I was recently interviewed by IEEE Spectrum for the article: The 5G Dilemma: More Base Stations, More Antennas—Less Energy?
Since 5G is being built in addition to the existing cellular networks, the energy consumption of the cellular network infrastructure as a whole will certainly increase when 5G is introduced. It is too early to say how much the energy consumption will grow, but even if the implementation would be vastly more energy efficient than before, we need to spend more energy to gain more in network capacity.
It is important to keep in mind that having a high energy consumption is not necessarily a problem. The real issue is that the power plants that power the cellular networks are mainly extracting energy from non-renewable sources that have a negative impact on the environment. It is the same issue that electric cars have – these are only environmentally friendly if they are charged with energy from environmentally friendly power plants. Hence, we need to keep the energy consumption of cellular networks down until cleaner power plants are widely used.
If you want to learn more about energy efficiency after reading the article in IEEE Spectrum, I recommend the following overview video (you find all the technical details in Section 5 in my book Massive MIMO networks):
Thank you for the great webinar.
In addition to the SDN architecture you mentioned, one other solution for reducing the circuit power consumption and achieving higher energy efficiencies in lower transmit powers is to implement most of the power consuming signal processing algorithms in data centers in the cloud instead of locally in the base stations. Another way to achieve higher efficiency is to shorten the training phase by incorporating non-orthogonal pilot sequences. This leaves more room for transmitting more data symbols during each channel realization which has the potential to significantly improve the energy efficiency.
The use of shorter non-orthogonal pilot sequences has not received much attention yet in the massive MIMO literature (there are plenty of work on non-orthogonal spreading sequences for CDMA). If you are targeting low spectral efficiency per user, you might be able to get a higher sum spectral efficiency by doing so (we saw this for the machines in the paper https://arxiv.org/abs/1808.09177) but I don’t believe it will have any an significant impact on the spectral efficiency or the energy efficiency. In fact, when using the state-of-the-art schemes such as M-MMSE, you can gain a lot from using longer pilots as shown in my book.