Thank you for bringing something to the discussion I can understand. I have been an electronics consultant for 30+ years, so these are old friends to me.
The pages you post are absolutely correct and are a classical example of how digital data can be corrupted by electrical noise. However that is far from the whole story...
This has more relevance at MHz and GHz frequencies where the pulses are very short and the difference between a logical high and low is only a small change in voltage. you might want to look up a book on using oscilloscopes for "ringing" and "overshoot" on square-waves. These are caused by stray capacitance and inductance in high-speed stuff - Even the copper tracks on the board cause problems when the speed is high and is why waveguides are used to transmit high frequencies as radio waves, rather than trying to funnel them through copper wire on satellite systems. Ringing and overshoot are present to some extent with all digital signals.
With these problems being ever-present with digital signals, we guard against them. Input gates are typically Schottky types with hysteresis loops to "recondition" the digital signal - Actually outputting a true voltage low when the input signal drops below a certain level and a true voltage high when the input rises just a little. This corrects the "rise time" of the digital signal and proper conditioning will also output a pulse of the correct width to prevent any errors further along. This creates a new digital signal in real time (i.e. on-the-fly). Sometimes even this simple signal processing is insufficient and a further safeguard of error-correction might be written in to the system, using something like error-correcting bits. Here you are correcting not single bits though, but using whole "words" to correct error(s) in the word's individual bits.
So while the pages you link to are quite correct, you have to realize that what they are showing are disaster conditions. To corrupt a digital signal that is running at even 100KHz, requires appalling levels of noise. No electronics designer worth their salt would produce anything like that. Particularly in audio gear. What are the S/N ratio for your pieces of equipment? Actually I bet they are spectacularly low. True that is quoted for the output RMS signal, but if the "upstream" circuitry was noisy, it would also be fed into the amps - So even the other electronics has to be electrically quiet to get your good output S/N levels, Now go back and look at those diagrams and tell me what S/N ratio they represent? My guess is about 50% to begin to cause data corruption...and you see my point, you might get that sort of S/N ratio in data transmissions over microwave and long distances, or at very high frequencies in electronics, but not in audio equipment as the signals are not fast enough and the environment is not electrically noisy enough. Ringing and overshoot are more likely to cause the odd error bit that will need correcting.
Further safeguards are the correct selection of the Nyquist sampling frequency. You already know something about this, even if you have not heard the name before. It is why we sample sound at just over twice the highest audible frequency (20KHz is the typically accepted upper limit, so the minimum sampling frequency for digital audio is usually taken as 44.1KHz if it is to have the full range and be unlikely to have error bits.)
There is something else that has not been discussed, or at least not that I have seen yet, When digital electronics have to switch voltages between zeros and ones, that can cause a small drop in the voltage locally on the board as current is drawn. This can cause problems for nearby circuitry and also be a possible cause of errors creeping in. But you have to remember that this takes so very little current, that even on a battery piece of equipment, there should be no significant drop in voltage as it's circuits turn on and off. The voltage changes and the emitted EMI/RFI can be measured with very sensitive equipment and this is where I think that graph comes from - Sony do indeed appear to have reduced the noise of the cards' circuits a little, but the question is, of what use is that? I personally can not envisage any difference in audio and wonder if it should not be aimed at other markets? Perhaps owners of camera might be more concerned about the odd mis-interpreted pixel, where their much higher frequencies and lower voltages are significantly more susceptible? Although if you know about camera sensors work, you will know the RAW data is full of incorrect pixels and a lot of processing power is actually needed to pull them back into line with the surrounding pixels to make a good picture. Perhaps the odd pixel error there is not going to bother them as they know even the best cameras are full of errors at the sensor level.
Remember that even if a bit was corrupted and did not get corrected, it could be the least significant bit and the sound produced would be off by 1 / (2^16) = 1/ 65536th of the frequency it was trying to reproduce (and that's only at 16 bits per sample, at 24 bits, it's 1/16777216). Assuming the worst possible case again, it was the most significant bit. Then the frequency would be wrong by almost half of it's value (say 1K5Hz, instead of 3KHz.) But remember the Nyquist sampling choice of over double the highest expected frequency - That means that error in the sound, would last for LESS than 1/20000th of a second, or again beyond the normally accepted upper human limit. For me, even the most elite ears there are would have to be listening to low-quality recordings on really cheap gear to have enough errors to be discernible even by them. I will NOT say they are deceiving themselves, because I cannot hear what they can hear. I do not understand how there could be any difference for them to hear in the first place, there is no scientific/electronic basis that I can think of that could produce any difference.
At the end of the day, I keep thinking of placebos - And what is wrong in that? Why have a 200MPH sports car when the limit on the roads is only 70MPH? - Because it is fun! - And that really is my conclusion. If you enjoy it and can afford the premium, then what is wrong with buying ultra-low noise memory cards? Whether there is even any difference to be heard or not is irrelevant, you just enjoy the Brinkmanship and specmanship.
