The term “Massive MIMO” has become synonymous with providing massive data rates in wireless networks, but this is not the technology’s only good trait. In this video presentation, which has also been given as an IEEE 5G webinar, I explain how Massive MIMO can enhance future cellular networks from many different perspectives.

5 thoughts on “Massive MIMO for 5G below 6 GHz”

Hi, I have read your paper recently named “Optimal Design of Energy-Efficient Multi-User MIMO Systems: Is Massive MIMO the Answer?.” You paper provided a detail analysis for the circuit power consumption which contain Channel estimation, Coding and Linear Processing. I want to know if this circuit power consumption model can be applied to the systems that combine the massive MIMO technology with the millimeter wave technology?

Thank you for the clarification. And this is a great blog to learn Massive MIMO.

Another question is also about the circuit power consumption in millimeter wave massive MIMO. If I use this circuit power consumption model at millimeter-wave massive MIMO, the huge number of antennas (128*8) and bandwidth (800MHz) will cause huge power consumption in linear processing. According to my calculation results, the power consumed by making matrix-vector multiplication per data symbol is the major part which could be more than 50 watts. This means the linear processing power consumption will be the major part in millimeter-wave massive MIMO. I don’t know whether this conclusion is reasonable.

You mentioned in “Optimal Design of Energy-Efficient Multi-User MIMO Systems: Is Massive MIMO the Answer?” that the computational efficiencies will increase rapidly in the future, so you did not consider the linear processing power in the article. As far as I know, the death of Moore’s law has been announced by the verification of Landauer limit and the influence of thermal noise. The failure of Moore’s law means that the increase of computational efficiency could be slow down and get a bottleneck eventually. What do you think of the linear processing power consumption in millimeter-wave massive MIMO? Should we ignore the linear processing power consumption in energy efficiency analysis?

While I think my power consumption model is fairly accurate, I am less certain of which values that the different parameters should have. The values that you find in the paper were all adapted from previous works in the time period 2010-2015. My new book “Massive MIMO networks” contains an updated set of parameter values and some future predictions. However, one of the reasons that we decided to make the simulation code openly available is to let people run the same simulations using other parameter values.

You are right that we cannot count on Moore’s law forever, but I don’t think that the digital signal processing complexity will be the practical bottleneck (at least not for the time being). Lund University has built a circuit that does the processing that you refer to using 50 mW: http://ma-mimo.ellintech.se/2017/04/12/real-time-massive-mimo-dsp-at-50-milliwatt/

They consider 20 MHz bandwidth, but with 800 MHz of bandwidth it will become 2 W. That is not much compared to other components in the system.

Hi, I have read your paper recently named “Optimal Design of Energy-Efficient Multi-User MIMO Systems: Is Massive MIMO the Answer?.” You paper provided a detail analysis for the circuit power consumption which contain Channel estimation, Coding and Linear Processing. I want to know if this circuit power consumption model can be applied to the systems that combine the massive MIMO technology with the millimeter wave technology?

Yes, the model is rather general and can applied for fully digital implementations in any frequency band.

Thank you for the clarification. And this is a great blog to learn Massive MIMO.

Another question is also about the circuit power consumption in millimeter wave massive MIMO. If I use this circuit power consumption model at millimeter-wave massive MIMO, the huge number of antennas (128*8) and bandwidth (800MHz) will cause huge power consumption in linear processing. According to my calculation results, the power consumed by making matrix-vector multiplication per data symbol is the major part which could be more than 50 watts. This means the linear processing power consumption will be the major part in millimeter-wave massive MIMO. I don’t know whether this conclusion is reasonable.

You mentioned in “Optimal Design of Energy-Efficient Multi-User MIMO Systems: Is Massive MIMO the Answer?” that the computational efficiencies will increase rapidly in the future, so you did not consider the linear processing power in the article. As far as I know, the death of Moore’s law has been announced by the verification of Landauer limit and the influence of thermal noise. The failure of Moore’s law means that the increase of computational efficiency could be slow down and get a bottleneck eventually. What do you think of the linear processing power consumption in millimeter-wave massive MIMO? Should we ignore the linear processing power consumption in energy efficiency analysis?

While I think my power consumption model is fairly accurate, I am less certain of which values that the different parameters should have. The values that you find in the paper were all adapted from previous works in the time period 2010-2015. My new book “Massive MIMO networks” contains an updated set of parameter values and some future predictions. However, one of the reasons that we decided to make the simulation code openly available is to let people run the same simulations using other parameter values.

You are right that we cannot count on Moore’s law forever, but I don’t think that the digital signal processing complexity will be the practical bottleneck (at least not for the time being). Lund University has built a circuit that does the processing that you refer to using 50 mW: http://ma-mimo.ellintech.se/2017/04/12/real-time-massive-mimo-dsp-at-50-milliwatt/

They consider 20 MHz bandwidth, but with 800 MHz of bandwidth it will become 2 W. That is not much compared to other components in the system.

Very thanks for the reply