5G – A new wireless network or computational paradigm?

The second day of the Johannesberg Summit was kicked-off by Jan Rabaey, UC Berkeley. Jan gave an interesting talk with a vision that we should see all devices around us not as passive sensors, but actually as computation and communication resources. As the memory and computational capacity of these devices keeps increasing for at least another decade, this should be a rich opportunity that we should exploit. Jan listed a number of conditions that need to be fulfilled to make wireless access and computational resources abundant.  The requirements on the devices, according to Jan, that are required to make this happen included (somewhat condensed by me) :

• Flexible radios
• Avoid all interference  (achieve by highly directional antennas at high frequencies)
• Considerable memory and computational resources

In the Q&A  I and several others, actually would add another important hurdle that needs to be overcome: improving our capabilities in distributed parallel computing.  Although we over the years have leveraged “Moore’s law” to create more and more computing power and memory (which is great for “linear” and structured problems), we have not become significantly better at managing how to partition large, tightly coupled problems and to distribute the over multiple processors without using lots and lots of bandwidth for inter-processor communication. This, I believe, is why multi-hop radio networks (MANET) have failed again and again over the last couple of decades to provide reliable communication at reasonable performance. It seems that the only reliable way we know,  is to exchange all the available data between all the nodes such that every node can compute the result on its own.  In many dynamic communication applications this often means that this “state information” exchange (e.g. Channel State Information, Network Topology Information,  etc.)  may, in fact, by comparable to or even exceed the “payload” data volumes!  I my opinion,  we need to improve our track record considerably in this field to make the “swarm paradigm” become useful.