From: legg@nospam.magma.ca   
      
   On Sat, 10 Sep 2016 12:19:34 -0700, Big Bad Bob   
    wrote:   
      
   >On 09/07/16 14:43, mick so wittily quipped:   
   >> On Tue, 06 Sep 2016 11:22:48 -0700, Big Bad Bob wrote:   
   >>   
   >>    
   >>> 3.  B+ voltage has a tendency to have VLF 'instability' in it.  As an   
   >>> example, let's say you use a 10-bit A:D to regulate the voltage,   
   >>> something like an ATMega processor might have built-in.  That means that   
   >>> a 350V power supply is off by at lest 0.35V no matter what, and probably   
   >>> twice that (or more) in actual practice.  The voltage will constantly   
   >>> vary around that inaccuracy, at a frequency that is related to the   
   >>> filtering capacitors, etc..  Let's say 2Hz.  I've seen that.   
   >>>   
   >>    
   >>   
   >> Surely, if your PSU has VLF instability it's been designed wrongly?   
   >   
   >nope.  it's inherent with switching power supplies.  It's due to the   
   >inaccuracies of the A:D converter.  10 bits is 1024:1.  So if you have a   
   >max volts of 450V (let's say), and you scale it so that 0-450 is 0-1023,   
   >then you end up with an inaccuracy of 0.45V.  THEN, the switcher is   
   >going to measure that voltage, and adjust output to "match".  All   
   >switchers do that, FYI.   
   >   
   >You would need an extremely sophisticated [and expensive] design to   
   >compensate for this, such as a PWM output that generates a reference   
   >voltage, and a differential A:D to compare the reference to the actual,   
   >etc. etc..  I've done that for other purposes, but not this one.   
   >   
   >You are still subject to A:D errors, regardless.  The best you can do   
   >with having a reference voltage is lower the amount of VLF, but it will   
   >NEVER actually go away.  You also have to take into consideration the   
   >transient response curve and the potential for overshoot, things like that.   
   >   
   >If it becomes more expensive than an iron transformer with LC filtering,   
   >it's not practical any more.   
   >   
   >   
   >you know, 'correct way' is rather arrogant.  the CORRECT way [in my   
   >view] is the method that uses the FEWEST components, is easiest and   
   >cheapest to build, and has the highest amount of flexibility.  That   
   >would exclude using up one of the PWM outputs to generate a reference   
   >voltage, when they're already being used for "other things"...   
   >   
      
   The bit error occurs at the conversion rate, well above the cutoff   
   frequency of the pwm filter. There's no reason why this should   
   generate a low frequency drift.   
      
   RL   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)