4

Does anyone know if the basic construction of a log amp (with a diode or BJT providing feedback from output to input) is in conflict with any ability to reach high slew rates?

I've been looking (octopart, digikey, mouser) for a logarithmic amplifier chip with a slew rate of at least 350 V/uS, preferably higher (2 kv/uS would be good). I'm seeing plenty of op-amps with slew rates in the Kv/uS range, and even a few instrumentation amps with respectable slew rates, but log amps with high slew rates don't seem to exist -- what I'm finding is log amp slew rates down around a few dozen V/uS; too slow by an order of magnitude.

Am I on the right track if, instead of looking for off-the-shelf log amps, I instead start with one of the high-slew op amps and use some sort of fast diode? Is that the limiting factor, or is there something else I'm missing?

The application is single-ended, multi-channel, DC most of the time with infrequent and random input pulses ranging from a few mV to a few volts, rise times of a few nS, output same but log scaled. The output gets fed into an A/D converter; one goal of the log scaling is to bring the low-amplitude pulses up into a more useful area of the ADC's range; another goal is to attenuate large pulses to prevent clipping.

(Starting some notes as possible answers to my own question...)

  • Breaking out my Art of Electronics; starting to think that what I'm really looking for is a log converter; since the diode's nonlinear curve is mainly what I'm after, a fast diode plus a high-speed op amp does seem to make sense. I'm worried about the noise from the diode, but don't remember how to quantify that. Will study.
  • There is another SE question that looks related to the idea of the fast diode: How to construct high speed logarithmic amplifiers
MarkU
  • 14,413
  • 1
  • 34
  • 53
stevegt
  • 211
  • 1
  • 4
  • various companies make RF log amps. What precision must you have? – analogsystemsrf Apr 18 '19 at 03:14
  • Spectrum Analyzer should tell you log amps are not limited to low frequencies – Tony Stewart EE75 Apr 18 '19 at 03:15
  • Spectrum Analyzers are superheterodyne receivers. – analogsystemsrf Apr 18 '19 at 03:17
  • High bandwidth log amps use something like a cascaded chain of amplifiers, not just a single amplifier. – Spehro Pefhany Apr 18 '19 at 03:40
  • As Spehro says the high BW ones are usually of the successive saturation type but they don't have good DC characteristics and are intended for AC signals. For a dynamic range of 1000:1 (60dB) why do you need a log amp? – Kevin White Apr 18 '19 at 03:45
  • @KevinWhite This is for one of those hard physics applications where the signal is DC the majority of the time, with a fast pulse from a detector once every few hours or days. The pulse profile matters, including for the low-amplitude pulses, and the log scale would help bring those small pulses up out of the ADC noise floor. I'll edit the question to clarify this a bit better. – stevegt Apr 18 '19 at 05:06
  • @analogsystemsrf I was looking at RF amps as well, but wasn't sure if they were appropriate for a mostly-flat DC signal with random infrequent pulses. Are you talking about baseband log amps, or something else? – stevegt Apr 18 '19 at 05:37
  • 1
    I think your main problem with the straightforward diode/amplifier approach will be that an amplifier stable at high signal levels will be sluggish at very slow signal levels. There may be some literature out there on ways to mitigate. – Spehro Pefhany Apr 18 '19 at 05:54

1 Answers1

2

Progressive-limiting Log amps might satisfy. Log conformance for many of these is +/- 0.5dB and accommodate a 60dB dynamic range.

Since you specify an input voltage source, most accept input voltage, and yield an output that is also voltage, in the ballpark of 20 mV/dB.
A big problem is output bandwidth. The fastest I see (AD8310) has rise time of 15 ns, delayed by 6 ns. However, bandwidth is not dependent on amplitude.

I take it that the occasional input pulse response above a somewhat noisy DC background is of most interest. A low-frequency offset compensation loop works in your favor here. To achieve a good output noise floor and good dynamic range, you might want to consider a high-pass filter at the input.

These log-converters respond equally to negative-going pulses as well as positive-going pulses. This non-linear response doesn't work in your favor - should your unipolar pulse overshoot, the negative-going overshoot is converted to a positive-going tail.

glen_geek
  • 23,591
  • 1
  • 22
  • 50