Fast Peak Detector

[PM3295]

This topic came up recently in a discussion. Design an even faster circuit without it being too complicated and expensive.

https://www.eeworldonline.com/perils-peak-detects/
https://www.analog.com/en/resources/technical-articles/ltc6244-high-speed-peak-detector.html
https://www.analog.com/en/resources/design-notes/peak-detectors-gain-in-speed-performance.html

The main problem with these types of circuits is the overshoot of the op-amp. There are various suggestions to deal with this as outlined in these articles.

I have simulated both of these circuits to get a feel of the performance. This may be an interesting topic for members to weigh in with their suggestions and comments.

Note: In the last example, I had to alter the values of C1 and R9 to get more acceptable results.

  

[LithiumOverdosE]

The first schematic is kind of alright but also lacks way of discharging its capacitor (unless the OPAMP has low input impedance, I didn't check out the datasheet).

I made a few fast peak detectors which performed quite well in real life (peak current measurement with rise time < 200 ns and impulse duration of 0.5-5 us), 

So, what you could do with the firsts schematic is the following.

1. Remove diode X4.

2. Add some LPF on the input if you're dealing with noisy signals. Something in the range of for example 33 Ohm / 100 pF should do the trick.

3. Use some fast rectifier diode. BAT42 or even better BAT17 should be adequate.

4. Select sufficiently fast OPAMP with low input offset, short settling times and with sufficiently high input impedance.
Something like OPA365 first comes to mind because IIRC I used them in some of those deisgns, but I'm sure there are other suitable ones as well.

5. Even if you're using R2R OPAMP it is a good practice to use bipolar power supply if you expect you'll be dealing with low level signals near 0.

6. The feedback resistor R2 can be of much lower value to improve the response of the circuit. In this case something in the range of 200-500 Ohm should be adequate enough.

7. Add some additional small feedback capacitor between the output and input pins of the input OPAMP. Something in the range of 2-10 pF should be fine.

8. I'm not sure what kind of signal you're planing to capture and measure but whenever possible you should use as small values of R3 and C1 as possible. Normally, I would expect to see R3 value in much lower range, something like 5-10 Ohm with C1 of much higher value like for example 10 nF.

9. Remove R1.

10. Provide active means of resetting/discharging the capacitor between successive measurements.  Some small MOSFET of BJT in parallel should do the trick but add some series resistance in collector i.e. drain of the switch to limit the peak discharge current from the capacitor.


As far as sims go, you can always do stepping of the feedback resistor and RC and determine the best values for your particular goals. You'll still have to test that in the real life but it should get you close enough.

[PM3295]

I made a few fast peak detectors which performed quite well in real life (peak current measurement with rise time < 200 ns and impulse duration of 0.5-5 us),

Can you show the working circuit you used?

[LithiumOverdosE]

Sure.
I had to draw it quickly from recollection but it should be accurate.

[PM3295]

Thanks. I simulated the circuit with the spice model provided on TI's website.

In the simulation, it is compared against one of the suggested designs. In the first instance, the circuit appears much faster than the reference circuit. The same overshoot problem requires a higher-value  series resistor to absorb this initial overshoot. I removed the input RC filter to get the fastest response signal for comparison purposes.

In the second plot, the overshoot is under control with the higher 33R resistor, but the response time is increased as expected.

The last plot shows the overshoot after the diode.

[LithiumOverdosE]

It all depends in how quick has to be response of the peak detector output and how much voltage offset can you tolerate.

So, one could use fast OPAMP with very short settling time as the input of the peak detector and take feedback directly after the rectifying diode.
Then you could use the second OPAMP only as simple non-inverting buffer and rely on low input/output voltage offset to reduce error.

[PM3295]

I played around with a couple design ideas based around using my favorite CFA option, and came up with this circuit that simulates with a rise time of just under 15 ns. The disadvantage is that it requires a small negative supply -1.25 V on the CFA.  

The design layout will be critical in an actual prototype to match this performance due to the large BW of the CFA where parasitic inductance and capacitance can kill the expectations.

If one can live with the slight ringing directly after D1, even faster rise times could be possible.

[optikon]

Nice work, can you attach the Simetrix project?

Thanks

[PM3295]

Sure. You may need to import the OPA2354.lib file to create the user op-amp required.

If for some reason, it is not transferring with the saved circuit file:
To install drag the lib file into the Command Shell area on the left side. You will see a popup to install the model.

[PM3295]

Here is the project file to run in MultiSim.

To run:

1/ Open project file
2/ Click single-shot on scope
3/ Click green run arrow on upper task bar

[PM3295]

Here is another approach using a fast comparator and a gated current source. This does not suffer from the saturation over-shoot problems when using op-amps. There is some error due to the 4.5 ns switching speed of the TLV3501. The error though is almost a constant value, which could be subtracted later downstream in hardware or software. This means that the rise time can be faster by making R3 smaller to increase the current limit, and subtracting the higher error value constant.

It can be noted that the Vpeak error is 0.15 V for both a 1 V or 2 V pulse input.

Using the faster TLV3601, you can shave off another 2 ns on the switching speed, resulting in a smaller 0.08 V error.

[PM3295]

So I got around getting the TLV3601 and build up a test board.

The measured results look very good. There is some disturbance on the measured yellow trace most probably caused by the scope probe on this sensitive node. The green trace is from the function gen terminated in 50 ohms.

Update:
After finding a suitable spring clip for the probe to reduce parasitics, the waveform looks much cleaner.

I also scoped the strobe pin showing rise/fall times close to the data sheet spec of 2.5 ns.

[PM3295]

Did some more work on this and modified the circuit by placing a 2.7 ohm in series with C1. This caused some instability, and I had to place another 1 nF (C4) in parallel with this resistor to quiet things down. This largely reduced the  final settling value error for both 1 V and 2 V pulse inputs. There is still some ringing regardless of my best probing attempts under these conditions. I will have to try using a fet input probe with minimal loading later to see if it reduces the ringing. This node seems very sensitive to probe tip loading.

[PM3295]

So I found a 1 GHz fet probe that only works on our Lecroy scope. This has only 1.8 pF of loading.

After connecting this up, it showed terrible oscillations around 550 MHz! One way of damping this, is to use a suitable ferrite bead. I inserted  two small beads in the leg shown in the updated circuit. This worked very well. The only remaining impulses seem to be coupling through from the strobe node and is of fewer concerns.

The lesson is, if you see strange behavior, your scope's limited bandwidth may be hiding bigger problems!