Episode 13: Distributed and Cell-Free Massive MIMO

We have now released the 13th episode of the podcast Wireless Future, with the following abstract:

Wireless devices normally connect to a single access point, deployed at one location. The access points are deployed sparsely to create large cell regions, each controlled by the nearest access point. This architecture was conceived for mobile telephony and has been inherited by today’s networks, even if those mainly transfer wireless data. However, future wireless networks might be organized entirely differently. In this episode, Erik G. Larsson and Emil Björnson discuss how one can create cell-free networks consisting of distributed massive MIMO arrays. The vision is that each user will be surrounded by small access points that cooperate to provide uniformly high service quality. The conversation covers the key benefits, how the network architecture can be evolved to support the new technology, and what the main research challenges are. To learn more, they recommend the article “Ubiquitous Cell-Free Massive MIMO Communications” and the new book “Foundations of User-Centric Cell-Free Massive MIMO”.

You can watch the video podcast on YouTube:

You can listen to the audio-only podcast at the following places:

12 thoughts on “Episode 13: Distributed and Cell-Free Massive MIMO”

  1. I have three questions:
    1. Can be Cell-Free Massive MIMO reality due to the synchronization challenge?
    2. Do we need to combine Cell-Free Massive MIMO with the other emerging technologies for better system performance or Cell-Free Massive MIMO can perform all at once?
    3. Is the user-centric a solution because linear signal processing does not work well in Cell-Free Massive MIMO?

    1. 1. Yes, I think so, but it will indeed be one of the implementation challenges.

      2. It depends on what you want to achieve. Cell-free Massive MIMO can be utilized in different frequency bands, sub-6 GHz as well as mmWave frequencies.

      3. No, the issue is rather that we cannot let all access points serve all users, because the complexity and fronthaul signaling will be too large. The goal with user-centric networking is that every user should be served by all its surrounding access points, but not those that are further way. There are no predefined cells, but every user is served by those access points that are needed to provide good service.

  2. Hi Emil,
    Thanks for ep. 13. I have two questions for you:

    1- I see that with the O-RAN concept being standardized and indeed a reality these days, the idea of distributed massive MIMO can be realized through O-RAN deployment where multiple RUs (connected to the same DU) can cooperate to simultaneously serve a group of users and therefore massive array gain can be achieved at DU side. Do you agree with this? If not, what differences do you see between the two concepts? Please elaborate on the similarities between the two concepts from your perspective.

    2- Do you possible know if any of 3GPP releases has been working on standardizing distributed massive MIMO concept? If so, can you please refer me to any parts of 3GPP standards focusing on making distributed massive MIMO a reality?


    1. These are good questions! I’m not an expert on the 3GPP standardization details, but as far as I know (and have been told by colleagues in the industry), distributed Massive MIMO can be built within 5G. The decoupling of the data and control plane imply that the user can connect to one AP, but then get served by a multitude of APs. The user device doesn’t need to know how the network is carrying out its transmissions – if there is one AP or many APs involved. One will most likely have to use the 5G features for reciprocity-based channel estimation, since the feedback-based estimation methods use codebooks designed for co-located arrays.

      The key challenge is to actually implement in a distributed Massive MIMO system in practice and deal with things like synchronization and resource allocation.

      O-RAN might be helpful in the sense that it specifies some protocols that can be utilized in the implementation (e.g., functional splits) to pass around information between APs and processing units, but one can certainly use proprietary solutions as well.

  3. Dear Professor Bjornson

    I want to know that in the simulations that you have done in your cellfree massive MIMO book. Do you change the location of APs in the different setups? By this question, I want to know that it is essential for APs to be in a fixed random location in different setups in simulation of the system.

    1. The simulations are done like this: We deploy the APs in one way (randomly or on a square grid) and then we measure the user performance at random user locations. We are repeating the experiment for different AP locations to see the differences.

      The purpose of considering random locations is to compare deployments in different parts of a city. For a particular setup, the APs will certainly not be moving around. So if you want to study a single AP deployment that is alright, but you need to be careful so that the conclusions are valid more generally.

      1. Dear Professor Bjornson

        Thanks for your attention. If I correctly understand your answer you say that In simulation of a cell-free massive MIMO system if we have for example 100 setups in the simulation and we deploy the APs randomly the location of the APs can vary during this 100 setups but I think even if we deploy APs randomly the location of them should not vary during this 100 setups and they should be in the fixed locations.

        1. The simulation methodology depends on what you want to achieve. Suppose you are simulating an area of 1 km x 1 km in your code.

          If you want to study a single deployment scenario of 1 km x 1 km, then you deploy the APs at fixed locations and consider 100+ random locations for the users.

          If you want to study a very large-scale deployment scenario (infinitely large), which is too much to simulate at once, then you can consider that every subarea of size 1 km x 1 km will have random AP locations. By simulating many such subareas, you will capture the random variations over the large-scale deployment scenario.

          The latter is what we do in the book.

  4. Dear Professor Bjornson

    Thanks for your informative answers.
    I have two question;
    1) Is it essential to use wrap-around topology for simulate a cell-free system? Can you explain about the importance of using it?
    2) If we use wrap-around topology for a cell-free network then the serving APs for different UEs will be different, Is it true?
    If so, is not it a problem in simulation?

    1. 1. If you don’t use wrap-around, then users close to the edges of the simulation area will experience less interference than those in the center of the area. This will lead to an overestimation of the achievable performance. The purpose of wrap-around is to simulate a finite-sized area but get performance results that reflects the typical performance in an infinitely large area.

      2. Yes, but this has nothing to do with the wrap-around topology. Each user should be served by the APs that are in its vicinity.

      If you are curious of how to implement these things, you can check out the simulation code related to my book on cell-free Massive MIMO.

  5. Dear Professor Bjornson

    In the simulation that you propose for your cell-free book you drop UEs in the simulation area one by one, Is this related to the correlation between the shadowing of UEs or there are other reasons?
    and If we consider UEs with uncorrelated shadowing then how the results will change? Will the results be very different from the case of consideration correlated shadowing?

    1. The reason that we drop UEs one by one is that the pilot assignment and cluster formation is based on that principle; in practice, users will connect to the network at different time instances.

      Uncorrelated shadowing would lead to less realistic results. If I had to guess, I think that the SEs will generally be improved since there is more macro diversity.

Leave a Reply

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