Tip of the Week: Speed up SpectreRF noise sims

This tip describes a way to dramatically speed up SpectreRF periodic noise sim's while including real bias current noise.  (The method also works with voltage noise, separate instructions are at the bottom of the message).

To be accurate, SpectreRF noise simulations should include the "real" bias current noise (as opposed to ideal current sources).  However, adding the real bias network (bandgap, bias current distribution, etc.) to your sim schematic can cause SpectreRF to slow to a crawl or even crash with insufficient memory, because of the increase in size and complexity of the test circuit.

A way to overcome this limitation while still using real bias noise is to use ideal current sources, while making use of the "Noise File Name" parameter that is available on these sources.  This parameter causes the source to read a text noise file which is then included as part of the noise simulation.  Instructions follow for generating the text noise file.

The easiest way to do this is to make a separate sim schematic with just the bias network feeding into ideal sources as loads.  (Note, step-by-step instructions that follow refer to the 5.1.41 release of IC tools and AWD waveform environment.)  One by one for each current value needed, do the following:

• Run a regular ac noise sim (SpectreRF is NOT needed for this step, since the bias network is not time-varying);
• Plot the SQUARED current noise through the ideal source load.  IT IS IMPORTANT TO USE NOISE POWER (I^2/Hz) rather than noise current ((I/rtHz), otherwise the table, when read, will give you back wildly exaggerated noise values. 
• Capture the waveform in the calculator using the calculator's wave function
• Use the calculator's printvs function to turn the wave into a text file.
• Use the print function in the popup table form to save the text data as a file.  IMPORTANT:  First convert to scientific notation using the Expressions-->Display Options form.
• Use your favorite text editor to strip out the three header lines that will appear at the top of the file you've saved.  (The file needs to be just the two columns of data, nothing else.)
• In your SpectreRF sim schematic, instantiate an ideal current source with the desired DC value for the current, and give the path to the noise file you have created in the "Noise File Name" field.
• Do this for each of your different current source values in the SpectreRF schematic, and you're done!
• The paranoid (a good quality in an analog designer) may wish to run a side-by-side ac noise sim with the real bias block and the ideal to make sure you get the same noise output from each (thus avoiding setup errors).
  

This sounds like a lot of work but it only took a half hour for me to do, and I had five separate current source values in my SpectreRF schematic.  The result was a SpectreRF sim that actually ran (quickly, I might add) rather than churning forever and then crashing with insufficient memory.

Sometimes it may be useful to work with voltage noise instead of current noise.  For example, you may have a DC voltage-generating block whose noise contribution you want to model without having to include the whole block in the SpectreRF sim.  In this case, you can follow a similar procedure to:  a) capture the voltage noise power (squared volts) from an ac noise sim of the block; b) convert it to a text file; c) include it using the "Noise File Name" in an ideal voltage source.

- Hugh

Originally posted in cdnusers.org by Hugh