64 or 128 Antennas?

After some successful trials, the first deployments of TDD-LTE with Massive MIMO functionality were unveiled earlier this year. For example, the telecom operator Sprint turned on Massive MIMO base stations in Chicago, Dallas, and Los Angeles last April.

If you read the press release from Sprint, it is easy to get confused regarding the number of antennas being used:

Sprint will deploy 64T64R (64 transmit, 64 receive) Massive MIMO radios using 128 antennas working with technology leaders Ericsson, Nokia, and Samsung Electronics.

From reading this quote, I get the impression that the Massive MIMO arrays contain 128 antennas, whereof 64 are used for the transmission and another 64 for the reception. That would be a poor system design, since channel reciprocity can only be exploited in TDD if the same antennas are used for both transmission and reception!

Fortunately, this is not what Sprint and other operators have actually deployed. According to my sources, the arrays contain 64 dual-polarized elements, so there are indeed 128 radiating elements. However, as I explained in a previous blog post, an antenna consists of a collection of radiating elements that are connected to the same RF chain. The number of antennas is equal to the number of RF chains, which is 64 in this case. The reason that Sprint points out that there are 64 transmit antennas and 64 receive antennas is because different RF chains are used for transmission and reception. The system switches between them according to the TDD protocol. In principle, one could design an array that has a different number of RF chains in the uplink than in the downlink, but that is not the case here.

So how are the 128 elements mapped to 64 antennas (RF chains)? This is done by taking pairs of vertically adjacent elements, which have the same polarization, and connecting them to the same RF chain.  This is illustrated in the figure to the right (see this blog post for pictures of how the actual arrays look like). As compared to having 128 RF chains (and antennas), this design choice results in lower flexibility in elevation beamforming, but the losses in data rates and multiplexing capability are supposed to be small since there are much larger variations in azimuth angles between the users in a cellular network than in the elevation angles. (This is explained in detail in Section 7.3-7.4 of my book). The advantage is that the implementation is more compact and less expensive when having 64 instead of 128 antennas.

15 thoughts on “64 or 128 Antennas?”

  1. I have yet to see anyone mentioning the power consumption of an antenna array whether 64, 128 antenna elements or higher. Can anyone shed light on this please?

  2. I have a small doubt on the antenna spacing of Massive MIMO panel in 2.5 GHz deployed by Sprint, where they mentioned 64 TRX. How is half-lambda-spacing achieved in a small device ~ near to laptop size?

    1. Sprint is using a 64-antenna base station for LTE. It can be used to serve any LTE device/phone. One of the beauties of Massive MIMO technology is that you only need advanced hardware at the base station, while the user devices can have a single antenna and very simple. (That said, every LTE device has at least two receive antennas.)

  3. I wanted to know optimal antenna spacing for massive MIMO. How does this antenna spacing affect the steering angle. Can you please give any reference to this. Thanks!

      1. Thanks Prof. Emil Björnson for your kind reply. I was trying to understand the reason for doing 64RF chains with 128 antennas by connecting 2 vertical antennas with the same RF. This no doubt will reduce the cost of the hardware, however comes with a cost of performance. Assuming TDD, we will not have full 128 antenna channel information due to similar antenna configuration on UL, and this limits the precoder design. Assuming MRT in the DL, where we will be matching to the DL channel, we will only have 64 antenna channel estimates and apply the same to the 128 antennas (the vertical antennas connected to same RF gets the same coefficient). Don’t this mismatch affect the performance of the DL?

        1. Yes, it will reduce the performance, but if the MRT beamforming weights for two vertically adjacent antennas would be nearly the same, then it shouldn’t hurt much to force them to be identical. In many deployments, the users’ signals arrive to the base station from roughly the same elevation angle, which leads to almost the same MRT weights for vertically adjacent antennas. This explained in the following white paper from Ericsson (see Figure 4):

          https://www.ericsson.com/en/white-papers/advanced-antenna-systems-for-5g-networks

  4. Hi sir, I am doing my research work on energy efficient cross layer optimization of multimedia data using fuzzy logic. I would like to ask can we use this approach in massive MIMO? One favor what topic I have to take in my PhD thesis.

    1. I haven’t worked with cross-layer optimization with fuzzy logic myself, but I suppose it can be applied along with many different physical-layer techniques. The rate formulas for massive MIMO that you find in the book Fundamentals of Massive MIMO can be utilized for cross-layer optimization. But if you are going to work on a PhD thesis, the main question you should ask is: What are the open questions that you want to answer? It is better to start from that perspective than to start with selecting the tools that you want to use.

  5. Sir, I want to design a MU MIMO system using a uniform planar array and optimize the training sequences. This is my PhD work on massive MIMO.

    1. You can read about channel modeling for uniform planar arrays in Section 7.3 of my book Massive MIMO networks (https://massivemimobook.com).

      As long as you use a set of orthogonal pilot sequences, it doesn’t matter what the sequences are. But you can find some examples in Section 3.1 of the book.

  6. Excellent book!!!! Looking at 5G specifically. 3 quick questions. With 64T64R…1) how many beams can the system support during the same symbol period? 2) Can different PRB’s (allocated to different users) be beamformed in different directions? 3) If performing 5G Carrier Aggregation, can the different carriers/bands be beam formed independently at the same time?

    1. 1) I don’t know the detailed of the NR standard. In theory, you can superimpose as many beams as you like. But in practice, you don’t want more than 64 beams. 2) Yes, and you can also assign the same PRB to multiple users, if each user is using a different well-separated beam. 3) Yes.

Leave a Reply

Your email address will not be published. Required fields are marked *