Cracking the Pilot Contamination Nut

When T. Marzetta introduced the Massive MIMO concept in his seminal article from 2010, he concluded that “the phenomenon of pilot contamination impose[s] fundamental limitations on what can be achieved with a noncooperative cellular multiuser MIMO system.”

More precisely, he showed that the channel capacity under i.i.d. Rayleigh fading converges to a finite limit as the number of base stations goes to infinity.  The value of this limit is determined by the interference level in the channel estimation phase. There are hundreds of papers on IEEEXplore that deals with the pilot contamination issue, trying to push the limit upwards or achieve higher performance for a given number of antennas. Various advanced mitigation methods have been developed to cure the symptoms of pilot contamination.

But was pilot contamination really a fundamental limitation to start with? In 2018, we published a paper called “Massive MIMO Has Unlimited Capacity” where we showed that there is an unexpectedly simple solution to the problem. You don’t need a sledgehammer to “crack the pilot contamination nut“, but the right combination of state-of-the-art tools will do. While I have written about this in previous blog posts and briefly mentioned it in videos, I have finally recorded a comprehensive lecture on the topic. It is 82 minutes long and was given online by invitation from Hacettepe University, Turkey. No previous knowledge on the topic is required. I hope you will enjoy it in small or big doses!

7 thoughts on “Cracking the Pilot Contamination Nut”

  1. Dear Sir,
    Sorry for disturbing you.
    I wish you could provide me with some details concerning the following question:
    In an FDD-based OFDM-MIMO system, channel estimation requires the insertion of what is called spectral nulls within RBs i.e when a reference signal is transmitted on any antenna at any given subcarrier, all other antennas must transmit nothing (spectral null) at the same subcarrier.
    Taking into consideration the overhead introduced by such arrangement, could you please Sir give me some insights on how these spectral nulls are optimally inserted in a massive MIMO-OFDM system.

    Thank you so much Sir.

    1. When transmitting a reference signal, you can map it to the antennas using any precoding vector. The receiver can then estimate the resulting channel (combination of precoder and physical channel). Normally, the references signals are transmitted from all antennas but are repeated using different precoding vectors, to beamform the signals in different physical directions. If you have 64 antennas, you will have to repeat the signal 64 times to beam form it in “all” directions. This is why channel estimation in FDD-based systems is rather inefficient as compared to TDD systems, where a single uplink reference signal is enough. This is explained in Section 1 of my book Massive MIMO Networks, https://massivemimobook.com.

  2. Hello Dr. Bjornson,
    Thanks very much for your post. I was just wondering if you have any ideas how the pilot contamination problem is dealt with in the current real massive MIMO products, e.g., Ericsson AIR 6468, Huawei AAU, Nokia Air Scale? Thanks.

    1. I believe the reference signals are allocated between cells to avoid pilot contamination to the extent possible. So far, the traffic load (number of simultaneously active user per resource block per cell) is so low that pilot contamination shouldn’t be a major issue.

Leave a Reply

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