When Will Hybrid Beamforming Disappear?

There has been a lot of fuss about hybrid analog-digital beamforming in the development of 5G. Strangely, it is not because of this technology’s merits but rather due to general disbelief in the telecom industry’s ability to build fully digital transceivers in frequency bands above 6 GHz. I find this rather odd; we are living in a society that becomes increasingly digitalized, with everything changing from being analog to digital. Why would the wireless technology suddenly move in the opposite direction?

When Marzetta published his seminal Massive MIMO paper in 2010, the idea of having an array with a hundred or more fully digital antennas was considered science fiction, or at least prohibitively costly and power consuming. Today, we know that Massive MIMO is actually a pre-5G technology, with 64-antenna systems already deployed in LTE systems operating below 6 GHz. These antenna panels are very commercially competitive; 95% of the base stations that Huawei are currently selling have at least 32 antennas. The fast technological development demonstrates that the initial skepticism against Massive MIMO was based on misconceptions rather than fundamental facts.

In the same way, there is nothing fundamental that prevents the development of fully digital transceivers in mmWave bands, but it is only a matter of time before such transceivers are developed and will dominate the market. With digital beamforming, we can get rid of the complicated beam-searching and beam-tracking algorithms that have been developed over the past five years and achieve a simpler and more reliable system operation, particularly, using TDD operation and reciprocity-based beamforming.

Figure 1: Photo of the experimental equipment with 24 digital transceivers that was used by NEC. It uses 300 MHz of bandwidth in the 28 GHz band.

I didn’t jump onto the hybrid beamforming research train since it already had many passengers and I thought that this research topic would become irrelevant after 5-10 years. But I was wrong – it now seems that the digital solutions will be released much earlier than I thought. At the 2018 European Microwave Conference, NEC Cooperation presented an experimental verification of an active antenna system (AAS) for the 28 GHz band with 24 fully digital transceiver chains. The design is modular and consists of 24 horizontally stacked antennas, which means that the same design could be used for even larger arrays.

Tomoya Kaneko, Chief Advanced Technologist for RF Technologies Development at NEC, told me that they target to release a fully digital AAS in just a few years. So maybe hybrid analog-digital beamforming will be replaced by digital beamforming already in the beginning of the 5G mmWave deployments?

Figure 2: Illustration of what is found inside the AAS box in Figure 1. There are 12 horizontal cards, with two antennas and transceivers each. The dimensions are 308 mm x 199 mm.

That said, I think that the hybrid beamforming algorithms will have new roles to play in the future. The first generations of new communication systems might reach faster to the market by using a hybrid analog-digital architecture, which require hybrid beamforming, than waiting for the fully digital implementation to be finalized. This could, for example, be the case for holographic beamforming or MIMO systems operating in the sub-THz bands. There will also remain to exist non-mobile point-to-point communication systems with line-of-sight channels (e.g., satellite communications) where analog solutions are quite enough to achieve all the necessary performance gains that MIMO can provide.

31 thoughts on “When Will Hybrid Beamforming Disappear?”

  1. What about the energy efficiency part? It seems to me that hybrid precoding is good candidate for energy-efficient transmission.

    1. That is possible, but I wouldn’t be so sure about that. Hybrid architectures require more complicated analog circuitry and this is often the most power consuming part.

  2. On the other side, I guess there is a misunderstanding about the costs of analog components and ADC/DACs. They presume that wide-band ADC/DACs would be very expensive, and analog phase shifters could be cheaper. I am not sure but, considering the architectures of these devices, high precision analog phase shifters operating at microwave frequencies could be much more expensive due to the high complexity analog circuitry and high-frequency switching elements.

  3. But hybrid systems have practical advantages not included in this discussion.
    * With hybrid systems you can make spatial filtering before the signal hits the ADC. It could be useful, e.g. when we have strong interferers in spatial directions (or with a spatial signature that can be filtered out), thereby reducing dynamic range problems.
    * With hybrid system less amounts of digital data will need to be sent from the remote radio heads (RRH:s), The huge amounts of digital data is as yet a quite unsolved problem as I understand it from companies.

    Thus, to me it is unclear where the coin will flip: maybe the advantages with hybrid systems outweighs the disadvantages (?)

    1. I agree with the first point. But I’m not sure how much of a problem this will be. If the same ADC resolution is used as in current systems, both the fully digital and hybrid system will be less sensitive to strong interferers – but the hybrid system will benefit even more.

      Regarding the second point, a RRH with fully digital transceivers can preprocess the signals before sending it to another point. It can reduce the dimensionality to match that of a hybrid system, but using a better processing than achieved by the hybrid system.

    2. “With hybrid systems you can make spatial filtering before the signal hits the ADC”

      This assumes that all analogue components prior to the ADC is perfect, which is not the case. Phase-shifters, LNA’s and other active components can be distorted just as an ADC can. I’m not sure how much of the robustness remains once these are introduced. Also, in a hybrid system you can’t post-process these effects in the same way as in a fully digital system as you throw away degrees of freedom in the analogue domain.

      “With hybrid system less amounts of digital data will need to be sent from the remote radio heads (RRH:s)”

      I agree with Emil here. Research on compression and decentralized processing can aid in solving this problem. Some work is already done, see for example the publications of C. Studer et al.

      https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=8114173

  4. While mmWave base stations cover much smaller areas with lower service reliability and much higher power consumption as far as I understand, why does the telecom industry attempt to develop fully digital transceivers in such high frequency ranges? What is the point in mmWave that I’m missing?
    Thank you.

    1. The main point with mmWave is to utilize huge bandwidths that are available in such bands. But support for mobility and spatial multiplexing of users become more complicated with hybrid beamforming. And analog components are usually more power consuming that digital components.

  5. Dear Prof. Dr. Björnson,

    In your post, you state that

    “With digital beamforming, we can get rid of the complicated beam-searching and beam-tracking algorithms that have been developed over the past five years and achieve a simpler and more reliable system operation, particularly, using TDD operation and reciprocity-based beamforming.”

    Could you provide some references for research showing how to get rid of the complicated beam-searching and beam-tracking algorithms, please?

    Thanks,

    Felipe

    1. The books “Fundamentals of Massive MIMO” and “Massive MIMO Networks” how to operate Massive MIMO systems with digital beamforming.

        1. I have looked for how to get rid of beam-searching and beam-tracking on both books but I had no luck at all. I´d like to understand how we can get rid of beam-searching and beam-tracking with digital beamforming.

          1. Those books explain how to operate Massive MIMO systems by exploiting channel reciprocity and uplink channel estimation. That is the way to get rid of beam-searching and beam-tracking, or rather the way to never have to add those algorithms to the system. They are simply not needed.

        2. All the chapters! If you read the books from the beginning, you can stop when you feel that you understand the general concepts.

          1. I agree with Emil.
            I think it may be based on the codebook. The codebook used for MIMO, may be the way to get rid of beam-searching and beam-tracking (scaning).

  6. Dear Prof. Dr. Björnson,
    Thank you for this insightful article.
    I have decided to work on Hybrid beamforming with MVDR algorithm for my college project
    But I am finding it difficult to find suitable implementation of the MVDR algorithm
    Please do refer to any materials which might clarify the same
    Thank you
    Manasa

  7. Dear Prof. Dr. Björnson,
    1) Why can analog beamforming not create multiple beams at time?
    2) How does hybrid beamforming reduces the number of RF chain?

    1. 1) There is only one RF input signal that is transmitted from all the antennas, with different phase-shifts (time delays). One RF input = one beam.

      2) In digital beamforming there is one RF input signal per antenna, thus one can create as many beams as there are antennas. But in many situations, it is better to transmit fewer beams to limit the interference between the beams. One can then get away with a number of RF chains that equal the number of beams that you want to transmit. In between the RF inputs and the physical antennas, there will be a phase-shifting network that phase-shift and combines the signals at each antenna.

      1. Thank you Dear Prof. Dr. Björnson,
        1) How many antennas are required to create a single beam in analog beamforming? What is the mathematics behind this?
        2) In digital beamforming, one antenna generates one beam, I understand that a beam is generated by the addition of multiple beams to get strong directed beam. I mean one beam for one user and one RF chain in such case. Am I right?
        3) In hybrid beamforming, digital precoder first applies phase shifting digitally then why is it required to use analog precoder again?
        Thank you for your help

        1. 1) Two radiating elements.

          2) No, one antenna is not generating a beam. To create a beam, you send the same signal from multiple antennas. This creates constructive interference in some directions and destructive interference in other directions. By adjusting the phase-shifts of the signals that you send from the individual antennas, you can control in which directions that constructive interference appear. That is beamforming. I recommend my video: https://youtu.be/xGkyZw98Tug

          3) The digital processing allow you to apply different phase-shifts at different subcarriers and also to transmit multiple signals at the same time.

  8. Dear Prof. Dr. Björnson, why deep learning is used for hybrid precoding instead of conventional methods?

    1. Since optimal hybrid precoding is hard to obtain using conventional methods, deep learning might be used to improve on the existing methods. I am not following the research in this area.

  9. Sir,
    Recently, a number of researchers have taken to hybrid beam forming, using machine learning (which requires training datasets, as in DNN (digital neural networks)). One presenter recently claimed in an IEEE Webinar that struggling with CSI / Pilots could be ignored, as one can forget about getting channel conditions, when using an E2E digital neural network, and get as good results, as one would from traditional methods. It looked like a new step… AFIK, neural networks, require creating datasets, and cannot be generic. So DNN could be restricted to a few use cases (e.g. BS with 64 Antennas for E2E case DL+UL). Am I wrong on this assumption?

    2. Researchers from Padova had earlier this year come out with a paper on spatial channel model, and simulated a MmWave massive MIMO network, using DFT->MRT-> EVD (to get the first Eigenvalue after MRT and MRC for DL and UL antenna weights) solution. They too have some upcoming papers on Hybrid Beam forming as well…

    3. These two approaches sort of indicate that hybrid beam forming is still a region of research interest in massive MIMO.

    4. Could you please give your observations, on the likely direction the beam forming for massive MIMO would take? Will it be hybrid beam forming for the next few years or change Totally to digital beam forming with multi-panel arrays, digital phase-shifters (aka Phased array Radars)… or to some ML-based DNN …

    1. 1. Whenever you design an algorithm, you should make use of all the available prior information. The training of neural networks require prior information (e.g., the training set), just as traditional methods might require prior information (e.g., statistics of unknown variables). To make a fair comparison between different algorithms, the same prior information should be considered. For example, if one wants to carry out channel estimation without pilots, then there are (semi-)blind algorithms for that, just as there are new learning-based methods. Channel estimation without explicit pilots is nothing new, but potentially the DNN-based algorithms will outperform the best man-made algorithms.

      2, 3. There is indeed a lot of research on hybrid beamforming and there are 5G products that make use of it (I mentioned the Ericsson Street Macro in Episode 5 of the Wireless Future podcast), but I’m convinced that digital solutions will take over in the coming years. I recently wrote a blog post about digital beamforming in handsets: https://ma-mimo.ellintech.se/2020/11/14/digital-millimeter-beamforming-for-5g-terminals/

      4. Digital solutions are used in sub-6 GHz bands and there are prototypes for mmWave frequencies, so it is only a matter of time before digital solutions are available in all the 5G bands. However, even when they are available, there will be deployment scenarios where they are not needed. One can then reduce cost and power by using simpler methods, such as hybrid or analog. We will discuss this in the upcoming Episode 6 of the podcast.

      1. Thank You So Much Sir,
        Looking Forward to Episode 6…
        2. To what level will future digital beam forming for MmWaves, could result in replacing a major chunk of software code, with embedded software?
        3. Would it hugely impact information theoretic approaches to Massive MIMO studies then?

        1. 2. Different algorithms will for sure have to be utilized. There is currently a trend towards moving the baseband processing to edge clouds (C-RAN) so that the software can be easily upgraded when the hardware is upgraded. There is always a tradeoff between efficient but fixed implementations on ASICs and less efficient but flexible implementations in data centers.

          3. There is plenty of information-theoretic studies of Massive MIMO with both digital and hybrid beamforming, so it is mainly a matter of selecting which results that apply for a particular implementation.

  10. Dear Prof. Dr. Björnson,
    Thank you for this insightful article.
    In my understanding, the hybrid precoding technology is very suitable for the Massive-MIMO technology of millimeter wave communication. I think that in low dimensional antenna arrays, such as 32 and 64 antenna arrays, it is purely digital precoding technology, because pure-digital precoding can directly preprocess each antenna, with good performance, even though the overall cost of the system is higher than the hybrid precoding of the same antenna array. However, the pure digital precoding technology does not have complex algorithms. We all know that those complex algorithms not only increase the computational complexity, but also lead to the nonlinear distortion of PA and LNA in the RF link hardware. In this way, the DPD algorithm will be introduced for processing, which is more complex. However, if the antenna array dimension of the millimeter wave system reaches 256-512 or even higher, I think it is necessary to use hybrid precoding technology to reduce costs. Even with complex algorithms and systems, a large number of RF links, and large bandwidth ADC/DAC, the cost of product development will be too high for the company, and users will not be able to afford this cost. Then our research will be meaningless. I don’t know if my idea is correct. Please help me and give me some suggestions. Thank you!

    Best wishes,
    Ke Chen

    1. My impression is that the academic literature often exaggerates the difference between analog/hybrid and digital beamforming by treating an “RF chain” as a component that one either has or hasn’t. When looking at the specific implementation details, the differences are smaller. An analog/hybrid implementation has fewer ADCs/DACs than a digital implementation, but instead requires phase shifters. The number of amplifiers is the same. The digital solution can get away with fewer quantization bits per converter since the total number of quantization bits is larger. With this in mind, I think we will eventually see digital implementations that are roughly as energy- and cost-efficient as hybrid implementations. The analog/hybrid implementations might, on the other hand, evolve as well to make use of e.g., metasurfaces so things are developing there as well.

      It will be interesting to see what happens in the future. My guess is that we will see both more antenna elements and more RF inputs/outputs, but the most efficient solution will be used in the UEs.

Leave a Reply

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