Python optimizer with LTspice
Written by Bob Adams Fellow Emeritus Analog Devices Inc. 2023
LTSpice_opt is a Python program that uses an iterative optimization approach to design analog filters. It is designed to be used in conjunction with the popular circuit simulator LTspice. It embeds an LTSpice simulation inside the two different optimizers (Particle Swarm and nonlinear least-squares), which are run sequentially.
It works as follows;
-
The user provides a target frequency and/or phase response in Python, and a circuit topology in LTspice with some initial component values.
-
The user provides a list of which circuit instances and/or parameter values the optimizer is allowed to vary.
-
The optimizer then iteratively adjusts those component values and/or parameters, using a 2-pass approach. In the first pass, a global search algorithm called "particle swarm" is used to find a solution that is reasonably close to the optimum solution. Since global algorithms are not based on computing gradients, this first pass is more likely to avoid the common problem of converging to a local minimum. This is followed by a non-linear least-squares phase where the component values are fine-tuned to find the minimum distance between the desired target response and the computed response. The complete process may involve hundreds or thousands of spice simulations as the optimizer completes its work. The Python program interfaces to LTspice using command-line control and file I/O.
-
Once the optimizer has finished, a new schematic is generated with the optimized component values. During the schematic generation process, each component value is quantized to a user-defined tolerance.
Why is this capability useful? Don't we already know how to design filters?
Traditional filter design uses standard circuit topologies such as Sallen-and-Key or multiple-feedback op-amp based active filters. In these cases, given a "standard" filter shape such a Butterworth or Chebychev, a high-order filter may be factored into 2nd-order sections, and an op-amp circuit can be used for each of those sections. This design procedure is quite straightforward and has not changed for many years.
However, there are many cases where this approach fails;
The desired filter shape is not a traditional shape such as Butterworth or Chebychev.
For example, the filter may need to compensate for some other part of the system that has a non-flat frequency response, while simultaneously attenuating other frequency regions.
The need to reduce power/area by combining multiple filter sections into a single op-amp circuit.
This leads to non-conventional circuit topologies that have very messy closed-loop formulas, and it becomes very difficult to solve for the component values.
The application operates at frequencies where finite op-amp gain-bandwidth degrades the frequency response.
Calculating the effects of finite gain-bandwidth on the frequency response is quite complicated, especially in cases where the gain/phase response deviates from the traditional single-pole model. By running optimizer simulations in LTSpice, the actual target op-amp may be included in the simulation. This yields a solution that inherently attempts to compensate for finite gain-bandwidth effects.
Download all files. Move all the Python ".py" files to a directory where you will run Python. Move the LTspice ".asc" files to a directory where you will store your schematics. See the README.pdf file for detailed instructions on how to run the Python program, and what other Python packages need to be installed.