LTSpice LM394 Log Converter

[mc172]

I'm messing around with a log converter (aka log amplifier) in LTSpice and am a bit stuck. I figured that a good place to start would be an application note or something so found this on the LM394 NPN "supermatched pair" datasheet:



The LM394 datasheet can be found here: http://www.sycelectronica.com.ar/semiconductores/LM394.pdf

When I put this circuit into LTSpice I get the following performance:



The red trace is Vout and the green trace is Vin. I have attached a ZIP file containing the LTSpice simulation.

The plot has Vin (V(n009)) as the horizontal scale, which is the input to the log converter and is a 1 Hz triangle wave going from 1 uV to 10 V. I expect to see a straight line (which would be the log of a triangle on a log scale) but I only see that for 3 decades. Has anyone got any idea why I only get log conversion after 10 mV?

I have built other variants such as the circuit in the SSM2212 datasheet and have the same problem - I can only get about 3 decades out of it. I am tempted to just build one and see if it works but I'd like to get the simulation working really.

[Ian.M]

The LM394 datasheet only claims four decades range for that circuit, and elsewhere it notes that most datasheet parameters are guaranteed from 1uA to 1mA Ic, which implies only a three decade range with guaranteed accuracy.

I don't have your transistor or OPAMP models, but doing a .dc sweep over a carefully chosen six decade range:

Code: [Select]

.dc dec V4 30u 30 50using the generic UniversalOpamp2 model with its default parameters and any small signal NPN, gives very close to a straight line for six decades - far better than I suspect you'll achieve in real life.  Remember all LTspice components default to being 0% tolerance, and most to being ideal.

[mc172]

Thanks for that. I just tried what you wrote in the code tags and indeed I get around 6 decades.

I was previously manually commanding a voltage source to generate a triangle, something like this:

Code: [Select]

PULSE(0.000001 10 0.1 0.4 0.05)

Do you know why doing a DC sweep gives different results to giving it a triangle? Not that it matters - I won't do that any more!

[Ian.M]

At the very beginning of the triangle wave, the transient response of the circuit is critically important.  I think it is compromising the result for very low Vin due to the brief period from when the up-ramp starts till it crosses  1mV.

[Ian.M]

Yep, confirmed.  Avoid the too rapid transition through the uV decades by using a behavioural voltage source to implement an exponential function for the input:

Code: [Select]

V=pow(10,7*time-6) and you'll get six decades with acceptable linearity same as the .dc sweep gave.

[mc172]

Thanks Ian, thanks for your time and effort. Much appreciated!