Phase and amplitude control and calibration — The 3GPP standard stipulates the structure of 5G signals.3 While specifying the methodology used to generate channels and signals, 3GPP does not dictate how to process the signals at the receiver. Those algorithms are left to the equipment designer. Likewise, 3GPP does not prescribe the algorithms used in the radio resource manager (RRM). The RRM is the entity in the base station that allocates radio resources to users to maximize cell capacity, coverage and user experience, by assigning RBs to users and controlling parameters such as modulation and error coding.

In mMIMO, the RRM also controls parameters such as the beamforming vectors. Some of the algorithms may assume specific beam shapes, including sidelobe levels to be produced after downloading the corresponding beamforming patterns to the RU. For this to be accurate, the actual amplitudes and phases of the radiating elements must not deviate significantly from the values defined by the beamforming vectors. The main lobes are relatively robust to amplitude and phase errors; simulations have shown deviations up to 5 degrees and 0.5 dB do not have a “visible” impact on the overall shape of a beam.

In a time-division duplex system, where the uplink and downlink time share the same band, the DU may use the reciprocal characteristics of the propagation channel. For example, the DU may use uplink channel estimates to derive downlink beam weight vectors. So the RU should ensure that channel reciprocity is not degraded in the transmitters and receivers. To keep a user free from interference from other users’ signals, the DU must be able to place notches in the direction of these signals, reducing them to 35 to 40 dB below the desired signal. If the notches are calculated assuming reciprocity, the phase and amplitude differences between transceivers must be less than 1 degree and a fraction of a dB, respectively.

As component parameters tend to change with temperature, voltage and age, precise closed loop calibration is necessary to maintain the required precision. The required calibration update speeds will vary with the deployment scenarios and geographies, so the mMIMO design should enable selecting among various accuracies and update rates.

Fronthaul — The fronthaul (FH) connects the DU to the RU. Generally, the RU and DU should use techniques to reduce FH bandwidth, as the bandwidth will drive the cost of the interconnect solution, i.e., the cost for cables, switches and transceivers increases with bandwidth. The O-RAN “Control, User and Synchronization Plane Specification” defines several compression techniques to reduce the FH traffic.4 For the user plane, various bit widths are specified, with modulation compression the most prominent technique, where the modulation function is shifted to the RU. The DU sends the raw unmodulated bits to the RU instead of the frequency domain symbols to be transmitted. The introduction of different sections in the U-plane enable sending only the symbols that are used over the FH interface.

The C-plane traffic includes updating the beamforming vectors. In 5G, these vectors can be updated as often as every orthogonal frequency division multiplexing (OFDM) symbol. For vector updates with every time slot, this can represent more than 30 percent of the FH traffic. Therefore, the O-RAN Alliance has introduced techniques for reducing C-plane traffic. The O-RAN standard enables storing the beamforming vectors in a database at the O-RAN RU using an index, where the stored beamforming vectors are retrieved from the database by referring to the corresponding index. Updating the beamforming vectors is also possible. The O-RAN standard also supports calculating the beamforming vectors in the RU. However, the technique is not well standardized, such that the DU may not know the actual result of the calculation—making this technique of limited use.

The O-RAN Alliance is defining interoperability profiles to enable RUs to be used with DUs from various vendors. It is important for an RU to comply with the selected interoperability testing profile to guarantee interoperability.

Programmability — mMIMO in 5G O-RAN systems is still relatively new and needs to mature in the field. Field experience after deployment will likely lead to adding functionality for the RU to improve system performance. As the cost of exchanging equipment in a cellular network is considerable, the equipment should be designed to support long lifetimes once deployed, i.e., a minimum of seven years. To achieve this, the RU must have inherent flexibility to be updated with new functionality, whether the software on the main RU controller or features in the data paths.

The O-RAN Alliance will continue improving FH performance by adding compression techniques that more efficiently use available FH bandwidth. One candidate is supporting semi-persistent scheduling (SPS) in the RU. By conveying SPS information to the RU, the scheduling information only needs to be signaled once. If the available FH bandwidth without this feature limits the update rate for the beamforming vectors, enabling SPS in the RU will free bandwidth and improve system performance. Other examples where updates will likely occur are improving the linearization in the DFE, reducing power consumption and improving temperature control.

Designing flexibility in the RU architecture enables manufacturers to introduce new technology as it becomes available and create derivatives tailored to various market needs. To update units already deployed in the field, the O-RAN Alliance has standardized field upgrades through the M-plane.

Security — To protect the infrastructure from attack, the RU must have security mechanisms, including authentication and integrity checks for software updates.

Power consumption — The power consumed by the RU adds to the operating expense of the network— with thousands of units consuming about 1 kW, the cost of energy is considerable. The power consumption of a mMIMO base station depends on the load, the instantaneous RF output power and the efficiency of the system. At full capacity, the power consumption is dominated by the PA and the efficiency of the transmit chains. While the efficiency of the PA is important, the losses between the PA and antenna must also be minimized, as well as the power consumption of the receive chains, the digital circuitry and the power regulators.

In most cases, the maximum load on the RU represents an extreme situation during peak hours of the day. The power consumption must also be optimized for typical and low load conditions. This is typically accomplished using RU power saving techniques like shutting off PAs, even shutting off complete carriers. Other than the RF power that gets radiated, the power consumed by the RU is converted into heat and needs to be efficiently transferred to the ambient environment while minimizing the temperature of the electronics. The power consumption drives the thermal design of the system, which adds to the size and weight of the RU.

Mechanical and Environmental

The size of an RU is a key requirement because the available real estate on the tower or at a pole is limited. In some cases, there is just enough space above the existing multi-band passive antenna to mount a 5G panel, provided it is not too tall.

Wind load is important because poles and tower structures are built and certified for a maximum wind load. Base stations are typically expected to remain operational in winds up to 150 km/h and to survive wind speeds of 200 km/h. The wind load of the RU is proportional to its surface area, i.e., panel size, and the form factor. Rounding the edges and using dedicated fins can reduce the wind load without changing the outer dimensions.

The weight of the RU determines the installation cost — how many technicians are needed to mount the equipment, possibly assisted with equipment like cherry pickers. In some cases, tower companies factor the wind load and weight into the rent, which contribute to an operator’s monthly expenses.

Other requirements common to all radio designs include the

  • Operating temperature range, typically -40°C to +55°C, with the output power reduced at higher temperatures to keep the unit operating reliably
  • MTBF, typically greater than 200,000 hours, which is a challenge because of the large number of components in the RU
  • Surge protection, to protect the RU from lightning strikes
  • Ingress protection, typically rated at IP65
  • Aesthetics.
Figure 2

Figure 2 AMD-Xilinx 64T64R RU hardware architecture.

O-RAN SPLIT 7.x mMIMO

To facilitate the deployment of mMIMO RUs for O-RAN, AMD-Xilinx has created reference designs and prototypes based on AMD-Xilinx IC technology (see Figure 2). As an example, Table 1 shows the design requirements for a 64T64R mMIMO RU covering 5G band n77 that has been implemented with the AMD-Xilinx architecture and chipset.

Table 1

The O-RAN FH interface, beamformer, physical random-access channel and sounding reference signal extraction are all mapped onto a single Versal™ VC1902 SoC. Versal ACAP is a fully software-programmable, heterogeneous computing platform that combines scalar adaptable and intelligent engines to achieve performance improvements up to 20x over the fastest FPGA implementations and > 100x over the fastest CPU implementations.6 The Versal device contains a powerful ARM® processor subsystem, programmable logic (PL), and AI engines. The AI engines are very long instruction word, single instruction multiple data vector processing engines, well suited for efficiently computing beamformer operations like matrix multiplication, singular value decompensation and, if needed, matrix inversion.7

The Zynq UltraScale+ RFSoC was designed primarily for RF applications. It integrates the key subsystems required to implement direct RF sampling transceivers. Significant investments have been made in high performance data converters using 16 nm FinFET CMOS technology. Each Zynq UltraScale+ RFSoC contains multiple GSPS analog-to-digital and digital-to-analog data converters. The converters are high precision, high speed and power efficient as well as highly configurable.

Figure 3

Figure 3 Component energy use in a 320 W 64T64R mMIMO radio using Xilinx SoCs and GaN PAs.

The latest version of the Zynq UltraScale+ RFSoC—called the Zynq UltraScale+ RFSoC DFE—uses dedicated logic for the digital functions often used in communications. They support the range of cellular applications, including indoor base stations for sub-6 GHz (FR1) and mmWave (FR2) bands, macro base stations and FR1 mMIMO RUs. The DFE’s dedicated logic functions are optimized, scalable and parametrizable using standard cell hard-blocks for computing, combined with PL to adapt the functions to different application requirements. The standard cell hard-blocks deliver performance typically only found with ASICs, while the PL offers the flexibility of an FPGA. With both functions, the Zynq UltraScale+ RFSoC DFE provides twice the performance of the previous RFSoC generation while consuming half the power.

The logic blocks are used for the filtering, digital up- and down-conversion (DUC and DDC), interpolation and decimation, crest factor reduction (CFR) and digital predistortion (DPD). Other logic blocks include the fast Fourier transformation often used for OFDM modulation, which is part of the RU because of the 7.2 functional split chosen by the O-RAN Alliance. Unused FPGA capacity is available on the RFSoC to add functionality, enabling new functions to be added to RUs deployed in the field.

Figure 4

Figure 4 EVM measurement of a 100 MHz bandwidth 256-QAM signal.

Figure 3 shows the relative energy consumption of the components of a 320 W 64T64R mMIMO radio using AMD’s SoCs and GaN PAs. 65 percent of the power is consumed by the analog components such as the PAs and drivers. 17 percent is consumed by the RFSoC DFEs, of which a significant portion is used for the analog-to-digital conversion and DFE functions. These are also found in ASIC implementations.

RU PERFORMANCE

AMD built and tested a prototype of this RU for the North American n77 band. The transmit, receive and beamforming performance were measured and compared to the 3GPP specifications. A DU emulator from Keysight Technologies was used to stimulate the RU using the O-RAN FH interface.

Figure 4 shows the measured performance of the RU with a 256-QAM, 100 MHz wide signal and 8.8 dB CFR. The measured RF output power met the requirement of 37 dBm (5 W) per port with good EVM quality, i.e., 2.6 percent for the physical downlink shared channel. The adjacent channel leakage ratio measured -49 dBc, confirming the digital predistortion algorithm linearizes the GaN PA sufficiently to meet the leakage requirements. Frequency and time alignment errors met 3GPP requirements, and the defined signal bandwidth was 97.3 MHz, the exact requirement.

Figure 5

Figure 5 OTA beamforming measurements of the antenna layer and RU.

Beamforming performance was measured over the air in an anechoic chamber using all 64 transceivers of the RU, and the results were compared with a measurement of the antenna layer (see Figure 5). The antenna was set to boresight, i.e., 0 degrees both horizontal and vertical, using a beamforming vector with uniform coefficients. Specified for a steering range of ±45 degrees, the two plots overlap from 0 to 30 degrees and show some divergence from 30 to 45 degrees.

SUMMARY

The O-RAN ecosystem is young. O-RAN systems are competing with the end-to-end options offered by the incumbent network equipment manufacturers. To achieve market acceptance, O-RAN solutions need to deliver equal or better performance at a cost advantage than the solutions offered by the established players.

The mMIMO RU adds uncertainty because it’s a new architecture with limited history. The installation cost for a mMIMO panel can only be justified if it is reliable and will stay on the mast for several years without coming down for updates or maintenance. Ironically, mMIMO performance will certainly improve with time, coming largely from software and algorithm improvements, so the current generation of hardware that is fielded must have the flexibility to adopt these new capabilities that will improve system performance.

The Xilinx UltraScale+ RFSoC DFE provides a direct RF sampling transceiver platform for mMIMO applications. It delivers ASIC-like performance with the flexibility of an FPGA and moderate power consumption. Measurements confirm 3GPP and O-RAN Alliance performance targets can be met with this SoC solution. By bringing this high performance and flexible capability to O-RAN, AMD-Xilinx hopes to accelerate market adoption of O-RAN and mMIMO RUs.

References

  1. V. Aue, “The Open RAN System Architecture,” Microwave Journal, Vol. 64, No. 11, November 2021, pp. 7082.
  2. O-RAN Alliance, Web: www.o-ran.org/.
  3. 3GPP RAN3 Technical Report TR38.803v1.1.0, Section 11, Figure 11.1.1-1, Web: www.3gpp.org/ftp//Specs/archive/38_series/38.801/38801-100.zip.
  4. O-RAN Fronthaul Working Group, “Control, User and Synchronization Plane Specification,” O-RAN.WG4.CUS.0-v04.00.
  5. 3GPP TS 37.145-1 V16.5.0 (2020-12), “3rd Generation Partnership Project; Technical Specification Group Radio Access Network; Active Antenna System (AAS) Base Station (BS) conformance testing; Part 1: Conducted conformance testing (Release 16),” 3GPP, Chapter 4, December 2020.
  6. Xilinx, “Versal: The First Adaptive Compute Acceleration Platform (ACAP),” WP505 (V1.1.1), September 29, 2020, Web: docs.xilinx.com/v/u/en-US/wp505-versal-acap.
  7. Xilinx, “Xilinx AI Engines and Their Applications,” WP506 (V1.1) July 10, 2020, Web: docs.xilinx.com/v/u/en-US/wp506-ai-engine.