N-key rollover test.

[IBI]

Sorry, I missed that it was for a G80, I was thinking it was a rubber dome keyboard. I don't know what the pins correspond to on the MX switch so I guess I'll have to read up on that.

If you're working off the back of the PCB that explains why it's the wrong way round - if you plan to do any others I'd flip it first before you start lettering.

[keyb_gr]

Whoa, that looks like a fair bit of work there. If I'm not mistaken, that should be an M-style matrix.

[huha]

Sorry, I missed that it was for a G80, I was thinking it was a rubber dome keyboard. I don't know what the pins correspond to on the MX switch so I guess I'll have to read up on that.

It's quite easy actually. The large solder pads correspond to the switch pins, the small ones to the "extra" that's installed. Depending on the switch, this can be nothing, a diode, a LED or a simple jumper. You can see what's installed by looking down on the switch; you'll recognize a LED instantly. Diodes are a bit harder to spot, but still quite easy, as can be seen on this picture. It doesn't really matter, though, as no current Cherry board seems to use diodes, save for the really good stuff the Koreans and Japanese are getting.
(You can't get it in Germany, though. They're OEM boards which Cherry are not allowed to sell here.)

If you're working off the back of the PCB that explains why it's the wrong way round - if you plan to do any others I'd flip it first before you start lettering.

That would have been a sensible idea, but I didn't think of it when I started. It's not that bad, the actual graphic representation of the matrix is quite uninteresting. But when you trace the traces, you can get the pin assignment, which is the real thing of interest. Think of it more as an art project, as the really interesting thing is written on a sheet of paper here--the G80-3000's key matrix.

Whoa, that looks like a fair bit of work there. If I'm not mistaken, that should be an M-style matrix.

I haven't taken an M apart yet, so I can't tell. Would be interesting to know, though. I'll try to polish my matrix table up a bit, make a nice spreadsheet and upload it here for anyone who's interested.

-huha

[keyb_gr]

I haven't taken an M apart yet, so I can't tell. Would be interesting to know, though. I'll try to polish my matrix table up a bit, make a nice spreadsheet and upload it here for anyone who's interested.

I haven't seen an actual M matrix either, but this one looks like the key blocking should be like that. Does this particular board accept QWER and SDFJKL but not QWAS?

[Red October]

Unicomp BO40B56 122-key Host-attached Terminal-Emulator keyboard, Model M architecture.  
Keys Accepted: 10
I suppose this is the highest practical maximum as that is all the fingers you have.  If you try to get cheeky with it and push an extra key with the heel of your hand, it will register 1-4 keypresses only, but will produce no spurious charcters.

[IBI]

Unicomp BO40B56 122-key Host-attached Terminal-Emulator keyboard, Model M architecture.  
Keys Accepted: 10
I suppose this is the highest practical maximum as that is all the fingers you have.  If you try to get cheeky with it and push an extra key with the heel of your hand, it will register 1-4 keypresses only, but will produce no spurious charcters.

So every combination of 10 you tried it accepted all of them? That's unusual for a Model-M based keyboard.

[zwmalone]

Remember, the Key # isn't the Max you could get, but the minimum. My Evolution can manage 8 keys on certain combos, but it's still a 2key board...

[o2dazone]

For WOW guys just mash your forehead against the keyboard and post a webcam shot.

Speaking of did you hear what they're doing for casual players? Emblems of Conquest in heroics and Triumphs for heroic dailies. Shame I quit before 3.1 came out...the ship has sailed, sorry Blizz

[alpslover]

For 6-key bragging rights try CTRL-QWAZX.

why is this particular key combination significant?

my cheap expo486 keyboard can do this combination, but it's far from being n- (or 6-) key rollover capable.

[alpslover]

I dunno - just picked it because it's a scenario you could have in a FPS and it's a combo of modifier and different rows.  So I'm surprised that combo works for you and others don't.  What ones don't and what board are we talking about??

i play fps's, but i always end up remapping keys.  for starters, i use esdf rather than wasd, so i wouldn't really know what action ctrl-qwazx does.

this is an 'expo486' branded keyboard, fcc id ifs80-2269-1.  as far as i can tell, it's 2-key rollover.  i.e., it cannot be guaranteed that any and every combination of 3 or more simultaneously depressed keys will register correctly.

For my $10 Logitech Media Keyboard (BN-52) it blocks any keys after CTRL-QW.  3-key.

one successful key combination is insufficient to prove rollover, but one unsuccessful key combination is sufficient to disprove rollover.  for it to be truly 3-key rollover, it needs to register any and all combinations of 3 simultaneously depressed keys.  but it's a little easier to prove the keyboard is not 3-key rollover by looking for a combination of 3 keys that fails.  if you can find a combination of 3 keys that won't register correctly on your logitech, then it's not 3-key rollover.

for example, this expo486 keyboard i'm using passes ctrl-qwazx, but fails on wefkop.  and fails on many other 6 key combinations as well, but a single failure is sufficient to prove that it's not 6-key rollover.  it's not even 5 or 4 or 3 key rollover because i've found combinations of 5, 4, and 3 keys that the keyboard will not register correctly.  but i haven't found any combination of 2 keys that will not work.  so it is a 2-key rollover keyboard in all probability.

[itlnstln]

It's like arguing the need for 12 key rollover when you have 10 fingers.

QFT. I never understood the mystique of NKRO when this is exactly the case. In gaming, where many people express the desire for NKRO, you generally have one hand on the mouse, so even the above statment doesn't apply. In that case, you only have five available fingers to use.

[itlnstln]

If Dilbert worked in a keyboard company.
 
Boss: We need N-Key Rollover in our next board! Everybody charges over $100 and puts a Hackers Special! label on it.
 
Dilbert: It costs $2.49 in parts to add true N-key.
 
Boss: Forget the true part, make it for 29 cents in parts.
 
Dilbert: Here it is, doesn't quite work for the following 2598 combinations.
 
Boss: Close enough. Marketing already put it on the box. Don't forget the pink WASD keys.

This sounds like it's pretty much it.  Until recently, I don't think too many people really investigated if NKRO claims were 100% true.  I think most people just took it at face value.

[alpslover]

QFT. I never understood the mystique of NKRO when this is exactly the case. In gaming, where many people express the desire for NKRO, you generally have one hand on the mouse, so even the above statment doesn't apply. In that case, you only have five available fingers to use.

that's true, but having n-key rollover is a guarantee that the keyboard will never fail on any combination of keys no matter how few or many fingers you have.

so while you may only realistically be able to depress a maximum of 5-10 keys at a time, your non n-key rollover keyboard may work fine for the key combinations you use now in the games you play now, but how do you know for sure that different key combinations you might need to use in games you play in the future will work?  n-key rollover guarantees they will no matter what the combination of keys is.

and frankly, it just doesn't cost a whole lot more money to implement n-key rollover.  diodes are not expensive.  it might matter for a $10 keyboard, but i think any keyboard that costs more than $75 should have n-key rollover.

[Rajagra]

QFT. I never understood the mystique of NKRO when this is exactly the case. In gaming, where many people express the desire for NKRO, you generally have one hand on the mouse, so even the above statment doesn't apply. In that case, you only have five available fingers to use.

I suspect that to guarantee ANY combination of 6 keys work, you need to make it NKRO capable, i.e. diodes in series with all keys. Of course you can still mess it up at the controller stage, which would be a shame after spending the money on the parts. Sadly I think many keyboard makers do exactly that.

The problem is laziness as much as money-grubbing. And if even people on this forum say NKRO is unnecessary, why should manufacturers bother to pull their finger out?

[talis]

I suspect that to guarantee ANY combination of 6 keys work, you need to make it NKRO capable, i.e. diodes in series with all keys. Of course you can still mess it up at the controller stage, which would be a shame after spending the money on the parts. Sadly I think many keyboard makers do exactly that.

I suspect you're right, either you have full n-key, or there exists a set of 3 keys that will cause an issue.

The problem with most keyboards isn't that the manufacturer is lazy, its the fact that you can't easily solder to a membrane, and therefore can't install diodes on every switch.  Without diodes, there's no easy way to achieve N-Key.  There are some techniques for doing this (flexible PCB type tech) but using that on such a large scale would add quite a bit to the price.

If the board is PCB based, then it comes down to price.  Adding diodes to every key probably adds $5-10 to the sticker price of a product ($1-2 to the production cost), at that point it's a question as to if the end consumer will pay 10-15% more for a keyboard for something they may not even know about (or if the group of people that understand the problem is large enough).

[talis]

Better than the current situations.  Paying 50% more for something that doesn't do what the box says it should do.

Exhibit A:  The ABS M1

I agree, just saying that's the choices that always come up in the arguments between Engineering and Marketing/Sales

[keyb_gr]

Thinkpad X60t internal keyboard (ISO layout): Accepts QWER, SDFJKL, fails QWAS and SDFJKL+Space. Close to a traditional IBM layout it seems, but possibly they had to do it in 16x8 with Win keys.

[1cewolf]

Microsoft Comfort Curve 2000
Interface: USB
Operating System: Windows 7
Max keys accepted: 6 (tested QWAS, QWEASD, IOPL:", SDFJKL, SHIFT+PL:"{)

Not bad at all for a $20 cheapie!

[Mr.6502]

I have an HP Compaq 6710 laptop.  The minimum you can press at the same time is 2.    So by the wiki definition that would mean its 3 key rollover?  

The controller for the keyboard seems to freak out though if you try to press any 3 nearby keys.  If you press 3 nearby keys at the sametime, instead of it blocking one, it blocks all of them and nothing comes out.

I'm not sure how common that is.  My MS Ergo keyboard would sometimes lose a keyup if I pressed and then released 3 buttons at the same time, resulting in one button staying held, but this is the first keyboard I've used where pressing just about any combination of 3 nearby letter or number keys at the same time resulted in nothing coming out.

[Mr.6502]

Yep, 3-key by the wiki definition.  It seems most keyboards are.

Thank MS and HP for saving a few pennies on diodes.

Diodes and decent mechanisms for the keys.  This laptop has one of those keyboards where you can't really press the spacebar on its ends, only in the middle.  And almost all the keys require being pounded down to ensure they register.  

The last keyboard I used that was this tiring was the Atari XE -_-

[keyb_gr]

The controller for the keyboard seems to freak out though if you try to press any 3 nearby keys.  If you press 3 nearby keys at the sametime, instead of it blocking one, it blocks all of them and nothing comes out.

Sounds like the rollover bug on the ABS M1. I'd guess it's ghosting suppression gone wrong - doesn't exactly speak for the keyboard controller's programming. Not like that would be surprising with the other properties you describe.

Strange to find this in what seems to be designed as an office workhorse. Besides, reviews indicate that the keyboard should actually be pretty good. A lemon, maybe? I'd bother the support.

[itlnstln]

As far as office use goes, I have never had any toruble with my ABSs.  I just don't really think that you press that many adjacent keys to ever see the difference.

[Mr.6502]

Sounds like the rollover bug on the ABS M1. I'd guess it's ghosting suppression gone wrong - doesn't exactly speak for the keyboard controller's programming. Not like that would be surprising with the other properties you describe.

Strange to find this in what seems to be designed as an office workhorse. Besides, reviews indicate that the keyboard should actually be pretty good. A lemon, maybe? I'd bother the support.

I don't think its a lemon.  Unfortunately we have 15 of them and they are all identical.  I guess it could have been a bad run of them or something.  One of them has a broken key so I'll be getting a replacement keyboard for it under warranty.  If its any better I may be ordering 14 more under warranty :-D

[microsoft windows]

I tried the N key rollover test on my M5-2 and it didn't accept any keys at all!

Keyboard type: Blue IBM logo, Lexmark Model M5-2 Trackball Keyboard, PS/2 Mini-DIN interface, made on 5 July 1994
Computer info:
OS: Windows 2000  SP4 all  latest updates
Model: Micron Clientpro CN from 2001 (looks like a big yellow box with a monitor on top)

[1cewolf]

Not MAX keys please. ONLY post minimum keys registered.  Try the ones in the wiki or in this post.

My bad. My Comfort Curve 2000 is a 3-key, then. It only seems to have trouble when you press down the center keys of the middle row, though.

The original Microsoft Natural I've got is a 3-key as well. It's such a crying shame considering that its fit and finish is light years beyond any of Microsoft's current models.

[alpslover]

i don't normally read the wiki, so i've only noticed this just now, but the wiki's definition of rollover appears to be incongruent with itself.  if rollover is 'the state that happens when one or more keys are being blocked', then by that definition, n-key rollover is a keyboard that blocks on n keys, n being an arbitrary number.  which is just another way of saying that the keyboard blocks on any key combination, and we know that that is not what an n-key rollover keyboard is.

in addition, the wiki mentions 'full n-key rollover/n-key rollover/no rollover'.  aren't 'no rollover' and 'full rollover' contradictory?  i've never seen any n-key rollover capable keyboard being described as 'no rollover', but i have seen them described as 'full rollover'.  which seems to indicate that 'rollover' is not the state of keys being blocked, but rather the opposite.

to me, a 3-key rollover keyboard is one that will not block on any combination of up to and including 3 keys (whereas by the wiki definition it would be 2 keys).  so an n-key rollover keyboard will not block on any combination of up to and including n keys, n being an arbitrary number, meaning it will not block on any key combination whatsoever.  this definition is logically consistent with itself as well as what we know an n-key rollover keyboard to be.

[keyb_gr]

I don't think its a lemon.  Unfortunately we have 15 of them and they are all identical.  I guess it could have been a bad run of them or something.  One of them has a broken key so I'll be getting a replacement keyboard for it under warranty.  If its any better I may be ordering 14 more under warranty :-D

It could be they are sourcing keyboards from different manufacturers. For the Thinkpad T60 series, apparently there are some made by NMB and others in China. BTW, the rollover hehavior might be according to US patent 4420744. Do any of the missing keys appear when one of them is released?

i don't normally read the wiki, so i've only noticed this just now, but the wiki's definition of rollover appears to be incongruent with itself.

This whole definition topic is a bit of a pain in the rear really, as the web is lacking good references.

I tried it with Google Books now, hoping to find something in dead tree format. Alas, it seems like the definition was anything but clear even in the olden days.
This book from 1980 says

Any technique that can correctly recognize a new key even though n - 1 keys are
already pressed and are still down is said to exhibit n-key roll-over. ...

Here it says

Two-key rollover will generate two keystrokes accurately if the second key is depressed before the first key is released.

This book claims:

There are two approaches to solving this problem — 2-key rollover and N-key rollover. In 2-key rollover, once a key is pressed, any other pressed key is not ...

It seems like "rollover" was also meaning the preservation of correct order, as shown here.
A relative of rollover is lock-out, i.e. 2-key lock-out means that once a key a pressed all the others are ignored.

OK, at least I found that the term 'N-key rollover' was coined by Maxi-Switch.

[Mr.6502]

My bad. My Comfort Curve 2000 is a 3-key, then. It only seems to have trouble when you press down the center keys of the middle row, though.

The original Microsoft Natural I've got is a 3-key as well. It's such a crying shame considering that its fit and finish is light years beyond any of Microsoft's current models.

Indeed.  I have 2 MS Natural Elites and love them.  They are the perfect keyboards for my hands and were both purchased used years ago and seem to be doing alright still (one of them started to occasionally drop a button press here or there but I still keep it around just in case).  But the 3-Key RO makes it tough to play games on them.  Those with N-key would be my personal ultimate gaming keyboard.

It could be they are sourcing keyboards from different manufacturers. For the Thinkpad T60 series, apparently there are some made by NMB and others in China. BTW, the rollover hehavior might be according to US patent 4420744. Do any of the missing keys appear when one of them is released?

With some key combinations missing keys appear when one or two of them are released.  Sometimes nothing appears, sometimes 2 or 3 appear.  There doesn't seem to be a case when more than 3 keys will appear when releasing buttons.  

I actually just tried my MS Natural elite via PS/2->USB adapter on this comp as well and it appears on this comp it drops ever 5th or 6th button press received through the adapter.  Its not something that happens on other comps so it would appear this laptop's keyboard is simply cursed.  Or maybe it has a Gremlin.

[IBI]

This whole definition topic is a bit of a pain in the rear really, as the web is lacking good references.

I tried it with Google Books now, hoping to find something in dead tree format. Alas, it seems like the definition was anything but clear even in the olden days.

Good research.

If you know the definitions on the wiki are wrong then go ahead and edit them, they were only my best guess at the definition from the patchy information availible on the web.

Personally I've kind of given up on the whole rollover terminology and feel that the best way to solve the problem would be to come up with a new and better terminology.

Ideally we want a simple diagram showing the matrix that makes it obvious which key combinations will block as most of the time we're interested in which specific keys can work together rather than the minimum number, but as an aid to those diagrams we should come up with a term that denotes the minimum number of keys you need to press down in any combination before one of those keys doesn't register so manufacturers who do implement higher than normal rollover numbers can easily advertise that.

The first one is by far the most important but a clear and immediately obvious term for the second should also be helpful.

Oh, and we also want some way to generate the diagram without having to open up the keyboard so if anyone could come up with the set of checks needed to determine it then I'm sure I, or someone else here, could make up a simple program to guide a user through those checks and produce the diagram.

[talis]

As far as I can tell, theoretically its safe to say :

- For every keyboard that doesn't have a hardware solution to ghosting (diodes, capacitive switches, etc) , there exists a combination of keys that can be pressed at the same time, up to the number of rows in the matrix, that won't cause ghosting.  As soon as a second key is pressed in the same row as one already being depressed ghosting may occur (depending on the state of the rest of the switches.

- For every keyboard that doesn't have a hardware solution to ghosting, there exists one or more combination of 3 keys that, when pressed together will cause ghosting.

- With a hardware solution, any number of keys can be pressed simultaneously, and correctly decoded by the controller.

- Its possible for the hardware designer to limit the above to ignore any key presses past x.

- The USB interface limits the number of keys that can be transmitted to the host to 6, and therefore any (standard, driverless) USB keyboard is limited to at most 6 simultaneous keys, regardless of the number of rows in the matrix, or any hardware solution to ghosting (however with a solution in place, you are always guaranteed 6 keys).

[IBI]

- For every keyboard that doesn't have a hardware solution to ghosting (diodes, capacitive switches, etc) , there exists a combination of keys that can be pressed at the same time, up to the number of rows in the matrix, that won't cause ghosting.  As soon as a second key is pressed in the same row as one already being depressed ghosting may occur (depending on the state of the rest of the switches.

'Hardware solution to ghosting' is meaningless without context, the statement applies to keyboards that use the matrix method of detecting keys in what's currently it's most common form.

The maximum number of keys is going to be equal to or less than the number of rows  or columns, whichever is smaller. I don't know if it's guaranteed to be equal to that though.
 

- For every keyboard that doesn't have a hardware solution to ghosting, there exists one or more combination of 3 keys that, when pressed together will cause ghosting.

This appears to be true for the most common method of matrix key registration.

- With a hardware solution, any number of keys can be pressed simultaneously, and correctly decoded by the controller.

This one is just meaningless stated as is.

- Its possible for the hardware designer to limit the above to ignore any key presses past x.

True, but obvious.

- The USB interface limits the number of keys that can be transmitted to the host to 6, and therefore any (standard, driverless) USB keyboard is limited to at most 6 simultaneous keys, regardless of the number of rows in the matrix, or any hardware solution to ghosting (however with a solution in place, you are always guaranteed 6 keys).

No, the USB specification allows six keys plus modifiers per packet so assuming your keyboard has the standard modifiers you can send ten or so keys.

From what I've gathered the issue is that USB only sends key down events and assumes that if a key isn't reported in the next packet that it's been released. I could be wrong though, but if you want to make definitive statements about the USB spec I'd read it first, I believe it's freely availible.

[talis]

'Hardware solution to ghosting' is meaningless without context, the statement applies to keyboards that use the matrix method of detecting keys in what's currently it's most common form.

The maximum number of keys is going to be equal to or less than the number of rows  or columns, whichever is smaller. I don't know if it's guaranteed to be equal to that though.

A necessary condition for ghosting is to have two keys in the same row pressed (that's what sets up the loop in the circuit), so its possible to find a combination of key presses up to the number of rows that will not ghost.  Unless I'm missing something, you can have any number of keys in the same column pressed without seeing a ghosting effect (provided there are no other keys pressed in the same row).

What I mean by hardware solution is either something like diodes, non-closure type switches, or something similar.  Its not a problem that can be solved in software, and still use a matrix layout.

With a hardware solution, any number of keys can be pressed simultaneously, and correctly decoded by the controller.

This one is just meaningless stated as is.

I just mean that if you correctly implement hardware anti-ghosting, that there is no limit on how many keys the controller can detect (obviously within the bounds of the matrix).  At this point the matrix design, and row/column layout are no longer set the bound on the number of keys the computer will ultimately see pressed.

The idea I was trying to get across is that with a standard matrix, with nothing in place to prevent ghosting, you cannot get more then 3-KRO, if a keyboard is seen to exhibit higher, its due to the fact that the keys being pressed are not in the same row (due to matrix design).   Ultimately the keyboard is still only 3-KRO.

If there is some form of anti-ghosting implemented in hardware, you will get up to n-key rollover bound by either the interface method, or the controller software design (or the number of keys in the matrix).

No, the USB specification allows six keys plus modifiers per packet so assuming your keyboard has the standard modifiers you can send ten or so keys.

From what I've gathered the issue is that USB only sends key down events and assumes that if a key isn't reported in the next packet that it's been released. I could be wrong though, but if you want to make definitive statements about the USB spec I'd read it first, I believe it's freely availible.

You are correct, my understanding was that low speed, interrupt transfer mode had a maximum of 8 bytes per packet, with 2 reserve bytes.   Further reading shows that one of the two reserve bytes does indeed carry the modifier keys, so USB caps at 6 non-modifier keys + up to 8 modifiers.  It still stands that in a USB keyboard with some form of hardware anti-ghosting, the cap on number of key presses is still bound by the transfer protocol.  Based on the definition of x-key rollover, USB still limits the keyboard to being 7-KRO because there exists a set of 7 keys that when pressed, will only show as 6.

(sorry, a few edits there).

[Rajagra]

For every keyboard that doesn't have a hardware solution to ghosting, there exists one or more combination of 3 keys that, when pressed together will cause ghosting.

This is an important point. It implies there are only three main keyboard types:

• Keyboards that make no effort to suppress ghosts and just report 4 keys being pressed instead of 3.
• Keyboards that suppress ghost keys AND the third key which created the ghost condition.
• Keyboards that can correctly handle any number of simultaneous keypresses. (USB limit aside.)

I've never used a keyboard as bad as the first type. The second type will have many variants, depending how intelligently the matrix is designed, and how cleverly they cope as keys are released. But at heart the bottom line is that "three keys suffice" to confuse them.

Following the ghost analogy, maybe the three types should be called Evil, Haunted, and Holy.

[IBI]

Can we not use the term ghosting please? It'd be better to reseve that just for the situation where you get keypresses registering for keys you didn't press. I generally to use blocking to describe this phenomina, but if you can think of a more appropriate term then we can use that instead.

A necessary condition for ghosting is to have two keys in the same row pressed (that's what sets up the loop in the circuit), so its possible to find a combination of key presses up to the number of rows that will not ghost.  Unless I'm missing something, you can have any number of keys in the same column pressed without seeing a ghosting effect (provided there are no other keys pressed in the same row).

I thought the condition required to cause ghosting was that a key had at least one key in the same row and at least one key in the same column pressed? I don't think there's anything special about the rows compared to the columns.

What I mean by hardware solution is either something like diodes, non-closure type switches, or something similar.  Its not a problem that can be solved in software, and still use a matrix layout.

It does seem correct that it's a hardware problem with the most common implementation of the matrix detection method.

I just mean that if you correctly implement hardware anti-ghosting, that there is no limit on how many keys the controller can detect (obviously within the bounds of the matrix).  At this point the matrix design, and row/column layout are no longer set the bound on the number of keys the computer will ultimately see pressed.

Yes, but any solution fixes the problem by definition. I think you're trying to include the point that it is possible to register all keys at once using some methods and there are no fundamental limitations.

[Mr.6502]

Would ghosting be the term for when a keyboard does not detect a keyup after a button is pressed?  So that it thinks the button is pressed until you press and release it again to reset it?

[Rajagra]

Would ghosting be the term for when a keyboard does not detect a keyup after a button is pressed?  So that it thinks the button is pressed until you press and release it again to reset it?

Imagine keys are laid on a grid with the grid lines being electrical paths, with signal levels being set/read by the keyboard controller chip.

Consider 4 keys that lie on the intersections of 2 rows and 2 columns.
If the keys are simple on/off electrical switches then:
If 1 key is pressed it can be detected easily.
If any 2 keys are pressed they can still be detected easily.
If 3 keys are pressed, there is a problem. The 2 rows and 2 columns are all shorted together. It is impossible for the controller to tell if 3 or 4 keys are pressed. Even if the controller assumes only 3 keys are being held, it can't work out which 3 are most likely.

The 3 keys being held down are real. The fourth one that the controller sees as being held down is the ghost key.

Ghosting is the phantom appearance of the fourth key press.

I was going to define "anti-ghosting" and "blocking" next, but each of those can be interpreted in different ways. So I'll quit while I'm ahead.

Above diagram taken from here: http://www.national.com/an/AN/AN-734.pdf

[Mr.6502]

I understood that already but I was mistakenly thinking there was a way to use ghosting to get the keyboard to miss a keyup.  My logic was off.

Keys staying pressed after being released is probably a result of anti-ghosting logic in the board or something.

[Jon8RFC]

I found this thread while trying to find some subjective tests on the Logitech G19.  For a quick background, I play fast-paced movement first person shooters and use the directional keys and surrounding keys contrary to most gamers, who use the WASD or ESDF configuration.  Once I began playing a game which required me to use up to five keys simultaneously (with one hand), I found that all keyboards were not created equal.  Fortunately, my Logitech Elite USB has served me well for many years and I've been unable to upgrade to a backlit keyboard which types well and will allow me to connect via USB and allow my odd key depressions.  Oddly, I've had better luck with cheap (read: less than $10) keyboards playing nicely and not one single "gaming" keyboard will accept my key combination variations using the directional keys...and it narrows the gap futher when I need backlighting and typing ease.

EDIT: oh, another requirement for a new keyboard is having the directional keys lined up with the rows of adjacent keys, and many newer keyboards have them lowered slightly...so it's been difficult, to say the least, to find an appropriate keyboard replacement/upgrade.  I almost purchased an extra one (and Wal-Mart had them on sale for just $25!) when a can of soda exploded on my first and had to buy a replacement, but I figured "nah, there will be a newer keyboard I'll buy if something happens to this one", but there aren't any that meet my needs!  The G15 version2 fails my directional keystoke tests, but I cannot recall if the G15/G11 version1 (blue backlight; folding LCD screen) failed my directional keystoke tests...both pass WASD/ESDF easily, but directional keys are where it's important to me.  I'm waiting for local stores to carry a display model of the G19 that I can plug my thumb drive into and use KeyScan on.

SO...here I am.  I noticed that y'all use that link for testing the rollover, but I've been using this neat little program for well over a year in my hunt for an acceptable keyboard, since it's easier & faster to fire up a dinky program to test the keystrokes rather than fire up my games (in-store, mind you) to see if the keyboard works properly.

I think y'all will find this a very helpful program with great feedback on keystrokes.  I can mash a bunch of keys at once, or progress my way up one at a time to find when it faults.  It's been a wonderful tool for me to test potential purchases and I hope it is of use for y'all as well =]

It's called "KeyScan" v0.9 by Digital Genesis.  Unfortunately, the download link is no longer working at http://www.digitalgenesis.com , but you're welcome to grab it from me if you like: http://jon8rfc.homeip.net/other/keyscan.exe
Send it over to virustotal.com ease your worries about using a downloaded program--I won't be offended =]  (if you haven't already heard of that site, there's another handy tip!)

For the above download, as a reference of character:  I've been active in the unofficial AIM tech-support community since about 2001.  If you ever removed the ads from your buddylist, modified the look of AIM, or even used one of the old add-ons--like AIM+, deadaim, middle_man, smilez, vario, messenger:mate, etc, or visited bigblueball.com--I was probably involved in one way or another, helping find solutions to your problems.

EDIT2: oh, and a great keyboard for you typists would be the Majestouch Tactile Touch, which I read a review of in my CPU magazine, which prompted me to do a search on "n-key rollover", since I finally found out a proper name for the annoying phenomenon I experience and was unable to describe to tech support when asking questions about keyboards I was interested in purchasing.  http://www.computerpoweruser.com/editorial/article.asp?article=articles/archive/c0909/10b09/10b09.asp&guid=

[keyb_gr]

It's called "KeyScan" v0.9 by Digital Genesis.  Unfortunately, the download link is no longer working at http://www.digitalgenesis.com

Actually, for me it worked.
Doesn't catch the extra key on ISO layouts though, and the detection seems to be character based rather than scancode based, so in the end it doesn't do very well with non-US layouts.

Oddly, I've had better luck with cheap (read: less than $10) keyboards playing nicely and not one single "gaming" keyboard will accept my key combination variations using the directional keys...

Interesting point. I think you're not the first one to notice that these cheap boards tend to be doing better than their fancier cousins. I'd guess that these may be using a standard IBM matrix layout or at least something close, and that isn't too badly designed. You can check the wiki for blocking patterns.

[Rajagra]

I cannot recall if the G15/G11 version1 (blue backlight; folding LCD screen) failed my directional keystoke tests...both pass WASD/ESDF easily, but directional keys are where it's important to me.

On my G15 V1, holding down the right Shift and/or Ctrl keys stops the inverted T direction keys from working well. E.g. Shift-Up-Right fails, Ctrl-Down-Left fails. Ironically, if you use left Shift and Ctrl, any combination of the four arrow keys works. Isn't that a kick in the teeth?!

[keyb_gr]

On my G15 V1, holding down the right Shift and/or Ctrl keys stops the inverted T direction keys from working well. E.g. Shift-Up-Right fails, Ctrl-Down-Left fails. Ironically, if you use left Shift and Ctrl, any combination of the four arrow keys works. Isn't that a kick in the teeth?!

I guess that happens when you try to improve rollover capability in one spot (cursor keys) without being able to enlarge the matrix. With a regular 2KRO board, every matrix layout is a compromise. Whether you've found a good one is best confirmed by field testing.

You know what would be useful? An algorithm that takes a blocking pattern and spits out a matching key matrix. Obviously you'll only catch a single column/row permutation but that usually does the job (otherwise you still have sort of a Rubik's Cube problem remaining). Basically one would have to start with two keys and then note the ones that are blocked. Then replace one of the two keys with a new one and check again. Might become somewhat tedious, but still better than opening up the board and tracing the matrix by hand.

[Rajagra]

You know what would be useful? An algorithm that takes a blocking pattern and spits out a matching key matrix. Obviously you'll only catch a single column/row permutation but that usually does the job (otherwise you still have sort of a Rubik's Cube problem remaining). Basically one would have to start with two keys and then note the ones that are blocked. Then replace one of the two keys with a new one and check again. Might become somewhat tedious, but still better than opening up the board and tracing the matrix by hand.

I was thinking the same thing, but doing it manually. If you start with 3 keys that don't all register you know they are on 3 corners of a rectangle in the grid. Release a key and find another that plays up, you've found another piece of the puzzle. Unfortunately it's not as trivial as it first seems, of the first 3 you don't know which one lies on the "join." I'm sure an algorithm is possible though.

Edit> Here's how to identify which of the 3 is on the corner: The 2 "diagonal" keys do not share any row or column. So they can only create (3-key) problems when pressed with the 2 keys that join them.

So test to see which 2 of the 3 original keys have that property. You now know the remaining key is the corner one, and have started solving the puzzle.

I've made that sound more complicated than it really is.

[IBI]

So test to see which 2 of the 3 original keys have that property. You now know the remaining key is the corner one, and have started solving the puzzle.

If two keys have no rows and columns in common then they cannot block with more than two other keys.

You may only be able to find one key and not two if you're at the end of a row/column.

If two keys block with more than two other keys then they must be in the same row or column.

Is there any difference between rows and columns when it comes to working out the matrix?

[keyb_gr]

Is there any difference between rows and columns when it comes to working out the matrix?

Don't think so.

An example on the Model M: SDX.
SX: Blocks >2 keys
SD: Blocks >2 keys
DX: Blocks 2 keys only

So D and X must be the opposing corners, and S shares a row with one and a column with the other. Which in fact is the case in this matrix.

Now which one should we replace by a new key in order not to mix up columns and rows (and if so, what should the new key be)?

EDIT: I think first we should complete the 2x2 matrix section by finding the 4th key, which is easy because it is one of the two blocked by those on opposing corners, more precisely the one that hadn't been in use so far. Still, how do we continue then?

What we could do: Of the known 2x2 matrix section, take two keys on the same column (or row). Then find another that is blocked by them, and repeat the "who's the corner" game. This should tell which row the new key belongs to. Then find the 4th key. Rinse, repeat.

When we can no longer find any new keys that are blocked, we must have exhausted two complete rows. Then we must go one row up (just one so we have some information already) and continue there.

This is really begging for automation, but then how many key pressing robots does one typically have floating around...

[keyb_gr]

BTW, now I'm pretty much convinced that a generous 3-key rollover (as I saw on my G80-1000HAD) is what you get when using diode-equipped switches with a well-implemented regular 2KRO microcontroller that checks for "4 in 2x2".

[Jon8RFC]

Late response, sorry.  Yes, the alternate n-key link y'all use for rollover testing is great.  The original link I saw referenced wasn't appropriate because it posts repeat keys without you being able to choose.

I'm a bit drunk right now, but I think I can still express my n-key frustration.  I'm using my buddy's G15 v1 and it does not pass my necessary arrow key combination, but it's better than the G15 fv2 from what I can recall from my testing.

So, officially, the G15 v1 does not pas n-key combination using arrow keys.  For some reason I remember that the G15v1/G11 passed, but it does not.  I have yet to test the G19.  I'll let y'all know when I test it...I would love a backlit keyboard for Quakecon 2009, but I'm not sure if I'll have one by then.


Great forum, by the way.  I love the coherent posts y'all have and it's one of the few niches I've found on the internet.  If you play Urban Terror, check evogc.com because the majority of the members are professionals (college degrees) over 25 years of age, are absolutely hilarious and have great senses of humor, and are great to hang out with over VOIP.

Thanks again for the input I've gotten from this specific thread.  If I can remember (based on the state I'm currently in), I'll hop back in before the 10th if I'm able to test a G19 and see how it accepts my inputs with the arrow keys.

[keyb_gr]

This, btw, is the NKRO test thread.

But true, one would think that there should be some results for these gaming boards to be found.

[keyb_gr]

They must have some crazy matrix routing in that Chicony KB-5312.
While the QWAS test shows that it is a 2-key rollover (of course - standard rubber dome), within one row you can press a lot of keys at once.
It accepts QWERTZUIOP, ASDFGHJKLÖ and YXCVBNM,.-, no kidding. Maybe even 1 more on the first two, but I ran out of fingers. If the 102nd key comes into play, it stops after
I think I know the trick: When holding down ASDF, a large part of the numpad is blocked, with only divide and enter still working. A Model M has less rollover capability in the alphanumeric part but does not exhibit numpad blocking in return.

Now I'm wondering whether I should really toss that one.

[LEXX911]

I'm maxing out 9 here with the Steelseries 7G.

Edit: Managed to do 31 with two palms here with the Steelseries 7G.

[keyb_gr]

I'm maxing out 9 here with the Steelseries 7G.

Given that this one has NKRO, it shouldn't choke on any combination though. The maximum number of keys will be limited by the USB HID protocol only.