was also found that covering the plastic window over the cliff sensor, which is recessed a half inch or more into the case bottom, drops the voltage to 30mv, more like a cliff than a closer proximity. Covering the case opening gives the same higher voltage as a normal floor. (So when disabling the cliff sensors, all three, for certain problem black floor materials absorbing the IR beam, cover only the case opening, not the recessed window.) The C07 error is listed as debris blocking the cliff sensor.

I had been concerned there was no resistor used between the TLC393 and the cpu input, and tried adding a 4.7K resistor to the input, with no effect. The input is high impedance analog, with the voltage across the 66K resistor on the cliff sensor only 2mv, on a normal cliff signal 0.567v.

Since no signal alteration could be found from inserting the mag sensor adapter into the cliff sensor circuit, I cannot identify the failure caused, and have to remove the adapter (which I no longer need anyway). Given how debris on the cliff sensor lowers the signal much like a cliff, contra usual proximity detector operation, how the C07 error is determined remains obscure. Perhaps that particular 30mv level is used but since no such effect is found from the mag sensor addition, I am not bothering to test that.
Possibly there has been some change or deterioration in the cpu or system board components, which is not possible to analyze. [it later seems to be the crude connector plug made, from additional voltage measurements]

[glnc222]

An alternative circuit using a dual OpAmp TS912A instead of the Comparator has been added to the top post in thread. Worked on breadboard but awaiting 2mm header delivery to make an improved plug for installation without cutting any existing original Samsung wires.
[edit] Now tested in the robot, with the mag sensor working, but apparently need to replace the Cliff Sensor, as it outputs a 496hz signal as expected, but does not respond to distance. No errors produced. This part is not a robot service part, but is found as an electronics part, Samsung KIR5006C distance sensor (proximity sensor), specified for the robots, available at Buyspares.com (special order, i.e. not stocked, but they order from Samsung).

[update] New circuit working with replacement sensor taking a few weeks special order and mail from British dealer. Some wrong voltages may have been due to reversing the sensor cable plug from wrong polarity labeling discovered when installing sensor, but luckily no damage. It's possible the old sensor is working but too much trouble to install (requires an external emitter signal, 4-pin type, so harder to test outside the robot as well).

[glnc222]

Feb. 2020. After running a dozen times with the new op amp circuit and new cliff sensor, the cliff sensor failed, outputting a ground, cliff signal. The robot stops on a C07 cliff sensor error. An attempt to eliminate the sensor was made by feeding the emitter signal 2v RMS 496hz back to the output line divided to the 0.6v RMS found on the other cliff sensors, using 47K/20K or 51K/22K resistor divider. The correct 0.6v was found in the output to the cpu, but the robot continued to stop with the error.
I am therefore replacing the robot with a refurbished R9350.
It might be possible to repair the 9250 (Euro model, not the U.S. R9350 with mag sensor) by replacing the cliff sensor once again, but there can be no guarantee.

[edit] perhaps a resistor between the sensor and the op amp input might help but I do not have any data sheet on the sensor or know enough of the engineering. It is possible the bypass did not work because of some timing difference between the emitter and the sensor output, but I would have to dig out the oscilloscope or logic analyzer to analyze more, some other time. Besides, the problem of the mod eventually failing would not be affected, so not especially worth while. The 9250 is five years old and a refurbished replacement pretty cheap, so got enough out of it.

[glnc222]

I think the clif sensors were damages just by plugging in cables backwards, often fatal to IC's, and the 9250 is being recycled on ebay for cost of shipping. Found and open box R9350 to replace.