Repository navigation
use dt=0 instead of dt=None (unspecified timebase) as default option for lti systems? #422
Description
I was wondering if somebody could tell me the motivation behind the current convention for default initialization of LTI systems to have dt=None? Using dt=None means an unspecified time base, allowing the system to be interpreted either as cont or discrete time. (I couldn't find any discussion of this but please let me know if I missed it).
A system with an unspecified dt seems to be an unusual object:
- printing out its transfer function shows 's' as the complex variable, suggesting it is a cont-time system
- but isdtime(sys) evaluates to true (but not if strict=True is specified) -- I thought this was a bug at first
- cont and discrete time systems have substantially different behavior, which seems to be at odds with the notion that an object should know how to handle itself
Have you thought about using a default of dt=0 (continuous time), and leave the dt=None case be a special case, to be instantiated only when needed? (Issue #419 made me think of this and is related)
Activity
ilayn commentedon Jun 22, 2020
It's a relic from Matlab with some very old historic reasons but survived under backwards-compatibility flow. Then apparently adopted also here. Internally matlab secretly inserts some default value (that I forgot) whenever dt is needed anyways. My take on this is that user should insert some random value if they don't care because they don't care or defaulys should map to dt=0.01 to be slightly realistic.
murrayrm commentedon Jun 22, 2020
I think I was the person who used dt=None. The motivation was a bit different: if someone didn't specify whether the system was continuous or discrete, we wanted to allow as much flexibility as possible. So, for example, if you combine a system without a timebase (dt=None) with one that has a time base (dt=0, dt=True, dt=0.1, etc) then the first system will assume the time base of the other.
This probably doesn't matter that much, though it is handy for systems that are constants (no states).
The current approach is documented here.
I would like to see a bit more discussion of this before we make a change, since it may break people's code.
sawyerbfuller commentedon Jun 23, 2020
Hi Richard, makes sense that dt=None that could be very useful for systems that are a static gain.
Could dt=None be a special case that is used by default when constructing such systems?
Looks like in matlab that the default is dt=0 (continuous time). with an option to set it to -1 to be unspecified if desired: https://www.mathworks.com/help/control/ref/tf.html#f4-209775_sep_mw_ec86807c-5450-4be4-ae9a-e85042c27676
I couldn’t find any examples of systems using an unspecified time base online after a quick search.
sawyerbfuller commentedon Jun 23, 2020
Does anybody here have code that would break if dt=0 were the default instead of dt=None?
murrayrm commentedon Jun 23, 2020
One other possibility is to set up a package wide configuration variable that sets the default system type. I've been thinking that as we make various changes to defaults and other parameters, we might create a function
use_legacy_defaults(version)that sets the defaults to whatever they were in a given release. That way if something breaks, you use calluse_legacy_defults('0.8.3')and everything works as in the current release.