Precision peak detector

The two types of peak detectors I am familiar with that are easy to use are log amps and diode detectors.

Log amps have a large dynamic range, which you apparently want in your application (1V to 1 mV =3D=3D aprox 60 dB dynamic range). Unfortunately, the output is lagarithmic, and it may make implementing it in your control loop impossible. If you have a control loop that can tolerate a microprocessor inside the loop (this means you will have a slower loop bandwidth) then this may work. If you are trying to shape each pulse with over the 1mV to 1 V range , the log detector might work, but you will need the reference signal to be a log equivalent to match the expected output from the log amp.

The second method is to use a diode detector. Unfortunately, diode detectors only have about 20 dB of decent linear dynamic range. Using a diode envelope detector may still work if the range of the signal is expected to move very slowly and you can adjust where the detector sits in the range. But if you are trying to shape each pulse with a high loop bandwidth loop, the diode detector method won't work either.

A full wave rectifier, using a fast op amp, followed by a sample and hold driven from the squared off input pulse might be one solution. A fast integrator after the fw rectifier, may be better, if the pulse width is constant. Integrate over the pulse width time, then sample and reset integrator...

Regards,

Chris

1. Do you need it to be that linear when the loop is open? If not, you can jack up the gain and let it hit the rails on the peaks, which makes the design easier. Some control applications work better when the error signal is clipped like that. (Tim Wescott could tell you more about that than I can.)
2. Do you have a timing reference for the pulses? If so, a gated integrator is the way to go. They have lower noise than diode-based peak detectors and have more predictable behaviour. (Op amp/diode peak detectors can have limit cycles and nasty overshoot artifacts.) This is the usual situation in a control system--you just want the feedback voltage to be stable until the next sample arrives. That's a gated integrator (or peak detector) followed by a zero-order hold.

As an alternative, if you know where the peaks are going to be, and they all have the same shape, you can use a sample-hold amplifier such as an LF398 to sample the peak value. That'll give faster loop response than a gated integrator/zero-order hold, but probably more DC offset due to needing a smaller capacitor.

1. Or do you really want a peak detector, i.e. one that will hold the peak value until a bigger pulse comes along, no matter how many periods that takes?

Cheers

Phil Hobbs

It sounds like you want a pair of "peak hold" circuits. Do you know before hand when the pulses will arrive? It makes the circuit a lot easier to make if you know the timing indirectly rather than having to take it from the pulses themselves. The basic topology is this:

U1 -------!+\\ D1 U2 ! >---->!----+-------!+\\ --!-/ ! ! >--- ! ! --!-/ ! ! ! ! ! ! ! ---------+ ! ! ! ------------------------------- ! GND---!!---------+---/\\/\\---o\\----(-V) C1 R1 SW1

There is an extra circuit near U1 to stop its output from moving after the pulse peak. For now we will ignore this.

U2 is often really just a JFET source follower circuit. It can have a gain less than unity because U1 servos the gain and offset out.

D1 is a low capacitance Schottky diode. It needs to be Schottky because charge storage in normal diodes leads to a nonlinear effect.

C1 is an expensive capacitor. Depending on how long it must hold, you likely can use an NPO/C0G ceramic.

R1 becomes sort of obvious when you know how fast C1 needs to be discharged.

SW1 is often just a small MOSFET. Bipolar transistors can also be used here if you are very careful about not saturating it too strongly.

I have made this basic circuit linear to about 12 bits.

scroll down

Yuri, I'm a bit confused by the bipolar' part of the peak detetor. Do you have pulses that may go either above or below ground? Is it just the size of the pulse that is important or do you need to know the sign also?

George H.

Yes, pulse comes from position sensitive detector (PSD) and can be positive or negative. The amplitude is function of light spot displacement from the center of sensor.

What you might want to start with, is an op amp tracker/follower feeding antiparallel diodes (large forward voltage preferred) into a tracking capacitor. When the peaks occur, the op amp output will slew very fast, you can sample/hold from the tracking capacitor when that fast-slew occurs. Just differentiate the before-the-diodes signal, into a window comparator, to trigger the sample event.

A second good approach, would be a Wilkinson A/D converter (which combines peak detect with digital conversion); you'd need two polarities of converter, or to split the signal and digitize both the inverted and noninverted copies with a pair of converters.

So do you know when the pulses are coming? If so a gated integrator (as suggested above by Phil H.) may be your best bet. (use to call them boxcar integrators.) Oh, can you average several of the pulses or do you need to respond to each one?

George H.

And the problem with A/D conversion and doing it digitally is...

Vastly simpler, cheaper, and better.

Depends on the loop performance you need. Digital is a crock for high accuracy feedback loops running at any sort of speed.

And high performance A/Ds are cheaper than, say, a HC4066 and a dual op amp? In what universe?

Cheers

Phil Hobbs

Thanks to all of you for discussion. I've found this old application note

Regards, Yuri.