[biz-engineer]

Warning: geek post.

I was curious about how IBIS actually works, and I had an unanswered question at the back of my mind for some years: how can sensor be moved in steps as small as 4.8 micron (e.g pixel shift).
Information found on the web doesn't give much technical detail. The best I found is this Olympus/Image Resource article from 2020 (INSIDE Image Stabilization: Olympus takes us on a geek’s tour of what makes IS work).
I assume Pentax camera use the same technology, more or less.

The sensor is moved by Lorenz force created by two sets of magnets and coils, and that kind of linear motor arrangement known as called "voice coil" technology.
I know how a servo loop works, thus I can understand how the sensor is moved to counter the effect of camera shake when the photographer is taking a picture "hand-held".

Now, how Pentax can move the sensor by exactly one square pixel step for pixel shift, applying a magnetic force (driving current pulses into the coils) is still a mystery.

More specifically, if there are 4.8 micron precise position references under the sensor (e.g Pentax K1), that would mean Ricoh need to develop a new IBIS module every time the camera sensor resolution changes, and doing so would be very costly!

When I run a sensor check on my K1, I put my ear close to the camera, I can hear some little noise (sounding like steps) while checks are being performed, I was wondering if the camera is calibrating sensor IBIS and steps?

Do you have knowledge about how precision of position is achieved with voice coil motors?

If step control of the sensor plate is unrelated to the pixel pitch of image sensor, then Pentax could implement sub-pixel pixel shift on my K1, just save 16 files on memory card, do it via firmware update right ?

[Digitalis]

Do you have knowledge about how precision of position is achieved with voice coil motors?

This is conjecture, I'm not a mechatronic engineer: I'd imagine they would use a closed loop servo system that can shift the sensor by very precise amounts: One instance where I know such technology is used is in high performance subwoofers from current manufacturers such as REL, Rythmik and Velodyne . These subwoofers typically use double voice coils within the motor assembly and closed loop servo control mechanisms connected to the amplifier. Through using very precise accelerometers* on the main voice coil and driver to correct for non-linear movement of the driver, the correction is done in a closed loop and thus it can react very quickly to non-linear movement. The system relies on feedback from the accelerometers and using this signal to ascertain the motion of the driver, when non-linear motion is detected: the servo system will modulate and correct the source signal through superimposing a correction signal looped through the amplifier through to a second voice coil [which is in a static position around the main voice coil in the motor assembly] and this signal modulation will cancel out any non-linear behavior from the driver to produce linear movement and thus reducing distortion**.


The IBIS system Pentax uses has multiple electromagnetic coils and closed loop accelerometers, it would be possible to use a similar method to move the sensor.



* Optical systems are also used where a light reflective strip is placed on the voice coil and is able to be read by optical sensors, the only problem with this method were modal harmonic oscillations that the optical system was blind to.
** These kind of systems are expensive and complex to engineer and implement: Certain speaker components must be a known quantity particularly the properties of the driver cone, driver suspension and the magnetic flux of the motor assembly as well as the complexity of the amplifier goes up as well. Most manufacturers such as SVS, KEF, and Klipsch use real time Digital signal processing to correct for non-linearities of the driver, these methods are effective albeit limited compared to a servo control system.

[photoptimist]

IBIS is based on very precise analog control of the voice coil's position within the strong magnetic field of a permanent magnet.(See @Digitlis' explanation) It's a variant of the same analog control loop used in hard disk drives that can enable the read/write head to follow a track on the disk that is less than 0.1 microns wide on the disk with levered voice coil movements on the order of 0.02 microns.

That control system does not have a "step size" per se although it does have some sort of tracking error figure that can be expressed in terms of step size.

For regular pixel shift to work, the tracking error must be a modest fraction of a pixel (maybe < 1/3 or 1/4 pixel). If the sensor shift was not accurate enough, then pixel shift images would tend to get strange odd/even row or column stair-step artifacts along linear edges.

Half-pixel, super-resolution shift is even more challenging. Not only is the required movement a half a pixel but: 1) the total number of movements is doubled or quadrupled (more opportunities for errors); 2) the duration of the sequence is doubled or quadrupled (worse accumulated drift in the control loop); 3) the image is calculated from differences of data (more amplification of errors).

[Bob 256]

I'm no expert on the Ricoh IBIS system, but a "voice coil" uses coil current to obtain very precise positioning. Small current variations are used for varying positioning in such a system. There would probably be a D to A converter involved somewhere, to convert a digital value to a current-drive value corresponding to a given camera sensor position. If the resolution of that D to A wasn't max'd out, then sub-pixel movements would be possible but it may be that such a system would operate at the single bit resolution of the D to A in which case no finer movement could be accomplished without major changes. Also, obtaining sub-pixel motion would at least require software modifications which probably couldn't be done in firmware. Accommodating other pixel pitches (in other camera models) would simply be a matter of scaling the stepper current so basically the same circuitry could be used.

One good example of commonly used "voice coil" positioning is in hard drives to position the head assembly over the tracks on the spinning discs. Feedback is used in this case which maximizes the signal output once a track is located, however it's all done with a highly controlled current and permanent magnets. Another example is the approach is used in the wire attachment process for integrated circuit manufacturing where positioning of a small "spot welder" is done at blinding speeds which zips the connections between the chip and the connections going to the outside world.

---

Last edited by Bob 256; 05-28-2022 at 02:21 PM.

[Digitalis]

It's a variant of the same analog control loop used in hard disk drives that can enable the read/write head to follow a track on the disk that is less than 0.1 microns wide on the disk with levered voice coil movements on the order of 0.02 microns.

I don't think modern* hard drives use analog control loops - when I was decommissioning a hard drive from 2001 I did note the presence of a Cirrus logic 4 channel 8-bit DAC chip on the controller board. The accelerometers may be analog but information the drive uses internally is digital**, when you look up S.M.A.R.T drive diagnostics the values are in binary. I'd imagine the Pentax IBIS system is able to operate in analogue to provide stabilization in a step-less manner, but also in stepped mode with discrete increments of movement down to a few µm.


*drives made anywhere between the late 90s and now
**Digital control makes sense considering how discrete data tracks are laid out on a CMR drive.

[biz-engineer]

It looks like, there is no discrete step reference anywhere that would limit the resolution of positioning for the image sensor. If the position is achieved by the amount and duration of current pulsed into a coil, then the limit is how precise how noisy is the electronic behind the coils. I imagined the sensor could be moved in sub-pixels steps diagonally to produce super-resolution images, and not necessarily via pixel shift " demosaicing", just saving a series of images on the memory card, each exposure slightly shifted, so that a super-resolution image can be created by external software.

[photoptimist]

I don't think modern* hard drives use analog control loops - when I was decommissioning a hard drive from 2001 I did note the presence of a Cirrus logic 4 channel 8-bit DAC chip on the controller board. The accelerometers may be analog but information the drive uses internally is digital**, when you look up of S.M.A.R.T drive diagnostics the values are in binary. I'd imagine the Pentax IBIS system is able to operate in analogue to provide stabilization in a step-less manner, but also in stepped mode with discrete increments of movement down to a few µm.


*drives made anywhere between the late 90s and now
**Digital control makes sense considering how discrete data tracks are laid out on a CMR drive.

You are absolutely right.

I meant to distinguish the difference in a control loop based on analog sensors (e.g., MEMS accelerometer, gyroscope, Hall-effect, back-EMF, etc.) versus one that uses digital sensors such as an encoder wheels. I'd also bet that the software in both disk drives and IBIS use algorithms that implement numerical analog control equations such as a PID controller.

The high-level commands (e.g., seeking a particular cylinder on a disk platter or moving the sensor for pixel shift or astrotracer) would compute a numerical goal for the mechanism location, digitally change a numerical value in the variables for the PID control equation and let the control loop do its thing.

The high-level system block diagram for the device would look like a fully analog control loop but the lower-level hardware diagram would have ADCs between the sensors and the controller and DACs between the controller and motion power electronics.

[biz-engineer]

a "voice coil" uses coil current to obtain very precise positioning. Small current variations are used for varying positioning in such a system.

Wikipedia talks about the lorenz force proportional to coil current, so when applying a current to the coil the motion would never stop unless the magnetic field seen by the coil changes so that the force drops to zero after the sensor moved reached its new position. So if the sensor is at position A (Xa,Ya), I see how the sensor plate move is initiated from A, but I don't see how the sensor is stopped and maintained at position B (Xb,Yb), unless there is a counter force to the sensor or unless the magnetic field from the magnet isn't a constant homogeneous field.

[Bob 256]

Wikipedia talks about the lorenz force proportional to coil current, so when applying a current to the coil the motion would never stop unless the magnetic field seen by the coil changes so that the force drops to zero after the sensor moved reached its new position. So if the sensor is at position A (Xa,Ya), I see how the sensor plate move is initiated from A, but I don't see how the sensor is stopped and maintained at position B (Xb,Yb), unless there is a counter force to the sensor or unless the magnetic field from the magnet isn't a constant homogeneous field.

This is dynamic system much like a motor (in fact it is sometimes termed a "motor"). If the current in the coil is reduced to and maintained at zero (somewhat like shorting the coil), it produces a braking force and will actually oppose any movement (although the braking action depends on the velocity). By controlling the magnitude and direction of the current, the force is controlled and positioning with it. It's not all that simple to just put a given current in the coil and move it to a given position (as you say, the force would then be constant). Usually there's some kind of feedback involved which allows the controller (which has control over the voice coil current) to know the position being controlled and vary the current as needed. As in a true "voice coil" in a speaker, the current can vary in both amplitude and polarity to get the desired positioning result.

[biz-engineer]

Usually there's some kind of feedback involved which allows the controller (which has control over the voice coil current) to know the position being controlled and vary the current as needed.

I see what you mean, current = force, a control can move the sensor to a position by computing dynamically the motion and velocity induced by that force, and modulating the current accordingly, and using a relative position sensor to correct for small differences in weight, friction, temperature and from camera to camera. It's just that Ricoh doesn't provide any detail about that sensor. Olympus say they use an hall effect position sensor, below the CMOS image sensor apparently. I've never opened a Pentax sensor module, but that would be interesting to see.

[slartibartfast01]

There seem to be many patents involving pixel shift, for example this one from Panasonic
https://patents.google.com/patent/US8253818B2/en
I wonder how Pentax came to use it.

[RobA_Oz]

The Pentax patent descriptions are as follows:
CN100523986C - Anti-shake apparatus - Google Patents
Google Patents
The second link is the earlier one, and precedes the Panasonic application.

[slartibartfast01]

The Pentax patent descriptions are as follows:

CN100523986C - Anti-shake apparatus - Google Patents

Google Patents

The second link is the earlier one, and precedes the Panasonic application.

Isn't that IBIS rather than pixel shift?

[Bob 256]

IBIS and pixel shift use the same "voice coil" actuator that this thread began discussing. The only difference is that IBIS uses computed information (derived from motion sensors) to drive the sensor actuator (and position) whereas, pixel shift uses predetermined shifts in the sensor position (up & down and sideways shifts). In IBIS, the sensor can actually be in motion (actually a fast series of small position changes or steps) while the shot is being taken, and there are 5 axes which apply to IBIS.

---------- Post added 05-30-2022 at 08:49 AM ----------

I see what you mean, current = force, a control can move the sensor to a position by computing dynamically the motion and velocity induced by that force, and modulating the current accordingly, and using a relative position sensor to correct for small differences in weight, friction, temperature and from camera to camera. It's just that Ricoh doesn't provide any detail about that sensor. Olympus say they use an hall effect position sensor, below the CMOS image sensor apparently. I've never opened a Pentax sensor module, but that would be interesting to see.

Correct. A Hall sensor is a sensor which can sense the strength of a magnetic field and used with a fixed magnet, it can sense the position of an object (used in a lot of "levitation" toys to sense the position, or more specifically the tilt, of the levitated platform). The fact that one might be used in Pentax's implementation makes perfect sense for use in getting position feedback information.

[RobA_Oz]

Isn't that IBIS rather than pixel shift?

Yes, but as far as I’m aware, pixel shift uses the same general control mechanism, just as horizon correction does. I could be wrong, but I thought that pixel shift was implemented in software, perhaps with hardware that evolved with finer tolerances in the later camera bodies.