A Multi-Service Oriented Multiple-Access Scheme For M2M Support in Future LTE
We propose a novel multiple-access technique to overcome the shortcomings of the current proposals for the future releases of Long-Term Evolution (LTE). We provide a unified radio access system that efficiently and flexibly integrates both traditional cellular services and machine-to-machine (M2M) connections arising from Internet-of-Things (IoT) applications. The proposed solution, referred to as multi-service oriented multiple access (MOMA), is based on a) establishing separate classes of users using relevant criteria that go beyond the simple handheld-IoT device split, b) service-dependent hierarchical spreading of the data signals and c) a mix of multiuser and single-user detection schemes at the receiver. Signal spreading in MOMA allows to handle densely connected devices with different quality-of-service (QoS) profiles and at the same time its flexible receiver structure allows to allocate the receiver computational resources to the connections that need it most. This yields a scalable and efficient use of the available radio resources and a better service integration. While providing significant advantages for key future communications scenarios, MOMA can be incorporated into LTE with a limited impact on the protocol structure and the signaling overhead.
The provisioning of IoT services is now widely seen in the telecommunications sector as one of the major features in the evolution of cellular systems. Indeed, having a unified cellular system capable of handling both IoT machines and handheld mobile devices would be greatly advantageous for both operators and users. Towards this goal, the design of an integrated radio access solution is a challenging problem. Major issues are the large number of IoT devices required to be simultaneously served and the difference between their traffic patterns/QoS requirements and those of mobile broadband services . Another issue is that IoT applications do not all have the same QoS and traffic characteristics , with the implication that future M2M-related system optimization should have inherent flexible support for several types of IoT services.
To address some of these challenges, the Third Generation Partnership Project (3GPP) has started to add M2M-type communications into the radio access subsystem of LTE starting from Release 12  by introducing a new user equipment (UE) category, namely Category 0 (Cat. 0). Cat. 0 devices are characterized by their low cost due to their by-design lack of support for high peak rates and multiple antennas. For Release 13 of the LTE standard, 3GPP is working on providing further cost reductions for M2M communications . The proposals that emerged within this work are referred to as clean-slate narrow-band cellular IoT (NB CIoT)  and most of them advocate orthogonal physical-layer transmission schemes that are a mixture of frequency division multiple access (FDMA) and time division multiple access (TDMA). This is the case, for example, of LTE for Machine-Type Communications (LTE-M) and Narrow-band LTE-M (NB LTE-M). Each of these two proposals introduces a new UE category, the so-called Cat. 1.4 MHz for LTE-M and Cat. 200 kHz for NB LTE-M . As their respective names indicate, these new UE categories restrict the device transceiver bandwidth to 1.4 MHz and 200 kHz, respectively. Other proposals focused on upgrading the LTE random access procedure for better support of massive M2M transmissions  by overcoming the so-called Physical Random-Access Channel (PRACH) overloading issue.
Limitations of M2M Proposals for Next LTE Releases
None of the existing M2M-related proposals for LTE is able to meet the following crucial requirements all at once.
Multi-class users/services: While most of the existing proposals treat IoT devices as a single class of users, not enough attention has been paid to the different QoS and traffic profiles within this class. Indeed, IoT services will not be limited to data collection from simple sensors and will not only emit small data packets [3, 5]. One example is mobile video surveillance which is expected to use medium to high-end devices that do not have battery life constraints . Moreover, there should be a distinction within the services running on handheld devices between mobile broadband applications and the applications that have traffic characteristics and data rate requirements resembling those typical of IoT services. In the latter group we have for example the messages generated by social networking and chatting applications. Significant gains in resource utilization efficiency are expected from a multiple-access scheme that treats the services with similar QoS profiles, whether running on handhelds or IoT machines, as belonging to the same user class.
Dense IoT deployment: The existing M2M proposals for LTE allow to enhance its IoT capabilities but their reliance on FDMA limits the number of simultaneous M2M connections to the (typically moderate) number of frequency sub-channels they reserve for IoT communications. There is a need to support a much larger number of simultaneously connected IoT devices (and possibly mobile services with low QoS requirements). This goal should be met without sacrificing the QoS of mobile broadband services.
Flexibility in resource assignment: In existing M2M proposals, there is no way to dynamically adjust the respective proportions of resources assigned to broadband services and IoT devices. This could lead to wasting resources or to denials of service depending on the current traffic demands. Moreover, these proposals have limited flexibility in resource allocation within the M2M frequency band which only comes from varying the number of sub-channels occupied by each device from within a limited number of possible values (6 in LTE-M). Any new multiple-access scheme should be more flexible in assigning resources to different classes and to users within each class.
Efficiency in resource utilization: Assigning orthogonal frequency sub-channels to devices with low QoS requirements is not the most efficient way to use the available radio resources. First, this will bound the maximum number of simultaneously connected devices by the number of available sub-channels. Second, it is known that the boundary of the multiple-access channel (MAC) capacity region is not achieved with orthogonal transmission schemes. Third, achieving robustness against timing and carrier frequency offsets in time and/or frequency division orthogonal schemes requires the use of guard intervals and/or bands around each sub-channel, further reducing their resource utilization efficiency.
In the sequel, we show that MOMA can overcome these limitations while being compatible with low-cost devices having simple transceivers and long battery life requirements. Indeed, both the (narrow) bandwidth values that were originally proposed for LTE-M (1.4 MHz) and NB LTE-M (200 kHz) as a way to reduce transceiver complexity are supported in MOMA.
Multi-Service Oriented Multiple Access
MOMA is a multiple-access scheme conceived for scenarios where users are grouped into different classes. It reveals all its potentials when the BS is equipped with a large number of antennas. In this article we assume that classes are defined based on users’ QoS requirements profiles. For example, we define classes of users as follows.
One maximum data rate (HD) class of users: Here, HD stands for high data rate. For the HD class the objective is to obtain a data rate as high as possible for simultaneous transmissions. Typically these users are associated with data-hungry applications on handheld devices such as video conferencing and media streaming.
Constant low-to-moderate data rate (LMD) classes of users: Here LMD stands for low-to-moderate data rate. The -th class, with , includes users requesting services with a relatively low or moderate data rate . In the sequel we assume that LMD classes are ordered from to with increasing target data rates. Services with low target rates could originate from applications running either on handheld devices (such as social messaging) or on machines (such as M2M light-duty data collection from smart meters and remote sensors). The same applies to services with higher target data rates such as moderate-quality live streaming from handheld devices and M2M heavy-duty data collection from mobile video surveillance machines. Class aims at accommodating the maximum number of simultaneous transmissions at the granted data rate .
Since what matters for HD users is maximizing their respective throughput, proper scheduling techniques will typically limit the number of simultaneous HD transmissions, exactly as in current wireless-communications standards. It is thus reasonable to assume that is small and that multiuser multiple-input multiple-output (MU-MIMO) techniques implemented on top of orthogonal frequency division multiple access (OFDMA) in the downlink and single-carrier frequency division multiple access (SC-FDMA) in the uplink can be used for HD/HD signal separation. We also propose the use of these frequency domain transmission schemes for the separation between the HD and the other user classes. This choice allows to maintain full compatibility with the LTE standard. As for the LMD classes, due to both their specific data rate requirements and the objective of massive M2M deployment, we propose to overload radio resources. Note that this overloading can be achieved, for instance, by MU-MIMO techniques operating also in the code domain. The way we propose to access the code domain is dubbed service dependent hierarchical spreading and can be thought of as a layered or hierarchical spreading with a class dependent overloading factor. The MOMA uplink transmission scheme is illustrated in the left-hand part of Fig. 1.
MOMA is based on service dependent hierarchical spreading. This new transmission scheme has the advantages of efficiently using available resources, being scalable with the number of connected devices and allowing flexible resource allocation among the different user classes.
MOMA can be easily integrated into LTE systems as it can be implemented on a sub-band of the LTE resource grid without affecting the legacy connections occupying the rest of the bandwidth. Furthermore, MOMA signals can be transmitted using modulation and coding schemes (MCSs) and transport block (TB) sizes that are taken from the LTE standard. Finally, MOMA can make use of some advanced features of LTE such as transmission time interval (TTI) bundling.
A narrow-band implementation of MOMA is possible, thus making it compatible with low-cost battery constrained devices.
MOMA exploits massive MIMO both to increase user multiplexing capabilities and to simplify the receiver structure. Thanks to these properties, low-complexity detection is used for most of the users in MOMA while the more complex detection schemes are only applied for the highest data rate classes.
Using MOMA entails a gain in coverage, making it advantageous for connecting devices deployed in remote or bad-coverage areas.
Several random access mechanisms with different degrees of signaling overhead are compatible with MOMA.
With these features, which are discussed in more detail in the sequel, MOMA can overcome the shortcomings of the current proposals for the evolution of LTE in supporting IoT. For instance, features 1, 4 and 6 are essential for enabling real massive M2M deployments.
Service Dependent Hierarchical Spreading
For the sake of clarity, we only consider the case from now on. The first LMD class () will be simply referred to as the LD class, where LD stands for low data rates, with target data rate . Similarly, the second LMD class () will be referred to as the MD class, where MD stands for moderate data rates, with target data rate .
To get a more precise description of MOMA, let be a code matrix (e.g. a Walsh-Hadamard matrix or a discrete Fourier transform matrix). In MOMA, the set of columns of matrix is divided into two disjoint subsets, namely matrices and with dimensions and respectively. Now assume that a maximum number () of simultaneously connected MD (LD) users are to be served within the current sub-frame. Since we want to overload the MD and LD radio resources, we will typically have and . Finally, since we want to guarantee higher data rates for the MD class as compared to the LD class, we impose .
Instead of assigning the orthogonal spreading codes to individual users, the (respectively ) columns of (respectively ) are simultaneously used in MOMA by the (respectively the ) users. Indeed, each MOMA transmitter applies as spreading code a linear combination of the columns of the code matrix corresponding to its class. The coefficients of this linear combination serve as a signature sequence to separate the signals of the users belonging to the same class. More precisely, the data symbols of each MD (LD) user are spread using one column of the product matrix () where () is an overloading matrix of dimensions (respectively ) whose columns are referred to in the sequel as the overloading sequences. In principle, () can be constructed by selecting () points from the surface of the -dimensional (-dimensional) complex sphere with radius 1. The resulting spread symbols of each user are then mapped to the elements of the OFDMA/SC-FDMA time-frequency grid elements that fall within the frequency band assigned to the MD and LD classes before being transmitted on the radio channel. Finally, since the BS is equipped with a number of antennas, we know from the literature  that the effective spreading gain of MD transmissions (respectively LD transmissions) is, roughly speaking, proportional to (respectively ). This intuition was confirmed by the analysis done in .
The main advantage of this multiple-access scheme is an efficient, scalable and flexible use of the available radio resources. MOMA efficiency is shown by its overloading of the available radio resources in order to connect a large number of IoT machines and of handhelds requiring low-to-moderate data rates. The scalability of MOMA with respect to increasing device densities is simply a matter of applying a larger value for the MD (respectively LD) overloading factor defined as (respectively ) and/or of employing a larger . MOMA flexibility is manifested by the ease with which the network can dynamically adjust the proportion of resources assigned to each class of devices/services and the degree to which these resources are overloaded within each class by means of simply updating the values of parameters , , and . Finally, by properly mapping MOMA signals to the LTE time-frequency resource grid, MOMA combines the benefits of both OFDM, e.g. robustness against timing errors, and frequency-domain spreading, e.g. the ability to harvest the frequency diversity of the channel and the robustness against carrier frequency shifts.
MOMA Integration into Future LTE
Let be the total system bandwidth and denote by () the bandwidth assigned to the HD (MD and LD) class such that . In order to apply MOMA in the future evolution of LTE, we need to set and , the spreading factor and the map of the spread data symbols to the OFDMA/SC-FDMA time-frequency grid. We also need to determine which new signaling messages could be needed and what effect MOMA could have on the LTE system protocols. First, let us recall how the available time-frequency resources are structured into sub-frames in LTE. The smallest item in the time-frequency grid in LTE is the resource element (RE) defined as one subcarrier within one OFDM symbol of a duration equal to 66.7 s. However, the basic unit for scheduling and resource allocation is the physical resource block (PRB), which is composed of REs in 14 consecutive OFDM symbols covering 180 kHz over 1 ms. The duration of the basic period of data scheduling in LTE, called a sub-frame, is also equal to 1 ms. Finally, the duration of one sub-frame is also referred to as the data transmission time interval (TTI).
MOMA on a 1.4 MHz Bandwidth
One possible implementation of MOMA provides that coincides with the frequency band occupied by the 6 PRBs that are destined for M2M communications in LTE-M. This implementation is illustrated in Fig. 2.
In this implementation, at the beginning of each TTI each active MD and LD transmitter extracts from its transmission queue enough data bits that, after coding and mapping, results in a number of data symbols equal to the size of one PRB. The motivation behind this choice is to retain full compatibility with LTE MCSs and TB sizes. Next, each data symbol is spread using service dependent hierarchical spreading with . The resulting spread symbol is finally mapped to 6 consecutive REs from one PRB as shown in Fig. 2.
MOMA on a 200 kHz Bandwidth
Another possible MOMA implementation is obtained by letting coincide with the one PRB reserved for M2M communications in NB LTE-M. Except for the difference in bandwidth and in resource allocation granularity, this implementation is not different from the 1.4 MHZ implementation.
MOMA with TTI Bundling
TTI bundling is a transmission technique that was originally proposed for coverage enhancement in delay-limited applications such as Voice-over-LTE (VoLTE) . It consists in allowing users to transmit the four redundancy versions (RVs) of their current codewords in one shot using four consecutive TTIs, instead of waiting for the BS acknowledgment (ACK) or negative acknowledgment (NACK) message after each RV transmission. When applied along with MOMA, the already-existing control messages and protocol structure that were introduced to support TTI bundling can be re-purposed in order to increase the multiplexing capabilities for IoT devices while getting a coverage enhancement gain. In a 1.4 MHz implementation of MOMA with TTI bundling, the MD and LD spreading code length is , where is the original spreading factor without TTI bundling. Each sequence resulting from spreading a MD or a LD symbol is mapped to consecutive REs from one PRB as shown in Fig. 3.
MOMA Receiver Complexity and Performance
We propose a receiver structure for MOMA that takes advantage of the large number of BS antennas. This massive-MIMO scenario, which is expected to be prevalent in next-generation cellular networks, proves to be advantageous for MOMA from both performance and receiver complexity perspectives. The proposed receiver is illustrated in the right-hand part of Fig. 1 and consists in performing the following two steps.
Spatial combining: When the BS has multiple antennas, linear receive combining in the uplink is typically utilized. When the number of BS antennas is large enough, maximum-ratio combining (MRC), which is a low-complexity scheme, has been shown  to achieve a spectral efficiency not far from that achieved with more involved linear combining methods thanks to the asymptotic (with respect to the number of BS antennas) orthogonality of users’ channel vectors in massive MIMO. We thus propose to apply MRC for the detection of both LD and MD spread signals on the chip level.
Code despreading: Since the target data rate for the MD class is relatively high as compared to , the number of devices that can be simultaneously served in that class is expected to be smaller than its LD counterpart. The BS can thus typically afford for this class the use of multiuser detection techniques such as successive interference cancellation (SIC). On the other hand, we will consider only single-user detection for the LD class. This choice is motivated by the need to maintain a reasonable detection complexity while serving a large number of LD users with low target data rates.
Interestingly, this simple receiver structure which does not involve any inter-class multiuser detection was shown  to achieve asymptotic MD/LD orthogonality even on fast-varying frequency-selective channels and even when the number of LD users grows to infinity. This is due to a favorable property of massive-MIMO channels. Indeed, as an effect of combining a large number of signals in the case of a rich-scattering propagation environment, the small-scale fading averages out over the array in the sense that the variance of the resulting scalar channel decreases with . This effect is known as channel hardening and is a consequence of the law of large numbers . Most importantly in our case, the frequency response of the effective channel is asymptotically flat and asymptotically constant over several consecutive OFDM symbols as illustrated in Fig. 4 when . The channel realizations used in this figure were generated using the Extended Type Urban (ETU) channel model .
In Fig.s 5 and 6, we plotted the number of MD and LD connections that can be simultaneously served with and without TTI bundling, respectively, as function of their respective target data rates ( and ) for both MOMA and LTE-M. The values of are taken from the range kbps while kbps. The higher value in these two intervals is dictated by the maximum per-link data rate achievable with orthogonal (thus underloaded) access schemes. From the figures we can notice the significant advantage of using MOMA as opposed to orthogonal-access NB CIoT solutions in terms of the capability of serving densely-deployed IoT devices. For instance, four times more simultaneous MD and LD connections can be served while both and are approximately at half their respective upper-bound values. Note that this performance has been obtained on doubly dispersive channels generated using the Extended Vehicular A (EVA) model  which is characterized with a relatively long delay spread and short coherence time.
Also note that the four times larger spreading gain resulting from TTI bundling allows to serve four times more MD and LD simultaneous connections while meeting their respective target data rates, in the same range as in the absence of TTI bundling.
The figures were obtained using 100 realizations of users’ distances to the BS randomly chosen in the interval m assuming that the BS is equipped with antennas and that users’ transmit power is equal to 23 dBm.
Coverage Enhancement with MOMA
The coverage measure adopted for LTE channels is the maximum coupling loss (MCL), defined as the difference in logarithmic scale between the maximum transmission power and the receiver sensitivity . A higher MCL value indicates that the transmitter-receiver distance can be made larger while still meeting the target received signal-to-noise ratio (SNR) (and hence the target block error rate (BLER)). This translates into better cellular service coverage.
As a consequence of the -long spreading used in MOMA, IoT users benefit from a coverage enhancement gain. This gain can be further increased if TTI-bundling is applied due to the higher processing gain (equal to ) resulting in this case from spreading over a larger number of resource blocks. Indeed the increase in MCL due to the use of MOMA is in the order of without TTI bundling and of with TTI bundling, corresponding to a gain of 7.78 dB and 13.80 dB, respectively.
Signaling and Random Access for MOMA
To implement MOMA, matrix should be known in advance to all MD and LD users, for instance in the form of a look-up table (LUT). Moreover, the resource allocation parameters and are dynamic and need to be broadcast by the BS. This can be done by making use of the LTE broadcast control channel (BCH). Note that the values of and can typically be kept fixed for a relatively long time as they only need to be changed when the traffic characteristics change. As for the overloading sequences, their allocation is intimately related to the kind of random access (RA) scheme we opt for.
Contention-free RA: In a contention-free access scheme, the BS is in full control of the assignment of the available radio resources to the active users in the cell area . In MOMA, this translates into the BS choosing which MD (respectively LD) user is assigned which column of the overloading matrix (respectively ). Note that these two matrices could be generated in advance, i.e. offline, for different combinations of the values of , , and and made available in the form of a LUT to the relevant users. Contention-free RA has the advantage of eliminating collisions among concurrent connections, and hence eliminating the need for collision detection and resolution. However, this comes at the price of a relatively large protocol overhead, especially in the case of a highly overloaded system. Contention-free RA is thus more suitable for systems with low-to-moderate user densities.
Contention-based RA: In a contention-based access scheme, we let MD and LD users compete within their respective classes for the overloading sequences. On one side this helps cut the protocol overhead thus making contention-based RA relevant for systems with high user densities. On the other side, it comes at the price of eventual collisions between concurrent uplink transmissions and ensuing retransmissions. A conflict/collision occurs when at least two MD or LD users choose the same overloading sequence, making it difficult for the BS to correctly detect their transmitted data symbols. In contention-based MOMA, there is no need to store LUTs corresponding to the overloading matrices as the overloading sequences could be locally generated when needed. In this case, collision resolution is left entirely to the higher network layers. Otherwise, the probability of collisions could be reduced by using preamble transmission as in LTE PRACH and/or by using contention transmission unit (CTU) messages  that have the overloading sequence as one of their fields.
Hybrid RA scheme: Interestingly, LD collisions that take place in contention-based MOMA cannot affect MD connections and both MD and LD collisions cannot affect HD (legacy) connections thanks to inter-class quasi-orthogonality in MOMA. This observation can be used to motivate the use of a contention-free RA scheme for HD and MD connections and a contention-based RA scheme for LD connections. Such a hybrid scheme has the advantage of reducing protocol overhead in systems characterized with a relatively high LD and a lower MD device density.
Note that in all three cases uplink synchronization is needed before initiating the RA procedure and the actual data transmission. As in LTE, this can be maintained with a timing advance procedure .
Conclusions and Perspectives
MOMA is a novel multiple access scheme compatible with massive MIMO, which can be integrated into the evolution of LTE to enhance its support for a wide range of services including M2M communications. MOMA is based on assigning, in a flexible and dynamic manner, different code resources and different degrees of resource overloading to different classes of users, each representing a different data rate requirement, a different service type and/or a different traffic pattern. Code assignment in MOMA is conceived in such a way that overloading the resources of the lower data rate classes would only slightly affect the higher data rate classes, dropping the need for wasteful guard bands and steep transmit filters for uplink transmission. Moreover, the different QoS requirements in the system can be satisfied in a flexible and efficient fashion by reserving higher-complexity detection schemes at the BS only for classes which need them. Finally, we showed that MOMA outperforms the other M2M-related proposals for LTE in most of the M2M-relevant performance measures while requiring comparable signaling and protocol overhead.
One research direction for MOMA consists in analyzing its performance using different MIMO channel models reflecting the diverse propagation environments and practical BS array configurations. Another research direction is conducting a higher-level assessment of MOMA using advanced models, as those introduced in , for the data traffic generated by the different classes of users. Finally, the integration into MOMA of other coverage enhancement techniques such as relaying is yet to be investigated.
Nassar Ksairi received the M.Sc. degree from CentraleSupelec, France, in 2006 and the Ph.D. degree from the University of Paris-Sud XI, France, in 2010. From 2010 to 2012, he was an Assistant Professor at the Higher Institute for Applied Sciences and Technology, Damascus, Syria. From 2012 to 2014, he was a Postdoctoral researcher at Télécom ParisTech, Paris, France. Since December 2014, he is a researcher at Huawei’s Mathematical and Algorithmic Sciences Lab, France.
Stefano Tomasin received his PhD in 2003 from the University of Padova, Italy, which he then joined since 2002, being now Associate Professor. He has spent leaves and sabbaticals at Qualcomm in California, the Polytechnic University in New York and the Mathematical and Algorithmic Sciences Laboratory of Huawei Technologies, in France. Since 2011 he is Editor of both IEEE Transactions of Vehicular Technologies and EURASIP Journal of Wireless Communications and Networking.
Mérouane Debbah entered the Ecole Normale Supérieure de Cachan (France) in 1996 where he received his M.Sc and Ph.D. degrees respectively. He worked for Motorola Labs (Saclay, France) from 1999-2002 and the Vienna Research Center for Telecommunications (Vienna, Austria) until 2003. From 2003 to 2007, he joined the Mobile Communications department of the Institut Eurecom (Sophia Antipolis, France) as an Assistant Professor. Since 2007, he is a Full Professor at CentraleSupelec (Gif-sur-Yvette, France). From 2007 to 2014, he was the director of the Alcatel-Lucent Chair on Flexible Radio. Since 2014, he is Vice-President of the Huawei France R&D center and director of the Mathematical and Algorithmic Sciences Lab.
-  L. Vangelista, A. Zanella, and M. Zorzi, Long-Range IoT Technologies: The Dawn of LoRa, in FABULOUS, Ohrid, September 2015, pp. 51-58.
-  The 3rd Generation Partnership Project (3GPP) Technical Report 36.888, ”Study on Provision of Low-Cost Machine-Type Communications (MTC) User Equipments (UEs) Based on LTE, V.12.0.0,” June 2013.
-  The 3rd Generation Partnership Project (3GPP) Technical Report 22.368, ”Service Requirements for Machine-Type Communications (MTC); Stage 1, V.13.0.0,” December 2014.
-  W. Guibene, K. E. Nolan, and Mark Y. Kelly, Survey on Clean Slate Cellular-IoT Standard Proposals, in CIT/IUCC/DASC/PICOM, Liverpool, October 2015, pp. 1596-1599.
-  A. Meader and P. Rost, The Challenge of M2M Communications for the Cellular Radio Access Network, in EuroView, August, 2011, pp. 1-2.
-  R. Ratasuk, N. Mangalvedhe, A. Ghosh, and B. Vejlgaard, Narrowband LTE-M System for M2M Communication, in VTC-Fall, Vancouver, September 2014, pp. 1-5.
-  M. Hasan, E. Hossain, and D. Niyato, “Random Access for Machine-to-Machine Communications in LTE-Advanced Networks: Issues and Approaches,” IEEE Communications Magazine, vol. 51, no. 6, June 2013, pp. 86-93.
-  G. Naddafzadeh-Shirazi, L. Lampe, G. Vos, and S. Bennett, “Coverage Enhancement Techniques for Machine-to-Machine Communications over LTE,” IEEE Communications Magazine, vol. 53, no. 7, July 2015, pp. 192-200.
-  S. V. Hanly and D. Tse, “Resource Pooling and Effective Bandwidths in CDMA Networks with Multiuser Receivers and Spatial Diversity,” IEEE Trans. Inf. Theory, vol. 47, no. 4, May 2001, pp. 1328-1351.
-  N. Ksairi, S. Tomasin, and M. Debbah, A Multi-Service Oriented Multiple-Access Scheme For Next-Generation Mobile Networks, in EUCNC, Athens, June 2016.
-  E. Björnson, E. G. Larsson, and T. L. Marzetta, “Massive MIMO: Ten Myths and One Critical Question,” IEEE Commun. Mag., vol. 54, no. 2, February 2016, pp. 114-123.
-  The 3rd Generation Partnership Project (3GPP) Technical Report 36.104, ”Evolved Universal Terrestrial Radio Access (E-UTRA); Base Station (BS) radio transmission and reception, V13.3.0,” Mars 2016.
-  K. Au, L. Zhang, H. Nikopour, E. Yi, A. Bayesteh, U. Vilaipornsawai, J. Ma, and P. Zhu, Uplink Contention Based SCMA for 5G Radio Access, in Globecom, July 2014, pp. 900-905.
-  F. Khan, LTE for 4G Mobile Broadband: Air Interface Technologies and Performance. Cambridge University Press, 2009.
-  D. Niyato, P. Wang, and D. I. Kim, “Performance Modeling and Analysis of Heterogeneous Machine Type Communications,” IEEE Trans. Wireless Commun., vol. 13, no. 5, May 2014, pp. 2836-2849.