Microwave Journal

The Open RAN System Architecture and mMIMO

November 14, 2021

Open radio access networks (O-RAN) are transforming mobile networks. O-RAN is about the disaggregation of the traditional RAN system into the radio unit (RU), distributed unit (DU) and centralized unit (CU) components and their hardware and software platforms.1 O-RAN fosters innovation by involving more manufacturers in the development of the RAN infrastructure, enabling new entrants to compete and disrupt the market if they can offer a competitive edge. Ideally, the O-RAN specifications will create a broad RAN supplier ecosystem, where operators can pick and choose components from different suppliers and not be bound to a single company. The disaggregation of hardware and software enables virtualization, meaning large parts of the network functions become virtualized and can be run on commercial off-the-shelf hardware or general purpose processors. Virtualization also enables “cloudification,” where many functions are hosted by multiple servers, typically bundled in one or more data centers.

Figure 1

Figure 1 Functional split between the CU and DU.

For O-RAN networks to succeed and become accepted, the standardization of interfaces and proven interoperability are the keys to success. 3GPP has investigated different functional split options between the CU and DU (see Figure 1).2 The O-RAN ALLIANCE has chosen 3GPP split option 2 for the interface between the CU and DU and split option 7 as the DU to RU interface. The centralization of the packet data convergence protocol (PDCP) layer enables scaling with the user plane traffic load. The O-RAN ALLIANCE has chosen the so-called split option 2-2, enabling the separation of the U-plane from the other planes, while having a centralized radio resource controller and radio resource manager. For the interface between DU and RU, the O-RAN ALLIANCE has chosen an intra-physical layer (PHY) split, i.e., between the low PHY and high PHY.

Figure 2

Figure 2 Network function partitioning into the CU, DU and RU.

The high-level functional partitioning into the CU, DU and RU is shown in Figure 2. The link between the RU and DU is referred to as the fronthaul, and the link between the CU and DU is referred to as the midhaul. Due to the various control loops within the system, different latencies can be tolerated. The most critical interface is the fronthaul, which typically tolerates latencies up to 160 µs. If a point-to-point connection is used to connect the RU to the DU, a distance between the RU and DU of up to 30 km can be supported.

As this article’s focus is the RU, the discussion is more on the O-RAN fronthaul interface and the corresponding architectural split. When selecting the fronthaul interface, the following aspects must be considered:

Transport Bandwidth — Referring to Figure 1, the required data rate reduces from the right (the option 8 interface between the PHY and RF) to the left. The chosen split provides a good compromise between flexibility and algorithmic differentiation, with a modest demand on the data rate.

Architecture Split — The split must reflect the intention of an O-RAN architecture: vendor neutral hardware and software. The radio’s performance is defined not only by the radio hardware, but also by the way the signals are processed. To be accepted by the market, an O-RAN system must deliver comparable performance to conventional single-vendor systems. The architectural split enables innovation and stimulates differentiation and, if possible, should not stipulate certain processing algorithms or preclude alternative processing techniques. The O-RAN ALLIANCE has chosen an interface that defines the radio hardware with clearly defined and understandable processing functions tightly controlled by the DU and its software.

Interoperability — Interoperability between different vendor systems is key for O-RAN to be adopted by the market. Therefore, the architectural split must provide an interface easily understood by any implementer, clearly described with no room for interpretation and rigorously tested for interoperability.

Figure 3

Figure 3 O-RAN architecture. Source: O-RAN Fronthaul Working Group.3

The O-RAN ALLIANCE has defined an interface referred to as the 7.2x split.3 In the 7.2x split, the O-RAN fronthaul interface resides between the resource element mapping in the DU and the time-frequency conversion in the RU, i.e., the inverse FFT (iFFT) and cyclic prefix (CP) addition in the downlink and CP removal and FFT computation in the uplink, respectively (see Figure 3). The dotted processing blocks in the figure are not mandatory for all RU categories. Precoding for certain RU categories can be done within the RU, in which case precoding in the DU is bypassed. For mMIMO radios, the interface foresees digital beamforming on the RU side. Digital beamforming is omitted for conventional radios, which typically have only a small number of transceivers. Additionally, the O-RAN ALLIANCE distinguishes between category A and B type O-RUs. The category B architecture supports MIMO precoding in the O-RU; category A does not. The category B O-RU provides support for modulation compression, a technique to reduce the fronthaul bandwidth by moving the modulation function to the O-RU.

In addition to the user traffic handled by the so-called U-plane of the fronthaul interface, O-RAN defines three other planes: the M-plane for handling management control data, the C-plane for handling near real-time control data and the S-plane for handling synchronization (frequency and time). The M-plane is primarily used for configuring the RU, reading out status information and handling errors and alarms.4 It is entirely based on the NETCONF protocol, an XML-based protocol to set and query the operation of a network device. It uses YANG as its data modeling language. Typical configuration data includes

  • Setting up the carrier (e.g., the center frequency, bandwidth and power)
  • Mapping the antenna layers
  • Fully resetting the software of the RU
  • Updating the O-RU software, as the M-plane supports downloading entire software images.

Typical parameters that can be queried concerning the O-RU state and general information are

  • Physical structure of the antenna radiating panel
  • Clock synchronization state
  • Fronthaul interface version and information about supported C- and U-plane section types and extensions
  • Boot state
  • O-RU alarms and performance counters, such as the number of packets received AND number of U-plane data packets received on time, late or corrupt.

The radiating panel is modeled as a rectangular array of equally distributed and independently controlled radiating elements. This information is useful for the O-DU to compute the beamforming weights used to form the desired beams. The beamforming weights determine the direction and shape of the beam. Especially in mMIMO systems, a different set of beamforming weights is typically used for every time transmission interval; however, changes can occur as often as every OFDM symbol.

As mentioned, the O-RAN ALLIANCE supports the vision of disaggregated hardware and software. Therefore, it has defined the radio (O-RU) to be directed by the O-DU, where the algorithms for channel estimation, weight computations and near real-time user scheduling reside. The traffic associated with the provision of beamforming weights can be substantial and may be the same order of magnitude as the user plane traffic. Hence, different means of reducing the traffic are defined in the O-RAN specification.

For a cellular network to work properly, the radio units must be synchronized with an accuracy of ±25 ppb in frequency and ±1.5 µs in time. The O-RAN specification defines several means of synchronizing the RUs to the network, with the predominant method for synchronization the IEEE 1588 protocol, also referred to as precision time protocol (PTP). PTP is based on measuring the time of arrival of IP packets. However, since the IP traffic may be subject to jitter, a relatively long observation time is needed to achieve the desired frequency accuracy. Therefore, O-RAN provides the option to make use of SyncE, which uses the line rate to convey the clock from the source (e.g., the O-DU or a switch) to the O-RU. IEEE 1588 has additionally defined hardware functions built into switches and routers that enable adjusting time stamps due to latencies introduced by those network functions. Since not all network elements may be equipped with such a function, the latencies may be difficult to estimate.

In some cases, the packet delay jitter introduced by the network may not be tolerable. O-RAN provides another methodology, to synchronize the O-RUs using GPS. In this case, the O-RU has a built-in GPS receiver to synchronize the O-RU to the accurate GPS clock and time (see Figure 4). Multiple RUs at a site receive the GPS signal from a common active antenna, and each O-RU has a GPS receiver. The O-RUs are connected to an O-DU that may be located at the site or up to 30 km away, e.g., in a data center, and the O-DU receives its time synchronization through the network from an external time server. Alternatively, it may be equipped with its own GPS receiver. As an alternative to GPS or as a fallback in the case of GPS synchronization failure, the O-DU may provide frequency and time synchronization to the O-RUs.

Figure 4

Figure 4 System using GPS at the RU.

Figure 5

Figure 5 Typical mMIMO antenna: full array (a) and subarray (b).

In some cases, cross-licensing agreements allow operators to use the spectrum of another operator to offer a joint service. This is referred to as multi-operator RAN (MORAN). From the O-RU perspective, two fundamental sharing architectures exist: If each operator uses its own O-DU, the O-RU must act as two independent RUs to the DU. If not, only the O-RU and the O-DU are shared between operators. MORAN is transparent to the RU. Such an architecture brings restrictions, such as time-division duplex (TDD) downlink and uplink transmission periods aligned over both operators’ networks, which must be carefully handled to avoid conflicts.


The O-RAN fronthaul interface supports both conventional radios, with two or four transceivers, or mMIMO radios. MIMO is a means to increase the capacity of a mobile network by using the spatial domain, where the “multiple input, multiple output” refers to the radio channel. Signals sent by multiple transmitters are received by multiple receivers. Assuming propagation conditions permit, advanced channel coding methods and signal processing algorithms enable separating the transmit signals from the receive signals. mMIMO applies when the number of single antenna terminals (i.e., number of users) transmitting at a given frequency and time is much less than the number of base station antennas receiving. TDD, reference sequences and feedback from the terminals to the base station help apply the same principles to the downlink.

In general, the more transceivers in a mMIMO system, the more users can be served over the same communication channel, assuming the propagation characteristics support user discrimination. The 3GPP standards provide up to 256 transceivers. However, as the cost and power consumption increase with the number of transceivers, practical configurations for a mMIMO base station range between 16 transmitters and 16 receivers (16T16R) and 64 transmitters and 64 receivers (64T64R).

The antennas in a mMIMO system are arranged in an array, where each antenna may consist of a subarray of antenna elements. A typical arrangement for a 64T64R panel is shown in Figure 5, a 12 × 8 antenna array (see Figure 5a) comprising 32 polarized subarrays. Each subarray is made of six antenna elements, where three antenna elements radiate with a negative polarization, indicated by “np,” and the other three antenna elements with a positive polarization, indicated by “pp” (see Figure 5b).

With conventional, passive base station antennas, antenna gain is an important factor. The antenna gain is defined as the ratio between the maximum radiated power at a specific angle and the power radiated by a hypothetical antenna transmitting the same total power isotropically, i.e., equally distributed over the full sphere. It is assumed that the hypothetical antenna radiates all the power without any loss between its antenna port and free space. For the same power at the antenna feed, the radiated power of a directional antenna at boresight will be greater than the corresponding isotropic antenna by the gain factor.

To enable comparison with conventional radio architectures, which have a remote radio head and a separate passive antenna, the 3GPP has defined a reference radio architecture for mMIMO. This reference architecture defines two reference points for conducted and radiated measurements,5 assuming a transceiver unit array connected to a composite antenna (see Figure 6). The transceiver unit array contains the transmitters and receivers, generates the modulated transmit signals and performs receiver combining and demodulation. The composite antenna consists of the radio distribution network and the antenna array, the interface between the transceiver unit array and the composite antenna is called the transceiver array boundary (TAB). The two points of reference defined by 3GPP are the TAB for conducted measurements and the far-field region for radiated measurements, also referred to as over-the-air measurements.

Figure 6

Figure 6 AAS radiated and conducted points of reference. Source: 3GPP.5

The RF output power of the transmitter limits both data rates and coverage. The output power typically refers to the power combined from all transceivers, which is the conducted power measured at the TAB. The corresponding measurement of radiated power is called the effective isotropic radiated power (EIRP), which includes the gain of the composite antenna. For example, a radio panel that provides 200 W (53 dBm) of RF power at the TAB and feeds an antenna array with 25 dBi gain at boresight will have an overall power measured at boresight of 78 dBm EIRP.

Figure 7

Figure 7 Typical azimuth antenna pattern for a user beam at 0°.

From antenna array theory, the far-field antenna array pattern of a linear array is the product of the single element pattern and the array factor (AF), assuming all antenna elements are the same kind, point in the same direction and are excited with equal power. The AF is the far-field radiation pattern of an array of isotropic radiators. Provided no coupling occurs between radiating elements, a linear array with 12 rows and eight columns and an element spacing of half-wavelength (λ/2) will have an AF of 96 (i.e., 19.8 dB). Coupling between the elements will reduce the AF and minimizing coupling can only be achieved by proper design; however, it becomes increasingly difficult when the element spacing de is less than a wavelength, particularly less than λ/2.

Sidelobes are another important characteristic of antenna arrays. The sidelobe level (SLL) is the maximum power from a sidelobe normalized by the strength of the main lobe. Alternatively, the inverse value, i.e., the ratio between the strength in the direction of the main lobe and the maximum sidelobe level, is often used and referred to as the sidelobe level suppression (SLS). In a mMIMO antenna, the actual beam pattern depends on the subarray radiation pattern and the amplitude and phase relationships between the subarrays, which are defined by the beamforming vector applied at the beamformer. The beamforming vector yielding the pattern with the lowest sidelobe levels typically uses different phase and amplitude values. For maximum output power, all the amplitudes need to be identical. To compare the radiation performance of different mMIMO active antenna units (AAU), only phase variations are allowed in the beamforming vector; the amplitudes are forced to be the same. To illustrate, Figure 7 shows a typical beam pattern with SLLs less than -16 dB.


The RU in a mMIMO AAU can be separated into its functions (see Figure 8). Some are only required once, while others, like the transceivers, are used multiple times. Each transceiver serves a single polarization of a radiation panel subarray. Among the functions used only once is the fronthaul interface connecting the O-RU to the O-DU, the beamformer, clock synchronization, management and control.

Figure 8

Figure 8 Typical RU architecture for a mMIMO AAU.

The fronthaul uses an Ethernet interface and is separated into the C-, U-, S- and M-planes. The M-plane interpreter and manager connects to the O-RU controller, which sets up and oversees the overall well-being of the unit. The O-RU controller measures the power consumption, temperature, output power and relative amplitude and phase accuracies for both receive (Rx) and transmit (Tx) to let the DU exploit channel reciprocities. It hosts multiple event counters and reports statistics, warnings and errors through the M-plane to either the O-DU or directly to the management system. The O-RAN ALLIANCE has standardized this as the service management and orchestration framework. S-plane packets following the IEEE 1588 PTP protocol are interpreted independently. PTP information is used to synchronize the O-RU clock to the network. As noted earlier, a built-in GPS receiver can be used as an alternative clock.

Figure 9

Figure 9 Elements of a digital transceiver.

The beamformer connects to NTx transmitters and NRx receivers. If NTx = NRx = N, which is usually the case, the beamformer is said to be connected to N transceivers, with all transceivers identical. Each transceiver consists of the digital transceiver and an analog front-end. For each layer, the beamformer performs two matrix multiplications for each subcarrier (SC). During transmit, the vector of resource blocks contained in each radio layer is multiplied by the Tx beamforming matrix to yield a vector of SCs for each transceiver. Likewise, during receive, each vector of SCs received from every transceiver is multiplied by the Rx beamformer to yield the SC vector for all layers. Figure 9 shows the functions of each digital transceiver, which contains the low PHY and digital front-end. The digital front-end encompasses front-end functions like filtering, gain settings and linearization in the digital domain, and the low PHY addresses time-frequency conversion and OFDM signal generation.

In the downlink direction, the SCs from the beamformer containing data in the frequency domain are transformed to the time domain within the iFFT. The OFDM signal is formed by adding a CP and passed to the digital front-end for digital up-conversion, comprising filtering, the frequency shift of the baseband signal and digital-to-analog conversion. For the downlink, the digital front-end includes crest factor reduction, which reduces the peak-to-average ratio of the OFDM signal, and digital predistortion (DPD), which linearizes the power amplifier. The uplink signals from the analog front-end are converted into the digital domain using analog-to-digital converters and digitally down-converted to baseband. With the down-converted signals, the CP is removed and the signal is converted to the frequency domain via the FFT. The OFDM symbols are then passed to the Rx beamformer. The low PHY also contains special functions for physical random access channel (PRACH) and sounding reference signal (SRS) filtering. The PRACH and SRS signals are passed to the fronthaul interface, and the O-DU post-processes the SRS and PRACH data.

The analog front-end (AFE) connects to the radiating panel and contains analog components like power amplifiers, filters, drivers and baluns and may contain switches and circulators. The AFE amplifies the Tx and Rx signals to and from the antennas. It must provide sufficient dynamic range for the Rx and Tx paths, isolate the paths and manage any noise introduced by the power amplifier stages. The radiating panel is designed to provide the required gain for the mMIMO system and the horizontal and vertical steering.


This article has provided a tutorial on the architecture of the O-RAN and the approach to standardizing the interfaces between the RU, DU and CU to achieve interoperability and provide an environment for new entrants and network innovation. mMIMO is one RAN implementation to improve data capacity in areas with high mobile traffic, and the article discussed the functional architecture of a mMIMO RAN meeting O-RAN specifications.

A future article will discuss the mMIMO architecture and performance parameters of the RU, applying the concepts to the design of two AAUs: for the North American band from 3.7 to 3.98 GHz and the European band from 3.3 to 3.8 GHz.


  1. O-RAN ALLIANCE, web: www.o-ran.org/.
  2. 3GPP, RAN3 Technical Report TR38.803v1.1.0, Section 11, Figure 11.1.1-1, www.3gpp.org/ftp//Specs/archive/38_series/38.801/38801-100.zip
  3. O-RAN Fronthaul Working Group, “Control, User and Synchronization Plane Specification,” O-RAN, WG4.CUS.0-v05.00.
  4. O-RAN Fronthaul Working Group, “Management Plane Specification,” O-RAN, WG4.MP.0-v05.00.
  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),” Chapter 4, 3GPP, December 2020.