Yeah EDN sucks, and they broke most of their links a long time ago anyway.
Seems there's a copy here (likely a dubious matter, but it's their responsibility not ours..):
https://www.radiolocman.com/shem/schematics.html?di=532995It's just an integrator with reset, a S&H to follow it, and a series of switches to make it do the thing.
Note that the level will decay slowly over time, between samples; charge is stored on C1 while the input is low, I think, and C2 when high. The use of CMOS switches and FET type op-amps makes this reasonable up to pretty fair duty cycles (like, using PP, PPS or C0G capacitors, and for less than 0.1% droop, a duty cycle below 1/1000 is probably fine, but maybe not 1/millions). For unlimited time between samples, better to also use the pulse to interrupt an MCU and run an ADC sample, thus preserving the information forever as digital bits free from analog decay. (Or, you know, anything made of equivalent logic, but the MCU is likely the most handy/accessible/compact solution. Minimal hardware would be something like, a parallel-output ADC with bus latch; those are bulky, and uncommon these days.)
Note also that minimum pulse widths must be observed; there will be some error in propagation time between switches (especially the one configured as an inverter gate), and the switches can only discharge/set the capacitors as fast as their resistance and switching times allow.
The mismatches should tend to manifest as nonzero intercept errors (charge injection, etc.), so can be subtracted out with a calibration step. Check that it doesn't vary with supply voltage or operating temperature, or if so, add those to the calibration table.
You do have the advantage that, resolution is perfectly continuous -- limited instead by noise performance (sampling aperture basically), and whatever other sources of interference might apply. Whereas measuring your pulses with a typical MCU might give 7-16 bits accuracy, depending on hardware and clock speed. Which, the upper end of that is pretty good honestly, but if you're constrained for other reasons (very cheap MCU?) you'll be stuck by low clock rate, and an analog method might not even be that impractical, in a real-world application. (It likely won't be fantastic in current consumption, though -- op-amps of reasonable speed (say >10MHz GBW?) tend to consume a few mA. But this could be mitigated in a number of ways, too.)
Tim