I had an interesting conversation with a respected colleague who expressed some significant reservations against massive MIMO. Let’s dissect the arguments.
The first argument against massive MIMO was that most traffic is indoors, and that deployment of large arrays indoors is impractical and that outdoor-to-indoor coverage through massive MIMO is undesirable (or technically infeasible). I think the obvious counterargument here is that before anything else, the main selling argument for massive MIMO is not indoor service provision but outdoor macrocell coverage: the ability of TDD/reciprocity based beamforming to handle high mobility, and efficiently suppress interference thus provide cell-edge coverage. (The notion of a “cell-edge” user should be broadly interpreted: anyone having poor nominal signal-to-interference-and-noise ratio, before the MIMO processing kicks in.) But nothing prevents massive MIMO from being installed indoors, if capacity requirements are so high that conventional small cell or WiFi technology cannot handle the load. Antennas could be integrated into walls, ceilings, window panes, furniture or even pieces of art. For construction of new buildings, prefabricated building blocks are often used and antennas could be integrated into these already at their production. Nothing prevents the integration of thousands of antennas into natural objects in a large room.
Outdoor-to-indoor coverage doesn’t work? Importantly, current systems provide outdoor-to-indoor coverage already, and there is no reason Massive MIMO would not do the same (to the contrary, adding base station antennas is always beneficial for performance!). But yet the ideal deployment scenario of massive MIMO is probably not outdoor-to-indoor so this seems like a valid point, partly. The arguments against the outdoor-to-indoor are that modern energy-saving windows have a coating that takes out 20 dB, at least, of the link budget. In addition, small angular spreads when all signals have to pass through windows (maybe except for in wooden buildings) may reduce the rank of the channel so much that not much multiplexing to indoor users is possible. This is mostly speculation and not sure whether experiments are available to confirm, or refute it.
Let’s move on to the second argument. Here the theme is that as systems use larger and larger bandwidths, but can’t increase radiated power, the maximum transmission distance shrinks (because the SNR is inversely proportional to the bandwidth). Hence, the cells have to get smaller and smaller, and eventually, there will be so few users per cell that the aggressive spatial multiplexing on which massive MIMO relies becomes useless – as there is only a single user to multiplex. This argument may be partly valid at least given the traffic situation in current networks. But we do not know what future requirements will be. New applications may change the demand for traffic entirely: augmented or virtual reality, large-scale communication with drones and robots, or other use cases that we cannot even envision today.
It is also not too obvious that with massive MIMO, the larger bandwidths are really required. Spatial multiplexing to 20 terminals improves the bandwidth efficiency 20 times compared to conventional technology. So instead of 20 times more bandwidth, one may use 20 times more antennas. Significant multiplexing gains are not only proven theoretically but have been demonstrated in commercial field trials. It is argued sometimes that traffic is bursty so that these multiplexing gains cannot materialize in practice, but this is partly a misconception and partly a consequence of defect higher-layer designs (most importantly TCP/IP) and vested interests in these flawed designs. For example, for the services that constitute most of the raw bits, especially video streaming, there is no good reason to use TCP/IP at all. Hopefully, once the enormous potential of massive MIMO physical layer technology becomes more widely known and understood, the market forces will push a re-design of higher-layer and application protocols so that they can maximally benefit from the massive MIMO physical layer. Does this entail a complete re-design of the Internet? No, probably not, but buffers have to be installed and parts of the link layer should be revamped to maximally use the “wires in the air”, ideally suited for aggressive multiplexing of circuit-switched data, that massive MIMO offers.
11 thoughts on “A case against Massive MIMO?”
Thanks Prof. Larsson for the information. Please throw some light on the ramifications of TCP/IP on bursty video traffic and how making change in the link-layer (with presence of buffers) and application and higher layer solve this problem in Massive MIMO. Thanks!
Bhupendra Kumar (IIT Delhi)
I think the point is just that the upper layer (or MAC/link layers) have to be engineered so as to make optimal use of the massive MIMO physical layer. Despite all papers on “cross-layer” designs over the last two decades, this does not seem to have been solved yet…
Excellent point on revamping upper layers for wireless networks, given that Massive MIMO can provide “wires in the air”. Perhaps some upper layer technologies for wireline become suitable?
Yes, maybe… again I think it is just a matter of engineering the MAC/link layers properly…
Hi, on the theme of indoor coverage and massive MIMO.
I work at an operator and can only agree that it is very difficult already today at relatively low frequency bands to penetrate buildings (where we have the vast majority of subscribers). In fact I would be happy if we found windows with as low penetration loss as 20 dB since in many cases the loss is even higher.
Looking at the 3.5 GHz band for NR, one of my major concerns is indoor coverage. Is my understanding correct that when it comes to pure SINR for users located deep indoor, beamforming/massive MIMO will not contribute so much, espescially not in UL?
Looking at the DL, there should be a gain from concentration of energy thanks to the narrower beam and maybe some gain thanks to channel hardening but in UL I can not find the corresponding gains or?
The only gain from massive MIMO/beamforming I can see in UL is interference rejection but in most cases UL load (interference) is very low so it will not be a a big change unless we have very high UL load.
Or is there something I have overlooked?
Thanks for your comment.
There is no difference in principle between UL and DL, as far as the MaMIMO beamforming gains and interference suppression capabilities are concerned. The coherent gain (numerator of the effective SINR) scales in the same manner with M (or M-K) in either case, see Fundamentals of Massive MIMO, chapters 3-4.
However the transmit power typically is lower on UL so the link budget may be worse. Hence, one is more likely to be limited by noise than interference on UL – which in turn may be a major effect! (In addition, in TDD, the channel estimates become noisy when the UL is noise-limited which in turn affects the DL beamforming.) I am not really sure about the numbers in practical deployments here.
Interesting point about the windows, yes it does seem that the physics is against indoor-to-outdoor coverage…
Based on our current experience in Switzerland, indoor penetration is below expectations. As we have very strict transmission power limits simply indoor penetration of LTE 1800MHz (2×2 or 4×4) is better than N78 band with mMIMO.
Sure its early release and devices need to be optimized too, but 25-30 dB loss on good quality Swiss windows is painful.
Regarding the outdoor to indoor penetration loss, my company works with repeater technology, in order to improve coverage and capasity in trains.
How will massive MiMo and polarisation of signals make a difference to MiMo repeater system, and can repeater technology also address the outdoor to indoor issues for buildings?
I have seen any specifications for MIMO repeaters, so I don’t know what such products are capable of. But repeaters can definitely be useful to alleviate outdoor-to-indoor penetration losses, if the signals canoe received at one side of the wall and retransmitted at other side of the wall.
I think the link budget improvements due to MIMO and repeater technology can be simultaneously achieved, but one might lose the ability to perform spatial multiplexing.
Given that the vast majority of users are situated indoors, and that the vast majority of networks are situated outdoors, there are very strong reasons why operators would want to make outside-in coverage work, even if it means trading off some capacity for improved coverage. Would be very interested in your view of ways operators can make this tradeoff. Assume some kind of supplemental UL or HPUE to improve UL coverage. What would happen on the DL? Ultra low modulation scheme? What kind of link is needed to maintain CSI and therefore the integrity of the MIMO beams? This is a hard but high stakes problem.
Indoor users are almost static, so selecting good MIMO beams shouldn’t be hard. Even if the uplink SNR is low, one can filter it over long time to get good estimates. Even grid-of-beams like (CSI-RS) approaches might work in these cases, since one can use a lot of time and reference signals to find good beams. So even if the SNR is low, I think it will be possible to obtain good CSI.
I don’t think that any particular modulation scheme is needed. Just conventional Massive MIMO theory + conventional modulation and coding schemes. These methods are operating close to the Shannon capacity bound. But even if we have good CSI, as motivated above, the SNR will remain to be low during the data transmission. We cannot cheat the underlying physics, but more antennas can be utilized to focus the signals better, to get a larger array gain.
If you want to provide really good indoor coverage, you need to deploy the network indoor. In Sweden there is WiFi almost everywhere which demonstrates that a large-scale deployment of indoor access points. If these access points can coordinate their transmission, we will get good indoor performance.