Dealing with power dissipation in 5G antenna design

Article By : Rik Jos

Many research challenges have to be overcome before 5G systems can be deployed in the field. The most important ones for the physical hardware are reduction of power consumption and development of low-cost assembly techniques for the array panels.

The emerging 5G wireless communications standard bears the promise of delivering more data to more customers at higher data rates than is currently possible – up to 1000 times more bandwidth by 2025, according to some forecasts. One way this will be achieved is by massive MIMO, using antennas made up of arrays of elements, driven by individual signals.
Such antenna arrays can create multiple signal beams. Massive MIMO can be applied in a scattering rich environment, i.e. in conditions where signals are scattered of buildings and other objects such that each user can be reached via multiple paths. At the location of the intended user signals from all these paths add constructively, enabling a high data rate. Away from the intended user the signals are not correlated and merely add to the background noise.
For operators to continue improving their services in this way, as we have come to expect, there will be a cost, particularly in the increased complexity and related power consumption of basestations and terminal equipment.
What's the issue? Multichannel phase shifting can be done in the analogue domain, by taking the transmit data stream, dividing it as many ways as there are elements in the antenna array, and then applying phase shifting to each of them (figure 1).

[EEIOL 2016JUN21 TA 01Fig1]*Figure 1: Phase shifting in the analogue domain for a 5G antenna array (Source: AMPLEON).*

This works, but is inflexible – it can only handle one data stream and therefore it can only generate one signal beam. If the system needs to handle multiple data streams and generate multiple beams from one array, we need to move to digital beam forming, as shown in figure 2, in which each element of the antenna array has its own transceiver and set of data converters.

[EEIOL 2016JUN21 TA 01Fig2]*Figure 2: Basic architecture of a digital beam forming array (Source: AMPLEON).
Greater complexity leads to greater power consumption, which needs to be controlled to reduce the environmental impact and operating-cost implications – and the cooling challenge.

Estimating the cooling needs of an antenna array
Take a 4×4 antenna array panel operating at 30GHz, in which the antenna elements are placed half a wavelength apart – that is, 5 mm. To use digital beam forming means each element needs 2 DACs (for I and Q), 2 ADCs, 1 PLL, 1 LNA, 1 PA, 1 transmit/receive switch and some amplifiers and other electronics, including filters. The circuits for each element should ideally be on one chip, for cost and size reasons. The antenna is then assembled by laying out 16 chips evenly across a panel, so that they have short connections to the antenna elements they are driving and the heat they produce can be evenly distributed, as in figure 3.

[EEIOL 2016JUN21 TA 01Fig3]*Figure 3: A 4×4 antenna array panel, with distributed elements per element and a central processing unit (Source: AMPLEON).*

If the power amplifiers have a peak output power of about 20 dBm each, and are built using the most advanced techniques available today, the total power dissipation in such a panel will be 3 to 4 W. This already assumes the use of data converters with limited bit depths, since research shows that we may need less resolution to deliver the same signal integrity to the receiver when using multiple antenna elements than when using a single antenna. However, the data converters still need to run at high speed to handle large signal bandwidths. The power amplifies account for about 75% of total dissipation when transmitting, since the efficiency at mm-wave frequencies is very low.
There are techniques, such as Doherty circuit architectures and envelope-tracking schemes, to improve PA efficiency, but they need elaborate digital pre-distortion, which itself consumes power, to achieve acceptable spurious signal levels. At some point, the benefits of these techniques are outweighed by the energy cost of implementing them, and there is no net gain. This 4×4 array example is very close to that cross-over point, so there is little point in applying such energy-saving PA architectures. Even in case of a simple, pretty linear class AB amplifier some kind of energy-efficient linearisation will be required anyhow. Fortunately massive MIMO systems will probably use TDD and the power amplifiers are only on part of the time.
In our example design, then, between 3 and 4 W of heat is being generated on a panel of 400 mm2. We want to cool it passively, for cost, energy consumption (in the fan), and reliability reasons. We can do this with an aluminium plate with cooling ribs, which has a cooling capacity of about 60 W/m2K. Assuming an ambient temperature of 60ºC (think of a basestation on a Middle Eastern rooftop in summer) and a temperature of 100ºC at the connection between the antenna panel and the transceiver chips, a quick calculation shows that we can cool 0.25 W/cm2 – or about a quarter of what the array needs. To dissipate the full 3.5 W will take a cooling panel of about 1400 mm2.
One way to fix this issue is to build a panel of the right size to cool the electronics, and then have it drive a separate, smaller antenna array panel. This may work for a 4×4 element array, but is impractical for arrays with tens or hundreds of antennas.
Building a sparse antenna array
One solution may be to use sparse array architectures, in which the distances between the antenna elements are much larger than the transmitted wavelength, but the array still should not generate unwanted side lobes or grating lobes next to the transmitted beam to avoid interference.
Antenna arrays with inter-element spacing equal to or larger than the transmitted wavelength λ give rise to unwanted grating lobes—if the elements are placed uniformly in a grid. If the elements are not spaced uniformly, as in the 150 element 'sunflower array' shown in figure 4, the average inter-element spacing can be larger, in this case 5λ, without suffering from grating lobes [1].

[EEIOL 2016JUN21 TA 01Fig4]*Figure 4: A non-uniform antenna array, modelled after a sunflower (Source: AMPLEON).*

To design a sunflower array, the elements are placed along a Fermat spiral (figure 5) so that with each turn, an equal amount of area is enclosed. The individual elements are positions on the spiral at multiples of the angle χ, which is 4π/(3+√5), the so-called Golden Angle.

[EEIOL 2016JUN21 TA 01Fig5] *Figure 5: A Fermat spiral (Source:AMPLEON).*

The sunflower array has an almost uniform power density, which simplifies cooling and makes effective use of the total aperture if all elements are excited to the same level. This arrangement does not generate unwanted grating lobes, and its beam angle can be steered as well as that of a dense array. This makes the sunflower array a good candidate for the kind of sparse arrays that will be needed in passively cooled 5G mm-wave antenna panels.
The 3 dB beam width of a sparse array is inversely proportional to the aperture size A, which means that using them to make cooling easier will mean accepting the design trade-off of a narrower beam. Using reasonable assumptions for an access point with a range of 150 m, simple calculations show that the maximum achievable beam width is probably 5 to 8 degrees – effectively a pencil beam.

Many research challenges remain before 5G systems will be deployed in the field. The most important ones for the physical hardware are reduction of power consumption and development of low-cost assembly techniques for the array panels.

Maria Carolina Vigano, Sunflower array antenna for multi-beam satellite applications, PhD Thesis TU Delft, 2011, ISBN 987-94-6113-030-3

About the author
Rik Jos holds a PhD in Physics from the University of Utrecht, The Netherlands. In 1986 he joined Philips Semiconductors in the development of RF technologies for power amplifier applications, where he was appointed a Philips Semiconductor Technology Fellow in 2002. Since 2004 he is also an adjunct professor at Chalmers University in Sweden. He has held leading positions in RF innovation in Philips and NXP Semiconductors. Since 2015 he is part of Ampleon in The Netherlands where he works on innovation of RF power technologies, especially wide bandgap semiconductors, and amplifier architectures, like switch mode power amplifiers. His current research activities focus on 5G mm-wave technologies and architectures.

Leave a comment