You are right, of course! What I meant was:



Which simply reflects the experience. Connecting the cap in the wrong manner leads to a much higher ripple and a considerably lower average V+. When connecting the cap to signal ground the charging current flows also through the input decoupling cap, which is discharged, when the V+ cap is charged. As also the generation of V- results in such a discharge of input cap, with an unsuited supply enormous potential dips can be seen, not necessarily in idle situation but especially with a long RS232 cable connected and during data communication.

The second circuit with the TDA2003 is nearly free from these effects, because the charging goes much slowlier than with the charge pump inside of MAX232. Also, such a doubler would nearly always be connected to a strong supply. But many datasheets don't tell, that a MAX232 should also see a strong supply, or at least an input decoupling cap, which is much higher (about ten times!) than the caps used in the charge pump. Which manufacturer would tell this, making his chip appear worse than others...

Again, on the paper it seems not to make any difference, whether you connect the cap to signal ground or Vcc. In both cases V+ should equal 2 x Vcc. But reality effects make the one methode appear highly unsuited, especially in worst case applications. That's why it's not recommended in datasheets.

Kai