Green-oriented multi-techno link adaptation metrics for 5G heterogeneous networks

Abstract

Fifth generation (5G) technical requirements turned toward end-to-end capacity increase with limited latency and constant network power consumption involve dedicated research topics. Multiple radio access techniques are one of the most promising paradigm safeguarding throughput increase and seamless connectivity in heterogeneous networks. Link adaptation metrics are then required to achieve radio link selection and multiple interface management. Following mobile network capacity increase trends, link adaptation metrics focused on spectrum efficiency maximization are implicitly sought, involving high network power consumption. This paper presents a novel link adaptation metric turned toward a power efficient radio link selection which warrants QoS and performs a fair comparison between independent air interfaces in a multiple radio access technique context. A procedure is proposed to practically evaluate the metric upon each air interface, thanks to the introducion of new normalized propagation selectivity parameters with respect to system parameters allowing a unified characterization of the propagation link reliability, tuning up the metric decision. Performance and transmit power gains resulting from the metric application are appraised upon wireless hot spot extensions covering 5GHz and 60GHz Wi-Fi technologies and mm-wave ultra-wide-band transmissions.

1 Introduction

Fifth generation (5G) technical requirements [1] turned toward 1000 time end-to-end capacity increase with latency up to 1 ms and optimized network power consumption involve new research topics. Multiple radio access techniques (multi-RAT) deployed in heterogeneous networks (HetNets) is one of the most promising solutions to improve network capacity and ensure seamless connectivity with a high quality of service (QoS). Multi-RAT management processing [2] performed with power efficiency (PE) link adaptation techniques and scalable spectrum management will ensure seamless connectivity in a high multi-user context, limiting multi-user interference, and guarantying QoS along with radio coverage. Innovative power efficient cooperative networks justify matter of necessity to develop new link adaptation metrics.

The paper introduces and evaluates a novel link adaptation metric turned toward PE radio link selection intended for multi-RAT cooperative networks. The present work has been initiated in the ICT-FP7 METIS project inquiring into new physical (PHY) waveforms and multi-RAT architectures for 5G (https://www.metis2020.com/) [3, 4]. Embedded solutions adapted to mm-wave overlay networks are currently outperformed in the ICT-FP7 Europe-Japan MiWEBA project (http://www.miweba.eu/#Project). MiWEBA addresses mm-wave ultra-broadband phantom cells integrated in mobile cellular networks and backhauling scenarios [5–7]. Multi-RAT architectures are examined, regarding novel power-efficient link adaptation metrics enable to switch between RATs [2] and their integration into innovative control/user (C/U) plane splitting architectures [8–10] in accordance with the 3GPP wireless local area network/Long Term Evolution-Advanced (WLAN/LTE-A) carrier aggregation study item [11]. In the GreenTouch Consortium [12] and MiWEBA project [8], new C-plane information exchanges steer small cell activation for U-plane transmissions.

Link adaptation techniques strive for spectrum efficiency (SE) optimization following throughput increase target settings for embedded 5G networks [13, 14]. Holistic solutions combine adaptive modulation and coding (AMC) at the physical layer and hybrid automatic retransmission request (HARQ) at the medium access control layer to improve SE and link reliability [15]. Differentiated link adaptation processing [16] is tackled to perform a personalized user HARQ adaptation. 3GPP and Wi-Fi physical layer procedures [17, 18] embrace a SE link adaptation processing for multi-RAT offloading [13]. Trade-off between SE and energy-efficient link adaptation processing is undertaken on LTE-A, addressing dynamic and static power consumption of digital baseband processing at the PHY layer [19].

Section 2 exposes the SE link adaptation techniques focused on throughput maximization bound with received signal strength and AMC to select transmission modes (TMs) in HetNets. Section 3 details the novel PE link adaptation metric proposed in the paper, which is denoted the Green Link Budget (GLB) metric, and accounts for assessment methods for software implementation in multi-RAT architectures. Section 4 deals with the performance and transmit power gains resulting from the GLB metric application upon MiWEBA test cases [5] covering mm-wave ultra-wideband (UWB) transmissions [20] and Wi-Fi extensions for indoor and hot spots. Conclusions of this paper detail perspectives of this work currently carried out in the MiWEBA project.

2 A focus on spectrum-efficient link adaptation processing

Spectrum efficiency based link adaptation techniques follow throughput increase target settings for embedded heterogeneous networks and mobile data offloading. These techniques result from several AMC figures as cited in Section 1, usually applied on a single air interface able to operate in several radio frequency bands combined with access network discovery and selection function (ANDSF) to control offloading between 3GPP and non-3GPP access networks [21]. Hence, in a single air interface context, link adaptation metrics select the highest possible data rate mode associated to the largest modulation order in accordance with a measured received signal strength indicator (RSSI) related to a maximum block error rate (BLER) threshold and look-up tables (LUTs) giving the correspondence between an equivalent signal-to-noise ratio (SNR) and the modulation and coding scheme (MCS) [13–16, 18, 19, 22].

An illustration is done for the Long Term Evolution-Advanced (LTE-A) downlink (DL) transmissions following 3GPP technical specifications [17, 23] and literature issues [16, 18, 19, 22]. In LTE-A, the evolved node base station (E-Node-B) makes the decision of the selected MCS for DL transmissions, using proprietary algorithms and a channel quality indicator (CQI) feedback reported by the user equipment (UE) to the E-Node B. As shown on Fig. 1, the E-node B forwards dedicated pilots to the UE in order to estimate RSSI on the DL. The UE converts the RSSI into an energy-per-bit-to-noise-power-spectral-density ratio, Ec/N₀, and determines a CQI index such that it corresponds to the highest MCS allowing the UE to decode the transport block with error rate probability not exceeding 10 % [18, 22]. From the BLER-versus-SNR plot, the SNR value for acceptable 10 % BLER is taken for each CQI value ranging from 1 to 15. The UE converts it into a 4-bit CQI word and forwards it to the E-Node B upon a reverse link using the channel quality indicator channel (CQICH) which supports forward link packet data operations. The E-Node B converts the 4-bit CQI word into SNR and uses proper CQI-to-SNR mapping tables to select MCS and throughput for DL. The E-Node B takes the decision of the MCS for the DL transmission which is close as possible as the UE CQI feedback.

Fig. 1

LTE-A AMC mechanism using CQI-to-MCS mapping

A UE with receiver of better quality may report better CQI and thus can receive downlink data with a higher MCS. The procedure is illustrated on the Fig. 1 giving referenced standards in the process. The correspondence between Ec/No, 4-bit CQI world, CQI index, and MCS parameters is compiled in the table using [18, 22] information. CQI-to-SNR mapping for the LTE-A is illustrated on Fig. 2. The left and right figures show that the CQI index hikes up with the modulation order M, data throughput and required SNR levels guarantying BLER targets.

Fig. 2

CQI value and modulation level correspondence, from [18]

An analytical expression is given in [19] to establish the relation between the SE, MCS, and power requirements upon DL LTE-A transmissions:

$$ \begin{array}{l}\mathrm{S}\mathrm{E}=\frac{1}{2}\frac{\left(1-\mathrm{BLER}\right){R}_c{N}_{\mathrm{cbps}}L}{10^{-3}}\hfill \\ {}M\cdot {R}_c={2}^{N_{\mathrm{cbps}}}{R}_c=\frac{1}{2}\frac{10^{-3}}{\left(1-\mathrm{BLER}\right)L}\cdot \left(1+\frac{\mathrm{EIRP}\cdot {g}_n}{B{N}_0}\right)\hfill \end{array} $$

(1)

R _c is the coding rate, N _cbps is the number of encoded bit per modulation constellation point, i.e., per subcarrier, and L is the number of subcarriers per resource block (RB) in the LTE-A subcarrier assignment. g _n is the average channel gain on RB, N ₀ is the noise power spectral density, and B is the bandwidth per RB. EIRP is the emitted isotropic radiated power. Equation (1) shows that, for a fixed BLER, the SE is linearly increasing with the number of information bits per subcarrier given by R _c N _cbps. The modulation order M linearly builds up with a transmit EIRP expansion and an equivalent SNR derived from the average channel gain g _n in a RB.

In IEEE802.11 technologies, RSSI is acquired during the preamble stage of receiving an 802.11 frame and the RSSI is adjusted accordingly to manufacturers with CQI index range set to 0 to 100.

The proposed link adaptation (LA) metric, exposed in the next section which is denoted the green link budget (GLB) metric, works with a targeted data throughput attached to the transported service and QoS, which may be delivered by several air interfaces embracing several MCS, variable bandwidths, and different types of baseband processing. The selection is steered toward a power efficient criterion which may exhibit variable SE levels.

3 The Green Link Budget metric model

The GLB metric, composed of two metrics α and β, is a multi-technology link adaptation (MT-LA) metric, which selects transmission modes (TMs) of various interfaces in targeting on a data throughput determined by the service to transport and a selection emphasizing transmission modes revealing lowest degradations at the PHY layer and the minimum propagation path-loss deduced from the received power measurements. TMs stand for baseband processing and radio frequency (RF) specifications at the PHY layer. In addition, the GLB metric performs a dynamic power control on selected TMs transacted by the β metric.

In order to compare independent interfaces with different multipath sensitivity levels as well as different radio frequency carriers and power regulation rules, the α metric of the GLB metric computes normalized link budget parameters with respect to idealistic cases given by additive white Gaussian noise (AWGN) performance at the PHY layer and free space path-loss connected to a perfect propagation link. The multipath power sensitivity S _M, expressed in decibel-milliwatts (dBm), is the minimum required received power at the PHY layer of a given TM which guaranties the QoS settled by a maximum BLER or bit error rate (BER) on data transmission.

The GLB metric differs from the SE link adaptation processing in working with a throughput range fixed by the concerned service instead of the highest possible throughput as performed by AMC processing (Section 2). It is the first element arguing on PE oriented radio link selection. Furthermore, the proposed procedure to estimate the metric detailed in Section 3.2, differentiates the line of sight (LOS) and obstructed and non-line of sight (OLOS/NLOS) propagation links, improving the selection with a conceivable transmit power reduction. The next subsection details the meaning and the mathematical description of the {α,β} metric.

3.1 The GLB metric model

The α-metric evaluates, for each TM in a dedicated deployment scenario, two relative degradation parameters, the multipath channel margin (MCM) derived from the PHY/MAC layer link level performance and the path-loss margin (PLM) deduced from the path-loss propagation signature characterized by obstacles obstructing the Fresnel zone. MCM expressed in decibels, provides, for a given TM, the exceeded SNR which is compulsory to generate the same BER/BLER as in AWGN. MCM may be alternatively estimated in either considering the difference between the multipath power sensitivity S _M and the AWGN power sensitivity S _AWGN or the required SNR difference between the multipath and idealistic case (SNR_AWGN) as expressed below:

$$ \begin{array}{ll}\mathrm{M}\mathrm{C}\mathrm{M}={S}_{\mathrm{M}}-{S}_{\mathrm{AWGN}}=\mathrm{S}\mathrm{N}\mathrm{R}-{\mathrm{SNR}}_{\mathrm{AWGN}}\hfill & \left(\mathrm{d}\mathrm{B}\right)\hfill \\ {}{S}_{\mathrm{AWGN}}={\mathrm{SNR}}_{\mathrm{AWGN}}+\underset{{}_{P_N}}{\underbrace{10\cdot \mathrm{Log}\left(kT{B}_w\right)}} + {L}_0\hfill & \left(\mathrm{d}\mathrm{B}\mathrm{m}\right)\hfill \\ {}{S}_{\mathrm{M}}=\mathrm{S}\mathrm{N}\mathrm{R}+\underset{{}_{P_N}}{\underbrace{10\cdot \mathrm{Log}\left(kT{B}_w\right)}} + {L}_0\hfill & \left(\mathrm{d}\mathrm{B}\mathrm{m}\right)\hfill \\ {}{S}_{\mathrm{ref}}=10\cdot \mathrm{Log}\left(k{T}_{\mathrm{ref}}{B}_{\mathrm{ref}}\right)=-114\hfill & \left(\mathrm{d}\mathrm{B}\mathrm{m}\right),\hfill \\ {}\mathrm{N}\mathrm{F}=10\cdot \mathrm{Log}\left(\frac{T}{T_{\mathrm{ref}}}\right),{B}_{\mathrm{ref}}=1\mathrm{MHz}\hfill & \left(\mathrm{d}\mathrm{B}\right)\hfill \\ {}{S}_{\mathrm{M}}=\mathrm{S}\mathrm{N}\mathrm{R}+\mathrm{N}\mathrm{F}+{S}_{\mathrm{ref}}+10\cdot \mathrm{Log}\left({B}_w\right) + {L}_0\hfill & \left(\mathrm{d}\mathrm{B}\mathrm{m}\right)\hfill \end{array} $$

(2)

As shown in (2), the power sensitivity is calculated from SNR and thermal noise power P _N; the amplifier non-linearity’s is neglected. The multipath power sensitivity, S _M,, translates the required power level to transmit data with a targeted BER associated to a TM delivering a data throughput D, in a multipath context. S _AWGN is the power sensitivity under AWGN channel related to the same TM and throughput D. S _ref is the thermal noise power with a reference noise temperature set to 290 K, k is the Boltzmann constant, NF is the noise figure affiliated to a noise temperature T, bandwidth (Bw) is the efficient transmission bandwidth, and Lo describes various wired connectivity losses involved by imperfect hardware connections at the transceiver and receiver sides.

Figure 3 illustrates the MCM values affiliated to two MCS deduced from link level simulations performed on the IEEE 802.11ac standard [24] in single input single output (SISO) configurations. MCM is deduced from the BER plot versus the SNR associated with MCS and QoS materialized by maximum BLER thresholds fixed by the QoS. The MCM#1 linked to QPSK 5/8 MCS is the difference in dB between the required SNR in multipath (MP-QPSK 5/8 curve on Fig. 3) and AWGN (AWGN-QPSK 5/8 curve on Fig. 3) along QoS#1 in conjunction with a BER set to 10⁻⁵. A similar illustration is given with MCM#2 (BER versus SNR plots with blue curve), considering 16-quadrature amplitude modulation (QAM) modulated signals with a forward error coding (FEC) rate set to three fourths (16 QAM 3/4) and a maximum BER adjusted to 10⁻³ (QoS#2 is representative of voice service, Fig. 3).

Fig. 3

The multipath channel margin definition

The path-loss channel margin (PLM) detailed in (5), measures, in dB, the path-loss excess involved by obstacles with respect to the free space path-loss without obstruction. PLM varies with the distance d between the transmitter and the receiver, in relation with multipath PL_MFS(d,fc) and free space path-loss PL_FS(d,fc) models expressed below:

$$ \begin{array}{l}{\mathrm{PL}}_{\mathrm{MFS}}\left(d,\mathrm{fc}\right)=P{L}_0\left({d}_0,\mathrm{fc}\right)+10n{\mathrm{Log}}_{10}\left(\frac{d}{d_0}\right) + \sigma \hfill \\ {}{\mathrm{PL}}_0\left({d}_0,\mathrm{fc}\right)\approx {\mathrm{PL}}_{\mathrm{FS}}\left({d}_0,\mathrm{fc}\right)\hfill \\ {}{\mathrm{PL}}_{\mathrm{FS}}\left(d,\mathrm{fc}\right) = 20\ {\mathrm{Log}}_{10}\left(\frac{4\pi }{c}\right) + 20{\mathrm{Log}}_{10}\left(\mathrm{fc}\right) + 20\cdot {\mathrm{Log}}_{10}(d)\hfill \end{array}. $$

(3)

In (3), PL₀(d ₀,fc) is a path-loss offset derived from path-loss measurements at a distance d ₀ between the transmitter and the receiver in which the attenuation may be assimilated to the free space path-loss when distance d ₀ is small. PL_FS(d,fc) is the free space path model computed from the Friis equation where c is the speed light, fc is the RF carrier and d is the distance between the transmitter and the receiver. The coefficient n is the exponential coefficient (logarithmic dependency in dB) dependency with the distance d between the transmitter and the receiver and σ is the standard deviation of measurements.

PLM may be either deduced from the path-loss difference between realistic case with obstructions and perfect propagation link, or the available received power difference between multipath, ARP_MFS(d,fc), and idealistic ARP_FS(d,fc) deduced from the free space path-loss. Received power levels are equivalent to the RSSI variations linearly increasing in dB with the EIRP, and depending on receiver antenna gain G _R and path-loss models as detailed in (4):

$$ \begin{array}{l}{\mathrm{ARP}}_{\mathrm{MFS}}\left(d,\mathrm{fc}\right) = \mathrm{EIRP} + {G}_{\mathrm{R}}-{\mathrm{PL}}_{\mathrm{MFS}}\left(d,\mathrm{fc}\right)\hfill \\ {}{\mathrm{ARP}}_{\mathrm{MFS}}\left(d,\mathrm{fc}\right) = \mathrm{EIRP} + {G}_{\mathrm{R}}-{\mathrm{PL}}_{\mathrm{FS}}\left({d}_0,\mathrm{fc}\right)-10n{\mathrm{Log}}_{10}\left(\frac{d}{d_0}\right)-\sigma \hfill \\ {}{\mathrm{ARP}}_{\mathrm{FS}}\left(d,\mathrm{fc}\right) = \mathrm{EIRP} + {G}_{\mathrm{R}}-{\mathrm{PL}}_{\mathrm{FS}}\left(d,\mathrm{fc}\right).\hfill \end{array} $$

(4)

The algebraic expression of MCM and PLM are given by

$$ \begin{array}{l}\alpha = {S}_{\mathrm{M}}-{S}_{\mathrm{AWGN}} + \underset{\mathrm{PL}\mathrm{M}}{\underbrace{{\mathrm{PL}}_{\mathrm{M}\mathrm{FS}}\left(d,\mathrm{fc}\right)-{\mathrm{PL}}_{\mathrm{FS}}\left(d,\mathrm{fc}\right)}}\hfill \\ {}\alpha = \underset{\mathrm{M}\mathrm{CM}}{\underbrace{S_{\mathrm{M}}-{S}_{\mathrm{AWGN}}}}+\underset{\mathrm{PL}\mathrm{M}}{\underbrace{{\mathrm{ARP}}_{\mathrm{FS}}\left(d,\mathrm{fc}\right)-{\mathrm{ARP}}_{\mathrm{M}\mathrm{FS}}\left(d,\mathrm{fc}\right)}}\hfill \end{array}. $$

(5)

PLM variations geared to the distance d, associated to test cases developed in Section 4, are illustrated on Fig. 4. PLM values are derived from path-loss models given in Table 1 at 5 and 60 GHz RF carriers. At 5 GHz, PLM is computed, considering omnidirectional antennas at the transmitter and receiver sides. At 60 GHz, the transmitter azimuth antenna pattern is set to 72° and two different receiver azimuth antenna patterns set to 60° and 10° are selected to increase the radio coverage, limit obstructions, and compensate path-loss offset acquired by the RF carrier. Figure 4 shows that PLM at 5 GHz increases with the distance in both LOS and NLOS because the exponential coefficient n is higher than at 60 GHz due to large antenna pattern (Table 1) translating obstruction occurrence affecting the propagation link. At 60 GHz, the difference between LOS and NLOS situations is highly marked. In LOS, PLM decreases with the distance meaning that multipath path-loss fast converge toward free space path-loss because of directive antenna use at the transmitter and receiver sides.

Fig. 4

Path-loss margin at 5 and 60 GHz

Table 2 Normalized propagation parameters

Table 1 Path-loss models at 5 and 60 GHz

The β-metric performs a dynamic power control at the receiver side based on an adjustment between the received power, ARP_MFS(d,fc), and the required power S _M of the various candidate TMs resulting from the α-metric decision. Selection is conjunctly finalized with α-metric minimization and β-metric maximization before the power control management resulting in EIRP adjustment by the way of β-metric refinements.

The algebraic expression of the β-metric is given by

$$ \begin{array}{l}\beta ={\mathrm{ARP}}_{\mathrm{M}\mathrm{FS}}\left(d,\mathrm{fc}\right)+{G}_{\mathrm{R}}-{S}_{\mathrm{M}}\hfill \\ {}\beta = EIRP+{G}_{\mathrm{R}}-\alpha -{S}_{\mathrm{AWGN}}-{\mathrm{PL}}_{\mathrm{FS}}\left(d,\mathrm{fc}\right)\hfill \end{array} $$

(6)

Figure 5 illustrates the {α,β} metric principles, derived from the general link budget assessments when considering a single TM denoted “TM1.” Radio coverage estimation is obtained when the available received power varying with the distance d converges to the multipath power sensitivity S _M which is a system dependant parameter. Plotted black circles on Fig. 5 give the equality between the available received power and the required power in both realistic and idealistic cases. The figure plots S _AWGN and S _M giving the MCM,1 value. In a similar way, the representation of the available received power in realistic and idealistic case (4) allows an estimation of PLM,1 varying with the distance. The β-metric varying with the distance d, is simply calculated with the difference between the available received power ARP_MFS(d,fc) and the multipath power sensitivity S _M represented on Fig. 5. The β-metric validity range is restricted to positive values. A power margin (PM) may be added to cover shadowing variations imposing a β-metric minimum threshold higher than PM dB. PM may differ rather we consider LOS or NLOS propagation links.

Fig. 5

The {α,β} metric representation for a single AI/TM

Figure 6 is in relation with the use case developed in Section 4.1 in which the user equipment may select two different access points (APs) using two different TMs exhibiting S _M1 and S _M2 multipath power sensitivity levels to forward the service. On Fig. 6, we assume the AWGN power sensitivity S _AWGN is similar for the two TMs. LOS and NLOS propagation conditions differ with the AP1 and AP2, respectively. The GLB {α,β} metric is evaluated on each TM/air interface (AI) associated to AP1 and AP2, respectively, supplying {α ₁,β ₁} and {α ₂,β ₂} metric variations. The AP selection is the one giving the lowest degradation, i.e., the minimum α-metric value and the largest β-metric value before transmit power adjustment applied on the selected AP by the α-metric. Distances may be different for these two links.

Fig. 6

The {α,β} metric representation with multiple LOS and NLOS links

3.2 Assessment methods of the GLB metric

The proposed method to practically measure the {α,β} metric, optimize β and refresh the selection, is centered on the MCM and PLM estimations using signaling frames and preamble headers associated to each interface and a LOS/NLOS discrimination thanks to the introduction of original normalized propagation selectivity parameters with respect to system parameters, leading to a refined radio link selection with the GLB metric.

CQI metrics usually use RSSI elements and BLER LUT supplying the correspondence between the BLER and the required SNR to select a MCS in a single AI (Section 2). The assessment method proposed for the GLB metric is similar, in providing, for each AI and TM, evolved LUT with a LOS/NLOS radio link identification, and the use of MCM and PLM outputs to compare independent interfaces. The problem statement is then to identify LOS/NLOS propagation conditions deduced from the channel state information. Normalized propagation parameters are then defined, characterizing the link reliability between different TMs and propagation conditions. These normalized multipath propagation parameters with respect to PHY system parameters point out on pre-established BER LUT differentiating LOS/NLOS criteria leading to required power levels in relation to BER/BLER targets. It then allows MCM, PLM, and α and β metric calculations upon LOS and NLOS conditions, respectively. RSSI values may provide elements used to evaluate PLM thanks to available distance information, known as EIRP and antenna characteristics given in PLCP headers of the concerned AI. In addition, the rice factor value deduced from the RSSI power distribution [25], may confirm the radio link obstruction degree. The power sensitivity is usually given in standards and may be registered in dedicated tables. The β-metric estimation may be deduced from α-metric assessments (6). The synoptic of MCM, PLM, and S _M estimation is illustrated on Fig. 7.

Fig. 7

The assessment procedure of the {α, β} metric and S _M

Normalized propagation parameters are derived from the equalization processing supplying the multipath channel state information (CSI), typically the equivalent baseband channel impulse response (CIR) of the propagation channel, h(t,τ), where h(t,ti), ti has to be written with the symbol policy translates time variations of complex amplitude of the channel, over time variant relative delay τi(t) of the CIR (7). The RMS delay spread, the coherence bandwidth with a correlation coefficient set to 0.5 and 0.9 are utilized to design normalized propagation parameters sensitive to LOS/NOS conditions able to point on appropriate BER look-up tables for any PHY layer systems.

The normalized delay spread diversity order O _DS is expressed in (8) as the ratio of the transmission symbol period T _SYM and the RMS delay spread σ _DS. The RMS delay spread σ _DS supplies the average standard deviation of multipath weighted by power probability γ _I of each delay i. The frequency diversity order O _FD in (8) is the ratio of the efficient Bw of the PHY system and the coherence bandwidth $\varDelta {f}_y^{\mathrm{cb}}$ of the propagation channel. The coherence bandwidth $\varDelta {f}_y^{\mathrm{cb}}$ is the frequency spacing for positive frequency components, providing a y factor decrease of the normalized average correlation function |RH(Δf)| magnitude of the channel with respect to no frequency deviation (Δf = 0). $\varDelta {f}_y^{\mathrm{cb}}$ is the coherence bandwidth with a correlation coefficient adjusted to y. P(t) is the instantaneous power of the propagation CIR h(t,τ). These parameters are expressed below

$$ \begin{array}{l}h\left(t,\tau \right)={\displaystyle \sum_{i=1}^Nh\left(t,{\tau}_i\right)\cdot \delta \Big(\tau -{\tau}_i}(t)\Big)\hfill \\ {}{\gamma}_i=E\left\{\frac{{\left|h\Big(t,{\tau}_i(t)\right|}^2}{P(t)}\right\},\kern1em P(t)={\displaystyle \sum_{i=1}^N{\left|h\left(t,{\tau}_i\right)\right|}^2}\hfill \\ {}{\sigma}_{\mathrm{DS}}=E\left\{{\sqrt{\frac{{{\displaystyle \sum_{i=1}^n{\tau}_i^2\left|h\left(t,{\tau}_i(t)\right)\right|}}^2}{P(t)}-\left(\frac{{{\displaystyle \sum_{i=1}^n{\tau}_i\left|h\left(t,{\tau}_i(t)\right)\right|}}^2}{P(t)}\right)}}^2\right\}\hfill \\ {}\mathrm{R}\mathrm{H}\left(\varDelta f\right)=E\left\{{\mathrm{FFT}}_{\tau}\left\{{{\displaystyle \sum_{i=1}^n\left|h\left(t,{\tau}_i(t)\right)\right|\Big)}}^2\right\}\left(\varDelta f\right)\right\}\hfill \end{array} $$

(7)

$$ \begin{array}{l}{B}_{\mathrm{cy}}=\varDelta {f}_y^{\mathrm{cb}}\Rightarrow \left|\mathrm{R}\mathrm{H}\left(\varDelta {f}_y^{\mathrm{cb}}\right)\right|=y\cdot \left|\mathrm{R}\mathrm{H}\left(\varDelta f=0\right)\right|\hfill \\ {}{O}_{DS}=\frac{T_{\mathrm{SYM}}}{\sigma_{\mathrm{DS}}},{O}_{\mathrm{FD}}=\frac{B_w}{\varDelta {f}_y^{\mathrm{cb}}\ }\hfill \end{array} $$

(8)

An illustration is given in Table 2 in a mixed indoor-outdoor use case. Normalized propagation parameters are evaluated upon the multi-rate filter propagation channel model [26] which resamples and filters the measured propagation channel, adapted to the PHY waveform of the system and simulated use cases. This model is applied at 60 GHz related to the ECMA-368 standard [20] transposed at 60 GHz and at 5 GHz in relation with the IEEE802.11ac/11n system [24]. Representative measurements are used at 60 GHz with a LOS, OLOS, and NLOS discrimination as inputs of the multi-rate filter propagation model. The reference TGac multipath channel model [27] channels D and E have been chosen to differentiate LOS and NLOS multipath signature on the propagation channel models representative of mixed indoor and phantom cell scenarios.

Normalized propagation parameters in the Table 2 highlight different values for LOS and NLOS situations. The O _DS decreases and the O _FD increases with the obstruction level of the radio link. The O _FD-0.5 is almost doubled when passing from LOS to NLOS. The introduced normalized propagation parameters, O _DS and O _FD − y with y taking two values {0.5, 0.9}, allow a LOS/NLOS discrimination upon independent air interfaces exhibiting different time-related PHY parameters.

As illustrated on Fig. 8, the power distribution within the time window of the frame may also provide a rice factor estimation [25] deduced from the RSSI distribution, leading to LOS/NOS discernment as a complement of the normalized propagation method. Pre-established LUTs of each scenario and available AI (terminal and BS) with LOS/NLOS separation are then used to extract S _M values of every considered TM/AI modes.

PLM is deduced from RSSI measurements, equivalent to the available received power ARP_MFS(d,fc) expressed in (5) and calculations of an ideal RSSI, i.e., ARP_FS(d,fc), deduced from the free space path-loss model. EIRP, receiver antenna gain G _R, and the distance d of the link which is furnished in AI headers.

Fig. 8

LOS/NLOS discernment using power distribution [25]

4 GLB metric performance in multi-RAT scenarios

4.1 Test cases

The GLB metric is evaluated in the context of small-cell hot spot deployment in ultra dense urban and large indoor areas with a distance range up to 40 m. The transmitter and the receiver are able to communicate with IEEE802.11ac/n [24] and multi-band UWB orthogonal frequency division multiplex (OFDM) interfaces [2, 20] operating, respectively, at 5 and 60 GHz. The first envisioned test case shows benefits of the GLB metric use instead of the SE 3GPP metric using directly RSSI levels to select TMs leading to the largest throughput TM (Section 2). The second test case compares different multiple input multiple output (MIMO) TMs using variable spatial stream number and bandwidth related to a single air interface, the IEEE802.11ac standard [24] and a throughput close to 80 Mbps. The last test case illustrates access point (AP) selection performed by GLB metric to establish a communication with the UE. Furthermore, the radio link propagation reliability varies with considered AP positions.

GLB metric versus SE CQI decision test case is a test case which compares the decision done by the SE maximization criterion as adopted in 3GPP and IEEE802.11n (Section 2) and the decision performed by the GLB metric applied on IEEE802.11 ad standard TMs dedicated to 60-GHz transmissions. For that purpose, we compare {α,β} metric variations along with different throughputs ranged from 1386 to 4158 Mbps. GLB values are deduced from link level performance carried out in [28]. SE-based decision would choose the highest throughput. The GLB metric performs the TMs selection in targeting throughput requirements and minimum α-metric values. Results are detailed and commented in Section 4.2.

The multiple TM MIMO test case considers a UE equipment connected to a single AP able to deliver and receive different IEEE802.11ac/n MIMO TMs under NLOS conditions delivering a data rate targeted to 80 Mbps. Spatial division multiplex (SDM) MIMO modes are compared together with a variable number of spatial streams, denoted NSS, MCS, and bandwidth size (20 or 40 MHz). Spatial time block coding (STBC) using Alamouti code is integrated in the evaluation. MIMO spatial division multiplexing SDM (NSS, N _TX, N _RX) represents the usual SDM technique considering N _STS time-spatial streams, N _TX, and N _RX antennas at the transmitter and receiver sides, respectively. STBC (NSS, N _TX, N _RX) refers to the Alamouti code implementation with a number of spatial stream (NSS = 1) lower than the number of time-spatial stream (N _STS = 2) to perform spatial redundancy processing.

The multi-RAT UE AP selection test case considers a UE connected to two separate AP. One AP, in a OLOS/NLOS propagation link with the UE, delivers two IEEE802.11ac TMs related to SISO and MIMO STBC TM which corresponds to the most outstanding TM selected in the multiple TM MIMO test case. The second AP, in LOS with the UE, generates 60 GHz MB-UWB signals associated with the ECMA-368 standard [20]. GLB metric is applied to perform AP selection combined with TM selection with a throughput ranged to 80 Mbps.

4.2 Performance

IEEE802.11ac [24] and ECMA-368 [20] TM performance are described in Table 3 for large indoor areas and outdoor pico-cell deployments. TM and link level parameters of the IEEE802.11 ad standard are detailed in Table 4. In these two tables, the required SNR values to achieve a BER close to 10⁻⁵ are deduced from link level performance in connection with the throughput of the TM. SNR values are deduced from BER plot SNRs. The noise figure, NF, used to estimate link budget parameters as detailed in (2) is set to 8 dB at 60 GHz and 10 dB at 5 GHz. Multipath and AWGN power sensitivity values as well as MCM are computed from (2), (3), and (4). MCM values are reported on Tables 3 and 4.

Table 3 The IEEE 802.11.ac and 60-GHz ECMA-368 link level parameters

Table 4 IEEE 802.11.ad OFDM link level parameters

The EIRP associated to the IEEE802.11ac modes operating at 5.25 GHz has been adjusted to 27 dBm and the receiver antenna gain has been set to 4 dB. At 60 GHz, the EIRP is identical, and receiver antenna gains for sectoral and directive antenna are fixed to 13 and 24.8 dB, respectively.

At 5 GHz, new path-loss models differentiating LOS and NLOS links are proposed that are derived from [27] and complementary measurements leading to new path-loss parameters described in Section 3.1 and listed in Table 1. At 60 GHz, measurements have been carried into large indoor areas leading to path-loss parameters differentiating receiver antenna diagram pattern as detailed in the Table 1.

GLB metric performance is discussed upon three test cases described in Section 4.1.

GLB metric versus SE CQI decision test case: IEEE802.11 ad link level performance has been evaluated [28] with a STBC Alamouti code setup in a multiple input single output (MISO) configuration with a single spatial stream (NSS = 1), two N _TX antennas, and a single receiver antenna N _RX. Simulated MCS and associated link level parameters are deduced from simulations (Table 4). GLB metric performance is evaluated upon several MCS delivering different throughput ranges. The SE-based link adaptation processing selects the highest throughput conditioned by the RSSI. GLB metric selection is focused on a throughput targeted by the service, assuming that several TMs may deliver the same throughput. Figure 9 shows that the MCM parameter (2) of the α-metric is sensitive to the FEC rate. MCM estimation may be assimilated α-metric variations with α arbitrary PLM value set in LOS and NLOS for a given distance d. It does not impact conclusions on α-metric variations along with variable throughput ranges. MCM value is lower for the MCS 16QAM 1/2 than for QPSK 3/4 despite a lower throughput delivered by the QPSK 3/4 MCS. This result is reported on both LOS and NLOS and is in favor of SE based AMC processing detailed in Section 2. If we consider the same modulation order, SE-based LA technique would choose the highest code rate and the GLB metric would choose the lowest code rate. Gains resulting from the GLB selection may lead to 1 dB on QPSK modulated signals. The 16 QAM 3/4 and 64 QAM 1/2 MCSs deliver the same throughput. It is interesting to see the MCS selection differ rather than we consider LOS or NLOS. In LOS, the 16-QAM 3/4 MCS is preferred to 64 QAM 1/2 and in NLOS a reverse selection is done. In LOS, lower throughput MCS is selected by the GLB metric decision. These results show the LOS/NLOS discrimination, in the case of LOS/NLOS transitions, is important in the MCS selection based on the assessment method proposed in Section 3.2.

Fig. 9

MCM variations with variable throughput at 60 GHz

The multiple TM MIMO test case compares MIMO spatial stream increase versus transmission bandwidth increase for the IEEE802.11.ac standard combined with two MCS, MCS_11n,10, and MCS_11n,26 (Table 3). Two TMs having the same MCS are compared, calculating four spatial streams combined with a 20-MHz bandwidth (SDM (4,4,4), QPSK 3/4) and two spatial streams combined with a 40-MHz (SDM (2,2,4), QPSK 3/4). In addition, a spatial redundancy technique performed by the STBC Alamouti code is compared to MIMO SDM and SISO transmission as detailed in Section 4.1.

Results of Figs. 10 and 11 show that it is more suitable to increase the bandwidth size rather than doubling the number of spatial stream to guarantee QoS and radio coverage and reduce the transmit power level through the β-metric variations. The α-metric gain is then set to 12 dB at a distance of 15 m. The STBC TM using the Alamouti code provides an additional gain set to 2–3 dB with respect to SDM (2,2,4) QPSK 3/4 TM. β-metric variations depicted on Fig. 11 translate quite similar gains as the α-metric. Gains may be translated in a transmitter power (Tx-power) reduction or a radio coverage extension as illustrated on Fig. 11 and (6). The power reduction given by ΔEIRP is deduced from the β-metric expression as follow:

$$ \begin{array}{l}\beta =\mathrm{EIRP}+{G}_{\mathrm{R}}-\alpha -{S}_{\mathrm{AWGN}}-{\mathrm{PL}}_{\mathrm{FS}}\left(d,\mathrm{fc}\right)\\ {}\varDelta \mathrm{EIRP}=\varDelta \beta +\varDelta {G}_{\mathrm{R}}+\varDelta \alpha +\varDelta {S}_{\mathrm{AWGN}}+\varDelta {\mathrm{PL}}_{\mathrm{FS}}\left(d,\mathrm{fc}\right).\end{array} $$

(9)

Fig. 10

α-metric variations on IEEE802.11ac TM in NLOS

Fig. 11

β-metric variations on IEEE802.11ac TM in NLOS

At a distance d set to 15 m, the power reduction read on the vertical axis may lead 5 dB between MCS_11n,10 and MCS_11n,26 modes (Table 3). Keeping constant the Tx-power, the radio extension, read on the x-axis for a β value close to 0, is then doubled. Regarding the β-metric variations, STBC provides lower gain than the SDM (2,2,4) TM following a higher S _M level of the STBC TM as reported on Table 3.

The multi-RAT UE AP selection test case considers one AP, named AP1, delivering an UWB signal associated with the 60-GHz ECMA-368 TM (Table 3) in LOS with two receiver antenna diagrams reported on PLM variations using parameters in Table 1 and results in Fig. 4. The second AP, named AP2, delivers two IEEE802.11ac/n signals over a NLOS propagation link: SISO with a 16-QAM 3/4 MCS and STBC MIMO with a QPSK 3/4 TM are considered with an equivalent efficient bandwidth set to 40 MHz (Table 3).

Results reported on Figs. 12 and 13 show that the 60-GHz transmission is much better than the two IEEE802.11ac TMs when the distance is higher than 2 m, showing that ECMA draws benefits from LOS conditions in the AI/TM selection. α-metric gain leads 14 dB for an identical distance d set to 15 m for the two APs. MIMO STBC TM versus SISO TM highlights gains up to 7 dB at 20 m when considering the MIMO STBC (1,2,4) mode with an identical bandwidth size for these two TMs. When comparing with 60 GHz UWB transmissions, additional mm-wave path-loss is compensated by better link level performance resulting from a large transmission bandwidth, taking advantage of multipath combination and LOS propagation link.

Fig. 12

α-metric variations of the multi-RAT UE AP selection test case

Fig. 13

β-metric variations of the multi-RAT UE AP selection test case

β-metric variations, illustrated on Fig. 13, exhibit power gain levels ranged from 10 to 20 dB when selecting a LOS 60-GHz radio link instead of a OLOS 5-GHz transmissions with distance ranging up to 40 m. This gain may be translated in a radio coverage extension when an equivalent Tx-power is set to the two APs. A higher 60-GHz receiver antenna gain, in accordance with Table 1 and (6), is reported on α-metric variations following PLM variations depicted on Fig. 4 and directly translated on β-metric variations.

5 Conclusions

This paper presents the performance gain of a novel power efficient multi-RAT link adaptation metric, denoted Green Link Budget metric, dedicated to multi-RAT network densification for future 5G networks. In the paper, mm-wave and Wi-Fi technologies for phantom cells with a maximum radio coverage of 40 m are considered. This metric, based on normalized link budget parameters, allows a fair comparison between independent interfaces and highlights benefits of mm-wave phantom cells with gains ranged from 10 to 25 dB on Tx-power level requirements to guaranty a radio coverage up to 40 m. The next step of the work will be to integrate multi-user distribution in the GLB metric assessments in order to evaluate and optimize multi-RAT end-to-end architectures. The second aspect is to pursue in a multi-user context, the integration of such metrics in emerging C/U plane splitting architectures [10] dedicated WLAN-LTE-A interworking including mm-wave components.