[SOLVED] High-speed voltage follower

[jackmorrison]

I'm building a specialized LED matrix display for neuropsychology (visual perception) research. The overall display is 64x64 LEDs, built up from 64 8x8-LED modules. These can't be time-multiplexed or PWM driven - the LEDs need to be controlled in parallel, with DC waveforms. We're looking at visual effects down to 250ns.

Each LED can be software-switched between two voltage levels for brightness control (I don't have the budget or room for 4096 current-control circuits), so I'm using a voltage follower circuit to keep the voltage constant regardless of load (number of LEDs switched on). The attached schematic shows the scheme, simplified a bit to show one channel, and representing one fixed + one switched LED. Actually there are 64 LEDs, each with a fast SPDT analog switch (NX3L4053) to select one of two sources. The op-amp is LM6142 (17MHz GBW) and transistor is ZXTN25012EZ (260MHZ fT).

This works except that the response time is too slow, about 2us rise time for a square wave input at LEDV1. Also, with a fixed voltage at LEDV1, switching from no load to one LED on takes about 10us. Switching from one LED to two LEDs is fine, I get a 250ns rise time. Adding a 1M fixed load (between Q5 emitter and ground) reduces the time to turn on the first LED to 3us, but also adds significant ringing to the square wave response. Using a faster op-amp (LT1813, 100MHz GBW) gives faster response, but there's significant overshoot on a rising edge input, and oscillation around 25MHz when driving an LED.

The circuit is built on a 4-layer PCB with ground plane. The farthest LED is about 3 inches from the amp circuit. Timing measurements are of actual LED emission, made with a photodiode and 200-ohm shunt resistor attached directly to a scope probe.

I'm hoping for some advice on how to adjust the circuit for better performance. Ideally, a square wave input (with switches fixed), or a DC level with switching, would give 250ns or faster rise/fall times with minimal overshoot and ringing.

Hopefully I haven't left out any important info. Thanks for any suggestions.

 

[FvM]

Although the feedback compensation can be optimized, you'll should be aware of a transient output response when switching the load. My main concern would be that voltage control of multiple LEDs is far from optimal control.

I would go for individual current amplifiers and all signal switching on the amplifier input side.

 

[schmitt trigger]

"We're looking at visual effects down to 250ns."

??????
This statement does not make a lot of sense. This is way to fast for the eye to perceive. Like 4 orders of magnitude.

 

[chuckey]

As TV and film refresh their picture ever 20mS, a bit faster, say 2mS would be fast enough. In your circuit, why have you C1 ? , it just slows the circuit down. If you arrange your 64 X 64 LEDs in a matrix then 64 switched current sources and 64 switched current sinks to actuate your rows/columns would do the job.
Frank

 

[jackmorrison]

I appreciate the responses. Maybe I wasn't clear enough about my goals, but I'm not trying to build a video display, which would be kind of pointless. The photochemical processes in the human eye operate on femtosecond time scales. **broken link removed** discusses recognition of microsecond-scale shape displays with a previous version of this LED display. Time multiplexing with a row-column matrix is not an option.

R1 and C1 were determined experimentally to prevent oscillation with the current op-amp.

Even if I could find a 4MHz current source and add 4096 of them, how does that help with the overshoot/ringing/response time issues?

 

[schmitt trigger]

Then the best is to use an amplifier specifically designed for video use.
Not because you will be displaying video....They have the fastest response available, low or no overshoot, and are capable of driving 75 ohm loads.

 

[FvM]

R1 and C1 were determined experimentally to prevent oscillation with the current op-amp.

You'll need at least a series resistor to C1 for a better compensation.

Even if I could find a 4MHz current source and add 4096 of them, how does that help with the overshoot/ringing/response time issues?

Mixes respectively confuses several points.
- using current rather than voltage output amplifier is preferable for exact intensity control
- overshoot/ringing is a matter of unsuitable amplifier compensation
- apart from the other points, switching the load of a high speed amplifier will cause unwanted signal transoens in any case
- switching the load impedance will affect optimal amplifier compensation

A specification of required LED current uniformity, current range and accuracy, pulse shape quality would be required for detail design.

I see that up to 4096 channel drivers might be beyond acceptable effort. But it has to be traded off against the performance of other solutions.

 

[mtwieg]

Even if I could find a 4MHz current source and add 4096 of them, how does that help with the overshoot/ringing/response time issues?

A current amplifier should be inherently more stable, since the capacitance of the LED will not cause lag in the feedback signal. A 4MHz transconductance amplifier should be relatively easy.

However you should be aware that the light output of typical LEDs (cheap ones meant as visual indicators) do not respond extremely fast to the applied current (believe me, I've tried to use cheap LEDs for optical communications). You should test your LEDs with a fast photodetector to make sure that the emitted light follows the applied current.

 

[chuckey]

A couple of points, how are you keeping the op-amp from saturating? If the LED is on and the input goes negative, the output of the op-amp goes negative, which via C1 cuts of the input transistor in the op-amp. The output transistor is cut off, but the spurious capacitance across the LED now can only discharge via the 1K, which will be slow.
Frank

 

[jackmorrison]

Then the best is to use an amplifier specifically designed for video use.

Interesting - A Digikey search for "video amp" shows some parts that didn't show up when looking for high-frequency "opamp". I'll experiment with that. Thanks!

- - - Updated - - -

- using current rather than voltage output amplifier is preferable for exact intensity control
- overshoot/ringing is a matter of unsuitable amplifier compensation
- apart from the other points, switching the load of a high speed amplifier will cause unwanted signal transoens in any case
- switching the load impedance will affect optimal amplifier compensation

No argument there. So I see your point that a per-LED driver is a better solution. Unfortunately I don't think it's practical for this system.

I've looked up some articles on opamp compensation and will try to educate myself better. Thank you.

 

[jackmorrison]

Under "video amps" I found MAX4392 (85MHz rail-to-rail). Switching to that, changing R1 to 100 ohms, and adding 100 ohms from minus input to ground, gives me a circuit that does what I need. The 2nd resistor or course changes the circuit to a gain of 2, which is a lot more stable (in spite of the part being described as "unity gain stable").

Once again, thanks for the advice.