Need a little design help on my latest project

[Charlie E.]

Hi Guys,
Ok, going to put myself in harms way, and ask for a little design
advice. I have been working on this project for a while now, and it
has gone through several iterations, and I keep having the same
problem!

First, the project: I am designing a small color reader for the
visually impaired. Really simple operation - put it against the
object you want to tell the color of, and it will say "RED" or
whatever the color is. Sensor is simply an RGB LED and a
phototransistor, buffered by some amps, and then digitized by a PIC24.
So, what's the problem? I can't get a stable reading. In normal
operation, this thing will run for about two seconds, and then be
turned off. To test, however, I run it in debug mode for hours. When
I first turn it on, and calibrate it to a white sample, I will get one
set of calibrations. Let it sit for about two minutes, and it starts
to drift. In about half an hour, I will have readings totally off the
scale.

So, why am I baring my soul to ya'll? I need your help identifying
where the gain drift is coming from, and some ideas on how to control
them. I have the schematic here:

http://edmondsonengineering.com/Documents/Rainbow color Reader Schematic.pdf

Basic description - MCU turns on an LED. The phototransistor is first
buffered by a non-inverting opamp with a gain of 2, and the signal is
split. Part goes directly to a PGA where it is first attenuated, and
then the PGA boosts it up. This gives me a calibration control to
deal with difference in output of the LEDs. The original and PGA
signal are added, and this is then applied to another non-inverting
opamp with a gain of 2. I also have one feed before this opamp to an
ADC input on the PIC.

Problems I have already solved:
First, each LED has a different output level. Red needs a gain of
around 2, BLUE a gain of around 5, and GREEN a gain of about 7. The
PGA was added to give me an adjustable gain from around 2 to 14, with
the two different taps into the separate input channels of the PGA.
This gives me 16 different gain levels to play with. Using the tap to
the second ADC channel, it actually gives me 32 different levels.

Right now, RED uses this lower gain channel, and is steady as a rock.
Part of this may be that my VCC is 3.3 volts, and only RED has a
forward voltage below this. Both GREEN and BLUE have forward voltages
of 3.4 volts.

So, potential problems? could it be that repeated use warms up the
GREEN and BLUE LEDs so that they become more efficient? Could the
power supply drift higher as it warms up? Could the opamps drift with
slight changes in temperature? Any advice ya'll can give will be most
appreciated.

Thanks,
Charlie

 

[Charlie E.]

What keeps the LED's at their _constant_ current?

...Jim Thompson

Hi Jim,
Basically, it is whatever the PIC outputs will put out. They are
rated at 18mA, but I suspect that they are putting out a whole lot
less...

Charlie

 

[Joerg]

Hi Guys,
Ok, going to put myself in harms way, and ask for a little design
advice. I have been working on this project for a while now, and it
has gone through several iterations, and I keep having the same
problem!

First, the project: I am designing a small color reader for the
visually impaired. Really simple operation - put it against the
object you want to tell the color of, and it will say "RED" or
whatever the color is. Sensor is simply an RGB LED and a
phototransistor, buffered by some amps, and then digitized by a PIC24.
So, what's the problem? I can't get a stable reading. In normal
operation, this thing will run for about two seconds, and then be
turned off. To test, however, I run it in debug mode for hours. When
I first turn it on, and calibrate it to a white sample, I will get one
set of calibrations. Let it sit for about two minutes, and it starts
to drift. In about half an hour, I will have readings totally off the
scale.

So, why am I baring my soul to ya'll? I need your help identifying
where the gain drift is coming from, and some ideas on how to control
them. I have the schematic here:

http://edmondsonengineering.com/Documents/Rainbow color Reader Schematic.pdf

Basic description - MCU turns on an LED. The phototransistor is first
buffered by a non-inverting opamp with a gain of 2, and the signal is
split. Part goes directly to a PGA where it is first attenuated, and
then the PGA boosts it up. This gives me a calibration control to
deal with difference in output of the LEDs. The original and PGA
signal are added, and this is then applied to another non-inverting
opamp with a gain of 2. I also have one feed before this opamp to an
ADC input on the PIC.

Problems I have already solved:
First, each LED has a different output level. Red needs a gain of
around 2, BLUE a gain of around 5, and GREEN a gain of about 7. The
PGA was added to give me an adjustable gain from around 2 to 14, with
the two different taps into the separate input channels of the PGA.
This gives me 16 different gain levels to play with. Using the tap to
the second ADC channel, it actually gives me 32 different levels.

Right now, RED uses this lower gain channel, and is steady as a rock.
Part of this may be that my VCC is 3.3 volts, and only RED has a
forward voltage below this. Both GREEN and BLUE have forward voltages
of 3.4 volts.

So, potential problems? could it be that repeated use warms up the
GREEN and BLUE LEDs so that they become more efficient? Could the
power supply drift higher as it warms up? Could the opamps drift with
slight changes in temperature? Any advice ya'll can give will be most
appreciated.

As Jim hinted, the first order of business would be to regulate the
current that goes through each LED. A resistor that drops only very
little voltage isn't going to cut it.

 

[Charlie E.]

As Jim hinted, the first order of business would be to regulate the
current that goes through each LED. A resistor that drops only very
little voltage isn't going to cut it.

So, my first stupid mistake was using a VCC of 3.3 volts with LEDs
that required 3.4! Not an easy problem to fix.

Makes me really wish that the PSOC1s had built in debug capability.
They could run at 5 VDC...

Charlie

 

[Joerg]

So, my first stupid mistake was using a VCC of 3.3 volts with LEDs
that required 3.4! Not an easy problem to fix.

Makes me really wish that the PSOC1s had built in debug capability.
They could run at 5 VDC...

It can be fixed. You'd need three step-up (boost) converters and a small
resistor in front to each and then regulate. If it has to be cheap you
could try to regulate with the uC. Alternatively one step-up and three
linear ones but LDOs are often rather buggy and if you boost to 6V or
more for a proper linear regulation you'll waste a lot of power.

Yet another option: Supply the PIC with 3.3V or whetever it needs and
the LED with more. But even with several volts of headroom a resistor is
not a constant current source. However, you can make a uC-controllable
current source with two transistors and two resistors per LED, no big
deal. AoE figure 2.25, just flipped around using NPNs and R1 goes to a
PIC port pin.

This will improve things massively. If that's still not low-drift enough
you can use an opamp current source.

 

[Joerg]

It can be fixed. You'd need three step-up (boost) converters and a small
resistor in front to each and then regulate. If it has to be cheap you
could try to regulate with the uC. Alternatively one step-up and three
linear ones but LDOs are often rather buggy and if you boost to 6V or
more for a proper linear regulation you'll waste a lot of power.

Yet another option: Supply the PIC with 3.3V or whetever it needs and
the LED with more. But even with several volts of headroom a resistor is
not a constant current source. However, you can make a uC-controllable
current source with two transistors and two resistors per LED, no big
deal. AoE figure 2.25, just flipped around using NPNs and R1 goes to a
PIC port pin.

This will improve things massively. If that's still not low-drift enough
you can use an opamp current source.

P.S.: If the cathodes are tied together the last idea only works if you
leave the current source up against positive, like in the AoE example.
Then drive either with a 5V uC or a logic level FET and add a resistor
from base to positive supply.

 

[Charlie E.]

Can't find any PIC with higher VCC? Most AVR can run up to 5.5V.

Ah, there's the rub. I need a PIC24 for the speech output routines,
and those are 3.3VDC only. A PIC18 might work, if I could get the
speech routines to function, but would require a whole different set
of development kits... :-(

Charlie

 

[Charlie E.]

P.S.: If the cathodes are tied together the last idea only works if you
leave the current source up against positive, like in the AoE example.
Then drive either with a 5V uC or a logic level FET and add a resistor
from base to positive supply.

Joerg,
Thanks for the advice. Yes, I had been concentrating so hard on the
amplifiers I never really considered the LEDs. In my mind, they would
just 'work' and I could then adjust accordingly. Didn't realize that
they would vary that much. Will have to look at maybe adding a higher
voltage, and go with the constant current drives for them. This does
need to be pretty accurate!

Charlie

 

[Joerg]

Joerg,
Thanks for the advice. Yes, I had been concentrating so hard on the
amplifiers I never really considered the LEDs. In my mind, they would
just 'work' and I could then adjust accordingly. Didn't realize that
they would vary that much. Will have to look at maybe adding a higher
voltage, and go with the constant current drives for them. This does
need to be pretty accurate!

Where is your VCC coming from? Regulator? If so, what's the minimum
voltage going into that regulator? If it is a battery that won't drop
below about 4.5V and has low load ripple (low source resistance, added
capacitors) fixing this part of the circuit would become fairly simple.

 

[Charlie E.]

Charlie E. wrote:

Where is your VCC coming from? Regulator? If so, what's the minimum
voltage going into that regulator? If it is a battery that won't drop
below about 4.5V and has low load ripple (low source resistance, added
capacitors) fixing this part of the circuit would become fairly simple.

Hi Joerge,
I only have two AA batteries, so voltage is only about 2.5-3.1 volts.
That was why I added in the power supply, to try and stabilize that
voltage. Most of the parts were pretty power tolerant, but I figured
(somewhat correctly) that the LEDs would be pretty voltage sensitive.

What do you think of this idea? Take an LED driver chip, like an
LM3519 to do the voltage step up and current control, and then three
fets to switch that current to each of the LEDs. Means a chip, a
small inductor and schottkey, a couple of caps, and three fets.
Shouldn't take up too much board space or budget...

Charlie

 

[Charlie E.]

Don't you want them to stay at a constant current?

...Jim Thompson

Yeah, I realize this now... ;-)

this is my first foray into photometry. Wanted something cheap and
easily reproducible. I just 'assumed' that an LED would give the same
output with the same drive...

Oh well, I guess the boss will just fire me!

Charlie

 

[Joerg]

Hi Joerge,
I only have two AA batteries, so voltage is only about 2.5-3.1 volts.
That was why I added in the power supply, to try and stabilize that
voltage. Most of the parts were pretty power tolerant, but I figured
(somewhat correctly) that the LEDs would be pretty voltage sensitive.

That will require switch mode conversion, no other choice.

What do you think of this idea? Take an LED driver chip, like an
LM3519 to do the voltage step up and current control, and then three
fets to switch that current to each of the LEDs. Means a chip, a
small inductor and schottkey, a couple of caps, and three fets.
Shouldn't take up too much board space or budget...

Nope, it ain't quite that easy. It doesn't have an external sense
resistor and, consequently, the "accuracy" to which it holds the current
is really horrid. Look at the Iout versus Vin, that's just not good
enough. If you want to use a chip (or three) you need to find one with
at least and external Rsense.

It is usually easier and less expensive to boost that voltage from the
two AA cells to 5V and add the analog current source circuits I
mentioned in my other post (one per LED section). The PIC could be
supplied directly from the AA cell if it's happy with 2.5V.

 

[Charlie E.]

That will require switch mode conversion, no other choice.



Nope, it ain't quite that easy. It doesn't have an external sense
resistor and, consequently, the "accuracy" to which it holds the current
is really horrid. Look at the Iout versus Vin, that's just not good
enough. If you want to use a chip (or three) you need to find one with
at least and external Rsense.

It is usually easier and less expensive to boost that voltage from the
two AA cells to 5V and add the analog current source circuits I
mentioned in my other post (one per LED section). The PIC could be
supplied directly from the AA cell if it's happy with 2.5V.

You are probably right. One problem is that the opamps for the
phototransistor (ambient light sensor) should also probably stay on
the battery, to prevent overdriving the ADC inputs for the PIC. There
are the other LED drivers with an external current sense resistor, so
will look into the tradeoffs involved...

Thanks!
Charlie

 

[Charlie E.]

Maybe 3xAA (or 4xAA), run uP off of two cells, then some little Mickey
mouse OpAmp mirror games to hold the LED currents constant?

Does uP have an on-board reference?

...Jim Thompson

Object is to have this as light and small as possible. Using AAs
instead of AAAs only because Serpak has their little M-6 enclosure
that fits right. Would weight too much with 3 or 4 AAs...

Charlie

 

[Charlie E.]

Could you run the whole thing from a 5 volt supply? Regulate that
down to 3.3 for the PIC and have the 5 volts for the LED's. (making
some sort of constant current source.)

George H.

Yes, that is pretty much what Joerge is suggesting. It looks like I
will pretty much have a two voltage system - 3.3 for the PIC and the
detector, probably also the speaker amp, and then 5.0 for the LEDs.

Charlie

 

[Joerg]

You are probably right. One problem is that the opamps for the
phototransistor (ambient light sensor) should also probably stay on
the battery, to prevent overdriving the ADC inputs for the PIC. There
are the other LED drivers with an external current sense resistor, so
will look into the tradeoffs involved...

The opamps can be handled with series resistors and, if necessary, BAV99
double-diodes. But if you have suitable opamps you might as well run
them off the PIC rail.

I'd really consider just one li'l boost converter that makes 5V. Tons of
those available. Then current sources for a clean control.

 

[Jon Kirwan]

Well, there's drive and there's drive. In the case of an LED, the drive
is current. And even then, it's not generally reasonable to expect that
a power conversion device (like an LED, that transforms electric power
into light) is really at all precise. Use an LED, or a motor, or a
solenoid, or whatever, 'in the raw', and you have to expect it's
performance to vary. The more efficient it is, and the less expensive
it is, the more you can expect it to vary.

This is a very important point. I've used LEDs where
precision was important in several different uses, where we
hoped for better but have learned a little in the process.

In one, we had them thermally stable (heated, by design) and
very precisely current controlled (better than 0.1%) and
still needed a long, 2-day bake-in testing process to cull
the "bad ones" which drifted like hell. We'd select only
those which showed a very gradual change near the end, but
that was perhaps 2% or so? The rest were tossed out. The
results were excellent, after that, and they definitely
served their purposes well. But ...

In another, which I'm currently doing something to correct
right now, the exact same part number from the same
manufacturer "worked great" with about 50% or better passing
manufacturing testing and then, a few months back, suddenly
that rate dropped to about 1% or so... and shipping nearly
came to an abrupt halt. Fixable, but parts change and if you
don't buy everything you need "today" you might find yourself
in deep trouble "tomorrow" if you depend upon specs that
aren't front and center on the data sheet. We didn't have a
choice here, as the whole thing is basically a bleading edge
FAB device. But still ...

LEDs, driven as well as you may wish, current wise, aren't
what you call "stable." Even adding thermal control doesn't
eliminate drift. Not even close, for some needs.

I think Charlie's method needs to be functional in the face
of, at best, modest current control, led variability, and
drift over time and ambient temps if it is going to be
practical. I think that can be done. But I think some
mathematics will be required here. I'd start with the idea
of a linear calibration matrix, 3x3 in size, created by a
quick calibration step and inverted and then applied to the
data he gathers from then one. Each measurement on his
system is a 1x3 vector. I've used such methods with
completely uncalibrated RGB LED products from OSRAM, to make
excellent use of them that is repeatable and cheaply
calibrated in less than a few seconds.

Jon

 

[Jon Kirwan]

then one

then on.

 

[Charlie E.]

The opamps can be handled with series resistors and, if necessary, BAV99
double-diodes. But if you have suitable opamps you might as well run
them off the PIC rail.

I'd really consider just one li'l boost converter that makes 5V. Tons of
those available. Then current sources for a clean control.

Actually, just found the MCP1252 series - switchable 3.3 or 5.0 boost
converters (charge pump) that don't need inductors! Two of these will
be cheaper than the inductor based solution I have now, and give me
both rails regulated...

Charlie

 

[Joerg]

Actually, just found the MCP1252 series - switchable 3.3 or 5.0 boost
converters (charge pump) that don't need inductors! Two of these will
be cheaper than the inductor based solution I have now, and give me
both rails regulated...

Careful, the 5.0V version doesn't have any oomph below 3V input and
AFAIK it quits once Vbat drops below 2.7V. I'd use a real boost
converter that can comfortably cover your battery voltage range all the
way to the end of discharge. That one little inductor isn't going to
kill ya