[PATCH RFC 0/3] input: rotary_encoder: use more than two gpios as input

[Rojhalat Ibrahim]

On Wednesday 02 December 2015 19:59:20 Uwe Kleine-König wrote:
> Hello,
> 
> On Wed, Dec 02, 2015 at 01:52:58PM +0100, Rojhalat Ibrahim wrote:
> > On Wednesday 02 December 2015 11:07:11 Uwe Kleine-König wrote:
> > > some time ago I worked on the rotary encoder driver to make it support
> > > more than two input lines. This is the (only slightly tested[1]) rebase of
> > > this series on top of 4.4-rc2 (from 4.1).
> > > 
> > > It would be great to get some feedback, especially (but not only) for my
> > > change to raumfeld.c.
> > > 
> > > Before Ezequiel's patch 3a341a4c30d4 ("Input: rotary-encoder - add
> > > support for quarter-period mode") we had a dt property
> > > "rotary-encoder,half-period" defined. It's presence meant that we had 2
> > > indents per period; it's absence that there is only 1. Ezequiel
> > > introduced rotary-encoder,steps-per-period instead when adding support
> > > for quarter-period mode (which has 4 indents per period).
> > > 
> > > Up to now (i.e. with two inputs) a period has length 4. Now with (say)
> > > four inputs a period has length 16 instead. Now how should this be
> > > modeled in dt? This series implements that I have to pass
> > > 
> > > 	rotary-encoder,steps-per-period = <16>;
> > > 
> > > now for "quarter-period mode" (i.e. 4 indents per 4 state changes[2]),
> > > but that feels unnatural. I'd prefer to set a property to <1> instead,
> > > meaning the device has an indent for each state change[2]. half-period
> > > mode would be <2> and full-period mode would be <4>. But I don't have a
> > > nice name for such a property and maybe it's easier to live with
> > > steps-per-period = <16>? What do you think?
> > > 
> > > Also note that these patches are not as technically versatile as
> > > possible. With 4 (n) input lines we could detect a quick rotation where the
> > > machine's latency only allows to sample after 7 (2^(n-1)-1) state
> > > changes. Currently this is not implemented, but can be done later.
> > > 
> > > Best regards
> > > Uwe
> > > 
> > 
> > AFAIUI the rotary encoder driver only handles incremental encoders not
> > absolute encoders. (So in fact the assumed rotary encoder could also be a
> 
> There is a device tree property "rotary-encoder,relative-axis" which
> switches between absolute and relative reporting. This is maybe badly
> named, but IIUC is there to differenciate between incremental and
> absolute encoders.
> 

IMHO the property switches between absolute and relative reporting but the
driver still handles only incremental encoders. Incremental encoder means that
the input lines tell you when the encoder moves and in which direction. So
by counting the steps you can generate an absolute value (which is what the
driver does). But this absolute value is always _relative_ to the position
at the time of start-up.

An absolute encoder allows you to determine its position by just looking at
the state of the input lines. There is no need to count steps.

> > linear encoder with an incremental interface.) Those encoders almost always
> > have an interface with two outputs (A, B) with a phase shift of 90 degrees
> > between them. So in this case we have 4 steps per period. Sometimes there
> > is only one line for 1 or 2 steps per period. But I have never seen or
> 
> I don't see how using only a single gpio would work for a rotary
> encoder. I suspect we use different vocabulary to describe our devices
> and so don't understand each other?! At least you cannot detect the
> direction of the movement with a single input line, do you?
> 

Correct. With a single input line you can only count steps upwards and have
no direction information. Sometimes that is sufficient.

> > heard of a device with more than 2 lines (except for the third output which
> > serves as a reference position index marker).
> > 
> > Do devices wirh more than two outputs actually exist?
> > Or is the purpose of supporting more than 2 lines something other than
> > connecting an actual encoder to them?
> 
> I have a real rotary encoder with 4 input lines and so 16
> distinguishable positions. It would be described as:
> 
> 	compatible = "rotary-encoder";
> 	gpios = <&gpio4 12 GPIO_ACTIVE_HIGH>, <&gpio4 11 GPIO_ACTIVE_HIGH>, <&gpio4 10 GPIO_ACTIVE_HIGH>, <&gpio4 9 GPIO_ACTIVE_HIGH>;
> 	rotary-encoder,steps = <16>;
> 	rotary-encoder,steps-per-period = <16>;
> 
> with the binding implemented in my patches.
> 
> The wikipedia article about rotary encoders[1] uses a device with 3
> input lines to explain gray encoding. So I didn't consider "my" device
> as exotic up to now.
> 

So you have an absolute encoder. From the state of the four input lines you
can determine the absolute position value. There is no need to count steps.
At least that's the original purpose of your device.

Of course it might be conceivable to use that kind of absolute encoder in an
incremental way and count the steps throughout multiple revolutions.
But why would you do that? The resolution of 16 steps per revolution is not
very good (incremental encoders often have hundreds or thousands of steps per
revolution). And the whole point of an absolute encoder is that you are able
to determine the current position at any time just by looking at the inputs
without having to count steps.

   Rojhalat


> Best regards
> Uwe
> 
> [1] https://en.wikipedia.org/wiki/Rotary_encoder
> 
>