Digital radio doesn’t have to be that bad; most new digital systems (satellite radio being one) are implementing an advanced modulation schema like 8-16-24-QAM or PSM which allows multiple bits per symbol, effectively increasing the bit rate. That, and combined with Viterbi code, approaches the Shannon limit for a channel capacity. Today is a matter of a few chips and a few lines of software, almost trivial.
You are speaking of the physical layer. At logical level, all that is available is 96 kbps. That can be used as a single channel using HE_AAC codec but it rarely is. Instead, the channel is subdivided as such depending on the station programming:
First line for example is two stations, each getting only 48 kbps!
At that rate, AAC codec severely limits audio bandwidth in order to keep compression artifacts in check. To compensate, a technology called Spectral Band Replication (SBR) is used to synthesize missing high frequencies. Modelling of the content is used as encoding time to provide hints to the decoder as to how to extend the bandwidth at playback time.
At first blush, SBR makes miraculous improvement compared to not using it. Without it, lack of high frequencies makes the sound quite dull. However, with a lot of content the extension of high frequencies can be grating and harsh.
Even at 64 kbps, SBR is needed to get to "CD" bandwidth as without it, the bandwidth likely will be 11 kHz.
With bandwidth being fixed with very low buffering opportunity, difficult parts of the content can get butched bad. Vocals often get impacted and sound bad to my ears especially in lower bandwidth channels.
All in all, it is a miracle that we can using advanced modulation cram so much data into the limited power and bandwidth we have. Fortunately mobile internet provides a great alternative. We have unlimited plan so I almost always am using that to stream high rate music and use FM broadcast for news and casual music listening.