Yes well the author also says right above that quote: "Another commonly used normalized measure of signal-to-noise ratio is Eb/N0" So the author appears to be of two minds on the topic. Which I understand, I'm in the same place. I do not think it is really appropriate to call Eb/No the same as SNR and I'm annoyed when sloppy figures show up in IEEE which label an axis "SNR" which is in fact actually "Eb/No". That said it is definitely a ratio of signal power to noise power with normalizations applied and it is unitless just like any given SNR is so you can see why people do that. Define signal to noise ratio specifically and I can then answer your question. The issue is no universally agreed upon definition of SNR. So yes, for a certain set of common definitions of SNR you absolutely can have a real life system with a negative signal to noise ratio. As stated in an earlier post one very common definition of SNR is signal power divided by noise power in a bandwidth equal to the instantaneous bandwidth of the signal. Using that common definition a wide variety of operational systems operate with negative SNRs. For instance a BPSK signal with a coding rate below 1/4 will operate below 0 dB SNR. This is common in satellite communications systems. Direct sequence spread spectrum systems can operate at extremely negative SNRs as well. GPS is probably the most common example but there are plenty of others including WCDMA modes used on some cellular phones, Zigbee networks that connect devices in smart homes and certain modes of WiFi. And also you can define SNR in ways more akin to Eb/No and depending on the fixed scaling factor you end up with in your normalization you may end up with the Shannon limit being defined as either negative dBs or positive dBs. You can see this in the document you linked where one normalization makes the Shannon limit 0 dB while the more common one makes it -1.59 dB. Sorry, it doesn't get anymore concrete than that.