Signetics (subsequently Philips) made a complete PPM RC system chipset for a number of years back in the 80's. The NE544 servo chip the NE5045 TX encoder and NE5044 decoder chip. The chip did exactly what was described above. It would only pass on signals between a preset "window" and discard the rest. It did not make any contribution except that the glitching was between 1 and 2 ms which helps the airframe nothing at all. Its been tried before on orbit radios of the sixties. Its old news. These conditions where you get a few good frames and a few bad frames simply does not happen in practice. Its close to being a on/off scenario.
I designed and build RX receivers and TX systems about 10 years ago using all of these scemes for use in high RF environments at long range . ie target drones for radar ACK ACK. I know what I am talking about. But dont believe me ask any communications engineer that works on actual data links.
Edgar, you have identified the problem exactly. Unless you have a history of good data you cannot eliminate a bad pulse within the "window" defined by simple timers, hardware or software. This scheme relies on the fact that the start of a pulse is valid. it may or may not be, completely negating the "error" detection.
And that is all there is to it. I am sure to dont want to see the mathematical model that proves this.
Andre Baird