Conventional mobile networks (a.k.a. cellular wireless networks) are based on cellular topologies. With cellular topologies, a land area is divided into cells. Each cell is served by one base station. An interesting question is: shall the future mobile networks continue to have cells? My quick answer is no, cell-free networks should be the way to do in the future!
Future wireless networks have to manage at the same time billions of devices; each needs a high throughput to support many applications such as voice, real-time video, high quality movies, etc. Cellular networks could not handle such huge connections since user terminals at the cell boundary suffer from very high interference, and hence, perform badly. Furthermore, conventional cellular systems are designed mainly for human users. In future wireless networks, machine-type communications such as the Internet of Things, Internet of Everything, Smart X, etc. are expected to play an important role. The main challenge of machine-type communications is scalable and efficient connectivity for billions of devices. Centralized technology with cellular topologies does not seem to be working for such scenarios since each cell can cover a limited number of user terminals. So why not cell-free structures with decentralized technology? Of course, to serve many user terminals and to simplify the signal processing in a distributed manner, massive MIMO technology should be included. The combination between cell-free structure and massive MIMO technology yields the new concept: Cell-Free Massive MIMO.
What is Cell-Free Massive MIMO? Cell-Free Massive MIMO is a system where a massive number access points distributed over a large area coherently serve a massive number of user terminals in the same time/frequency band. Cell-Free Massive MIMO focuses on cellular frequencies. However, millimeter wave bands can be used as a combination with the cellular frequency bands. There are no concepts of cells or cell boundaries here. Of course, specific signal processing is used, see  for more details. Cell-Free Massive MIMO is a new concept. It is a new practical, useful, and scalable version of network MIMO (or cooperative multipoint joint processing) [2, 3]. To some extent, Massive MIMO technology based on the favorable propagation and channel hardening properties is used in Cell-Free Massive MIMO.
Cell-Free Massive MIMO is different from distributed Massive MIMO . Both systems use many service antennas in a distributed way to serve many user terminals, but they are not entirely the same. With distributed Massive MIMO, the base station antennas are distributed within each cell, and these antennas only serve user terminals within that cell. By contrast, in Cell-Free Massive MIMO there are no cells. All service antennas coherently serve all user terminals. The figure below compares the structures of Cell-Free Massive MIMO and distributed Massive MIMO.
|Distributed Massive MIMO||Cell-Free Massive MIMO|
 H. Q. Ngo, A. Ashikhmin, H. Yang, E. G. Larsson, and T. L. Marzetta, “Cell-Free Massive MIMO versus Small Cells,” IEEE Trans. Wireless Commun., 2016 submitted for publication. Available: https://arxiv.org/abs/1602.08232
 G. Foschini, K. Karakayali, and R. A. Valenzuela, “Coordinating multiple antenna cellular networks to achieve enormous spectral efficiency,” IEE Proc. Commun. , vol. 152, pp. 548–555, Aug. 2006.
 E. Björnson, R. Zakhour, D. Gesbert, B. Ottersten, “Cooperative Multicell Precoding: Rate Region Characterization and Distributed Strategies with Instantaneous and Statistical CSI,” IEEE Trans. Signal Process., vol. 58, no. 8, pp. 4298-4310, Aug. 2010.
 K. T. Truong and R.W. Heath Jr., “The viability of distributed antennas for massive MIMO systems,” in Proc. Asilomar CSSC, 2013, pp. 1318–1323.
47 thoughts on “Cell-Free Massive MIMO: New Concept”
Absolutely, that’s a great idea for 5G. How to identify the density of APs for a particular areas, with the use of Free Cell Massive Mimo?
Good question! The density of APs depends on the density of the users of that area. I do not have a precise value of the ratio between the number of APs (M) and the number of users (K) now. This ratio depends on what you want. If you want a high spectral efficiency per user, then M/K should be large. But I believe M/K < 10, for most of the cases.
how cell free Massive MIMO will be implemented in future, frequency re-use is core concept in the cellular technology.similarly the resources (spectrum and energy) are limited for cellular networks .how synchronization issues across APs are handled in cell-free communication
As in collocated massive MIMO, cell-free massive MIMO offers favorable propagation channels, and hence, all users can share the same time-frequency resource with not very high inter-user interference. In cell-free massive MIMO, synchronization can be done locally. For example, at a given time slot, some APs are chosen as the level-1 APs, and the remaining APs are considered as level-2 APs. Each level-1 AP is connected to at least another level-1 AP. All level-2 APs get synchronized by their neighboring level-1 APs. This method is called the hierarchical method.
Have you worked before with a prototype to prove the concept cell-free Massive MIMO (not Matlab)?
Is there channel hardening in cell-free massive MIMO? with channel hardening meaning that the effective channel gain is practically deterministic, which makes downlink pilots redundant. If there is no channel hardening, is that a problem for cell-free massive MIMO?
In cell-free massive MIMO, the channel still hardens, but is not as good as the channel in collocated massive MIMO (in term of the hardening property). In fact in , the signal processing at the users is done relying on the channel hardening property. The system performance is still very good. Furthermore, we can send downlink pilot via beamforming training scheme to improve the system performance. Note that the corresponding channel overhead for this downlink pilot scheme does not depend on the number of APs. So the channel hardening is not a problem for cell-free massive MIMO. More detail about this can be found here:
– Giovanni Interdonato, Hien Quoc Ngo, Erik G. Larsson, and Pal Frenger, “How much do downlink pilots improve cell-free Massive MIMO?,” IEEE Global Communications Conference (GLOBECOM), 2016. Link: https://arxiv.org/abs/1607.04753
What is the consideration of applying beamforming/precoding at Access Points instead at CPU?
In , we proposed to use beamforming at the APs in a distributed manner, not at the CPU. The CPU is used for exchanging the data and power control coefficients.
Yes in  I mean what is the advantage of doing beamforming at AP? By doing so we send K times signal over the backhaul which is increasing the load of the backhaul, don’t we?
You are right that by doing beamforming at the APs, the CPU has to send K signals to each AP over the backhaul links. However, its backhaul requirement is still lower than the backhaul requirement by doing beamforming at the CPU, when the number of APs (M) is large. More precisely, if beamforming scheme is performed at the CPU, then all APs have to send the channel estimates to the CPU. There are MK such coefficients for each coherence interval. So the backhaul requirements increase. More importantly, this makes the system unscalable (with respect to M).
P/S: it is worth to compare the backhaul requirements for two cases (beamforming at the APs and beamforming at the CPU) when M is not very large.
What is the difference between the cloud radio access network (C-RAN) and the cell-free massive MIMO system in this case? As both of them has the same structure, I mean multiple access points or RRHs connected to a CPU.
C-RAN is an architecture that moves the baseband processing from the access points to “the cloud”. Cell-free massive MIMO means that many distributed access points are serving the users by coherent joint transmission. These two methods can be used together, but one can also use C-RAN with conventional base stations and/or cell-free without C-RAN.
Hi. I study about cell-free massive MIMO. My thesis is about equipping each AP with more than one antenna. How can help us in cell-free Massive MIMO?
I think more antennas per AP can help us to serve more user and it’s better for energy efficiency and spectrum efficiency.
Hi! The pros and cons of having multiple antennas at each AP in cell-free Massive MIMO are discussed in the following paper: https://arxiv.org/abs/1710.00395
You will basically get more channel hardening and having more antennas is always better from a spectral efficiency perspective. However, if you have to choose between 100 single-antenna APs and 25 four-antenna APs, the single-antenna alternative might be more beneficial since you get more macro diversity (shorter distances between a typical UE and its nearest APs). When it comes to energy efficiency, more antennas and/or APs are not always beneficial. The throughput gains must compensate for the increased energy consumption.
I recently started reading a few papers to understand the basics of CF Massive MIMO. Is there any hot topic in research that hasn’t been addressed so far in CF Massive MIMO?
Yes, there are plenty of open problems in CF Massive MIMO. I recommend you to read the list of open problems in the following overview paper: https://arxiv.org/abs/1804.03421
As far as I understand, CF mMIMO is a large-scale version of nwMIMO / CoMP JT, which essentially can be viewed as a hybrid of conventional nwMIMO and mMIMO. That is, the cooperating BSs are divided into clusters via dynamic-cell clustering (a.k.a. user-centric clustering) and cooperation is restricted within each such cluster to facilitate cooperation / data sharing, as in conventional CoMP JT. The main difference is that the cooperation clusters in CF mMIMO are much larger, and this “massiveness” (in terms of the total number of transmit antennas – recall that CoMP JT is modeled mathematically as a composite MIMO BC) brings in the channel hardening and FP phenomena that we notice in the conventional mMIMO paradigm (at least, in the extent that this is possible by the fact that the antennas are not colocated but they are distributed instead). On the other hand, there are known tradeoffs from the CoMP literature in using larger vs. smaller cooperation clusters. Do these apply here as well?
A somewhat related question: CoMP did not meet the expectations, but as far as I know this is because the LTE-A implementations utilized network-centric clustering which suffers from OCI and provides poor sum-SE, especially for heterogeneous user distributions. So, what would be the difference in performance and complexity between conventional CoMP JT with DCC and CF mMIMO (which, by definition, utilizes DCC as far as I understand – the “cell-free” terms implies that, I believe), i.e., what would be the difference between small clusters (conventional nwMIMO) and large clusters (CF mMIMO)?
As you said, CF mMIMO is a particular form of CoMP JT with user-centric cooperation clusters and only data-sharing between the access points. This is not a new approach (I considered something similar in the papers “Cooperative Multicell Precoding: Rate Region Characterization and Distributed Strategies with Instantaneous and Statistical CSI” and “Optimality Properties, Distributed Strategies, and Measurement-Based Evaluation of Coordinated Multicell OFDMA Transmission” from 2010-2011), but it differs from the type of CoMP that was considered in LTE-A. For example, the “larger vs. smaller cooperation cluster” tradeoff is something that is created by the network-centric clustering approach in LTE-A. What CF mMIMO has contributed with is a renewed focus on CoMP JT with user-centric cooperation clusters, including how to deal with channel estimation and resource allocation.
Many of these things are discussed in the overview paper “Ubiquitous Cell-Free Massive MIMO Communications” that is available on Arxiv: https://arxiv.org/abs/1804.03421
Thank you for your feedback! I will certainly take a look on these papers!
I have a couple of questions regarding joint transmission in CoMP, which is somehow related with cell-free massive MIMO, and I was hoping that you might clarify these things.
1. Since in joint transmission each scheduled UE is served by two or more base stations, should these UEs have more than one antenna?
2. Related with the previous question: Do these base stations that serve jointly a specific UE send the same data symbol to it or different data symbols? (In order to make this question more clear, let’s give a small example. Assume that there are two base stations, BS1 and BS2 and serve a user, let’s simply call her/him UE, by sending a single symbol each. Do they send a symbol s1 and a symbol s2, respectively, or they transmit the same symbol (let’s say, s)?)
1. It is enough to have one antenna. Most papers on cell-free massive MIMO consider that case, but that is mainly for analytical tractability. Practical UEs will most certainly have more than one antenna (LTE phones have two receive antennas…)
2. Both options are possible. Sending the same symbol is called coherent transmission and sending different symbols is called non coherent transmission. Coherent transmission is preferable but requires phase-coherency between the access points. If N access points transmit one phase-coherent symbol, the received power is proportional to N^2. If N access points transmit N different symbols the received signal power is proportional to N.
CF appears a good concept. But won’t the power required to coordinate these antennas in a densed 5G network not overwhelm the system and result in increased power consumption? What about the prospect of UE battery life in a CF scenario?
Making CF network scalable in the sense of being implementable in dense networks is indeed a challenge. This is why CF is designed so that every access point only requires local channel state information that it can estimate from uplink pilots. It is only data signals that need to be conveyed over the fronthaul.
My decreasing the propagation distances between users and access points, the transmit power can be reduced. That will hopefully lead to longer UE battery life.
Some of the aspects that you raise are discussed in this new survey paper: https://arxiv.org/pdf/1910.00092.pdf
Many thanks Emil Björnson. Fantastic survey paper
Hi, I work on cell free massive MIMO. I am interested in the max-min SINR problem. Can you help me please and give an axe of research. thanks
The max-min power optimization problem has been solved in a sequence of papers:
H. Q. Ngo, A. Ashikhmin, H. Yang, E. G. Larsson, and T. L. Marzetta, “Cell-free Massive MIMO versus small cells,” IEEE Trans. Wireless Commun., vol. 16, no. 3, pp. 1834–1850, 2017.
E. Nayebi, A. Ashikhmin, T.L. Marzetta, H.Yang, and B.D. Rao,“Precoding and power optimization in cell-free Massive MIMO systems,” IEEE Trans. Wireless Commun., vol. 16, no. 7, 2017.
M. Bashar, K. Cumanan, A. G. Burr, M. Debbah, and H. Q. Ngo, “On the uplink max-min SINR of cell-free massive MIMO systems,” IEEE Trans. Wireless Commun., vol. 18, no. 4, pp. 2021–2036, 2019.
I’m not sure if there are any remaining open problems around this.
Currently downlink conjugate beamforming max-min power control is solved by “Second Order Cone Programming” (see [1, 2] in Emil’s message above), which is too slow for practical implementation. So a faster algorithm is needed.
I have a question for my research.
How does Cell-free massive MIMO differ from Artemis pCell SDR RAN?
They developed a DIDO system by using beamforming spaced antennas in 2011. See here https://www.artemis.com/products
pCell is a practical implementation of Network MIMO where the computations are made in the cloud. Cell-free Massive MIMO is way of analyzing network MIMO systems and most papers are considering that most of the signal processing is done near the antennas instead of in the cloud. If you have a look at the paper [R] below, then pCell is an implementation of Level 4 while Level 2 and Level 3 is what most often considered as Cell-free Massive MIMO in the literature:
[R] Emil Björnson, Luca Sanguinetti, “Making Cell-Free Massive MIMO Competitive With MMSE Processing and Centralized Implementation,” IEEE Transactions on Wireless Communications, https://arxiv.org/pdf/1903.10611
Hi, which decoding schemes can be used for cell free massive MIMO?
Basically the same as in cellular massive MIMO, either implemented centrally for the entire system or on a per-access-point basis. I recommend this paper: https://arxiv.org/pdf/1903.10611
What is the difference between cell-free massive MIMO and user-centric CoMP-JT?
Cell-free massive MIMO is to large extent user-centric CoMP JT. But it is advisable to not use the term CoMP in this context since most people associate that with the network-centric implementation that 3GPP tried to standardize. The name comes with a lot of baggage and bad experiences. So cell-free massive MIMO is a fresh start where we make use of the best ideas from the CoMP development and combine it with knowhow that was developed in the massive MIMO development.
Thanks for sharing this amazing work!
I have read a few articles on cell-free massive MIMO and I wish to simulate basic simulations that show the system performance. For example, the number of users that can be supported per each access point. How can I do that without considering optimization problems, (only system-level simulations)?
One starting point might be to have a look at the simulation code from some existing papers on the topic: https://cell-free.blogspot.com/p/source-codes.html
Optimization problems only need to be solved if you want to optimize something particular, such as the transmit powers. However, if you select the parameters in a reasonable but suboptimal way, you don’t need to optimize anything.
Is cell free massive MIMO is practically design anywhere in any lab ?
Yes, here is a startup that did it a few years ago: https://www.artemis.com/pcell
Dear Dr. Emil
Could cell-free massive MIMO be an efficient technology that allows to meet the ultra-low latency requirement in Tactile Internet networks (1 ms) ?
Cell-free Massive MIMO is mainly about increasing the worst-case data rates by spreading out the antennas. The latency is more related to other aspects than that. In particular, we cannot increase the speed of light, thus 1 ms means that information can at most travel 300 km. The propagation delay from the user to the base station will only be a small portion of that. However, the symbol time is inversely proportional to the bandwidth, so larger bandwidth means lower latency. I think the largest delays are created in the core network and when processing data.
Thank you very much for the prompt answer
If I understand correctly, in a cell-free massive MIMO network, there is no need to have disjoint spatial areas wherein only a specific frequency is used, right?
Let’s say, in a conventional cell-based network, with a frequency reuse factor of K=7, one would have to create patterns with K=7disjoint frequency bands, F1, F2, …, F7. In contrast, in cell-free networks, the whole frequency chunk F1-F7 is used across all the network, right?
Yes, that is correct. The same is also true in cellular Massive MIMO systems. The pilot sequences are transmitted with reuse factors to enable channel estimation within each cell with limited inter-cell interference, but after that, each cell can use all frequencies and the users are separated spatially using the many antennas.
Thank you Prof, I would like to ask about solving optimization problem for power control in cell-free Massive MIMO using the polyblock algorithm (PA) and branch-reduce-and-bound (BRB) algorithm for max-min and sum-rate, respectively. Is there any advanced algorithm than can be used in distributed and centralized power allocation?
The max-min problem is quasi-convex so it can be solved to optimality without the need for global optimization algorithms such as PA and BRB. When it comes to sum-rate maximization, I believe the algorithms described in my first book “Optimal Resource Allocation in Coordinated Multi-Cell Systems” can be used directly. I actually had cell-free systems in mind when writing that book, even if the “cell-free Massive MIMO” terminology had not been coined yet. The main building block of the PA and BRB algorithms is that one must be able to solve the feasibility problem of determining if a certain set of SINRs can be achieved or not. As long as that problem can be solved efficiently, the algorithm can be applied. It is the same feasibility problem as in the max-min fairness algorithm.
The main issue with both PA and BRB is that the complexity is very high. It isn’t reasonable to have more than 6 users.
I am a Research scholar. I want to work on Cell free Massive MIMO interference issues. Please refer some papers on this topic.
I think the best starting point is our textbook on the topic: https://arxiv.org/abs/2108.02541