1
Random Access Protocols for Massive MIMO
arXiv:1606.02080v1 [cs.IT] 7 Jun 2016
Elisabeth de Carvalho, Member, IEEE, Emil Bj¨ornson, Member, IEEE, Jesper H. Sørensen, Member, IEEE, Petar Popovski, Fellow, IEEE, Erik G. Larsson, Fellow, IEEE
Abstract—5G wireless networks are expected to support new services with stringent requirements on data rates, latency and reliability. One novel feature coming with 5G is the ability to serve a dense crowd of devices, calling for radically new ways of accessing the network. This is the case in machine-type communications, but also urban environments or hotspots. In those use cases, pilot sequences are in shortage as the number of devices is much larger than the sample duration of the channel coherence interval. This article motivates the need for random access to pilot sequences by the devices and shows that massive MIMO is a main enabler to achieve fast access with high data rates, and delay-tolerant access with different data rate levels. Three pilot access protocols along with data transmission protocols are described, fulfilling different requirements of 5G services.
I. I NTRODUCTION There is a growing consensus (3GPP, METIS, ITU-R) that 5G wireless networks must support three generic categories of services: • Enhanced Mobile BroadBand (eMBB), with very high data rates as the central feature; • Massive Machine Type Communication (mMTC), with massive numbers of rather simple machine-type devices; • Ultra-Reliable Low Latency Communications (URLLC), with very low latency and extremely high robustness. Within each service category there can be specific services with additional requirements; for example, eMBB services that require low latency are referred to as Tactile Internet applications. While the sheer number of devices is an explicit part of the mMTC service definition, it also plays a significant role in eMBB services. There are several key scenarios where a large crowd of users is served in a limited spatial region, such as a shopping mall, stadium, or open air festival [1]. In addition, macro-cells that cover large spatial regions will remain important to provide continuous availability of mobile eMBB services in dense urban scenarios, where dynamic crowds appear along streets and highways [1]. These are some of the most challenging scenarios that 5G networks need to manage, in order yield ubiquitous connectivity. In this article, we introduce the term cMBB (crowd MBB) to denote the distinct class within eMBB intended for use in E. de Carvalho, J. H. Sørensen, and P. Popovski are with the Department of Electronic Systems, Aalborg University, Aalborg, Denmark (email:
[email protected],
[email protected]). E. Bj¨ornson and E. G. Larsson are with the Department of Electrical Engineering (ISY), Link¨oping University, Link¨oping, Sweden (email:
[email protected],
[email protected]). This work was performed partly in the framework of the Danish Council for Independent Research (DFF133500273), the Horizon 2020 project FANTASTIC-5G (ICT-671660), the EU FP7 project MAMMOET (ICT619086), ELLIIT, and CENIIT. The authors would like to acknowledge the contributions of the colleagues in FANTASTIC-5G and MAMMOET.
crowd scenarios, and describe efficient methods for devices to access the network in various crowd scenarios. As described below, a key technique is decentralized assignment of pilot sequences, based on random access. Next, we describe the motivation for random access to pilots and the distinctive role played by massive MIMO (multiple-input multiple-output). A. Why Random Access to Pilots? Channel State Information (CSI) is necessary at the receiver to enable coherent communication. CSI is particularly important at the access points in crowd scenarios, since conventional scheduling and power control algorithms are insufficient to manage many simultaneous device connections. The use of smart antenna arrays at the access point is thus desired to also manage interference in the spatial domain—by means of spatial multi-user beamforming where each beam is tailored to the CSI of the corresponding device. The CSI acquisition is enabled by the transmission of pilot sequences. This classical mechanism becomes particularly challenging in the use cases with large numbers of users: cMBB and mMTC. However, the limitations for CSI acquisition are different for these two services. Assuming a simple protocol with orthogonal pilot sequences, in cMBB the number/duration of the pilot sequences is limited by the coherence time of the channel due to mobility. On the other hand, in mMTC, the devices are generally assumed to be quasi-static, but the number of pilot sequences is rather limited by the power budget of the devices and the fraction of power that can be spent on pilot transmission. When the mMTC devices are mobile or in a dynamic environment, mobility might also become a limiting factor. Regardless of the reason, those restrictions put a fundamental limit on the number of pilots that can be shared by the devices and, as a consequence, the number of devices is much larger than the number of available orthogonal pilot sequences.1 This is the key motivation for devising new efficient and scalable pilot assignment protocols for cMBB and mMTC. The characteristics of the wireless traffic are important when determining the access method for the pilots. In mMTC, each device is sporadically active. In the initial access phase, an active device connects to the access point, identifies itself and establishes a coarse synchronization [2]. In principle, the access point can pre-allocate the pilot sequences to the devices as they connect. However, in view of the intermittent user 1 Any number of non-orthogonal pilot sequences can be generated to give each device a unique pilot sequence. Instead of being interfered by devices using the same pilot, each device will be partially interfered by a much large set of devices. This interference will be substantial, potentially larger than in the case of orthogonal pilots, thus the use of non-orthogonal pilots does solve the problem. However, it transfers some of the issues from the MAC-layer to PHY-layer interference mitigation, which is not the subject of this work.
2
activity, this becomes impractical. The same observation is valid for cMBB in non-streaming internet applications, where short periods of activity alternate with long periods of silence. With high device density and intermittent traffic, it becomes infeasible to have a dedicated pilot allocation per device. A natural solution is a protocol for Random Access to Pilots (RAP), where a device that wishes to transmit or receive data randomly selects a pilot sequence from a predefined set. RAP may lead to pilot collision, which is essentially pilot contamination, and the access protocol should have efficient mechanisms for dealing with collisions. B. 5G Services and Pilot Shortage # of (orthogonal) Pilot sequences Mobile
(Quasi)-Static
# of pilots limited by channel coherence time
# of pilots limited by transmit power
C. Massive MIMO as an enabler of 5G services and Random Pilot Access
eMBB cMBB
URLLC
mMTC
# of devices
Reserved Pilots
CROWD PILOT SHORTAGE ZONE Centralized pilot assignment
cMBB: The central feature remains the data rate, but the number of devices is much larger than the number of pilots. The traffic is intermittent so that coordinated pilot allocation becomes impractical. An option for pilot reuse is to allocate a small set of orthogonal pilots to each cell and use large reuse distances. The assignment in each cell is done by RAP. • mMTC: In this category the number of devices is in the order of 100000 or more per access point. Each device is only sporadically active, such that it is necessary to use RAP to connect. The focus of this article is on the random access mechanisms for scenarios in the pilot shortage zone; for example, cMBB and mMTC. From the pilot access point of view, there are two major differences between cMBB and mMTC: (1) the number of devices which is orders of magnitude larger in mMTC and (2) the characteristics of the downlink traffic. In eMBB, the downlink traffic volume generally dominates the uplink traffic volume while it is the opposite for mMTC. •
Random Access to pilot sequences
Fig. 1: 5G services as a function of the number of devices (horizontal axis) and pilot sequences (vertical axis). When the number of devices is no larger than the number of orthogonal pilot sequences available, centralized pilot assignment is possible. In a crowd scenario, the number of devices is much larger than the number of orthogonal pilot sequences. In this pilot shortage zone and with intermittent traffic, random access to pilot sequences is a viable solution. In Figure 1, a schematic representation of the 5G services is provided. The services are positioned as a function of the number of devices (horizontal axis) and the number of pilot sequences available (vertical axis). They are described below. • URLLC: Due to the requirements on high reliability, a small number of pilots should be exclusively allocated for URLLC transmissions. In order to mitigate the detrimental effects of small-scale fading, the channel hardening that beamforming from large antenna arrays can provide is desirable. To harness this effect the channel estimation should be sufficiently precise and frequent, such that when critical low-latency data arrives it can be sent with very high reliability. • eMBB: The central feature is the high data rate. The number of devices is moderate and can be larger than the number of pilot sequences. Coordinated pilot assignment remains possible without a large overhead. An efficient inter-cell pilot reuse plan can be implemented to suppress pilot contamination.
Massive MIMO is a key ingredient of 5G wireless networks, due to its capability to provide very high data rates in lower frequency bands [3], by greatly improving the cell spectral efficiency (measured in bit/s/Hz/cell). Massive MIMO is often described as a system with many antennas at the access points, which serve a number of devices that is much smaller than the number of antennas. Such operational conditions bring extra spatial degrees of freedom that lead to two physical phenomena [4]. First, channel hardening appears which result in beams that are asymptotically stable and immune to small-scale fading; that is, they create an almost deterministic (scalar) communication channel to each device. Channel hardening ensures a stable rate and improves communication latency. To achieve channel hardening it is critical to have accurate CSI for each served device. Second, the high array resolution improves the ability to separate users spatially, which results in an asymptotic decorrelation of the beams to the devices. Effectively, each device gets an exclusive, focused data beam and does not suffer from neither small-scale fading nor interference. The interference between users that utilize the same pilot is, however, hard to suppress, which makes pilot contamination a key characteristic in massive MIMO. For all the good reasons listed above, massive MIMO plays a crucial part in providing the very high data rates targeted in eMBB. The requirements in cMBB calls for an alternative massive MIMO design where its high multiplexing capability is exploited to serve simultaneously a crowd of users. The enhanced spatial resolution of massive MIMO enables serving devices that are closely spaced. Moreover, as pointed out in [5], the highest spectral efficiency is attained by aggressive multiplexing where the asymptotic decorrelation of the user beams is not attained, meaning that each user receives a relatively low spectral efficiency (but the cell spectral efficiency is extraordinary high). Another important advantage of the high multiplexing capability is that it naturally decreases
3
the communication latency by allowing simultaneous service to many devices. Table I summarizes the gains brought by massive MIMO in the 5G services. Massive MIMO enabling features Large array Gain
5G Service Features
High user spectral efficiency High cell spectral efficiency Latency
✔ ✔ ✔
High device density Low power devices Ultra-reliability Random access protocols
✔ ✔ ✔
Large spatial multiplexing capability
✔ ✔ ✔
✔
Channel hardening
✔
✔ ✔
TABLE I: Massive MIMO enabling features for 5G services and random pilot access.
device activity can be expressed using binomial distributions with parameters K, τp or pa . A. SUCR protocol: RAP for data transmission without pilot collisions Since pilot collisions lead to pilot-contaminated interference, which is hard to suppress by signal processing, it is desirable to protect the data transmissions from uncoordinated pilot collisions. This can be achieved by a protocol where RAP is only used for initial access, while coordinated pilot assignment with a pilot reuse factor of, say, 3 or 4 is utilized in the subsequent payload data transmissions [5]. Such a protocol is described in [6], with the specific goal of reengineering the access mechanisms to achieve data transmission conditions that resembles those assumed in the seminal massive MIMO paper [4]; that is, to unleash the huge spectral efficiencies that the technology promises. The RAP protocol proposed in [6] assigns a certain pattern of time-frequency blocks for random access for connecting devices. The operation of each such block consists of four steps, as illustrated in Figure 2a.
Step 1: Each connecting device transmits one of the τp pilots at random using full power. The uplink pilot transmissions These changes in the physical layer form a remarkable basis are not time-synchronized in this initial access, but the for the massive MIMO technology to become perhaps the sinvariations are typically within the cyclic prefix so a pilot gle most important enabler of 5G services. However, those ascollision occurs when two devices use the same pilot. sets have to be judiciously incorporated in the communication Step 2: The access point is unable to separate colliding devices, protocols and require a mindset shift when designing MACbut it can utilize the pilot to estimate the sum of the layer protocols. Several essential assumptions that are applied channels that the pilot has propagated over. This estimate for resource allocation in 3G or 4G become questionable. For is utilized in Step 2 to beamform a downlink pilot example, user scheduling becomes unnecessary due to the fact response to the corresponding devices. The estimationthat every user in massive MIMO can utilize the full bandwidth based beamforming provides a total array gain of M , and use it through a nearly deterministic communication which is shared between the users proportionally to their channel, while being spatially separated from the other users. respective path gains. To unleash the full potential of massive MIMO, a MAC-layer Step 3: The colliding devices utilize the downlink beamformed redesign for eMBB and cMBB services is advisable. When pilot to measure which share of the array gain that it comes to mMTC, a MAC-layer redesign is even required each one has achieved. This estimate is very accurate in since these services pose entirely new problems, which cannot massive MIMO, thanks to the channel hardening. Each be handled with conventional networks or small-scale MIMO device can now distributively detect the collision and technology. how strong its path gain is compared to the contenders. In the sequel we describe efficient access mechanisms to A strongest-user collision resolution (SUCR) decision a limited set of pilots in cMBB and mMTC. Pilot selection rule is proposed in [6] to resolve the collision by only is performed at the device using RAP and CSI is acquired at letting the device with the strongest path gain repeat the the access point based on the uplink pilots. A time-division pilot in Step 3. duplexing (TDD) system is targeted where channel reciprocity Step 4: The vast majority of the pilot collisions were resolved in is used to obtain downlink CSI estimates directly from the Step 3 and the devices that transmit collision-free pilots uplink CSI estimates. can now be identified and be assigned a dedicated pilot that can be used for payload data transmission. II. R ANDOM ACCESS TO P ILOTS AND DATA We refer to this approach as the SUCR protocol. The key T RANSMISSION benefit over the random access in legacy systems, such as LTE The access point is equipped with M antennas and the number of orthogonal pilot sequences is denoted as τp . The total number of devices is denoted by K, where K is much larger than τp in cMBB and mMTC services. In RAP, each user transmits with probability pa and selects one pilot sequence uniformly at random. The quantities related to the probabilistic
[2], [7], is the decentralized collision resolution that manages 90% of the collisions [6]. Centralized resolution mechanisms can be added on top of the SUCR protocol to take care of the remaining collisions. Note that the SUCR protocol exploits the natural variations in path gains that occur in practice, due to different distances and shadowing. This stands in contrast
4
1. Random access pilot 2. Beamformed pilot response User devices
3. Decentralized collision resolution and pilot repetition 4. Admit users to send data
Access point
Average Number of Access Attempts
(a) 10
clearly outperforms the baseline protocol. An alternative to the SUCR protocol, which uses power variations to resolve collisions, is to detect and utilize timing variations between the devices to resolve pilot collisions. One such approach is described in [8] and it also benefits from the high resolution provided by the many antennas in massive MIMO. In another related work [9], an aggressive pilot reuse plan (not relying on random access) is proposed leading to pilot contamination. A coded pilot scheme allows detection of pilot collisions at the base station where downlink data is solely transmitted to collision-free users.
8
B. E-RAPiD protocol: random access to pilot and data transmission with pilot collisions
System is fundamentally overloaded
6
4 SUCR protocol Baseline
2
0 0
2000
4000
6000
8000
10000
12000
Number of Inactive UEs
(b)
Fig. 2: Data transmission without pilot collisions is achieved by assigning dedicated pilots to the accessing devices. (a) The SUCR protocol for such pilot assignment. (b) The protocol can resolve collisions in the RAP procedure.
to LTE that attempts to mitigate these variations by power control. We illustrate the SUCR protocol’s ability to resolve collisions in RAP in a crowd scenario with a varying number of devices K ∈ [100, 12000], M = 100 antennas at the access point, and τp = 10 pilots. These devices are uniformly distributed in a hexagonal cell and each one decides to access the network with 0.1% probability (i.e., pa = 0.001). The average cell-edge signal-to-noise ratio (SNR) is 0 dB and shadow fading is taken into account, see [6] for details. If an accessing device is not admitted immediately, then in the upcoming blocks it will make a new attempt with probability 0.5. After 10 failed attempts, the access is considered to be denied by the network. Figure 2b shows the average number of attempts that a device makes, as a function of K. The SUCR protocol handles up to K = 8000 devices without noticeable delays. Notice that at K = 10000 there is on average K · 0.001/τp = 1 device per pilot, meaning that the network is fundamentally overloaded. Nevertheless, the average access delay is small because an astonishing 90% of the devices are still admitted to the system. This behavior remains also for K > 10000. Figure 2b also shows a baseline protocol where pilot collisions are not resolved at the devices, but only by making new attempts at random time instants. This protocol generally requires many more access attempts. The functionality begins to break down at around K = 3000 and for K = 10000 only 1.5% of the devices are ever admitted. The SUCR protocol
The SUCR protocol is an efficient mechanism to resolve pilot collisions and protect data transmission from the interference induced by pilot contamination. It is well suited for cMBB with low latency requirements. In mMTC, the requirements significantly differ. A large majority of the traffic originates from uplink transmission where a device transmits sporadically and at random time instants. Data transmission is rather insensitive to delays and the data rates are low. Those two properties can be exploited to devise an alternative joint pilot and data transmission protocol. 1) Joint pilot and data transmission protocol: Uncoordinated pilot selection gives intra-cell pilot contamination whenever a collision occurs. The basic idea of the second protocol is to randomize the effect of pilot contamination over multiple transmission slots. The protocol is depicted in Figure 3a in a simplified example. A transmission slot is limited by the channel coherence time and coherence bandwidth of the channel. The duration of a transmission slot is denoted as τu . A block fading model where the channel for each device takes an independent value in each transmission slot. The protocol relies on three main features: • Pilot hopping: in each transmission slot, each active device selects (pseudo-)randomly a pilot sequence from a codebook of orthogonal pilots. Note that CSI needs to be estimated in each transmission slot. • Ergodic data transmission: for each device, the codeword to be transmitted is divided into multiple parts that are transmitted according to the device activation pattern. • User discrimination: from a single transmission slot, it is impossible to discriminate between colliding devices. However, the whole series of pilot sequences selected by a device across all transmission slots provides a unique identifier that can be used for appropriate merging and decoding. For an asymptotically large number of transmission slots, one codeword is affected by an asymptotically large number of channel fades and interference events. Interference includes the interference caused by pilot contamination: for a given device, in each of its active transmission slots, contamination comes from a different random set of devices, so that asymptotically the device is affected by all possible sets. Maximum ratio combining (MRC) is used at the BS in which case interference also comes from devices that use different pilots. Relying on
5
Device 1
1
Device 2
2
Device 3
1
Device 4
3
⋯
3
⋯
3
⋯
3
⋯
in [11], we provide bounds that are loose but that follow the variations of R and lead to a viable optimization. Those modified bounds are the basis for the following heuristics: •
Transmission slot
•
(a) 120 M=400 M=100
Sum Rate [bit/s/Hz]
100
80
1.0 bit/s/Hz per user
60
x2
40
20 100
0.5 bit/s/Hz per user
150
200
250
300
350
400
450
500
One third of the slot is used for pilots: τp = τu /3. √ The average number of active users is pa K = x M τu , where the scalar x depends on the variations of the device path gains. The larger the variations, the larger x.
When power control is possible at the device side, the solution does not depend on the channel characteristics. Furthermore, √ this solution leads to a scaling of the sum rate as M τu . Figure 3b shows the bound R for a scenario with K = 800 devices as a function of the transmission slot duration. The channel path gains vary uniformly at random around a fixed ¯ The performance is value β¯ with a maximal gap of 0.25β. limited by interference and is quite insensitive to the SNR. For τu = 300, the average rate per user is 0.5 bit/s/Hz for M = 100 and 1 bit/s/Hz for M = 400. For τu = 300, the average number of active users is around 60 for M = 100 and around 140 for M = 400.
=u
(b)
Fig. 3: (a) E-RAPiD protocol: simplified example with four devices and two pilot sequences {s1 , s2 }. CWDi (j) denotes the portion of the codeword transmitted by device i during transmission slot j. (b) Bound R on the sum rate for K = 800, for M = 100 and M = 400 antennas as a function of the transmission slot duration.
the ergodic properties of this transmission process, it becomes possible to define for each device a reliable data rate. The transmission protocol is called Ergodic Random Access to Pilot and Data transmission (E-RAPiD). E-RAPiD is suitable for low-power devices thanks to the large array gain of the massive array. Furthermore, the diversity provided by channel hardening is a simpler alternative to conventional solutions relying on multi-hop transmissions and path diversity to combat fading. 2) Uplink sum rate and transmission optimization: The performance of E-RAPiD can be characterized using a lower bound on the uplink sum rate. Starting from the assumption that the number of active users and the number of contaminators are fixed, a first bound is derived considering uncorrelated Rayleigh fading channels and MMSE channel estimation at the BS [10]. The performance of E-RAPiD is determined by computing the expected value of this first bound with respect to the joint distribution of the number of active users and the number of contaminators. The resulting sum rate bound R is tight provided that the number of antennas at the BS is large. Details are contained in [11]. The performance bound R is a function of M and the transmission slot duration: the larger those quantities, the more devices can be multiplexed. R is also a function of the device activation probability, pa , and the number of pilot sequences, τp . To maximize the sum rate, one can optimize pa and τp . R is not amenable to a low-complexity optimization. Instead,
C. C-RAPiD Protocol: coded random access to pilot and data transmission with pilot collision resolution The third access protocol targets mMTC scenarios, like the E-RAPiD protocol. It also uses pilot hopping for joint pilot and data transmission. However, instead of dividing the codeword into multiple parts, which are transmitted in separate transmission slots, the codewords are replicated in each transmission slot within a predefined duration, termed a frame. The duration of a frame is in general lower than ergodic transmissions require, which means this protocol is targeting more delaysensitive applications in the mMTC category. The main goal of codeword replication is to use the successful decoding of a data transmission in one transmission slot to cancel the interference brought by replicas in other transmission slots. On the one hand, transmitting multiple replicas increases the intra-cell pilot contamination, on the other hand it provides multiple opportunities to successfully decode the codeword. This is a key trade-off in this protocol and it makes careful selection of pa , and thereby the amount of replicas within a frame, very important. We call the protocol Coded Random Access to Pilot and Data transmission (C-RAPiD). The interference cancellation in C-RAPiD is reminiscent of belief propagation decoding of codes on graphs, hence in recent literature the technique has been termed coded random access [12], [13]. The existing work in this area is based on the assumption that the channel coefficients remain constant within a frame, such that interference cancellation can be performed across transmission slots. Conventionally, this limits the duration of the frame to the minimum coherence time among the users, which in many practical systems is a strong limitation. Remarkably, this limitation vanishes when the technique is applied to massive MIMO systems. The channel hardening provided by massive MIMO makes the frame length limited only by large-scale fading. The signal processing performed at the access point consists of two steps:
1) Maximum ratio combining: The received uplink pilot and data signals are processed using MRC, which are constructed from the contaminated estimates achieved during the pilot transmission phase. MRC transforms the received signals from linear combinations with smallscale fading coefficients (the channel coefficients) to linear combinations with large-scale fading coefficients (the channel path gains). Note, that this transformation is only possible due to the channel hardening and beam decorrelation brought by massive MIMO. See [14] for the details on the MRC operation. 2) Successive interference cancellation: The linear combinations achieved through MRC represent a system of equations, which we wish to solve in order to obtain the data messages. It should be noted that the access point has no a priori knowledge of the random activity and pilot choices of the devices. Hence, the system of equations cannot be solved using, e.g., Gaussian elimination. Instead we employ successive interference cancellation (SIC). Initially, the access point decodes the messages from users that were lucky to avoid pilot collision. Embedded in the uplink data is the random activity and pilot choices of the device, which allows the access point to locate all the replicas of the same packet sent by that device. This enables the access point to cancel the interference caused by these replicas. Potentially, this enables the access point to decode additional data messages, whereby the iterative SIC procedure continues. The efficiency of the SIC procedure strongly depends on the distribution of the plurality of pilot collisions. In coding terminology, this is referred to as the degree distribution. For a given number of devices, K, we can adjust the degree distribution through the choice of τp , pa , and the frame size. We present a numerical comparison between the proposed C-RAPiD protocol and two references; scheduled massive MIMO (SMM) and ALOHA. In SMM, each device is allocated a unique pilot sequence in a particular transmission slot. Hence, transmissions are fully scheduled and SMM therefore serves as an upper bound to random access schemes. ALOHA is the conventional approach to random access, where users randomly select a pilot sequence and a time slot and only collision-free transmissions, i.e., signals with degree one, contribute to the throughput. We apply the channel model from Section II-B with deterministic power control, such that all devices experience an SNR of 10 dB. We assume that a channel code is applied at the physical layer with coding rate R and QPSK modulation. For all protocols, τp , pa and the frame size have been numerically optimized for maximum throughput, defined as the number of successfully recovered data messages per time slot, when accounting for fractional rate loss from channel coding and pilot transmissions. For the value of τp only powers of two have been considered. Smoother performance curves would have been achieved if all integer values had been considered. Fig. 4 shows throughputs for all protocols at both R = 0.5 and R = 1. As expected, the performance of all protocols increases with M . However, in the case of R = 0.5, the
Throughput [messages per time slot]
6
60
SMM R = 0:5 SMM R = 1:0 C-RAPiD R = 0:5 C-RAPiD R = 1:0 ALOHA R = 0:5 ALOHA R = 1:0
50 40 30 20 10 0 100
200
300
400
500
600
700
800
900
1000
M
Fig. 4: Comparison of throughputs at R = 0.5 and R = 1 and optimized values of τp , pa and the frame size.
ALOHA protocol experiences a saturation of the performance at roughly M = 200, whereas the C-RAPiD protocol continues to increase. The reason is that the ALOHA protocol can only benefit from the increased SINR until the point that degree one signals are decoded with high probability. This happens at a lower value of M when a channel code is applied, due to the increased resilience towards errors. C-RAPiD is able to further benefit from massive MIMO and the channel code, due to improved efficiency of the SIC. C-RAPiD achieves 45% of the throughput of scheduled operation with M = 400 and R = 0.5, while ALOHA achieves 33%. At M = 1024, the performance of C-RAPiD is increased to 61%.
III. C ONCLUSIONS AND P ERSPECTIVES Massive MIMO is currently one of the most compelling technology for 5G wireless networks. The operation in TDD and the resulting uplink-downlink reciprocity renders the system entirely scalable with respect to the number of access point antennas, leaving channel coherence (device mobility) as the only remaining, fundamental limiting factor. Fastmoving devices result in short coherence and room for fewer orthogonal pilots in each cell. In this article, we addressed crowd-eMBB and M2M/IoT scenarios in which there exist many more devices in a cell than there are unique orthogonal pilots, and where devices periodically or sporadically want to access the network, without prior coordination with the access point. Specifically, we saw how the abundance of spatial degrees of freedom, and the presence of channel hardening, in massive MIMO facilitates efficient resolution to resolve colliding transmissions, even in case the colliding packets use the same pilots. This brought the central conclusions of the article: •
•
first, massive MIMO is a fundamental enabler for crowdeMBB scenarios, sensor networks, IoT and M2M communications; second, the creation of an efficient standard for wireless networks based on massive MIMO technology will require a complete re-design of the multiple-access layer.
7
R EFERENCES [1] M. Fallgren, B. Timus et al., Deliverable D1.1: Scenarios, requirements and KPIs for 5G mobile and wireless system. ICT-317669-METIS, 2013. [Online]. Available: https://www.metis2020.com/ [2] E. Dahlman, S. Parkvall, and J. Skold, 4G: LTE/LTE-Advanced for Mobile Broadband, 1st ed. Academic Press, 2011. [3] E. G. Larsson, F. Tufvesson, O. Edfors, and T. L. Marzetta, “Massive MIMO for next generation wireless systems,” IEEE Commun. Mag., vol. 52, no. 2, pp. 186–195, 2014. [4] T. L. Marzetta, “Noncooperative cellular wireless with unlimited numbers of base station antennas,” IEEE Trans. Wireless Commun., vol. 9, no. 11, pp. 3590–3600, 2010. [5] E. Bj¨ornson, E. G. Larsson, and M. Debbah, “Massive MIMO for maximal spectral efficiency: How many users and pilots should be allocated?” IEEE Trans. Wireless Commun., vol. 15, no. 2, pp. 1293– 1308, 2016. [6] E. Bj¨ornson, E. de Carvalho, J. H. Sørensen, E. G. Larsson, and P. Popovski, “A random access protocol for pilot allocation in crowded massive MIMO systems,” IEEE Trans. Wireless Commun., 2016, submitted, Available: http://arxiv.org/pdf/1604.04248. [7] M. Hasan, E. Hossain, and D. Niyato, “Random access for machineto-machine communication in LTE-advanced networks: Issues and approaches,” IEEE Commun. Mag., vol. 51, no. 6, pp. 86–93, 2013. [8] L. Sanguinetti, A. A. D’Amico, M. Morelli, and M. Debbah, “Random access in uplink massive MIMO systems: How to exploit asynchronicity and excess antennas,” in IEEE Globecom, 2016, submitted. [9] O. Y. Bursalioglu, C. Wang, H. Papadopoulos, and G. Caire, “RRH based massive MIMO with “on the fly” pilot contamination control,” in Proc. IEEE ICC, 2016. [10] H. Ngo, E. Larsson, and T. Marzetta, “Energy and spectral efficiency of very large multiuser MIMO systems,” IEEE Trans. Commun., vol. 61, no. 4, pp. 1436–1449, 2013. [11] E. de Carvalho, E. Bj¨ornson, E. G. Larsson, and P. Popovski, “Random access for massive MIMO systems with intra-cell pilot contamination,” in Proc. IEEE ICASSP, 2016. [12] E. Paolini, C. Stefanovic, G. Liva, and P. Popovski, “Coded random access: How coding theory helps to build random access protocols,” IEEE Commun. Mag., vol. 53, no. 6, pp. 144–150, Jun. 2015. [13] G. Liva, “Graph-based analysis and optimization of contention resolution diversity slotted aloha,” IEEE Transactions on Communications, vol. 59, no. 2, pp. 477–487, February 2011. [14] J. H. Sørensen, E. de Carvalho, C. Stefanovi´c, and P. Popovski, “Coded pilot access: A random access solution for massive MIMO systems,” IEEE Trans. Wireless Commun., 2016, submitted, Available: http://arxiv.org/pdf/1605.05862.
