The article defines dynamic range as the signal to noise ratio of the sensor, measured using the difference (divider ratio really) between fully saturated sensor and no signal at all (pure noise). How does that tranlate in number of stops? One extra stop is twice as big a ratio (before applying the log formula). Both the counting in stops and the counting in dB allow one to get a representative idea without using numbers that spiral out of sight exponentially (literally!) :? ;-) It's a different, but similar way of counting though.
Well depth must be the level at which the sesnsor is saturated, which in colors would be pure white, but remember the sensor is monochromatic: it's the color filter in front of it, in the Bayer pattern that allows color to be captured. On the other end, the noise level is when nothing but the noise is registered (i.e. pitch black), and it looks to take very few electrons per sensor site... The rest is your typical signal to noise ratio formula, which is indeed the dynamic range, spelled differently.
The arcticle goes on showing how many bits are necessary to quantize that dynamic range with various sensors, and as long as enough bits are used, also shows that the lowest order bit(s) may actually be meaningless that they are just quantizing the noise: of course, the noise is not always represented by full number of bits either, which is a concept that many people have a hard time getting their mind around...
Thanks for this link Chris: quite interesting read instead.