Proof-of-Concept of a Millimeter-Wave Integrated Heterogeneous Network for 5G Cellular

The fifth-generation mobile networks (5G) will not only enhance mobile broadband services, but also enable connectivity for a massive number of Internet-of-Things devices, such as wireless sensors, meters or actuators. Thus, 5G is expected to achieve a 1000-fold or more increase in capacity over 4G. The use of the millimeter-wave (mmWave) spectrum is a key enabler to allowing 5G to achieve such enhancement in capacity. To fully utilize the mmWave spectrum, 5G is expected to adopt a heterogeneous network (HetNet) architecture, wherein mmWave small cells are overlaid onto a conventional macro-cellular network. In the mmWave-integrated HetNet, splitting of the control plane (CP) and user plane (UP) will allow continuous connectivity and increase the capacity of the mmWave small cells. mmWave communication can be used not only for access linking, but also for wireless backhaul linking, which will facilitate the installation of mmWave small cells. In this study, a proof-of-concept (PoC) was conducted to demonstrate the practicality of a prototype mmWave-integrated HetNet, using mmWave technologies for both backhaul and access.


Introduction
The fifth-generation mobile networks (5G) are expected to be deployed by 2020 and to be a primary source of Internet connection for the year 2020 and beyond. Driven by the rapid growth in the number of mobile-connected devices and the emergence of new services, research and development on 5G have been expanding [1,2]. The Working Party 5D (WP5D) of the International Telecommunication Union Radiocommunication sector (ITU-R) has established a work plan for International Mobile Telecommunications (IMT) for 2020 and beyond (IMT-2020) [3]. The 3rd Generation Partnership Project (3GPP) has initiated studies that aim to make functional specifications for 5G available in 2019 [4]. The leading 5G use case is enhanced mobile broadband, closely followed by critical and massive machine type communications used for wireless sensors, meters or actuators [5]. It is anticipated that 45 trillion networked sensors will be deployed in 20 years [6]. 5G should consider the ability to support a massive number of connected devices and various types of traffic. band only became a reality a few years later with the development of a complementary metal oxide semiconductor (CMOS) radio frequency integrated circuit (RFIC) with low-phase noise [26], built-in self-calibration [27] and an RF phase shifter with low power consumption [28]. The latest wireless standard for the 60-GHz band is IEEE 802.11ad (also referred to as WiGig), which was standardized in 2012 [29,30]. Intel (Santa Clara, CA, USA) then released the first WiGig chipset supporting 7 Gbps for docking stations, and in 2014, Qualcomm (San Diego, CA, USA) released a processor combining the LTE and WiGig standards [31,32]. The first WLAN AP product with the WiGig standard is scheduled for release in 2016 [33]. Samsung (Suwon, South Korea) and Deutsche Telecom (Bonn, Germany) have recently demonstrated 5G on a smartphone integrating 60-GHz technologies [34].
Although the regulations for mobile use have not yet been defined, research and development on mmWave frequencies other than a 60-GHz band is currently being undertaken. Samsung and Qualcomm have investigated the 28-GHz band [34][35][36], and Huawei (Shenzhen, China) and Nokia (Uusimaa, Finland) have been working on the 73-GHz band [37,38], mainly for 5G outdoor access. Two United States cellular operators, Verizon (Basking Ridge, NJ, USA) and T-Mobile (Bellevue, WA, USA), have applied for test radio licenses for the 28-and 38-GHz bands [39], and Ericsson (Stockholm, Sweden), Intel, Samsung, Qualcomm and Nokia are planning to join these tests. As well as mmWave access, much work is being conducted on mmWave backhaul technologies for 5G. InterDigital (Wilmington, DE, USA) has investigated 60-GHz meshed backhaul for small cell BSs in which a self-organizing function is realized by a phased array [40]. NEC (Tokyo, Japan) has released an E-band (71-86 GHz) transceiver supporting 10-Gbps backhaul to enable centralized radio access networks (C-RANs) [41]. This range of activity demonstrates the key role of mmWave technology in boosting the data rates of 5G systems. However, to date, no integrated PoC has been developed for 5G, which combines mmWave access, mmWave backhaul and mmWave antennas. To the best of the authors' knowledge, this contribution is the first to demonstrate a PoC for an mmWave-integrated cellular network.
The rest of this paper is organized as follows. Section 2 introduces the system architecture and the components of the mmWave-integrated HetNet. Section 3 discusses state-of-the-art mmWave antenna technologies for backhaul and access. Section 4 presents our PoC of the mmWave-integrated HetNet. Finally, Section 5 presents our conclusions.

Control/User Plane Splitting for HetNet
In CP/UP splitting [14,23,42], also referred to as "phantom cell" or "soft cell", the CP and UP are handled separately. In a CP/UP-split HetNet, the mobile stations (MS), also referred to as user equipment (UE), are capable of maintaining a logical link with both the macrocell and multiple small cells. This capability, called dual connectivity (DC), allows the CP and UP to be handled individually and flexibly. The single CP also enables efficient, centralized radio resource control (RRC), for example mobility control or beamforming (BF). DC affects UE peak power consumption because the UE has to activate multiple RF devices at the same time. However, the higher throughput makes the download time shorter, allowing the device to go to the sleep state soon after the download has been finished. In [43,44], it is indicated that average UE power consumption with two frequency carrier aggregation does not exceed that with single frequency carrier. Furthermore, a 99.7% reduction of UE power consumption could be achieved via a combination of traffic offloading and dynamic on/off [45].
To achieve CP/UP splitting, the macrocell and small cells are connected via a backhaul interface used for exchanging control information and user data and are controlled by the same RRC entity as shown in Figure 1. The backhaul, as the intermediate link between the cells and the centralized network, is therefore an essential part of the CP/UP-split HetNet. The use of mmWave technologies in the backhaul is discussed in Section 3. The DC and CP/UP splitting feature was introduced in Release 12 of the 3GPP LTE standard [42]. In Release 13, it was extended to support DC between the LTE and an IEEE 802.11 wireless local area network (WLAN) [46,47]. The detailed architecture of LTE-WLAN CP/UP splitting in Release 13 is discussed in Section 2.3. The CP/UP splitting concept must be a key enabler of HetNet.

60-GHz mmWave WLAN
At the World Radiocommunication Conference 2015 (WRC-15), ITU-R identified portion(s) of the frequency range between 24.25 and 86 GHz to be considered for allocation to IMT-2020 [48]. However, the 3GPP LTE standards before Release 13 were designed for frequency bands below 6 GHz [49] and are not optimized for the mmWave frequency bands. In contrast, IEEE 802.11ad WLAN [29], also known as WiGig (wireless-gigabit) [30], addressed the unlicensed 60-GHz frequency band, which has at least 5 GHz of continuous bandwidth available in many countries [50]. WiGig uses 2.16 GHz-wide continuous bands per channel for transmission and is capable of multi-Gbps transmission. WiGig supports three distinct modulation methods: spread-spectrum, singlecarrier (SC) and orthogonal frequency-division multiplex (OFDM) modulations. The spreadspectrum modulation is dedicated to the transmission of control messages. SC modulation is mandatory and is suitable for low-power consumption devices because of the low peak-to-average power ratio (PAPR). OFDM modulation is optional, suitable for relaxed power-consumption devices and more robust against multipaths than SC modulation. An example of the modulation and coding schemes (MCS) for WiGig is shown in Table 1. The physical (PHY) layer achieves a maximum data rate of almost 7 Gbps, and the next generation IEEE 802.11ay 60-GHz WLAN [51] aims to support a maximum throughput of at least 20 Gbps on the medium access control (MAC) layer by spatial multiplexing or further improvement of MAC efficiency.

60-GHz mmWave WLAN
At the World Radiocommunication Conference 2015 (WRC-15), ITU-R identified portion(s) of the frequency range between 24.25 and 86 GHz to be considered for allocation to IMT-2020 [48]. However, the 3GPP LTE standards before Release 13 were designed for frequency bands below 6 GHz [49] and are not optimized for the mmWave frequency bands. In contrast, IEEE 802.11ad WLAN [29], also known as WiGig (wireless-gigabit) [30], addressed the unlicensed 60-GHz frequency band, which has at least 5 GHz of continuous bandwidth available in many countries [50]. WiGig uses 2.16 GHz-wide continuous bands per channel for transmission and is capable of multi-Gbps transmission. WiGig supports three distinct modulation methods: spread-spectrum, single-carrier (SC) and orthogonal frequency-division multiplex (OFDM) modulations. The spread-spectrum modulation is dedicated to the transmission of control messages. SC modulation is mandatory and is suitable for low-power consumption devices because of the low peak-to-average power ratio (PAPR). OFDM modulation is optional, suitable for relaxed power-consumption devices and more robust against multipaths than SC modulation. An example of the modulation and coding schemes (MCS) for WiGig is shown in Table 1. The physical (PHY) layer achieves a maximum data rate of almost 7 Gbps, and the next generation IEEE 802.11ay 60-GHz WLAN [51] aims to support a maximum throughput of at least 20 Gbps on the medium access control (MAC) layer by spatial multiplexing or further improvement of MAC efficiency. The latest semiconductor technology brings 60-GHz communication to small handsets. For example, silicon-based CMOS technologies are implementing integrated system-in-package systems including mixers, LNAs and PAs in the mmWave frequency bands [14]. Figure 2 shows examples of IEEE 802.11ad/WiGig-compliant communication devices. The latest semiconductor technology brings 60-GHz communication to small handsets. For example, silicon-based CMOS technologies are implementing integrated system-in-package systems including mixers, LNAs and PAs in the mmWave frequency bands [14].

LTE-WLAN Interworking and Aggregation
The access network discovery and selection function (ANDSF) [53-56] is a core-network (CN) function that controls UP data offloading between 3GPP RATs (e.g., LTE) and non-3GPP RATs (e.g., WLAN). ANDSF is located in the evolved packet core (EPC), which is part of the 3GPP CN. ANDSF delivers to the UEs the auxiliary information for RAT discovery and selection on the basis of UE context, such as location or IP flow. On the basis of the delivered policies, the UE selects an appropriate RAT to connect with the network. Figure 3a shows an example of the ANDSF-based LTE-WLAN interworking architecture known as CN-level interworking. In this LTE network, the UE is connected to a base station (BS) called evolved Node B (eNB), via an LTE air interface, which is called a Uu interface (radio interface between the Universal Terrestrial Radio Access Network and the UE). When the traffic from an external packet data network (PDN), for example the Internet, is handled by the Uu interface, the traffic is transferred via the PDN gateway (PDN GW), the serving gateway (S-GW) and eNB to the UE. For offloading the traffic to WLAN, two schemes are available:

LTE-WLAN Interworking and Aggregation
The access network discovery and selection function (ANDSF) [53-56] is a core-network (CN) function that controls UP data offloading between 3GPP RATs (e.g., LTE) and non-3GPP RATs (e.g., WLAN). ANDSF is located in the evolved packet core (EPC), which is part of the 3GPP CN. ANDSF delivers to the UEs the auxiliary information for RAT discovery and selection on the basis of UE context, such as location or IP flow. On the basis of the delivered policies, the UE selects an appropriate RAT to connect with the network. Figure 3a shows an example of the ANDSF-based LTE-WLAN interworking architecture known as CN-level interworking. In this LTE network, the UE is connected to a base station (BS) called evolved Node B (eNB), via an LTE air interface, which is called a Uu interface (radio interface between the Universal Terrestrial Radio Access Network and the UE). When the traffic from an external packet data network (PDN), for example the Internet, is handled by the Uu interface, the traffic is transferred via the PDN gateway (PDN GW), the serving gateway (S-GW) and eNB to the UE. For offloading the traffic to WLAN, two schemes are available: •  ANDSF is an efficient and scalable mechanism for controlling the UE behavior of RAT discovery and selection from the network. The UE can be authenticated from both LTE and WLAN networks with a single subscriber identity because the EPC and WLAN can exchange authentication, authorization and accounting (AAA) information via a 3GPP AAA server. However, ANDSF cannot control UP offloading in a hyper-dense HetNet for two reasons. First, the initial design of Release 8 ANDSF was intended to provide policies based on a static coverage map [57] and not to be responsive to real-time traffic or radio network conditions. Later releases of ANDSF were extended to provide quasi-real-time statistics, such as RAN congestion. However, it is difficult to handle real-time changes in a dense network because the ANDSF entity is separated from both eNB and WLAN AP in the  ANDSF is an efficient and scalable mechanism for controlling the UE behavior of RAT discovery and selection from the network. The UE can be authenticated from both LTE and WLAN networks with a single subscriber identity because the EPC and WLAN can exchange authentication, authorization and accounting (AAA) information via a 3GPP AAA server. However, ANDSF cannot control UP offloading in a hyper-dense HetNet for two reasons. First, the initial design of Release 8 ANDSF was intended to provide policies based on a static coverage map [57] and not to be responsive to real-time traffic or radio network conditions. Later releases of ANDSF were extended to provide quasi-real-time statistics, such as RAN congestion. However, it is difficult to handle real-time changes in a dense network because the ANDSF entity is separated from both eNB and WLAN AP in the network. Second, the mobility anchor point (PDN GW) is too distant from the eNB and WLAN AP in the network. Frequent inter-RAT handover in a dense network consequently increases the signaling load in the EPC, which lengthens the handover interruption time. Moreover, the enhanced features deployed in eNB, such as mobile-edge computing (MEC) [58], cannot be used via WLAN access.
An RAN-level LTE-WLAN integration architecture [46, 47,59] is therefore proposed and is shown in Figure 3b. In this architecture, the LTE eNB and WLAN AP are directly connected for both UP and CP via a coordination interface that directly transfers real-time traffic and radio information. The mobility anchor point resides in the eNB, minimizing interruptions during handover. This architecture enables CP/UP splitting between the LTE and WLAN since a single CP entity in eNB (i.e., RRC) can handle both LTE and WLAN UPs.
The LTE-WLAN integration architecture was proposed and discussed during 3GPP standardization. As a result, two types of LTE-WLAN integration architectures [46,47] are defined in Release 13: LTE-WLAN aggregation (LWA) and LTE-WLAN radio level integration with IPsec tunnel (LWIP). Figure 4 shows the protocol stacks of these architectures. network. Second, the mobility anchor point (PDN GW) is too distant from the eNB and WLAN AP in the network. Frequent inter-RAT handover in a dense network consequently increases the signaling load in the EPC, which lengthens the handover interruption time. Moreover, the enhanced features deployed in eNB, such as mobile-edge computing (MEC) [58], cannot be used via WLAN access.
An RAN-level LTE-WLAN integration architecture [46, 47,59] is therefore proposed and is shown in Figure 3b. In this architecture, the LTE eNB and WLAN AP are directly connected for both UP and CP via a coordination interface that directly transfers real-time traffic and radio information. The mobility anchor point resides in the eNB, minimizing interruptions during handover. This architecture enables CP/UP splitting between the LTE and WLAN since a single CP entity in eNB (i.e., RRC) can handle both LTE and WLAN UPs.
The LTE-WLAN integration architecture was proposed and discussed during 3GPP standardization. As a result, two types of LTE-WLAN integration architectures [46,47] are defined in Release 13: LTE-WLAN aggregation (LWA) and LTE-WLAN radio level integration with IPsec tunnel (LWIP). Figure 4 shows the protocol stacks of these architectures.   The main difference between the architectures is the layer of the forwarded packets. In LWA, the eNB forwards packet data convergence protocol (PDCP) data units (PDUs) to the WLAN AP. The PDCP PDUs, which are encapsulated in the LWA adaptation protocol (LWAAP), can be sent via WLAN. In this architecture, the PDCP features can be utilized in WLAN access. For example, security protection on WLAN can be simplified because the PDCP provides external ciphering and an integrity protection mechanism. However, the WLAN infrastructure needs to be updated to support WLAAP. In LWIP, the eNB forwards PDCP service data units (SDUs), i.e., IP packets, to the WLAN AP. The forwarded IP packets are encapsulated in the Generic Routing Encapsulation (GRE) protocol and are secured using IP security (IPsec). The setting of an LWIP bearer is therefore more complicated  The main difference between the architectures is the layer of the forwarded packets. In LWA, the eNB forwards packet data convergence protocol (PDCP) data units (PDUs) to the WLAN AP. The PDCP PDUs, which are encapsulated in the LWA adaptation protocol (LWAAP), can be sent via WLAN. In this architecture, the PDCP features can be utilized in WLAN access. For example, security protection on WLAN can be simplified because the PDCP provides external ciphering and an integrity protection mechanism. However, the WLAN infrastructure needs to be updated to support WLAAP. In LWIP, the eNB forwards PDCP service data units (SDUs), i.e., IP packets, to the WLAN AP. The forwarded IP packets are encapsulated in the Generic Routing Encapsulation (GRE) protocol and are secured using IP security (IPsec). The setting of an LWIP bearer is therefore more complicated than that of LWA. However, LWIP can be introduced into the existing WLAN infrastructure because the GRE and IPsec tunnels are transparent to the existing infrastructure.
Both LWA and LWIP are controlled by eNB based on UE measurement reporting of radio information, so that their handover failure can be reduced dramatically as reported in [59].

mmWave Beamforming Antenna and Backhaul
Highly directional antennas are required to compensate the millimeter-wave propagation, which has a higher path loss than lower band communication systems, like LTE or Wi-Fi. These highly directional antennas should also be steerable to support access and a reconfigurable backhaul link. A range of concepts and designs for such antennas have been proposed [18][19][20][21]. In the next two subsections, we discuss the most practical approaches.
Different requirements are placed on the mmWave antennas in backhaul and access. For small cell backhauling, medium distances of several hundred meters in static configurations are most important, while access links require fast beamsteering to track mobile users.
The key performance indicators for backhaul links are the data rate, power consumption and cost. The cost can be divided into capital expenditure (CAPEX) and operational expenditure (OPEX). While CAPEX is driven by the price of the device and the cost of installation, OPEX is driven by power consumption, licensing costs and required servicing.

Passive Reflectarray Antennas
Our implementation of a static mmWave backhaul link was based on the use of passive reflecting structures [20]. These so-called reflectarrays comprise an array of specially-designed dipole elements on a printed circuit board (PCB), with a ground metallization layer on the opposite side [60]. Through the proper design of the dipole structures, the phase of the locally-reflected electromagnetic wave can be influenced, and a collimating effect similar to that of a parabolic dish can be obtained. The dipole elements can be made polarization dependent, allowing the reflectarray to use different focal points for the orthogonal polarizations. Furthermore, the shape of the collimated beam can be controlled by the phase configuration. The aperture and gain of a reflectarray can be adapted to match specific needs, in the same way as a parabolic dish. The application of well-understood PCB technology allows both the cost and weight to be kept low.
We combined an integrated mmWave transceiver with on-chip antennas in a reflectarray structure, with the goal of increasing the antenna gain and therefore the link distance that could be achieved. Figure 5 shows our design. The integrated transceiver chip was installed at the position of the feed antenna and facing towards the array.
The integrated transceiver can be, for example, based on the IEEE 802.11ad standard, operating at 60 GHz. RF feeding structures and their associated losses can be minimized, as the on-chip antenna requires only a low gain. As the proposed solution is intended for static links, a mechanical fine alignment can be implemented to reduce the installation cost.
A prototype of the static backhaul is shown in Figure 6. The transceiver was based on an IEEE 802.11ad modem and was connected to an embedded PC acting as both the controller and router. A standard Gigabit Ethernet connection was available. Average and peak power consumption of the can be influenced, and a collimating effect similar to that of a parabolic dish can be obtained. The dipole elements can be made polarization dependent, allowing the reflectarray to use different focal points for the orthogonal polarizations. Furthermore, the shape of the collimated beam can be controlled by the phase configuration. The aperture and gain of a reflectarray can be adapted to match specific needs, in the same way as a parabolic dish. The application of well-understood PCB technology allows both the cost and weight to be kept low.
We combined an integrated mmWave transceiver with on-chip antennas in a reflectarray structure, with the goal of increasing the antenna gain and therefore the link distance that could be achieved. Figure 5 shows our design. The integrated transceiver chip was installed at the position of the feed antenna and facing towards the array. The integrated transceiver can be, for example, based on the IEEE 802.11ad standard, operating at 60 GHz. RF feeding structures and their associated losses can be minimized, as the on-chip antenna requires only a low gain. As the proposed solution is intended for static links, a mechanical fine alignment can be implemented to reduce the installation cost. A prototype of the static backhaul is shown in Figure 6. The transceiver was based on an IEEE 802.11ad modem and was connected to an embedded PC acting as both the controller and router. A standard Gigabit Ethernet connection was available. Average and peak power consumption of the prototype was 14 Watts and 20 Watts, respectively. Since the IEEE802.11at power over Ethernet (PoE) standard allows up to 25.5 Watts (0.6 A × 42.5 V) of power delivered at power devices [54], the entire device could be supplied with PoE. The achievable net TCP/IP throughput of the point-to-point prototype was around 1.7 Gbps, and the round trip IP latency was below 1 ms. The achievable range was in the order of several hundred meters using a reflector with a 20-cm edge length. The data rate could be partitioned arbitrarily between uplink and downlink, as the underlying protocol operated in TDD mode. However, the wired Ethernet interface limited the speed in each direction to around 1 Gbps. Full throughput was therefore achieved when traffic was symmetrical in both directions.

Highly Directional Steerable mmWave Antennas
Phased antenna arrays are a well-established and proven technology for electronically-steerable antennas. However, highly directional arrays with large apertures may encounter a range of implementation problems, such as heat dissipation, feeding line losses and placement/installation difficulties. These are especially critical for small-form factor millimeter wave antennas. In [21], two novel solutions were proposed: the modular antenna array (MAA) architecture and the lens array antenna (LAA). The first allows the creation of a large aperture, high gain antenna array from smallphased antenna array (PAA) modules (see Figure 2b). Hybrid beamforming [61] can be realized in the MAA [62] through coarse analog phase-shifting in the RFIC circuits and fine digital beamforming in the baseband (BB) (see Figure 7a). In this design, an arbitrary number of single modules (eight in this case) is connected to a single baseband and control block.
This proposed design may substantially decrease the feeding line power losses, as they are of a short length in each subarray, and the common feeding is performed at an intermediate frequency (IF). Mass production of unified inexpensive modules further reduces the cost of the large-aperture antenna. The total output power is equal to the sum of the individual single modules and can therefore be scaled according to the specific needs.
The LAA architecture uses a different approach to achieve high directivity with a phased The achievable net TCP/IP throughput of the point-to-point prototype was around 1.7 Gbps, and the round trip IP latency was below 1 ms. The achievable range was in the order of several hundred meters using a reflector with a 20-cm edge length. The data rate could be partitioned arbitrarily between uplink and downlink, as the underlying protocol operated in TDD mode. However, the wired Ethernet interface limited the speed in each direction to around 1 Gbps. Full throughput was therefore achieved when traffic was symmetrical in both directions.

Highly Directional Steerable mmWave Antennas
Phased antenna arrays are a well-established and proven technology for electronically-steerable antennas. However, highly directional arrays with large apertures may encounter a range of implementation problems, such as heat dissipation, feeding line losses and placement/installation difficulties. These are especially critical for small-form factor millimeter wave antennas. In [21], two novel solutions were proposed: the modular antenna array (MAA) architecture and the lens array antenna (LAA). The first allows the creation of a large aperture, high gain antenna array from small-phased antenna array (PAA) modules (see Figure 2b). Hybrid beamforming [61] can be realized in the MAA [62] through coarse analog phase-shifting in the RFIC circuits and fine digital beamforming in the baseband (BB) (see Figure 7a). In this design, an arbitrary number of single modules (eight in this case) is connected to a single baseband and control block. The use of single modules without lenses (D2D) allowed transmissions with the highest data rate in SC mode (MCS #12: 4.7 Gbps physical layer rate, 16-QAM with coding rate ¾; see Table 1) at distances up to 15 m. The application of a lens at the transmitter side achieved the maximum data rate (4.7 Gbps), making 16 QAM modulation practical for outdoor communications at distances of about 30-35 m, which can be appropriate for millimeter wave access links.
A second experiment investigated the backhaul link on the base of the two LAA antennas.  The outdoor field trial setup for LAA is shown in Figure 9. This study demonstrated that the LAA is a practical high-gain steerable millimeter wave antenna for use in future millimeter wave overlay networks (D2D, access and backhaul) within the framework of 5G communication systems. This proposed design may substantially decrease the feeding line power losses, as they are of a short length in each subarray, and the common feeding is performed at an intermediate frequency (IF). Mass production of unified inexpensive modules further reduces the cost of the large-aperture antenna. The total output power is equal to the sum of the individual single modules and can therefore be scaled according to the specific needs.
The LAA architecture uses a different approach to achieve high directivity with a phased antenna array. The array aperture is increased by using a dielectric lens with a special geometry [63][64][65] that focuses the single module phased array beam into a single plane (the elevation, for example), allowing steering in another plane (the azimuth). For efficient azimuth plane beamsteering, a lens array antenna with an elliptic-toroidal geometry is proposed [21] with the emitting-phased antenna array module placed on the backside (Figure 7b). In this antenna design, the antenna aperture is determined by (a) the vertical dimension of the lens and (b) the horizontal dimension of the phased antenna array. The number of array elements in the vertical plane may be significantly lower than is required for a phased antenna array of the same gain. Effective heat dissipation may be achieved by placing a radiator on the back of the lens array antenna.
The LAA prototype developed in the framework of the MiWEBA project has an aperture (lens height) of 112 mm and a gain of up to 27 dBi. To demonstrate the feasibility of the proposed LAA in access and backhaul applications, real packet transmissions with different MCSs were performed under field conditions. Three experimental scenarios were investigated: device to device (D2D) linking (single PAA module to single PAA module transmission), access linking (LAA to single PAA module) and long-range backhaul (LAA to LAA).
The use of single modules without lenses (D2D) allowed transmissions with the highest data rate in SC mode (MCS #12: 4.7 Gbps physical layer rate, 16-QAM with coding rate 3 4 ; see Table 1) at distances up to 15 m. The application of a lens at the transmitter side achieved the maximum data rate (4.7 Gbps), making 16 QAM modulation practical for outdoor communications at distances of about 30-35 m, which can be appropriate for millimeter wave access links.
A second experiment investigated the backhaul link on the base of the two LAA antennas. The outdoor field trial setup for LAA is shown in Figure 9. This study demonstrated that the LAA is a practical high-gain steerable millimeter wave antenna for use in future millimeter wave overlay networks (D2D, access and backhaul) within the framework of 5G communication systems.
In LAA, we have used one antenna array module with 2 × 8 elements including the RFIC chip, and one BB module for digital signal processing. The total power consumption was about 4-6 Watts (from direct measurements) depending on regime of work (RX or TX). For the MAA prototype, we have used eight antenna array modules (each with 2 × 8 elements and one RFIC chip) and one, common, BB module for digital signal processing. Thus, the total power consumption of the eight-module MAA was estimated as 20-25 Watts. Therefore, both the LAA and MAA equipment could be power supplied with PoE.
linking (single PAA module to single PAA module transmission), access linking (LAA to single PAA module) and long-range backhaul (LAA to LAA).
The use of single modules without lenses (D2D) allowed transmissions with the highest data rate in SC mode (MCS #12: 4.7 Gbps physical layer rate, 16-QAM with coding rate ¾; see Table 1) at distances up to 15 m. The application of a lens at the transmitter side achieved the maximum data rate (4.7 Gbps), making 16 QAM modulation practical for outdoor communications at distances of about 30-35 m, which can be appropriate for millimeter wave access links.
A second experiment investigated the backhaul link on the base of the two LAA antennas. Figure 8 gives a screenshot of the constellation diagrams obtained for MCS #12 transmissions at 100, 150 and 200 m. It was demonstrated that the backhaul link with two LAA gave the highest data transmission rate of 4.7 Gbps at distances up to 200 m, with a negligible packet error rate. The outdoor field trial setup for LAA is shown in Figure 9. This study demonstrated that the LAA is a practical high-gain steerable millimeter wave antenna for use in future millimeter wave overlay networks (D2D, access and backhaul) within the framework of 5G communication systems. In LAA, we have used one antenna array module with 2 × 8 elements including the RFIC chip, and one BB module for digital signal processing. The total power consumption was about 4-6 Watts (from direct measurements) depending on regime of work (RX or TX). For the MAA prototype, we have used eight antenna array modules (each with 2 × 8 elements and one RFIC chip) and one, common, BB module for digital signal processing. Thus, the total power consumption of the eight-module MAA was estimated as 20-25 Watts. Therefore, both the LAA and MAA equipment could be power supplied with PoE

Proof of Concept Implementation and Integration
An overview of our PoC is given in Figures 10 and 11. The PoC hardware included the backhaul and access equipment. IP-based wireless backhaul linking was established using a reflectarray antenna with an integrated WiGig transceiver. LAA was used for access. A HetNet access network was used, with an access link provided in both 2-GHz LTE and 60-GHz WiGig. CP/UP splitting was

Proof of Concept Implementation and Integration
An overview of our PoC is given in Figures 10 and 11. The PoC hardware included the backhaul and access equipment. IP-based wireless backhaul linking was established using a reflectarray antenna with an integrated WiGig transceiver. LAA was used for access. A HetNet access network was used, with an access link provided in both 2-GHz LTE and 60-GHz WiGig. CP/UP splitting was introduced in the HetNet. A multi-RAT CP application implemented in LTE eNB managed UP data offloading between the LTE and WiGig.
An overview of our PoC is given in Figures 10 and 11. The PoC hardware included the backhaul and access equipment. IP-based wireless backhaul linking was established using a reflectarray antenna with an integrated WiGig transceiver. LAA was used for access. A HetNet access network was used, with an access link provided in both 2-GHz LTE and 60-GHz WiGig. CP/UP splitting was introduced in the HetNet. A multi-RAT CP application implemented in LTE eNB managed UP data offloading between the LTE and WiGig.

LTE/WiGig-Integrated HetNet Prototype
Our PoC hardware for the mmWave-integrated HetNet used an LTE eNB as the macrocell and WiGig APs as the small cells. Figure 12 is a photograph of our PoC hardware. The LTE macrocell comprised an eNB baseband unit (BBU) and a remote radio head (RRH) connected via an optical fronthaul based on the European Telecommunications Standards Institute (ETSI) Open Radio Interface (ORI) standard [66,67]. This is capable of transferring LTE baseband I/Q signals with a 50% compressed data rate. The WiGig AP was connected to the LTE eNB via Ethernet. The UE had LTE modem and WiGig devices were implemented. The WiGig device is the shown in Figure 2 and supported MCSs up to Index #9, as shown in Table 1. (a) (b) Figure 11. Hardware set-up of HetNet PoC.

LTE/WiGig-Integrated HetNet Prototype
Our PoC hardware for the mmWave-integrated HetNet used an LTE eNB as the macrocell and WiGig APs as the small cells. Figure 12 is a photograph of our PoC hardware. The LTE macrocell comprised an eNB baseband unit (BBU) and a remote radio head (RRH) connected via an optical fronthaul based on the European Telecommunications Standards Institute (ETSI) Open Radio Interface (ORI) standard [66,67]. This is capable of transferring LTE baseband I/Q signals with a 50% compressed data rate. The WiGig AP was connected to the LTE eNB via Ethernet. The UE had LTE modem and WiGig devices were implemented. The WiGig device is the shown in Figure 2 and supported MCSs up to Index #9, as shown in Table 1. Figure 13 shows the protocol stack of the PoC. This was similar to LWIP, but with the security protections omitted for simplicity. Security is an important feature of mobile networks, but was not the focus of the current PoC. We will return to this feature in future work. An LTE-WiGig-integrated CP (multi-RAT CP) application was implemented on top of the CP protocol stack. The multi-RAT CP monitored the real-time traffic load and radio link conditions, such as the received signal-strength indicator (RSSI) of both the LTE and WiGig. The measured link condition on UE was reported to the eNB by CP signaling via LTE. On the basis of the monitored information, the multi-RAT CP on the eNB managed the handling of the UP data. The UP data path was switched between LTE and WiGig by Open vSwitch (OVS) [68]. When the WiGig was used, the multi-RAT CP controlled activation and authentication of the WiGig. PDCP SDUs were sent via the WiGig or LTE link. Since the interface between the LTE-eNB and WiGig-AP was IP-based, various Layer-2 media can be used for the backhaul, including the 1, 2.5, 5 or 10 Gigabit Ethernet or mmWave backhauls presented in Sections 3.1 and 3.2.

LTE/WiGig-Integrated HetNet Prototype
Our PoC hardware for the mmWave-integrated HetNet used an LTE eNB as the macrocell and WiGig APs as the small cells. Figure 12 is a photograph of our PoC hardware. The LTE macrocell comprised an eNB baseband unit (BBU) and a remote radio head (RRH) connected via an optical fronthaul based on the European Telecommunications Standards Institute (ETSI) Open Radio Interface (ORI) standard [66,67]. This is capable of transferring LTE baseband I/Q signals with a 50% compressed data rate. The WiGig AP was connected to the LTE eNB via Ethernet. The UE had LTE modem and WiGig devices were implemented. The WiGig device is the shown in Figure 2 and supported MCSs up to Index #9, as shown in Table 1.  Figure 13 shows the protocol stack of the PoC. This was similar to LWIP, but with the security protections omitted for simplicity. Security is an important feature of mobile networks, but was not the focus of the current PoC. We will return to this feature in future work. An LTE-WiGig-integrated CP (multi-RAT CP) application was implemented on top of the CP protocol stack. The multi-RAT CP monitored the real-time traffic load and radio link conditions, such as the received signal-strength indicator (RSSI) of both the LTE and WiGig. The measured link condition on UE was reported to the eNB by CP signaling via LTE. On the basis of the monitored information, the multi-RAT CP on the eNB managed the handling of the UP data. The UP data path was switched between LTE and WiGig by Open vSwitch (OVS) [68]. When the WiGig was used, the multi-RAT CP controlled activation and authentication of the WiGig. PDCP SDUs were sent via the WiGig or LTE link. Since the interface between the LTE-eNB and WiGig-AP was IP-based, various Layer-2 media can be used for the  To demonstrate the CP/UP splitting concept, we evaluated the performance of LTE-WiGig handover. The experimental setup is shown in Figure 14. The WiGig station (STA) device was attached to a linear actuator and moved horizontally at a constant velocity of 83 mm/s. The vertical distance between the AP and STA antennas was set to 200 mm, so that the STA moved into and out of the coverage area of the WiGig. Switching of the UP data path was based on the availability of the WiGig link and the reported RSSI. If the WiGig link were available and the RSSI exceeded the To demonstrate the CP/UP splitting concept, we evaluated the performance of LTE-WiGig handover. The experimental setup is shown in Figure 14. The WiGig station (STA) device was attached to a linear actuator and moved horizontally at a constant velocity of 83 mm/s. The vertical distance between the AP and STA antennas was set to 200 mm, so that the STA moved into and out of the coverage area of the WiGig. Switching of the UP data path was based on the availability of the WiGig link and the reported RSSI. If the WiGig link were available and the RSSI exceeded the threshold RSSI th , the UP data was offloaded to the WiGig link. Otherwise, it was carried on the LTE link.

PoC Applications
Two applications were selected to showcase the benefits of the mmWave-integrated cellular networks proposed by MiWEBA: "Context Aware Caching" and "Small Cell On/Off".
The setup for "Context Aware Caching" is shown in Figure 16. It comprises a dual-connectivity user device, a macro base station (LTE eNB) and an mmWave small cell (mmWave AP). The client runs a video streaming application while on the move. A context aware cache entity, lying as a middle layer between the video application and the network connection, performs the data requests, using the connectivity status (mmWave connection available or unavailable) as context information. The video file (served as multiple small chunks in Hypertext Transfer Protocol Live Streaming (HLS) format) is fetched from the content server and made available to the video player. When within the coverage of an mmWave small cell, the cache pre-fetches and stores the video file chunks locally at  Figure 15 shows the transition of the instantaneous UP throughput during an LTE-WiGig handover. The downlink UP data were transmitted from the eNB in TCP using Iperf [69]. The handover decision threshold RSSI th was set to −60 dBm, and the throughput was measured in the MAC layer. From the figure, it can be seen that the data path was switched smoothly, depending on the UE location, and that a throughput of more than 1 Gbps was achieved when the UP path was switched to WiGig.

PoC Applications
Two applications were selected to showcase the benefits of the mmWave-integrated cellular networks proposed by MiWEBA: "Context Aware Caching" and "Small Cell On/Off".
The setup for "Context Aware Caching" is shown in Figure 16. It comprises a dual-connectivity user device, a macro base station (LTE eNB) and an mmWave small cell (mmWave AP). The client

PoC Applications
Two applications were selected to showcase the benefits of the mmWave-integrated cellular networks proposed by MiWEBA: "Context Aware Caching" and "Small Cell On/Off".
The setup for "Context Aware Caching" is shown in Figure 16. It comprises a dual-connectivity user device, a macro base station (LTE eNB) and an mmWave small cell (mmWave AP). The client runs a video streaming application while on the move. A context aware cache entity, lying as a middle layer between the video application and the network connection, performs the data requests, using the connectivity status (mmWave connection available or unavailable) as context information. The video file (served as multiple small chunks in Hypertext Transfer Protocol Live Streaming (HLS) format) is fetched from the content server and made available to the video player. When within the coverage of an mmWave small cell, the cache pre-fetches and stores the video file chunks locally at the maximum achievable rate. For this purpose, the cache creates a database of all chunks in the currently playing video. A status is associated with each chunk indicating whether it is locally available, currently downloading or not locally available. The most recently played chunk is stored as a pointer and used to decide which chunk is downloaded next. When outside the range of a small cell, the cache fetches the content on a just-in-time basis to reduce the load on the LTE link. From the user's perspective, the content is available seamlessly. We demonstrated [70] "Context Aware Caching" by using the PoC and confirmed that the downloading rate was boosted dramatically while an mmWave link was available. user's perspective, the content is available seamlessly. We demonstrated [70] "Context Aware Caching" by using the PoC and confirmed that the downloading rate was boosted dramatically while an mmWave link was available. The second demonstration of PoC was the ability to control and optimize the operation of the network resources in terms of energy efficiency by using the developed multi-RAT CP [71,72]. The centralized RAN (C-RAN), which is LTE eNB in this case, processes context information on its mmWave small cell base stations, including the number of connected user devices and their traffic demand. Depending on the load situation, the C-RAN turned off small cells via the multi-RA CP to save energy. As the user devices were always connected to the CP, the small cells could be immediately reactivated to meet traffic demand. Figure 17 shows a snapshot of small cell on/off and its energy efficiency. This is a result of a system level simulation of the mmWave-integrated cellular networks; however, the function used in this simulation can be realized by the PoC hardware shown in Section 4.1. The left of Figure 17 shows the status of small cell on/off, and the right shows the energy efficiency of the network measured by (bps/W). In the left, the distribution of UEs (blue circles) is not uniform, and there are several hotspots whose locations change in space and time. In this application, the C-RAN implemented in LTE macro eNBs (indicated by yellow stars in the figure) collects traffic demands from UEs. Based on this information, the C-RAN dynamically controls the status of small cell BSs as on (filled red circle) or off (unfilled red circle) to maximize the energy efficiency. In the case of on, the C-RAN immediately switches the UP to small cell BSs as described in Section 4.1. The right of Figure 17 shows the realized energy efficiency by changing time in a day. The performance is calculated with two different frequencies of 60 GHz and 3.5 GHz. Since the coverage (footprint) of 60 GHz is much smaller than 3.5 GHz, there is higher impact of small cell on/off in 60 GHz (purple line) than 3.5 GHz (green line). It is also clear that the performance of energy efficiency maximization (purple line) in the 60-GHz band has much better performance in terms of energy efficiency compared to that of system rate maximization (yellow line) in 60 GHz without the control of small cell on/off. In this demonstration, we just showed a snapshot of "Small Cell On/Off"; however, the details of optimization algorithm and simulation results can be seen in [71,72]. The second demonstration of PoC was the ability to control and optimize the operation of the network resources in terms of energy efficiency by using the developed multi-RAT CP [71,72]. The centralized RAN (C-RAN), which is LTE eNB in this case, processes context information on its mmWave small cell base stations, including the number of connected user devices and their traffic demand. Depending on the load situation, the C-RAN turned off small cells via the multi-RA CP to save energy. As the user devices were always connected to the CP, the small cells could be immediately reactivated to meet traffic demand. Figure 17 shows a snapshot of small cell on/off and its energy efficiency. This is a result of a system level simulation of the mmWave-integrated cellular networks; however, the function used in this simulation can be realized by the PoC hardware shown in Section 4.1.
The left of Figure 17 shows the status of small cell on/off, and the right shows the energy efficiency of the network measured by (bps/W). In the left, the distribution of UEs (blue circles) is not uniform, and there are several hotspots whose locations change in space and time. In this application, the C-RAN implemented in LTE macro eNBs (indicated by yellow stars in the figure) collects traffic demands from UEs. Based on this information, the C-RAN dynamically controls the status of small cell BSs as on (filled red circle) or off (unfilled red circle) to maximize the energy efficiency. In the case of on, the C-RAN immediately switches the UP to small cell BSs as described in Section 4.1. The right of Figure 17 shows the realized energy efficiency by changing time in a day. The performance is calculated with two different frequencies of 60 GHz and 3.5 GHz. Since the coverage (footprint) of  60 GHz is much smaller than 3.5 GHz, there is higher impact of small cell on/off in 60 GHz (purple line) than 3.5 GHz (green line). It is also clear that the performance of energy efficiency maximization (purple line) in the 60-GHz band has much better performance in terms of energy efficiency compared to that of system rate maximization (yellow line) in 60 GHz without the control of small cell on/off. In this demonstration, we just showed a snapshot of "Small Cell On/Off"; however, the details of optimization algorithm and simulation results can be seen in [71,72].

Conclusions
In this contribution, we disclosed a piloted proof-of-concept (PoC) of an mmWave-integrated heterogeneous network (HetNet) utilizing mmWave technologies both for backhaul and access for integration into cellular networks. In the HetNet, CP/UP splitting was implemented to allow user plane access in the mmWave small cells, while retaining a reliable control plane connection to the macro cell. Passive high-gain antenna technology in mmWave was applied for wireless backhauling of the small cells. A lens antenna array with steerable beam was used for mmWave access. This PoC was conducted to demonstrate the practicality of the mmWave-integrated HetNet. Successful CP/UP splitting between the conventional LTE and mmWave WiGig was demonstrated, and seamless connectivity was achieved between the different technologies. The high-throughput performance of the long-range backhaul link using a passive reflectarray was also confirmed. Although the mmWave access targets high-speed data communication in the first phase of 5G, the proposed architecture may be applied to future ultra-real-time and low latency sensor networks as required in autonomous driving that will be realized in the later phase of 5G. The achieved results will be contributed respectively to 3GPP, IEEE, ITU and ETSI. designed the passive reflectarray and backhaul prototype. Kei Sakaguchi provided literature surveys on the latest mmWave PoC. Andrey Pudeyev and Alexander Maltsev designed and conducted field experiments of LAA prototype. Richard J. Weiler, Ingolf Karls and Kei Sakaguchi led the PoC in the MiWEBA project as coordinators.

Conflicts of Interest:
The authors declare no conflict of interest. The funding sponsors had no role in the design of the study; in the collection, analysis or interpretation of data; in the writing of the manuscript; nor in the decision to publish the results.

Conclusions
In this contribution, we disclosed a piloted proof-of-concept (PoC) of an mmWave-integrated heterogeneous network (HetNet) utilizing mmWave technologies both for backhaul and access for integration into cellular networks. In the HetNet, CP/UP splitting was implemented to allow user plane access in the mmWave small cells, while retaining a reliable control plane connection to the macro cell. Passive high-gain antenna technology in mmWave was applied for wireless backhauling of the small cells. A lens antenna array with steerable beam was used for mmWave access. This PoC was conducted to demonstrate the practicality of the mmWave-integrated HetNet. Successful CP/UP splitting between the conventional LTE and mmWave WiGig was demonstrated, and seamless connectivity was achieved between the different technologies. The high-throughput performance of the long-range backhaul link using a passive reflectarray was also confirmed. Although the mmWave access targets high-speed data communication in the first phase of 5G, the proposed architecture may be applied to future ultra-real-time and low latency sensor networks as required in autonomous driving that will be realized in the later phase of 5G. The achieved results will be contributed respectively to 3GPP, IEEE, ITU and ETSI.

Conflicts of Interest:
The authors declare no conflict of interest. The funding sponsors had no role in the design of the study; in the collection, analysis or interpretation of data; in the writing of the manuscript; nor in the decision to publish the results.