Log amp - why can't I stabilise it?

The "trouble" with transistors is they have gain - additional gain over the opamp's already high gain, leading to oscillation, as we discuss in our book. It's easy to fix with a capacitor in the right place. But you've got an opamp in the way, driving the capacitor with opamps from both ends. Eliminate opamp 1, and you'll be OK.

I haven't read the Gibbons and Horn 1964 IEEE Transactions on Circuit Theory paper (not lately anyway), but I'm sure the extra amplifier they show, which you used, is not an opamp, but a transconductance amplifier, etc., which the feedback capacitor can deal with.

Oh, and one other thing. Those mcp602 opamps you've picked have degraded bandwidth and closed-loop phase margin into high-Z loads (see datasheet figure 2-14), or when driving capacitive loads. I suggest you add load resistors to the opamp output, e.g., two 4.7k resistors, one to the supply and the other to ground, etc.

Dear All,

a quick thanks for the responses so far...

...gotta go shop for the long weekend, back on Tues and I'll consider your eminent replies fully then,

regards, Colin

Well, with the higher input current rises also the BW and the phase margin decreases. You also have to rectify the input voltage, a negative input voltage is NOT allowed. C is worthless with the follower before, you can just drop opamp1 and the composite gain can be reduced further by increasing the emitter resistor to

try

or possibly

martin

Dear All,

Martin: The price of the AD or similar part is prohibitive - =A310 vs. 30p for the opamps. The Thatcorp part needs at last +/-4V I only have +4V.

Ban: The input does not go -ve, it goes about 100mV below the 1.2V ref. input and TR1 provides a clamp to limit output swing in the opposite phase from that which I am measuring. The huge advantage of a.c. coupling the input is that no offset trimming of these cost effective opamps is necessary to achieve the low currents I need. The frequency response at low levels of a single opamp solution is insufficient with the feedback capacitor necessary to achieve stability at high levels, hence the two amp approach. I don't understand your comment about cutoff current and Ut, could you elaborate? I also don't follow the dif. amp. idea, do you have a circuit? I am used to seeing dif. amps. with matched resistors not with nonlinear elements in the feedback path.

Win: Looks like I'm going to need that paper to get a full circuit rather than a crude sketch. I've tried the resistors, should have thought of that one myself, it's helped in other circuits... .=2E.but not in this case! There's a very slight improvement with a single opamp with a pull up only, with two opamps there's no improvement. I also tried with opamp 1 adding a classic capacitive load enhancement which I'd already got on my test board :) |\\ ---+ \\ | >---\\/\\/\\/- + +--- / | | | |/ | | |---||--+ | | | +------\\/\\/\\/-----+

again no joy.

My current thought is to use a smoothed precision rectifier feeding into a DC coupled logamp. It means I have to pay a =A32 for a zero offset type opamp but the log amp will no longer have to swing anything like as fast,

regards, Colin

Nice ASCII art, but I prefer normal-size caps, and the big resistor rectangles obscure the drawing. Doesn't this look tighter, cleaner:

. .-----------------. .-------------------. . | T2 | | T1 | . | 10p |/ \\| 100p | . o---||-. .--| |-. .-||---o . Vin | | | |> -o GND '-----)-- 33k -o-< | . GND --|+/ | | \\+|-- GND . o------- R3 -------o . | | . o R4 . Vout | . === . -Vt(R3+R4) R2*Ve GND . Vout = -------- ln ------- . R4 R1*Vref

Dear Ban,

ah.....that Ut!

I had only posted the relevant part of my circuit, that which does not go fast enough at low levels. The other half of a transistor pair with reference current and gain stage follows the posted part.

I'll try your topology as it is different from mine, although won't it still be slow at low levels?

regards, Colin

Vbe= -Ut ln (Ve/(Ies*R1)) Ies is the current I'm talking about. It increases 10-fold from 20 to 50°, which gives an almost useless drift.

.-----------------. .-------------------. | | | | | ||10p |/ \\| 100p|| | o---||-. .--| T2 T1 |-. .-||---o | || | | |> -o '-----)--|___|-o-< | R2 GND -|+/ | | 33k \\+|-GND |/ | | \\| | ___ | o--------|___|-----o | R3 | o .-. Vout | |R4 | | Vout=-Vt(R3+R4)/R4 * '-' | ln(R2*Ve / R1*Vref) === GND (created by AACircuit v1.28 beta 10/06/04

As you can see we do not have this term any more, only Vt, which can be neutralized with a +3300ppm/K thermistor.

Much better indeed. And how nice the equations, THX. You really earned the ASCII award here.

Dear Robert,

TIA for the paper, maybe available when I return 'coz....

...I am holidaying for a fortnight, from the w/e, yippee!

Also again thanks for all other assista-nts (or is that -nce?),

regards, Colin Smith

A paper by a physics prof at SLAC 15 or so years ago shows how to get good bandwidth ovst a 9-11 decade range. First, i will see if i can find it, then scan it, and then figure how the hell to send it to abse. So far, i get nasty messages from OithRink stating that it will not send applications to Usenet - and *all* attachment types are considered as "applications".

I read in sci.electronics.design that Robert Baer wrote (in ) about 'Log amp - why can't I stabilise it?', on Wed, 31 Aug 2005:

You may be able (e.g. if it's not in columns) to scan it to Word and then save as plain text. It would make a long article, but ....

It sounds worth the effort.

It is mostly a schematic, so going into Nerd would be counter-productive. But i have to find it first; not in my master archives, so must trek to the "bonded storage"...

I finally had a chance to look in my "bonded storage" for that article about speeding up a log amp over a wide current range. Bad news: did not find it. It was over 20 years ago,, but this is what i remember about the scheme. Used 2 op-amps; the input opamp was primarily for keeping the logging transistor E-C junctioon at zero volts, and the second one was to give the gain for the feedback. Each one had its own compensation, and the resultant Bode plot (as i remember) was something like this:

-----------\\ \\ \\ \\--------- \\ \\ \\ \\ I cannot vouch for the accuracy of this, but this may be useful. The paper was about improved and/or wideband logamps, published by some physics PHD at Stanford Linear Accelerator Center. If anyone knows someone that works there, ask that person to do a physical search in the SLAC library. I do know someone that works there, but he is never available and never calls back; the Pope is easier to communicate with.

Dear Robert,

thanks for looking, 'tis frustrating when you're near enough to smell the solution! I had a search around the SLAC website for some time but could not find a reference to such a paper.

Your description sounds very much like the ciruit with which I started this thread,

regards, Colin

Hello boys, I'm baack

Robert Baer wrote:

I'm looking at the circuit in your original post.

You say it is unstable. What freq is the oscillation?

Is there any significant C loading at Vout? Even a scope probe might be too much.

Mark

Dear Mark,

thanks for the input

The oscillation is 625kHz.

I've today also been able to reduce the oscillation for the first time by the circuit I mentioned here for opamp 1:

AND a series resistor between opamp 2 output and the feedback capacitor.

However so far when I get it stable at higher input currents it is once again as slow or slower than a single opamp solution! I will try that again with scope probe isolation and see if the cap can be reduced,

regards, Colin