I've always been intrigued by LVDTs since I learned that they can be used to measure millionths of an inch displacements. With the advent of cheap computing with good A/D, perhaps it's time to add a DIY LVDT to my project list.
You can find pretty cheap resolvers on eBay (I have a couple of Singer units that I paid $30 for). I'm sure you can also find LVDTs. Resolvers & LVDTs only differ by motion type: resolvers are rotary and LVDTs are translational.
At least that way you don't have to do the "annoying" part of the project, which is likely to be the mechanical aspects of winding and placing the coils.
I built a product that basically simulates the process: it takes an analog or digital input and outputs sine & cosine signals that look like position information for a motion controller that expects an LVDT or resolver.
I've been thinking about using the magnetic and mechanical design of an LVDT in a different application: a high-reliability keyboard with a four-dimensional scan matrix to reduce the number of electrical lines required.
For a conventional keyswitch-matrix multiplexed keyboard with 81 keys, you need 18 GPIO lines, 9 row lines and 9 column lines. Even with Charlieplexing, I believe you need 13 GPIO lines to get to 81 keys. (½(14·13) = 91.) Keyswitch matrices are also mechanically and chemically delicate; a spill of solvent, battery acid, or sometimes even saltwater can damage the keyswitches, and they do not work underwater unless the keyboard is hermetically sealed. Such seals have to be flexible and are regularly flexed during usage, so they usually fail after only a few years. Some keyswitch contacts were often made of metal, which suffers oxidation over time resulting in keyswitch failure; many current keyswitches instead use contacts made of graphite-filled rubber, which doesn't form a solid oxide surface layer. (Keyswitches also generally require debouncing, though I suspect this is less of a problem with the graphite-filled rubber contacts.)
Capacitive keys avoid contact bounce and oxidation, but tend to suffer even worse from submersion because of the high electrical permittivity of water. They are also more sensitive to electrical noise.
By contrast, a differential-transformer key mechanism would permit an 81-key keyboard with only 12 GPIOs, high EMI immunity, and extreme mechanical robustness.
Each key contains a differential transformer, similar to an LVDT but without any attention given to linearity. When the key is not depressed, the core in the differential transformer is at its nulled position, where a pulse of current through the primary will produce exactly canceling voltages across the two opposing secondaries. But when the key is depressed, the core is substantially displaced, so that the net voltage pulse induced across the two opposing secondaries is significant.
Submersion poses no problem for the mechanism, because the magnetic permeability of water is basically the same as air or vacuum, so water filling the tube around the core is not a problem. As the TE page explains, the same is true of things like high-pressure hydraulic oil and even low-temperature molten metals. The mechanism would not work if you submerged it in a ferrofluid, or if you heated the core past its Curie point, but that is not much of a problem in most practical environments.
The four-dimensional multiplexing works as follows. There is a 3×3 primary-winding matrix and a 3×3 differential secondary-winding matrix. Each of the 9 primary-winding-matrix cells has the primary windings of 9 different keys in it, each of which belongs to a different cell of the secondary-winding matrix. These 9 primary windings in a single primary-winding-matrix cell are preferably in parallel. By pulling one of the three row lines of the primary-winding matrix high and pulsing one of its three column lines low, while maintaining the other 4 row and column lines tristated, you send a pulse of current through those 9 primary windings.
Similarly, each of the 9 secondary-winding matrix cells contains the 9 opposing-series-wound secondary-winding coils, in parallel, one for each of the 9 primary-winding-matrix cells. So each of the 81 keys represents a unique combination of a primary-matrix cell and a secondary-matrix cell.
I'm not yet entirely clear on how to scan the secondary-coil matrix for a given primary cell. It would be fairly straightforward if you had an electromechanical relay for each row, a diode on the anti-series secondaries of each key, and a sense resistor to ground on each column: the voltage induced on a depressed key connected to an open-circuited row would not be able to draw any current from its open-circuited row, so it would drive no current through its column's sense resistor, which would therefore remain at ground, and the open-circuited row line would be driven below groun...
In any practical application, the number of GPIOs simply wouldn't be an issue. Transistors are as cheap as sand.
What would you foresee as the application for a keyboard like this? It sounds like Hall-Effect switches would work just as well and cost significantly less.
Thinking about it a bit more, if it were true that "in any practical application, the number of GPIOs simply wouldn't be an issue. Transistors are as cheap as sand," nobody would multiplex keyboards or LEDs at all. And it's true that it's feasible at this point to put a driver chip on every LED; the WS2812 is a common chip that you can hook up into long daisy-chains into which you shift a bunch of digital binary data to tell them how to drive one RGB LED each. And you can do the same thing with keys on a keyboard, putting one microcontroller on every key and connecting them all to a common bus or a token-ring-like bucket brigade.
But people still do multiplex lots of LED matrices and keyboard matrices.
Since an LVDT operates on electromagnetic coupling principles in a friction-free structure, it can measure infinitesimally small changes in core position. This infinite resolution capability is limited only by the noise in an LVDT signal conditioner and the output display's resolution. These same factors also give an LVDT its outstanding repeatability."
LVDTs are fun! I bought one for a project a few years ago (since abandoned), and have been slowly designing my own readout electronics for them. Three revisions later, and the performance is very good - to the point that I've bought more increasingly precise measurement gear to benchmark how it's performing. Currently, I'm testing it against some good glass scale linear encoders, and a capacitive gauge with a single-digit nm noise floor.
Interesting. What applications are you using it for? (If you are okay discussing it.) diy calipers? Active feedback to a more complex mechanical system? As part of an optical system? It would be interesting to hear how these are being used. Especially from someone experience using/building them.
I was using it to build a dilatometer - an instrument for measuring thermal expansion curves of materials. Theoretically simple, in that you take a sample of something and measure how long it is while sweeping the temperature around. In practice, you need very stable ~um measurements and lots of care to make sure all other length changes around the sample cancel.
In the real-world, they're often used for precision gauging for in-process metrology.
Nuclear reactors (at least, some – no idea on all models) use them for measuring control rod height. Accurate, precise, reliable, dead-simple. Can’t ask for more.
So you can get one for 6 at AliExpress or >100 at western vendors... might be worth playing with the AliExpress one I guess. Wonder what the difference is.
I have a cheap Chinese electronic caliper that is also quite accurate. I wonder if there any sensors based on it and what they would be called (EDIT: linear magnetic/capacitive/optical encoders).
LVDT-like sensors on AliExpress seem quite expensive, but maybe I'm not looking in the right place.
18 comments
[ 2.7 ms ] story [ 55.4 ms ] threadDifferential Transformer https://arxiv.org/abs/2410.05258
At least that way you don't have to do the "annoying" part of the project, which is likely to be the mechanical aspects of winding and placing the coils.
I built a product that basically simulates the process: it takes an analog or digital input and outputs sine & cosine signals that look like position information for a motion controller that expects an LVDT or resolver.
For a conventional keyswitch-matrix multiplexed keyboard with 81 keys, you need 18 GPIO lines, 9 row lines and 9 column lines. Even with Charlieplexing, I believe you need 13 GPIO lines to get to 81 keys. (½(14·13) = 91.) Keyswitch matrices are also mechanically and chemically delicate; a spill of solvent, battery acid, or sometimes even saltwater can damage the keyswitches, and they do not work underwater unless the keyboard is hermetically sealed. Such seals have to be flexible and are regularly flexed during usage, so they usually fail after only a few years. Some keyswitch contacts were often made of metal, which suffers oxidation over time resulting in keyswitch failure; many current keyswitches instead use contacts made of graphite-filled rubber, which doesn't form a solid oxide surface layer. (Keyswitches also generally require debouncing, though I suspect this is less of a problem with the graphite-filled rubber contacts.)
Capacitive keys avoid contact bounce and oxidation, but tend to suffer even worse from submersion because of the high electrical permittivity of water. They are also more sensitive to electrical noise.
By contrast, a differential-transformer key mechanism would permit an 81-key keyboard with only 12 GPIOs, high EMI immunity, and extreme mechanical robustness.
Each key contains a differential transformer, similar to an LVDT but without any attention given to linearity. When the key is not depressed, the core in the differential transformer is at its nulled position, where a pulse of current through the primary will produce exactly canceling voltages across the two opposing secondaries. But when the key is depressed, the core is substantially displaced, so that the net voltage pulse induced across the two opposing secondaries is significant.
Submersion poses no problem for the mechanism, because the magnetic permeability of water is basically the same as air or vacuum, so water filling the tube around the core is not a problem. As the TE page explains, the same is true of things like high-pressure hydraulic oil and even low-temperature molten metals. The mechanism would not work if you submerged it in a ferrofluid, or if you heated the core past its Curie point, but that is not much of a problem in most practical environments.
The four-dimensional multiplexing works as follows. There is a 3×3 primary-winding matrix and a 3×3 differential secondary-winding matrix. Each of the 9 primary-winding-matrix cells has the primary windings of 9 different keys in it, each of which belongs to a different cell of the secondary-winding matrix. These 9 primary windings in a single primary-winding-matrix cell are preferably in parallel. By pulling one of the three row lines of the primary-winding matrix high and pulsing one of its three column lines low, while maintaining the other 4 row and column lines tristated, you send a pulse of current through those 9 primary windings.
Similarly, each of the 9 secondary-winding matrix cells contains the 9 opposing-series-wound secondary-winding coils, in parallel, one for each of the 9 primary-winding-matrix cells. So each of the 81 keys represents a unique combination of a primary-matrix cell and a secondary-matrix cell.
I'm not yet entirely clear on how to scan the secondary-coil matrix for a given primary cell. It would be fairly straightforward if you had an electromechanical relay for each row, a diode on the anti-series secondaries of each key, and a sense resistor to ground on each column: the voltage induced on a depressed key connected to an open-circuited row would not be able to draw any current from its open-circuited row, so it would drive no current through its column's sense resistor, which would therefore remain at ground, and the open-circuited row line would be driven below groun...
What would you foresee as the application for a keyboard like this? It sounds like Hall-Effect switches would work just as well and cost significantly less.
But people still do multiplex lots of LED matrices and keyboard matrices.
Since an LVDT operates on electromagnetic coupling principles in a friction-free structure, it can measure infinitesimally small changes in core position. This infinite resolution capability is limited only by the noise in an LVDT signal conditioner and the output display's resolution. These same factors also give an LVDT its outstanding repeatability."
Related: https://en.wikipedia.org/wiki/Linear_variable_differential_t...
>"A counterpart to this device that is used for measuring rotary displacement is called a rotary variable differential transformer (RVDT)." (https://en.wikipedia.org/wiki/Rotary_variable_differential_t...)
So many don't buy an LVDT, I guess.
In the real-world, they're often used for precision gauging for in-process metrology.
LVDT-like sensors on AliExpress seem quite expensive, but maybe I'm not looking in the right place.