How to improve this precision rectifier by further decreasing cross-over transition times?

Question

The Circuit

The rectifier comes from TI's TIDU030: Precision Full-Wave Rectifier, Dual-Supply application note.

My "benchmark" used as a baseline for evaluation is as shown below, powered from +12V/-12V, with a 4u6/50V tantalum across U2:4-U2:8.

^{simulate this circuit – Schematic created using CircuitLab}

R2=R3=R4, and the rectifier has gain of +/-1. The architecture is nice as it keeps both op-amps operating closed-loop, without saturation.

Anytime U1's output polarity changes, it has to slew two diode drops. This introduces artifacts near zero. The duration of those artifacts grows as the amplitude of the AC signal on the input decreases. This is mentioned and measured in the application note. It's one of the drawbacks of this architecture.

The transition waveforms I've measured for a 1kHz square wave of 1Vpp, 100mVpp and 10mVpp amplitude, are:

Note that the lowest amplitude waveform has a 10us/div time base.

The simulated waveforms from the app note look quite similar:

As a partial fix, D1 and D2 could be RF Schottky types with lower forward voltage drop, decreasing the slewing U1 has to do on zero crossover.

The rectified 1Vpp sine wave:

The character of aberrations in the rectified signal is very similar to that shown in the app note for 1kHz 50mVpp rectified output:

Note the time scale: The shape/slew rate of the aberrations is same in both circuits, in spite of the app note using quite a bit faster op-amp.

What both circuits share in common is the 1N4148 diode.

It seems that this circuit under-utilizes the op-amp.

I had to use averaging to obtain the traces, just as the authors of the app-note had to: otherwise, modern scopes without preamps are too noisy for such pretty pictures.

This is to assure anyone trying to reproduce the plots: without averaging they look like a big noisy mess, especially at the 20mV/div and 5mV/div.

The Question

Is there some way to fundamentally mitigate the effects of forward voltage drop across the diodes (making it effectively "very small"), so that the transition times would be improved, essentially turning the diodes into ideal diodes?

I'm looking for improvements that don't change the op-amp type. Those improvements will also improve performance with better op-amps, by lowering the slew-rate requirements of the op-amp to be more in line with what the signal itself requires, rather than diode slewing.

I've been experimenting with adding synthetic voltage sources to "undo" the voltage drop on the diodes, but perhaps there are other known approaches to such problems.

I assume that there may well be several ways to do it, e.g. depending on the supply vs. signal voltage span, etc.

Answer 1

I have thought of these methods. You need to experiment to see if any of these is good enough for you.

• Use a comparator and switch to switch a signal depending on whether it is negative or positive. There are a few signals to switch that may achieve the purpose, for example switching between the normal and inverted inputs, or switching one of them on or off with twice the amplitude. There are chips like the AD8037 with comparators and an op amp designed for this purpose. Read its datasheet for more about this method. Figure from datasheet.

• Convert to digital with an ADC, rectify digitally, and convert back.

• You can make a half-wave rectifier by using a CMOS rail to rail output opamp whose negative power rail is connected to ground. Then sum with input to get a full-wave rectifier. CMOS opamps don't suffer from saturation, so they may recover faster. But I don't know whether this method really works. If any of you tried this, please leave a comment.

^{simulate this circuit – Schematic created using CircuitLab}

• You can switch an opamp on or off with the comparator method if the opamp supports gain setting or being switched off. The late Jim Williams used the LT1228 to measure the settling time of 20-bit DACs, as described in Application Note 120.

• In your original circuit, the problem is that the opamp slews too slowly when its output is near ground. When an opamp doesn't have enough output capabilities, maybe a buffer can help. There is an inexpensive buffer that has high slew rate and gain bandwidth product, and better yet has high gain when output is around ground. It is the CMOS digital inverter. Unbuffered ones like the 74HCU04 have only one stage, making its gain fall off with one pole, which means they are about as easy to compensate as opamps. But be aware that these have high output impedance in the linear region, so add a buffer at the output of the entire precision rectifier if there is a large load. This usage of CMOS inverters is described in The Art of Electronics: The x Chapters.

• Wire up a simple opamp follower. Connect a resistor from its output to ground. Depending on input polarity, current through this resistor is supplied by the either VCC or VEE of the opamp. That means VCC and VEE are half-wave rectified versions of the input if you regard them as current outputs. Make sure the supply pin is connected to a low impedance like the emitter of a transistor to prevent nasty problems.

Answer 2

Ideal diode controllers switch on FETs, which requires slewing much more than 0.7V usually.

For this to work fast, ultimately you again need some kind of fast-slewing op-amp.

You might aswell replace the op-amps in the original circuit with faster types to solve the same problem. Schottky diodes also help if you're after smoother crossover.

Answer 3

Using opamps (which are loop compensated) and diodes to switch loops will always have some type of unnecessary delay as some signal transitions through 1 or 2 diode drops.

Instead, use 2 opamps: One as a buffer (+1); the other as an inverter (-1). Each opamp will remain linear and in closed loop. Use a comparator (with small hysteresis ?) to select the +1 or -1 signals depending on the polarity of the -1 output. Be careful that the BW of the +1 will be higher than the -1; it might need some adjustment.

Answer 4

Effects of Diode Type

The diode type does affect the results.

There are other tweaks to be made, and I'll be editing more results into the answer as I get more measurements done.

1N4148

This is a small-signal switching silicon diode, and is the benchmark.

1kHz square wave transition transient

Traces for 1.0Vpp, 0.9, ..., 0.1, 0.09, 0.08, 0.07, 0.06, and 0.05Vpp.

1N5711

This is an RF microwave rectifier Schottky diode.

1kHz square wave transition transient

Traces for 1.0Vpp, 0.9, ..., 0.1, 0.09, 0.08, 0.07, 0.06, and 0.05Vpp.

1kHz sine and triangle wave transients

3BX81B B-E junction

This is the base-emitter junction of a Germanium NPN transistor.

1kHz square wave transition transient

Traces for 1.0Vpp, 0.9, ..., 0.1, 0.09, 0.08, 0.07, 0.06, and 0.05Vpp.

It is interesting that the transition delay is getting shorter with decreased peak-to-peak voltage.

1kHz sine and triangle wave transients

These junctions are much leakier, though, and introduce significant distortions, including 20% amplitude error: