Overclock.net › Forums › Components › Mice › Discussion of targeting technologies
New Posts  All Forums:Forum Nav:

Discussion of targeting technologies

post #1 of 130
Thread Starter 
Discussion of targeting technologies

What is a good mouse?
What is tracking?
Which sensor is good?
Which is difference between sensors?

Some people choose mouse only by amount CPI, thinking that higher means better. Some people are sure, that only max tracking speed matters. Some people think, that sensors must be only LED based. Some people say, that sensor doesn’t matter at all, shape is the only priority. Some even strongly believe than gamepads are better than mice.

Let’s look at this from different perspective and try to investigate, how mouse sensors work.

First of all, accuracy theoretically can’t be higher than 100%. Sensor can’t reproduce hand movement “better” than it was. All it can do is try to track it. Technology can’t have strengths, only flaws and limitations. So let’s compare flaws and limitations to choose a technology which has less of them. Let’s look, which one which gives best accuracy with least limitations.

Ball mice Spoiler (Click to show)

Ball mouse use ball, two perpendicular chopper wheels and light sensors to detect movement. When ball rolls, chopper wheels roll as well, opening and closing way to beams of light. By this sensors measure amount of movement. When there is enough cohesion between ball and wheels, sensor detects movements perfectly accurate. Resolution may be changed by changing chopper wheels.


Limitations of this type of sensors are related with mechanics: friction and inertia. If there is not enough cohesion, ball and chopper wheels may slip. With movements not along one axis, cohesion may be lower, ball may slip, so X or/and Y sensitivity may be lower. If movement speed is too high, ball may jump, and movement won’t be tracked.

Tracking surface should be stable/hard and should give enough friction for ball to roll. Mouse and tracking surface should always be clean to have stable friction.

If these problems could be eliminated or reduced, we could have a simple and accurate mouse.
ICS (Avago/Pixart Image correlation sensors) Spoiler (Click to show)

MLT04, A3090, S3988, AM010, PMW3310, PMW3366, A9500, A9800 and most others modern mouse sensors are image correlation sensors (ICS).

ICS include source of light (LED or laser) and array of photodiodes. Source of light illumines tracking surface, light reflected from surface goes to photodiodes.


Photodiodes make small photos of tracking surface and compare them. Framerate is usually variable, 1 000 - 12 000 FPS. Photodiode array usually is 30x30 pixels, which means photos as well are 30x30 pixels. First ICS had photodiode array of 19x19 pixels (MLT04), largest photodiode array is 36x36 pixels (PMW3366).



This is a photo of A3090 sensor photodiode array.




Judging by photo of A3090 photodiode array, photodiodes cover about 5% of surface. On this picture photodiodes are highlighted with green, spacing with red.


Knowing it’s size (about 1.5 mm) you can count it’s optical/native resolution. Optical resolution of modern ICS is about 400-800 DPI. More detailed calculations (higher CPI) can only be achieved by guess-work (interpolation/upscaling). This is an example of upscaling.




With each frame ICS receives frequency response characteristic of brightness of light, projected from tracking surface area to photodiode array. Then ICS correlates frequency response characteristic of frames. Basically, we are dealing with frequency analysis.

Photodiode measures average level of brightness from all light, projected from tracking surface area to photodiode surface. One photodiode signal is equal to one sample per frame. This is an example of which part of information sensor can see per frame.

If part of light projected from tracking surface is not in focus, light will be scattered not linearly, but depending on optics.

It will have negative effect on accuracy of sample. By removing focus, projected image will change in unpredictable way. In this case, received information will look like this.

This is difference.

If image is not in focus, part of information will be smoothed, but high frequency details influence won’t be eliminated. High frequency details will have different influence depending on position.

Photo diodes can’t receive all information about surface. As each photodiode reports one average result per sample, we can’t know light allocation inside photodiode surface and light allocation between photodiodes. If formula for the allocation is inharmonic, we can’t accurately re-create surface information. In this case we don’t know what was inside photodiodes and between photodiodes, and we can’t re-create this information. It results misinterpretations like spatial aliasing, Moiré pattern and high frequency noise.
CREATOR: gd-jpeg v1.0 (using IJG JPEG v62), quality = 90
Which results noise, unpredictability and inaccuracy of result.

Aforesaid means that:
- if frequency response characteristic of surface is inharmonic, it can’t be measured accurately
- if harmonic frequency response characteristic of surface has higher frequency, than 2 pixel steps, it can’t be measured accurately
- if harmonic frequency response characteristic of surface has lower frequency, than pixel array, it can’t be measured accurately



Also, photodiodes form vertical and horizontal lines. Basically, it means that ICS not only have given optical resolution, but even optical direction. At different angles pixel density is different. Different directions have different sample rate.

Angle – Samples surface on chosen direction
0.00° - 22.45%
45.00° - 12.64%
26.57° - 7.96%
18.43° - 5.81%
33.69° - 4.68%
14.04° - 4.59%
11.31° - 3.80%
36.87° - 3.31%
21.8° - 3.28%
30.96° - 2.88%
38.66° - 2.56%
...
5.71° - 2.05%
...
1.91° - 0.72%



At higher speed, same distance on surface will be represented with lesser amount of frames. This will result lesser signal/noise ratio, more errors and worse tracking quality. In other words, higher movement speed result lesser accuracy. It results unstable tracking and speed errors at higher speeds.



If tracking surface has difference in height, in other words, is not completely flat, not every level of height may be in focus, which will result lower tracking accuracy.

Also, every level of height will have different projection on photodiode array, which will lead to distortion between frames and lower tracking accuracy. This means that tracking surface for ICS should be as flat as possible.



LED based sensors look at surface strictly from above, and laser based sensors look at surface at different angle. Optical resolution of LED based sensors depend on distance to tracking surface. With moving closer to surface (for example: pressing mouse button or increasing pressure on cloth mouse pad) optical resolution will increase, which will increase mouse sensitivity for about 5-10%.


Laser based ICS with such movement will report cursor movement in the direction of laser beam. Also, laser based sensor will track vertical movements.

It means, that to work properly, any ICS should be used on hard surface.



As sensor is optical and it measures texture surface, tracking quality strongly depends on characteristics of tracking surface. First of all, surface must contain information, that sensor could read, so surface needs to have contrast texture. To make it possible to re-create texture by samples, formula for the light allocation must be harmonic, with minimal frequency of 2 native pixels and maximal frequency of whole pixel array. This is an example of such texture.



This texture may be re-create from samples with bi-cubic filtering. Such surface will provide accurate tracking even at high resolution (about 5x hardware resolution without jitter).



Sadly, on any other surface with inharmonic texture and relief tracking accuracy will be very low, much lower than offer any other technology, specified in this text.



Among other things, ICS have specific flaw related with noticeable lowering of accuracy with sensor rotation. It’s related with arrangement of photodiodes in pixel array.
PTE (Philips TwinEye Doppler shift sensors) Spoiler (Click to show)

Doppler shift effect is the change in frequency of a wave for an observer moving relative to its source. So basically, it means that if source of light moves, phase of reflected light will change.


In way similar to sonar, TwinEye emits laser beam to the surface and measures phase of reflected light.

Sensor receives analog information, which is then digitized by built in ADC.


This approach doesn’t have fixed resolution (DPI), information about speed and time may be divided in any amount of CPI. PTE measures fractions of reflected laser beam wave, it means accuracy is only limited by wavelength of laser (857 nm or 1 / 30 000 inch), quality of digitizing and characteristic of tracking surface.

There is no spacing between pixels (laser beam has no spacing in it), which means even smallest movements may be tracked. Errors may be only at level of measurement of phase of reflected laser and digitizing (not at level of optical resolution).

Also, PTE measures speed instead of distance, so it will receive same amount of information about tracking surface not depending on speed. It means PTE tracks slow and fast movements equally accurate.

There are two lasers, one measures movement along X-axis and second along Y-axis. They are completely independent, it eliminates angle errors.


Both lasers are directed to surface at angle of 60 degrees. It means they measure not only horizontal mouse movements, but also vertical. So, when you raise the mouse, or soft tracking surface bends while you press buttons, sensor may report movement. It’s also knows as Z-axis bug.

This design has another weakness. There is only one laser beam which is tracking motion, so if there is anything on its way, sensor won’t be able to measure anything. For example if there is a hair under the sensor, it’s likely that one axis won’t track at all.

Accuracy doesn’t depend on direction and speed. However, official data sheet shows how PTE’s sensitivity depends on movement speed:

Pretty stable at most speeds, but at extremely low speed (0.1 mm/s) sensitivity is lower for some reason. Seems to be ADC or math error. Maybe buffer is too short or there is not enough digits after dot. It's not a problem if you use like 400-800 CPI in Windows, but when you switch to 4 000 CPI, it becomes noticeable.

0.1 mm/s is about:
1.57 count/sec at 400 CPI
3.15 count/sec at 800 CPI
6.30 count/sec at 1 600 CPI
12.60 count/sec at 3 200 CPI

To avoid Z-axis bug surface should be hard. For stable tracking surface should be as flat and even as possible. Also, it should reflect and scatter light, or be matte in other words.

With proper surface and cleanness, this sensor may provide decent accuracy. Slight speed errors are there, though they are predictable, and I guess in theory they may be compensated/improved in next sensor models.
Graphics tablets Spoiler (Click to show)

Graphics tablet differ in principle of operation, but main idea is same. Tablet is divided on zones. Each zone corresponds one pixel on screen. With pen/mouse movement, new location is calculated and cursor on screen moves to corresponding location.

This approach leaves no place for angle errors and speed errors. Resulted cursor location will always be predictable.

It’s the only technology that works properly even on soft surface.


If graphics tablets would be adopted for games, it could give better results than anything else.
Flaws comparison Spoiler (Click to show)
1. Tracking on soft surface.

Ball.
Friction is too low and unstable.

LED based ICS.
1. Tracking quality is worse than on hard surface with proper texture. Soft surface texture is unstable and distorts under pressure.
2. Under vertical pressure surface bends, sensor moves closer to surface, optical resolution and sensitivity increase. So any fast movement with pressure will have (about 5-10%) higher optical resolution/DPI/sensitivity.

Laser based ICS.
1. Tracking quality is worse than on hard surface with proper texture. Soft surface texture is unstable and distorts under pressure.
2. Under vertical pressure, cursor moves along laser direction.

PTE.
1. Tracking quality is worse than on proper hard surface. Soft surface is unstable and distorts under pressure.
2. Under vertical pressure, cursor moves along lasers direction. Amount of cursor displacement may be bigger than in case of laser based ICS.

Graphics tablet.
May be used with cloth mouse pad.

2. Displacement of mouse

Ball.
When mouse lifts off, cursor moves in direction of chopper wheels.
After displacement, cursor moves back roughly to previous location.

ICS.
After displacement, cursor slightly moves, usually in direction of LED light.
Amount of cursor displacement depends on LOD.

Laser based ICS.
When mouse lifts off, cursor moves along laser direction.
After displacement, cursor moves back roughly to previous location.
Amount of cursor displacement depends on LOD.

PTE.
When mouse lifts off, cursor moves along lasers direction.
After displacement, cursor moves back.
In earlier versions of PTE amount of cursor displacement were dependent of lifting and lowering speeds, which might usually result noticeable cursor displacements.
In newer versions of PTE cursor moves back roughly to previous location, amount of cursor displacement mainly depends on LOD.

Graphics tablet.
After mouse/pen displacement, cursor moves to location, corresponding to new mouse/pen location.
Software related problems in first-person shooters.

3. Speed related accuracy

Ball.
Can’t track fast movements.

ICS.
Higher movement speed result lesser is tracking quality.

PTE.
Tracking quality doesn’t depend on movement speed, however, extremely slow movements have less CPI.

Graphics tablet.
Tracking quality doesn’t depend on movement speed.

4. Small movements tracking

Ball.
Accurate tracking.

ICS.
Unpredictable errors compared with size of optical pixels. Depends on movement direction.

PTE.
Predictable, however, extremely slow movements have less CPI.

Graphics tablet.
Accurate tracking.

5. Angle error

Ball.
Predictable. Related with friction and cohesion.

ICS.
High unpredictable errors. Depend on movement direction.

PTE
Result of X and Y errors.

Graphics tablet.
Accurate tracking.


6. High sensitivity

Ball.
May be achieved by changing chopper wheels.

ICS.
Interpolation. More CPI on box usually means more smoothing, angle snapping and other path correcting algorithms. This is needed to hide noise at higher CPI.
Using CPI higher than native DPI not recommended.

PTE
Calculates from speed and time info. CPI may be set to any value.

Graphics tablet.
Depends on type of graphics tablet.
Conclusion Spoiler (Click to show)
Sadly, ball mice don’t receive evolution now, in spite of their high accuracy and low cost.

But the least accurate ICS received most expansion, and most users suffer from their inaccuracy and unpredictability. Majority of modern day users haven’t even seen and tried other technologies.

PTE got bad reputation from incompetent users for so called Z-axis bug. As result, sensor is not that popular, and most of mice with it have path correction directed to reduce vertical tracking on cloth. But these algorithms degrade tracking quality. There are only a few mice with PTE and without such algorithms.

Graphics tablets worth much more than average mice. But at this moment it’s the most promising technology. But it doesn’t get enough evolution because of it’s high price and specific reputation.

---

If you have thoughts, suggestions or critique, please leave them in responses. OP will be corrected if it’s needed. I’d like to keep it as objective and accurate as possible.

Special thanks to @qsxcv and @daniel0731ex for given information and corrections.
Edited by Above8 - 7/15/15 at 12:12pm
post #2 of 130
Very informative thread, thanks.
Quote:
Originally Posted by Above8 View Post

Discussion of targeting technologiesPTE got bad reputation from incompetent users for so called Z-axis bug. As result, sensor is not that popular, and most of mice with it have path correction directed to reduce vertical tracking on cloth. But these algorithms degrade tracking quality. There are only a few mice with PTE and without such algorithms.
Could you name a few? Would the Razer Spectre be one of them?

Also, have you tried using a graphics tablet in an FPS? What happens if you do? I imagine there are still other disadvantages such as polling rate or input lag.

Why do you not consider lower cpi on slow movements with PTE sensors a limiting factor? Wouldn't ICS ideally be better if tracking could be made very stable up to unrealistic movement speeds?
post #3 of 130
Quote:
To work properly, any ICS must be used on hard surface.

Such sensor has no theoretical potential to track hand movement accurately.
these vague statements are your opinion
Quote:
One of the main problems of modern ICS is spacing between photodiodes. Sensor can’t see all the surface.
this is only true if the image projected by the lens is in very good focus, and if there is significant high frequency detail

here are real images of what an ics sensor (3366, led) sees
Quote:
Originally Posted by qsxcv View Post

BBEAgNX.png
slightly shifted:
csw9Yx6.png

Edited by qsxcv - 7/5/15 at 12:24pm
main
(15 items)
 
old
(14 items)
 
 
CPUMotherboardGraphicsRAM
4770k maximus vii impact nvidia gtx 970 crucial ballistix tactical 16gb 
Hard DriveCoolingOSMonitor
crucial mx100 noctua nh-c14 windows 7 ultimate sony cpd-g520 
KeyboardPowerCaseMouse
kbp v80 matias quiet silverstone sx500-lg ncase m1 v3 logitech g100s with mcu replaced by teensy2.0 
Mouse PadAudioAudio
allsop raindrop xl chord mojo hifiman re-600 
CPUMotherboardGraphicsRAM
i7 920 evga x58 sli le galaxy gtx 460 crucial something 3x1gb 
Hard DriveCoolingOSMonitor
intel 330 180gb scythe kotetsu windows 8.1 pro sony cpd-g520 
KeyboardPowerCaseMouse
logitech k120 silverstone st75f-gs nxzt h440 evga torq x5 
Mouse PadAudio
allsop raindrop mobo 
  hide details  
Reply
main
(15 items)
 
old
(14 items)
 
 
CPUMotherboardGraphicsRAM
4770k maximus vii impact nvidia gtx 970 crucial ballistix tactical 16gb 
Hard DriveCoolingOSMonitor
crucial mx100 noctua nh-c14 windows 7 ultimate sony cpd-g520 
KeyboardPowerCaseMouse
kbp v80 matias quiet silverstone sx500-lg ncase m1 v3 logitech g100s with mcu replaced by teensy2.0 
Mouse PadAudioAudio
allsop raindrop xl chord mojo hifiman re-600 
CPUMotherboardGraphicsRAM
i7 920 evga x58 sli le galaxy gtx 460 crucial something 3x1gb 
Hard DriveCoolingOSMonitor
intel 330 180gb scythe kotetsu windows 8.1 pro sony cpd-g520 
KeyboardPowerCaseMouse
logitech k120 silverstone st75f-gs nxzt h440 evga torq x5 
Mouse PadAudio
allsop raindrop mobo 
  hide details  
Reply
post #4 of 130
Cool, another "optical is wholly inaccurate" thread in disguise.
post #5 of 130
Why do you keep posting about this "optical sucks PTE rocks" thing? Do you have some sort of hidden agenda?
post #6 of 130
Quote:
Originally Posted by Crizzl View Post

Why do you keep posting about this "optical sucks PTE rocks" thing? Do you have some sort of hidden agenda?

No he's not, just read the info because he's explaining the whole concept of what a mouse does so instead of you deriding him because he didn't become THE chosen fanboy of your type of sensor.

Geez you clowns think you know how a mouse works, then you start abusing others who step up here to explain it in detail, which some should so that their concepts can be understood by the many that visit this place.

Instead of kicking someone constantly for an opposing view maybe you should try and understand were he is coming from.
post #7 of 130
well it's clear that he's extremely biased against ics sensors.
quite of bit of stuff is vague/inaccurate/misleading
main
(15 items)
 
old
(14 items)
 
 
CPUMotherboardGraphicsRAM
4770k maximus vii impact nvidia gtx 970 crucial ballistix tactical 16gb 
Hard DriveCoolingOSMonitor
crucial mx100 noctua nh-c14 windows 7 ultimate sony cpd-g520 
KeyboardPowerCaseMouse
kbp v80 matias quiet silverstone sx500-lg ncase m1 v3 logitech g100s with mcu replaced by teensy2.0 
Mouse PadAudioAudio
allsop raindrop xl chord mojo hifiman re-600 
CPUMotherboardGraphicsRAM
i7 920 evga x58 sli le galaxy gtx 460 crucial something 3x1gb 
Hard DriveCoolingOSMonitor
intel 330 180gb scythe kotetsu windows 8.1 pro sony cpd-g520 
KeyboardPowerCaseMouse
logitech k120 silverstone st75f-gs nxzt h440 evga torq x5 
Mouse PadAudio
allsop raindrop mobo 
  hide details  
Reply
main
(15 items)
 
old
(14 items)
 
 
CPUMotherboardGraphicsRAM
4770k maximus vii impact nvidia gtx 970 crucial ballistix tactical 16gb 
Hard DriveCoolingOSMonitor
crucial mx100 noctua nh-c14 windows 7 ultimate sony cpd-g520 
KeyboardPowerCaseMouse
kbp v80 matias quiet silverstone sx500-lg ncase m1 v3 logitech g100s with mcu replaced by teensy2.0 
Mouse PadAudioAudio
allsop raindrop xl chord mojo hifiman re-600 
CPUMotherboardGraphicsRAM
i7 920 evga x58 sli le galaxy gtx 460 crucial something 3x1gb 
Hard DriveCoolingOSMonitor
intel 330 180gb scythe kotetsu windows 8.1 pro sony cpd-g520 
KeyboardPowerCaseMouse
logitech k120 silverstone st75f-gs nxzt h440 evga torq x5 
Mouse PadAudio
allsop raindrop mobo 
  hide details  
Reply
post #8 of 130
Quote:
Originally Posted by Elrick View Post

Instead of kicking someone constantly for an opposing view maybe you should try and understand were he is coming from.

Coming from Mr. "If you find a mouse in Logitech's current gaming peripheral lineup you must be a crack addict" :-)
post #9 of 130

nice write up +rep!

post #10 of 130
very interesting - will now look at my mouse in a different light - i still like the idea of not having to clear the fluff out of the old ball mice biggrin.gif

and yeah! hava rep for my 1st detailed mouse read thumb.gif
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Mice
Overclock.net › Forums › Components › Mice › Discussion of targeting technologies