LED backlighting suggestions (4 in CC config, 3.2-3.4V 20mA)

[Zero999]

Good point about the speed of the LM317.

I wouldn't have thought he would PWM at 20kHz.

How about the LM334?

I didn't bother with a model for the simulation so I made one with a behavioural current source. Here's a practical example.

[Benta]

This is getting a bit exotic, somehow.
Sure, you can throw silicon at the design, but the two basic discrete circuits still stand. And at around 30 cents (hobbyist pricing) you can't do it cheaper or simpler.
Temperature stability or PWM precision?

That's the decision. I'd go for the 2nd choice every time. The first can be compensated with the PWM (every system has a temperature sensor these days).

Cheers.

[Zero999]

I'm just having a bit of fun. In all honesty, if it were my design, I'd opt for the circuit posted by Eugene. It's simple and the the 5V rail should be stable enough to make plain old current limiting resistors a viable option. I'd select high enough resistor values to ensure the maximum current rating of the LEDs isn't exceeded at the upper tolerance of the 5V rail (5.5V is reasonable) and lower forward voltage of the LEDs: 3.2V.

R = V-V_F/I_F = (5.5-3.2)/0.02 = 2.3/0.02 = 115R

It makes no sense to use 115R, an E96 value, so the closest E24 value of 120R can be used instead.

Of course the forward current will be approximately halved on the lower end of the tolerance of the 5V rail, which we can assume to be 4.5V, but 10mA should be enough and the user can simply adjust the brightness setting to compensate.

[Benta]

I've removed my last posts due to wrong information.
I'm unsure what happened here. I'm certain that the data sheet I found on the display showed four separate anodes and four separate cathodes. Perhaps I found the wrong data sheet? No idea, and I can't solve it. Sorry.

[Nominal Animal]

After a lot of simulation, I've decided to start practical experiments by building the following circuit:


The VLED is provided by Microchip MCP4801 SPI (/CS, SCK, MOSI) DAC with an internal 2.048V reference, with the output varied between 0 and 2.048V only when the display is turned on or off; it will stay at 2.048V during normal operation.  So, for the simulations I describe below, VLED=2.048V.
 
At 15°C, the LED current is 18.5mA, and current over Q6 6.0mA.
At 25°C, the LED current is 18.8mA, and current over Q6 6.1mA.
At 65°C, the LED current is 19.8mA, and current over Q6 6.4mA.

So, the model suggests about 5% drift upwards at 40°C ambient rise, which I consider acceptable.

Replacing Q5 (BC857C, PNP) with BC847C (NPN) and adjusting the per-LED resistor to 74 Ohms, I can even invert the temperature dependence: 15°C/19.4mA, 25°C/18.9mA, 65°C/17.0mA, giving 10% downwards drift in current at 40°C ambient rise.

I don't exactly trust those simulations (mostly because I constructed them; I don't doubt the Nexperia SPICE models), but it does give me a solid place to start testing; I already ordered a heap of components I can experiment with. 

[Benta]

Looks fine to me, your results seem OK.
From a design point of view, I find the Q6 current unnecessarily high. but that's your decision. With the BC857C you have an h_FE of over 220 to play with.

[Nominal Animal]

From a design point of view, I find the Q6 current unnecessarily high.

True!  I do not have a good grasp of the design criteria to determine what a suitable current would be.

You did mention 2mA as the absolute minimum, so I thought 6mA ought to be safe, but not too much.

(I do not know how you arrived at that, but would like to know for sure.  The only thing I can see in the Nexperia BC847C datasheet is that with V_CE=5V and I_C=2mA, h_FE is specced to be at least 420. And that below 3mA, the collector-emitter saturation voltage is flat.  Do these matter in practice?
At or below 20mA, BC857C has h_FE > 400.  If we use 200 so we have a reasonable margin, then the base current per PNP is max. 0.1mA, right?  So, my theoretical understanding says that even with four PNPs, 1mA over Q6 would suffice.  But would it, in practice?)

How exactly do you determine a suitable practical collector current in this kind of a circuit?

If I switch R5 and R6 to 470 Ohm, current over Q5 and Q6 would be almost exactly 3mA, without meaningful changes in e.g. thermal behaviour.
With 1 kOhm, about 1.3mA.

[MikeK]

When driving a transistor as a switch the gain (beta) used is usually 10 or 20, and this can be found in the datasheet where it gives the conditions for saturation voltage.  For the BC847 I'm reading Ib=0.5mA and Ic=10mA...that's a gain of 20.  So if you're expecting a collector current of, say, 20mA, a base current of 1mA is enough to do it.

[Benta]

When driving a transistor as a switch the gain (beta) used is usually 10 or 20, and this can be found in the datasheet where it gives the conditions for saturation voltage.  For the BC847 I'm reading Ib=0.5mA and Ic=10mA...that's a gain of 20.  So if you're expecting a collector current of, say, 20mA, a base current of 1mA is enough to do it.

This circuit is running 100% linear, so totally irrelevant. Sorry.

[Benta]

From a design point of view, I find the Q6 current unnecessarily high.

You did mention 2mA as the absolute minimum, so I thought 6mA ought to be safe, but not too much.

(I do not know how you arrived at that, but would like to know for sure.  The only thing I can see in the Nexperia BC847C datasheet is that with V_CE=5V and I_C=2mA, h_FE is specced to be at least 420. And that below 3mA, the collector-emitter saturation voltage is flat.  Do these matter in practice?
At or below 20mA, BC857C has h_FE > 400.  If we use 200 so we have a reasonable margin, then the base current per PNP is max. 0.1mA, right?  So, my theoretical understanding says that even with four PNPs, 1mA over Q6 would suffice.  But would it, in practice?)

How exactly do you determine a suitable practical collector current in this kind of a circuit?

If I switch R5 and R6 to 470 Ohm, current over Q5 and Q6 would be almost exactly 3mA, without meaningful changes in e.g. thermal behaviour.
With 1 kOhm, about 1.3mA.

The 2 mA I mentioned were for a slightly different design using B-types and different base drive, so let's leave that aside.

But basically: it's all about experience, a bit of gut feeling and a lot of design iteration. With the BC847C857C type you have plenty of gain (it drops a bit when operating at -55 C )
At a LED current of around 20 mA, a base current for the driver transistors of 4 x 25 uA = 0.1 mA is reasonable. That's what Q6 has to pull through its collector in addition to the bias current through Q5. At 6 mA, this corresponds to 1.6%. At 2 mA it's 5%.
That's one part of the "error current" in the Q6 emitter resistor (which sets the reference for the full circuit). The second part comes from the Q6 base current (which is negligible).
Now start adding in other error sources such a V_BE variations (both between devices and over temperature), resistor tolerances etc., and you'll see that the Q6 current is the very smallest part of all this.
There's absolutely nothing "wrong" about your 6 mA and absolutely nothing "right" about my 2 mA. It's a design choice/balance.

The target of my original circuit, now modified to the current one, was to have as close a matching between the LED currents as possible for an even backlight.
Absolute precision was never the target.
And I'm pretty certain, that the backlight LEDs are much worse matched than the currents from this circuit.

Saturation voltages are irrelevant here.

[Zero999]

When driving a transistor as a switch the gain (beta) used is usually 10 or 20, and this can be found in the datasheet where it gives the conditions for saturation voltage.  For the BC847 I'm reading Ib=0.5mA and Ic=10mA...that's a gain of 20.  So if you're expecting a collector current of, say, 20mA, a base current of 1mA is enough to do it.

This circuit is running 100% linear, so totally irrelevant. Sorry.

True, the figure quoted by MikeK is for the saturation region. The original poster's design is working in the linear region. The data sheet specifies an h_FE of about 500, when V_CE = 5V. It'll be a little lower with V_CE = 3V, but not by much. I_B  for Q6 will probably be around the 12µA mark.

[RoGeorge]

I didn't read all the previous messages, have the MOS-FET + coil (SMPS - Switching Mode Power Supply) solution been considered instead, so the energy won't be wasted as heat dissipated in the BJT transistors and the limiting R/68 ohms?

A SMPS can be set as a DC current source, that would be great for a non-flickering back light.  PWM might get nasty beats with the display's own refresh rates.

There might be dedicated ICs with I2C or alike, to control a LED intensity without flicker, in DC.

For a discrete components design, see if any of these SMPS topologies seem appealing:
https://www.ti.com/lit/ml/sluw001g/sluw001g.pdf  Brief
https://www.ti.com/seclit/ug/slyu036/slyu036.pdf  Handbook

[Benta]

I didn't read all the previous messages

Do that first, before posting something generic about switching converters, please. We all know how to do an internet search.

[Nominal Animal]

I didn't read all the previous messages, have the MOS-FET + coil (SMPS - Switching Mode Power Supply) solution been considered instead, so the energy won't be wasted as heat dissipated in the BJT transistors and the limiting R/68 ohms?

Even when dissipating the difference in the transistors and resistors, the design is still 67% efficient (producing at most 136mW of waste heat).  Diminishing returns: I'm balancing design simple enough for a hobbyist, easily adjustable and customizable, component count, and overall price here.  Since this is wall-wart powered and not battery-powered/portable device –– it is a display for routers and other Linux SBC-based appliances ––, the waste heat is an easily managed problem, compared to the other facers.

A SMPS can be set as a DC current source, that would be great for a non-flickering back light.  PWM might get nasty beats with the display's own refresh rates.

I'm using an 8-bit DAC with an internal reference – MCP4801T-E/SN –, not PWM.  I have a physical potentiometer as a voltage divider to control the actual brightness, with the DAC providing smooth turn-on and turn-off effects.  These are personal choices not dictated by the circuit, though.  Similarly, I want four separate resistors, one of each LED, in case one of my displays has unmatched LEDs.  (Unlikely, but possible; since these are all one-off designs for my own purposes and not something I intend to make in batches, I definitely want to spend some extra for flexibility and configurability.)

The circuit shown above takes a 0 - 2.048V control voltage; I omitted the circuit I'm using to produce that control voltage.  (Right now I'm traveling, and don't have the circuit with me.)

There might be dedicated ICs with I2C or alike, to control a LED intensity without flicker, in DC.

Yes, especially switched-capacitor pumps like TI LM27951 or LM2794, but they're unobtainium right now.

[RoGeorge]

For a DAC, you may want to consider our senses about intensity are logarithmic, while a DAC has linear output, therefore most of the values will translate to almost full light, and only a few very small DAC values will get a good dim.  You may want to experiment first, and see if an 8 bits DAC is enough to implement a logarithmic output. 

For example, this is a 1 minute showing LEDs browsed with 8 bits linear dim.  Note how most of the time they are mostly on, dimming happens linear for the average current, but not for the eye:



Video is from this demo:  https://hackaday.io/project/6356-delta-sigma-versus-pwm
https://github.com/RoGeorge/Delta-Sigma_versus_PWM/blob/master/main.c

Since you already have an MCU and a few pins to control the DAC, I would just pulse those DO to dim the LEDs instead (and remove the DAC).  To minimize flicker, I wonder if a C in parallel with the LED will worth, thinking about this as we speak, didn't calculate/simulate, anything.  I mean, you have the series resistor, then the LED, but in parallel with the LED, add a capacitor to store some energy for the periods when the pulse from the MCU is off.  Tempting?


With the risk of getting boo-ed even more, I'll prefer just a logarithmic (mechanical) potentiometer in series with the LED's and nothing more , not sure if such a potentiometer (with a small enough value) could be found.

[Nominal Animal]

For a DAC, you may want to consider our senses about intensity are logarithmic

Like I said, it's only used to ramp the backlight on and off smoothly.  Similarly, even though the potentiometer is linear (even logarithmic and antilogarithmic ones are usually just piecewise linear), as it controls the overall brightness, its response and dynamic range is much more important than the overall curve.

There is also the fact that the luminosity-current curve of the LEDs is definitely non-linear.  Eight bits – or in my case, a bit over seven effective bits, since the current turns off at about 0.7V control voltage, leaving 2.048V-0.7V ≃ 1.35V effective control range – is enough even in linear form to achieve a nice smooth ramp-in and ramp-out; that is, the resolution suffices at both ends, regardless of the ramp function I choose.  Most likely a fifth or seventh-degree polynomial with fixed coefficients, as the MCU I'm using (Teensy 4.0, i.MX RT1062) has hardware single-precision floating point support.

(I'm very familiar with psychovisual effects, having done real research into both sonification and visualization wrt. representations of molecular simulations.)

With the risk of getting boo-ed even more, I'll prefer just a logarithmic (mechanical) potentiometer in series with the LED's and nothing more , not sure if such a potentiometer (with a small enough value) could be found.

They tend to be piecewise linear.  However, I think you can compensate both the LED luminosity-current curve and the human psychovisual response by using the potentiometer as a variable resistor, as the next-to-ground part of a voltage divider; possibly with a fixed resistor in parallel to the voltage divider.

However, there is no need for the potentiometer to precisely match the inverse of the perceived luminosity.  It is much more important that at every possible position, there is immediate response.  (A mathematician would say that if the perceived response is the same at all points, then the potentiometer precisely matches the inverse of the perceived luminosity.. the thing is, with human perception, a difference of 10% - 20% is insignificant here, and even 50% in the difference of the immediate response is acceptable here, so the overall curve can be quite far from the inverse of the perceived luminosity!)

[RoGeorge]

I see now this thread has settled down 2 monts ago before I replied (arrived here from a PM link, and didn't check the last post's date, my bad, sorry, shouldn't have necropost), but since that already happened ...

Took a brief read and this seems to be the current version, right?



If so:

1. - I wouldn't have used a REF3333.  That's a very stable reference, a total waste for the purpose (and in the current chips shortage), especially because Teensy already has a 3.3V stabilized voltage.  Use the existing 3.3V from Teensy's 3.3V stabiliser as reference.

2. - Eugene gave the proper solution, 1 transistor to control all 4 LED's.  4 transistors controlled all at once doesn't make much sense to me.

3. - If I must have 1 transistor/LED, at least I would control them individually, so the ON pulses will be the same duty factor, but shifted in time, so to quadruple the flickering frequency.

4. - You were saying something about observing in simulation the switching regime.  Indeed, when driven into saturation, switching off a BJT will lag behind the control signal.  Have to either avoid entering saturation (easiest with a clamping diode to collector), or to remove the charges from the base in a "forced" way, so to turn off the transistor faster (can be done many ways, the simplest would be an RC parallel circuit in series with the base).

5. - I would prefer switching MOSFETs than BJTs.

6. - VLED signal is output from the DAC, yet tied to the cursor of a potentiometer and to the base of Q1.  You were talking in the OP about PWM, so are the LED pulsing or controlled in DC?  If pulsing, then who's pulsing them, if DC, then who's stabilising the current?  Controling LEDs by voltage doesn't work well, there are to bigger variations with temperature and from one batch of LEDs to another to count on Voltage vs Light intensity charts.  You said you want pulsed so to not change the spectrum of the light, then you seem to control it with DC, and in voltage?  What am I missing?

7. - AFAIK (disregarding light intensity with the die temperature, which are not really visible do the eye), the forward current I_f vs V_f is an exponential, and Light intensity (I think flux, not sure) is directly proportional with I_f.  The eye perceived flux is logarithmic.

8. - If I were you I would choose a display with less power hungry LEDs.  The ones you have will need typical 70 million A (MA) according to page 11 of the datasheet https://www.buydisplay.com/download/manual/ER-TFT028A2-4_Datasheet.pdf



Anyways, if you made the circuit already and it works upon your wish, then well done  and please don't waste time to clarify the above points I didn't understood from the 4 pages.

Posting again only because you just wrote something else that I'm very interested in:

I'm very familiar with psychovisual effects

Does this happen to cover any effects of very low frequency blinking lights (less than 10Hz)?

Asking because I did once a blinking toy with two (monochrome) LEDs, and the effects to the brain were very unexpected.  It was possible to perceive full colors (which were other colors than the LED's color) with the eyes closed (light from the LEDs through the eyelid), and other visual illusions like patterns depending of the blinking type, some people were reporting sleepiness, others nausea, etc.

That experiment was very interesting, then forgot about it and it stayed dormant in my bucket list researches for about 15-20 years now, and you just summoned it again. 

[Nominal Animal]

Took a brief read and this seems to be the current version, right?

Nope; it's something like the following:

4 transistors controlled all at once doesn't make much sense to me.

While they are usually well enough matched, I just want the option to slightly adjust the balance between the LEDs, in case they are not.

You said you want pulsed so to not change the spectrum of the light, then you seem to control it with DC, and in voltage?  What am I missing?

No, I did not.  I said that I want the display to be driven with constant current, and that I was aware of the slight change in the light spectrum depending on the current.  I did not say that I wanted to do PWM (or PDM), only that it was one option I was considering.  Any slight bias in the backlight spectrum is irrelevant, because this is an information display, not something that attempts to reproduce photographs or similar.

In the very first post, I wrote that "I'd also like to have MCU brightness control, maybe via PWM or something (although I'll only use it for smooth on-off transitions)."  That is, the backlight brightness is only controlled by the MCU when turning it on or off.  At all other times, either the backlight is completely off, or it is on at whatever maximum brightness the end user wants it.  Having the physical brightness control be a potentiometer not under MCU control is simple and robust, which I like; it also makes it easier to both simulate and breadboard-test the design.

I was hoping that having this thread here might help other hobbyists investigating how they want to control their LED backlights.  Some might note that I chose particular transistors which I could easily both simulate and obtain (I now have 100 of BC847C,215 and BC857C,215 each from Nexperia); I tried to de-emphasize my own choices, and emphasize the reasons why one might make particular choices, so that others with different design criteria could benefit from this thread, too.  So, if one reads this thread just to see what I was going after, it can definitely look odd.  Do not forget, I learned a lot about practical circuits (especially driving LEDs) in this thread, too, so progression is to be expected!

Does this happen to cover any effects of very low frequency blinking lights (less than 10Hz)?

No, I focused more on color-depth perception, edge/outline thickness in relation to perception of structure and connectedness, and such.  Basically, anything that could be useful in trying to answer the question "How do I control the information in this illustration, so that it gives the desired intuition in the viewer?"  You see, making "nice" or "photorealistic" pictures of atoms and molecules gives completely incorrect intuition about what the illustration attempts to describe, and controlling the information conveyed – by minimizing superfluous information (so cel-shading instead of photorealistic reflections and shadows) and controlling the visual cues that generate the intuition.  There is a lot of empirical knowledge about this in visual arts.  For sonification, I wanted to find methods that would complement the visual information (and thus be useful to those without problems in vision) like "tracers"; there are already quite a lot of ways of describing statistical data, distributions and curves.

I'm personally very prone to visually induced headaches by blinking lights –– I particularly hate bicyclists that use blinking headlights: if I'm tired, and it is dark outside, the blinking ones can induce me to vomit.

It was possible to perceive full colors (which were other colors than the LED's color) with the eyes closed (light from the LEDs through the eyelid), and other visual illusions like patterns depending of the blinking type, some people were reporting sleepiness, others nausea, etc.

Yes; human visual perception being such that if you look at a picture for more than a few seconds, looking at a constant background makes you perceive the original picture in negative.  There are many such effects, some dealing with depth perception (including color), others dealing with structure and outline and persistence (like how you can easily see through a picket fence from a moving car, even though the spaces between the pickets are only a small fraction of the width of the pickers)... and then there is the sleight of hand, or controlling the focus of attention.  To blow your own mind, I recommend you watch the selective attention test from 1999 by Daniel Simons and Christopher Chabris.

[RoGeorge]

To blow your own mind, I recommend you watch the selective attention test from 1999 by Daniel Simons and Christopher Chabris.

He, he, nice one indeed, know that one from when it first came to YouTube.  I failed that test! 

Back then I was not watching YouTube at 2x speed (speed shift was not yet an option), and I remember I was looking at that video only, and paying attention to it.  So I payed full attention and still failed.

That video was one of the clues that consolidated one of my pet theories about what a conscious mind most probably is:

...
The brain is an Oracle, it can predict the near future.  We wouldn't be able to do the simplest thing, like walking a room, without predicting the outcome of each and every step in advance.  This is happening in an automated way, and we are not aware of it.  We think we "see" the surrounding, but experiments tells we rather simulate the surrounding.  The data stream coming from the eyes is only adjusting the simulation so it won't diverge too far.  The brain is continuously simulate what we "see" or "feel" in advance.

Mind is nothing but a driven illusion.
A very brief simulation, in regards to the complexity of the objective reality that surrounds us.
...

And that is what I think self-awareness/conscience is:  it's the continuous simulation, the driven illusion we keep adjusting to match reality.  This ability is not specific to humans only.  Most of the other animals having a brain are obviously conscious and self-aware, too, just like us humans.  I'm not yet sure about more exotic lifeforms, e.g. a bee-hive, or an ant-colony, or a fungi-network.

Oh, and about those blinking LEDs, I could see the colors, too, and I have no aversion against blinking lights.  It was something else.  Back then it was quite trendy, I think it was first introduced by the same guy that did the TVBgone, an IR remote control gadget able to shut down almost any brand of TV, good to shut them off in shops, small bars, hotels lobbies, etc. (back than it was always a TV on somewhere in public spaces continuously brainwashing everybody around, very annoying).  I digress.

Will search again for the timing of those blinking LEDs, and post it for fun (not here, sorry for the OT). 

My interpretation back then was the pulses were producing some sort of a Heaviside steps, and that was full in many other frequencies than the usual, in regards to the normal nervous activity rhythms, so the pulses were ringing in the nervous paths giving birth to colors and patterns.

Would like to test the hypotheses by using pulses with soft edges instead, like a Gaussian with the same energy as the square pulse (because the Fourier of a Gaussian is also a Gaussian shape, so no harmonics).  I would expect for the colors and patterns illusions to not form when the LEDs are blinking with Gaussian instead of square pulses.

[Nominal Animal]

Would like to test the hypotheses by using pulses with soft edges instead, like a Gaussian with the same energy as the square pulse (because the Fourier of a Gaussian is also a Gaussian shape, so no harmonics).  I would expect for the colors and patterns illusions to not form when the LEDs are blinking with Gaussian instead of square pulses.

Perception is more complicated than that, as we can perceive extremely short pulses of light, without "decomposing" them into lower frequency components somehow.

However, there is something there, because soft-edged pulsing (roughly sinusoidal) gives me very little nausea (typically none, except around 10-15 Hz) compared to hard-edged pulsing at the same frequency, while the duty cycle does not seem to matter much (although 50% is the easiest to tolerate).

I suspect the root phenomena is deeper in the human brain, something related to motion perception, instead.  (That could explain the nausea, too, as a dissonance between visual stimuli and the vestibular system in our ears.)

[RoGeorge]

we can perceive extremely short pulses of light

Have you noticed how much more sensitive is the peripheral vision in regards to faint-light blinking or movement?

Especially during night, even the faintest light makes you turn the head there (i.e. the blinking blue light of the heartbeat "alarm armed" LED in a car parked 100m away somewhere at the lateral edge of the field of view), only to see nothing when looking directly. 

For some strange reason only the peripheral vision can spot such a faint blinking light.  When looking directly to the car, you see nothing blinking.  Then when you turn the head away, you see it blinking again, very annoying! 

[RoGeorge]

About the OT, opened a new project:
https://www.eevblog.com/forum/projects/visual-hallucinations-induced-with-two-blinking-leds/


Regarding the 4 transistors controlled the same, one for each LED

While they are usually well enough matched, I just want the option to slightly adjust the balance between the LEDs, in case they are not.

Indeed, LEDs from the same batch are so well matched that, in commercial LED light bulbs, the LEDs are often connected in parallel without any balancing resistor, and there is no visual difference in intensity, not even at very low currents and over a wide range of temperatures.

A visual example to illustrate how well they match in practice, 3 groups of 2 parallel LEDs (from a former mains light bulb).  Note how each of the two LEDs that are in parallel looks identical in light intensity even at very low light levels (where the eye is able to spot any differences more easily):



Just in case the LEDs might not be very well matched, a small balancing resistor in series to each LED might be enough for balancing, while they are all controlled from one transistor.

[Nominal Animal]

we can perceive extremely short pulses of light

Have you noticed how much more sensitive is the peripheral vision in regards to faint-light blinking or movement?

Yes, this is well known phenomenon.  (I can see the flicker in most fluorescent lights here (50 Hz mains, so 100 Hz flicker) in my own peripheral vision, and it annoys the heck out of me.)

The entire vision setup in humans is weird.  First, we have four types of receptor cells in our retina, one dedicated for low-light conditions ("scotopic", most sensitive in the blue-green region, around 555 nm), and the three for color perception.  Second, the retina is not uniform: there is a rather small spot of dense receptors (fovea), with fewer receptors outside that spot.  The brain automagically keeps track of objects and movements outside the sharp vision cone, so that we think we perceive everything in sharp focus.  Third, the eyeball is not static nor does it always move smoothly: they "twitch" (saccades, or saccadic movement) to compensate for the slowness of the visual processing in the human brain.  (Human eyes are only attached to six muscles each, plus the nerve bundle to the brain.)  Fourth, while those color receptors are for red, green, and blue, our brains actually processes them somewhat as "red-green" (blue-green to magenta) and "blue-yellow" (yellow-green to violet).  The color receptors also react faster than the low-light receptors.

So, there are at least three completely different facets to human vision: physical, due to properties of the eyeball and cells involved; visual, due to the processing chain in the brain; and perceptive, due to the 'interpretation' of the former two in the brain.

It is notable that many devices have switched from blinking leds to "breathing" leds, with roughly sinusoidal intensity curves, exactly because the abrupt on-off-on transitions are so taxing/annoying to us humans.


This does tie into this thread in that it explains why some humans are more susceptible to being annoyed by certain visual effects, but more importantly, why one might choose a different backlighting method depending on the purpose of the device.

In my case, this display is intended to convey state information (about current wired and wireless networks) to non-technical users in a relaxed, usually not very brightly lit common room or media room or such; definitely not an office.

If I were to use PWM, I'd need it to be above 1 kHz or so, just to be sure I avoid perceptible flicker.  However, I also hate audible coil/capacitor whine, so I'd prefer to push the frequency above human hearing, to say 30 kHz or above.  At those frequencies, we already have effects like even a very short pulse on the transistor base yielding a relatively long pulse of conductivity, which limits the minimum on-time (as I discovered earlier in this thread).  Using the transistors in their linear region does mean the difference between the supply voltage and the forward voltage of the LEDs is wasted as heat (in the transistors and the resistors), but avoids the aforementioned issues completely, plus makes it easy to control the LED current with a control voltage basically linearly; which in turn allows me to use a DAC for fade-in and fade-out, and a physical pot to control the steady-state brightness, as that seems to me to be the optimum for this particular use case.

The one quirk in this design is that I specifically wanted to be able to easily adjust the current to each LED (if possible), in relation to each other, to account for any imbalance in light output at the same current between the LEDs.  Here, the 68 ohm resistors will actually be pairs of 0805/0603 resistor footprints, so that if I do need a bit of adjustment, I can do so by putting two resistors in parallel.  For example, a 100 ohm in parallel with a 200 ohm one yields 1/(1/100+1/200)=200/3≃67 ohm.  I do not expect to need to do this, but since I'm making only one or two of these, I want the ability to do so if needed, rather than having to order another display (which typically takes weeks to arrive).