|
|
| Fig. 1: A typical 5G cellular tower. (Source: Wikimedia Commons. |
From 2015 to 2018, the global mobile network data traffic, as measured in exabytes (1 EB = 1018 bytes), has increased from 20 EB to over 160 EB. [1] This steep rise in transmitted information has driven a growing demand for higher data throughput for cellular base stations. To accommodate this surge, the average throughput of cellular base stations has increased markedly, from 120 MB (1 MB = 106 bytes) under 4G LTE to 1,963 MB under 5G NR. [4] However, the amount of information that one could transfer per channel is not infinite and it always costs some amount of energy.
For a naive example, let's consider a phone call over a single wire. Suppose that the caller can either keep at a quiet tone (denoted as '0') or a much louder tone (denoted as '1'). However, in real life, there is always some noise in the wire such that a nominally quiet tone can sound louder. Thus, '0' can sound similar to '1'. In this case, '0' could be occasionally flipped into '1' due to the presence of noise. A simple way to circumvent this error is to increase the bandwidth. Instead of only using a single line, one could make the call with multiple lines and hopefully most of them agree with each other and are correct. The other way is to shout harder and make '1' sound much louder than '0'. The downside is that it takes more power to make the call.
The example above illustrates the Shannon-Hartley theorem which states how much information could be transferred given a channel bandwidth and background noise level [2,3]
| C | = | B log2(1 + S/N) |
Here, C is the channel capacity in bits per second which tells how much information can be faithfully transmitted, B is the bandwidth of the channel in Hz, and S/N is the signal-to-noise ratio in the communication channel. One of the key ideas of the Shannon-Hartley theorem is that every extra bit of information comes with a cost. Naively, one can gain capacity only by either expanding bandwidth or increasing signal power. Modern cellular networks follow the same logic. 5G achieves higher data rates by accessing wider spectrum and using more sophisticated beam-forming to boost the effective signal-to-noise ratio. With 5G increasing its throughput, the key question becomes whether these gains are achieved with greater energy efficiency or at greater energy cost.
|
|||||||||||||||
| Table1: Power consumptions of typical 4G and 5G macro base stations. [4] |
Let's now imagine there are a bunch of callers trying to make calls using the same communication channel. Compared to one single caller, the channel will appear noisier due to interference from multiple signals like many people talking in the same room. These interfering signals behave like noise, reducing the effective SNR and reducing channel capacity. Modern coding techniques like MIMO full-duplex and NOMA help mitigate this interference, but they cannot eliminate it entirely. [5,6] Therefore, there is a fundamental tradeoff here. More users and more data traffic will lead to more interference and noise in the communication channel. To maintain the same effective throughput, one would need more power for complex decoding or higher bandwidth signal generation. The practical consequence of this tradeoff is partially reflected in the energy consumption of the cellular base stations.
A cellular base station is the place where information exchange happens. It receives radio signals transmitted from user devices and amplifies and relays them to the networks central system, while also transmitting data from the network back to users. Each of these base stations relies on a tower (shown in Fig. 1) where high-frequency radio wave transceivers are mounted to emit and receive signals. Each of these signal exchanges incurs an energy cost. Table 1 compares the total power consumption and data throughput of typical 4G LTE and 5G NR base stations. The total power represents the energy required to operate a base station over time, whereas the throughput indicates the amount of data it can process per second. We can thus use this information to calculate the amount of energy it costs to process and transmit 1 MB of data:
| 5G NR energy per MB | = | 4297 Watts 245 MB/s |
= | 17.5 Joule/MB |
| 4G LTE energy per MB | = | 1100 Watts 15 MB/s |
= | 73.3 Joule/MB |
In terms of energy per data, 5G NR costs less energy to process and transmit the same amount of data as compared to 4G LTE, but it does not mean 5G NR uses less energy in an absolute sense. Because 5G typically uses higher frequency radio waves, the signal attenuates much faster than 4G. Hence, achieving the same coverage area requires a greater number of 5G base stations. The attenuation for a radio frequency wave can be written as
| Attenuation | = | β | = | α L f |
where α is the attenuation coefficient and its magnitude is determined by the medium through which the waves propagate. For air, α = 1.64 dB/(MHzcm). L is the distance and f is the wave frequency. We can use this relation to compare attenuation between 5G and 4G (using common frequencies for each: [7,8]
| β(5G NR) β(4G LTE) |
= | f(5G NR) f(4G LTE) |
= | 3.5 × 109 Hz 800 × 106 Hz |
≈4.3 |
Roughly speaking, because 5G signals weaken around four times faster than 4G signals, they can only travel about one-quarter as far before their strength drops below usable level. If we assume each base station covers a roughly circular region, one would need ~16 more 5G base stations to cover the area of one 4G station, and each 5G station consumes ~4 more power (Table 1). Therefore, while 5G is often claimed to be more energetically efficient per bit, it does not necessarily consume less total energy.
Although 5G base stations reduce the energy required to transmit a given amount of data compared to 4G, the significantly larger number of base stations needed for coverage results in a higher overall energy demand. If increasing network speed always requires raising frequency, the industry will continue building more towers and consuming more energy. While ICT accounted for only about 4% of global electricity usage in 2020, this raises questions about scalability and the ultimate limits to data traffic density. [9]
© Bingcheng Suo. The author warrants that this work is the authors own and that Stanford University provided no input other than typesetting and reference guidelines. Permission is granted to copy, distribute, and display this work in unaltered form, with attribution, for noncommercial purposes only. All other rights are reserved by the author.
[1] "Ericsson Mobility Report," Ericsson, June 2025.
[2] C. E. Shannon, "A Mathematical Theory of Communication," Bell Syst. Tech. J. 27, 379 (1948); ibid. p. 633.
[3] R. V. L. Hartley, "Transmission of Information," Bell Syst. Tech. J. 7, 535 (1928).
[4] C.-L. I, S. Han, and S. Bian, "Energy-Efficient 5G For a Greener Future," Nat. Electron. 3, 182 (2020).
[5] F. Rusek et al., "Scaling Up MIMO," IEEE 6375940, IEEE Signal Process. Mag. 30, No. 1, 40 (2013).
[6] Z. Ding et al., "Application of Non-Ortthogonal Multiple Access in LTE and 5G Networks," IEEE 7842433, IEEE Commun. Mag. 55, 185 (2017).
[7] C. Sudhamani et al., "A Survey on 5G Coverage Improvement Techniques: Issues and Future Challenges," Sensors 23, 2356 (2023).
[8] S.-J. Seah et al., "The Study of Shadowing Effect For LTE and 5G Networks in Suburban Environment," Int. J. Microw. Wirel. Technol. 16, 2 (2024).
[9] J. Malmodin et al., "ICT Sector Electricity Consumption and Greenhouse Gas Emissions - 2020 Outcome," Telecommun. Policy 48, 112791 (2024).
