The “Massive MIMO” name is currently being used for both sub-6 GHz and mmWave applications. This can be very confusing because the multi-antenna technology has rather different characteristics in these two applications.
The sub-6 GHz spectrum is particularly useful to provide network coverage, since the pathloss and channel coherence time are relatively favorable at such frequencies (recall that the coherence time is inversely proportional to the carrier frequency). Massive MIMO at sub-6 GHz spectrum can increase the efficiency of highly loaded cells, by upgrading the technology at existing base stations. In contrast, the huge available bandwidths in mmWave bands can be utilized for high-capacity services, but only over short distances due to the severe pathloss and high noise power (which is proportional to the bandwidth). Massive MIMO in mmWave bands can thus be used to improve the link budget.
Six key differences between sub-6 GHz and mmWave operation are provided below:
|Deployment scenario||Macro cells with support for high user mobility||Small cells with low user mobility|
|Number of simultaneous users per cell||Up to tens of users, due to the large coverage area||One or a few users, due to the small coverage area|
|Main benefit from having many antennas||Spatial multiplexing of tens of users, since the array gain and ability to separate users spatially lead to great spectral efficiency||Beamforming to a single user, which greatly improves the link budget and thereby extends coverage|
|Channel characteristics||Rich multipath propagation||Only a few propagation paths|
|Spectral efficiency and bandwidth||High spectral efficiency due to the spatial multiplexing, but small bandwidth||Low spectral efficiency due to few users, large pathloss, and large noise power, but large bandwidth|
|Transceiver hardware||Fully digital transceiver implementations are feasible and have been prototyped||Hybrid analog-digital transceiver implementations are needed, at least in the first products|
Since Massive MIMO was initially proposed by Tom Marzetta for sub-6 GHz applications, I personally recommend to use the “Massive MIMO” name only for that use case. One can instead say “mmWave Massive MIMO” or just “mmWave” when referring to multi-antenna technologies for mmWave bands.
32 thoughts on “Six Differences Between Massive MIMO for Sub-6 GHz and mmWave”
There is a stereotype that mm-wave channels are sparse (which in fact has been validated by some measurement trials). This has stimulated a lot of interest into compressed sensing schemes, PCA algorithms etc for channel estimation at mm-wave frequencies. Yet, very recently, there is a widespread belief that mm-wave channels may experience 6-8 clusters of rays with tens of rays within each cluster. 3GPP is currently trying to standardize this.
We are working on that at the moment.
Good point! I think what one can say for sure is that, in a given scenario, mmWave channels will be less rich than sub-6 GHz channels, due to less diffraction and more blockage at mmWave frequencies. Then how “sparse” the channels will be depends on the sceanrio. I think it is wise that 3GPP is not relying on the existence of sparsity.
You are most welcome to share references to recent measurement campaigns!
Can we regard Massive MIMO in mmWave as beamforming Massive MIMO? As in sub-6 GHz the aim is to increase the efficiency of loaded cells but in mmWave, the aim, even for a single user, is to focus the beam to avoid losses due to path loss, i.e. beamforming.
Yes, that is correct.
If you have one user, beamforming is fairly easy since you only need to point the beam in roughly the right direction. If you spatially multiplex many users, beamforming is more challenging since you want to point the beams to avoid unnecessary interference between users. That is one reason why hybrid beamforming is ok for mmWave applications, but not a good choice for sub-6 GHz applications.
Small cells do not necessarily mean a small number of users! Think of an auditorium, library, or cafeteria, stadium, etc., settings. The number of co-channel users can still be “large.” 🙂
Thanks for the great input! I realize that I could have defined my “simultaneous user” terminology more precisely. Let me explain what I meant.
I certainly agree that the number of users can be large in a small area. However, one should bear in mind that there is usually a large difference between the number of users that reside in a cell and the number of users that are active in a given time slot (and thus must be multiplexed).
Suppose you have 250 people in an auditorium. In the visionary METIS scenario “TC2: Dense urban information society”, each person generates one packet of 30 Mbyte per minute. With 1 Gbit/s, it takes 0.24 seconds to transmit that packet. The total time it would take to transfer 250 packets (one per user) is 1 minute. Hence, you will on average only have one simultaneously active user… This was the hidden reason behind in my “simultaneous user” terminology 🙂
That said, I agree that there are plausible small-cell scenarios where you need to spatially multiplex users, so it would be great if mmWave communications support SDMA. But I don’t think it is the main feature or benefit of the technology.
This is a great article to compare the differences. Yeah, terminologies and use cases could be mixed up sometimes.
There is also another terminology that always comes together with massive MIMO; that is, multi-user MIMO. According to your analysis, is it right to say that multi-user MIMO (spatial multiplexing?) ability is not efficient for mmwave?
Multi-user MIMO (a.k.a. spatial multiplexing or space-division multiple access) can certainly be used in mmWave bands. In fact, it would be a great technology due to the large number of antennas that one can pack into a small area!
My concern is that mmWave communications have a short range and troublesome link budget. Today, most cellular base stations have at most one active user at a time. If you reduce the range, there will be even fewer users active per TTI, except in some very special scenarios with very many users that require very high data rates simultaneously.
When the link budget is bad, FDMA is also a convenient alternative to multi-user MIMO, since when you divide the frequency band equally between two users, each one gets a 3 dB increase in SNR.
I’ve noticed that many people believe that beamforming is done at mmW frequencies but not at lower ones, and this is used to differentiate mmW and lower frequencies.
This comparison, however, is inaccurate because one can do beamforming at any frequency, as long as a multi antenna system is used.
Perhaps it’s more accurate to state that it is easier to achieve sharper beamforming (pencil beams) at mmW compared to lower frequencies thanks to the fact that we can have more antennas at mmW frequencies for the same footprint. This in turns means better spatial multiplexing at mmW frequencies.
MmWave is indeed ideal for spatial multiplexing, at least if we can achieve a fully digital implementation. We just need to find deployment scenarios where there are sufficiently high traffic demands from many users to utilize the fully capacity of the technology! In the first iterations of 5G, I think that single-user transmission will be the key use case for mmWave technology.
Nice post Emil!!
Let me also call your attention on an invited paper that we recently published on ZTE Communications:
S. Buzzi and C. D’Andrea, “Massive MIMO 5G cellular networks: mm-wave versus mu-wave frequencies,” ZTE Communications, special issue on 5G New Radio, Vol. 15, No. S1, pp. 41-49, 2017
Surprisingly, our paper also points out six differences between massive MIMO at sub-6 Ghz and at mmWave frequencies, even though they of course are not exactly the same as the ones that you point out.
I am summarizing them here:
Difference # 1: mm-waves may be doubly massive (i.e. many antennas also on the mobile device)
Difference # 2: Analog (beam-steering) beamforming may be optimal
Difference # 3: The rank of the channel does not increase with the number of antennas
Difference # 4: Channel estimation is simpler at mmWaves
Difference # 5: Pilot contamination can be less critical
Difference # 6: Antenna diversity/selection procedures may be less effective
I would appreciate receiving your thoughts on this!
I was not aware of your paper, but is very interesting!
Since you asked for my thoughts, here you go:
Difference # 1: I agree, at mmWave frequencies, you can also use many antennas in the terminal.
Difference # 2: Yes, in single-user, line-of-sight communication, analog beam-steering is sufficient. At least if the bandwidth is not more than, say, a few 100 MHz, so that the beam squint effect can be neglected.
Difference # 3: The rank of the channel will at least be smaller the higher the frequency is.
Difference # 4: There is indeed more structure in mmWave channels. Whether or not that allows for simpler channel estimation remains to be proved, since you need to learn the structure to utilize it. If a hybrid analog-digital implementation is used, then it might be harder to learn the structure.
Difference # 5: I agree, mmWave systems might be noise-limited and then pilot contamination is not an issue.
Difference # 6: I am not a proponent of antenna selection procedures, but yes, with less richness in the multipath propagation there is less diversity to extract.
Thank you for the wonderful comparsion.
#1. You mention at the last difference that it is feasible to implement full digital Beamforming in the transceiver at Sub-6 GHz, however doesn’t the number of antennas increase the RF chains to use full digital beamforming, and hence more feasible to use a Hybrid Beamforming in both cases.
#2. When hybrid beamforming is implemented with limited RF chains, I hardly can notice any improvements in Spectral efficiency (at Sub-6 GHz) with the increase in the number of transmit antennas.
I apologize if I am missing something, any references would certainly help my studies.
#1. It might be easier to implement hybrid beamforming but, as compared to full digital beamforming, it will create a lot of other practical problems. For example, it becomes more difficult to estimate channels (i.e., to steer the analog beams in the right directions) and to cancel interference between users when you cannot control every antenna element.
#2. I think you are essentially answering the question yourself. If you add antennas without adding RF chains (hybrid beamforming), the gain will be small. If you add antenna and RF chains jointly (full digital beamforming), you get the maximum gains.
Very interesting post!
Can you elaborate on your point above that hybrid beamforming has lot of practical problems and why it is not preferred according to you?
Thanks in advance!
This is something that we write more about in the following paper:
There are two main points: 1) It is not obvious that (fully connected) hybrid is actually easier to implement than fully digital, as explained in the paper. 2) Fully digital implementations of mmWave technology are already on its way. There were experiments shown at conferences this year. The future is digital, and now it seems to come even quicker than I first believed. Hybrid leads to lower performance and requires complicated beam searching and beam tracking, which is not needed with fully digital implementations that exploit reciprocity.
When we are saying that mm-wave experience few propagation paths, does that mean in mmWave we don’t need to do multi-path mitigation techniques such as equalization?
Does it mean that because of using beam-forming there is no frequency selective fading as well?
If there is only one propagation path, there will be little frequency selectivity (but it still exist due to the change of wavelength over the frequency band, which effectively changes the inter-antenna distances measured in wavelengths).
However, you need to build the system to be capable of handling multiple paths (direct path plus a few reflections).
Per “3GPP TR 38.913 document” it’s like macro-cells also gonna be used at mmWave band, will the first case in the differences between sub-6 GHz and mmWave operation then shrinks down to the mobility of users?
I appreciate if you let us know your comment on that.
I wrote the six differences to be as distinctly different as possible. One can certainly use sub-6 GHz spectrum for small cells with low-mobility users (we have done that for decades…). It is also possible to make a macro cell deployment using mmWave spectrum, but the coverage will be quite bad.
The 38.913 document describes both sub-6 GHz and mmWave spectrum for the “dense urban” scenario, while mentioning that such a deployment will contain both a macro layer and a micro layer. The document says that one can consider any frequency range in any layer for evaluation purposes. However, I think that such an evaluation will lead to the conclusions that sub-6 GHz should be used in the macro layer and mmWave in the micro (small cell) layer.
Hello, Thanks for your article, one point which still not clear is that, what is the main reason we can’t make Digital beamforming for mmwave and why can’t we add more RF Chains like 64 or 120 RF Chains, what is main limitation ?
I think there are two main reasons: The first reason is that the mmWave technology is new and therefore the first implementations are using simplified hardware. Over the next five years, we will likely see a transition towards digital beamforming also in mmWave. The second reason is that mmWave technology is mainly intended for transmission to one user at a time, at least in the first deployments. One can then get away with simpler beamforming that is not as fine-tuned as when we also need to cancel interference using methods such as zero-forcing.
Thanks Emil for this great post, one point about mmwave.
1) Why Multi user MIMO is not supported?
2) Is it true that in mmwave communications, you have to split the antenna panel and you can’t use the full panel like it is the case in 3.5GHz? Why is that?
1) It can be supported but since the coverage area is small in mmWave bands, the chance of having multiple users that request the maximum data rate at exactly the same time is much smaller than in conventional frequency bands. Hence, it is not the first feature that is going to be implemented.
2) No, you don’t have to split the antenna panel like that, and I would discourage from doing that since you loose in beamforming capability. But some simplified implementations might be using that approach.
Thanks for your reply,
Last question, you have up to 8 RF Chains in mmWave Massive MIMO products, that is why it is only max 8T8R antennas?
This 4T4R or 2T2R mmWave products depends on number of antennas?
Yes, 8T8R means that there are 8 RF chains for transmission and 8 RF chains for reception. This is what people in academia would call 8 antennas, but each RF chain is often connected to multiple “antenna elements” so there is some confusion of what an antenna really is.
Great article, one question
In Massive MIMO, Multi user MIMO can benefit from up to 16 layers for Pairing, but also in 3.5 GHz we can have up to 8 SSB beams as per 3GPP in FR1.
But how can we support multi user MIMO with pairing of 16 layers with only a max of 8 beams, 16 layers doesn’t mean 16 beams?
I don’t know the details of the NR standard, but I think that the SSB beams are used to search for users in different directions without knowing their channels. They are not used for data transmission.
If you transmit 16 layers during data transmission, you will have to determine 16 beams for that. Those beams should be adapted to the user channels to limit the interference between the users.
Hi sir, I’m working on massive mimo receiver in which i’m using LLL-ZF/LLL-MMSE detection technique. My question is can we apply a favorable propagation check in LLL algorithm before size reduction, assuming that due to favorable propagation the two selected vectors for size reduction might be nearly orthogonal and if true jump to lovasz check condition otherwise size reduction.
I haven’t worked with that algorithm myself. I think it makes sense that one should only consider the most dominant interferers when designing the detection technique. However, I’m not sure if favorable propagation is the best metric to determine which interferers to consider. It is better to consider the size of inner products of channel vectors without any normalization.
Sir, how do I calculate the size of an inner product. Does this mean that the two selected vectors should be like size of (h1’g1) and with what value do I have to compare the size value.
An inner product is also known as dot product: https://en.wikipedia.org/wiki/Dot_product
You need to vectors of the same size (dimension) and the result is a scalar.