By Ana García Armada, PhD, Professor at Universidad Carlos III de Madrid
While driving on a highway in Europe (as a passenger), I tried my smartphone’s 4G-LTE connection and the best I could get was 30 Mbps downlink, 10 Mbps uplink, with latency around 50 msec. This is not bad for many of the applications we use today, but it is clearly insufficient for many low latency/low jitter mobile applications, such as autonomous driving or high-quality video while on the move.
At higher speeds, passengers of ultra-fast trains may enjoy the travel while working. Their 4G-LTE connections are often good enough to read or send emails and browse the internet. But would a train passenger be able to have a video conference call with good quality? Would we ever be able to experience virtual reality or augmented reality in such a high mobility environment?
How to achieve intelligent transport systems enabling vehicles to communicate with each other has been the subject of several papers and reports as per Reference . Many telecommunications professionals are looking to 5G for a solution, but it is not at all certain that the IMT 2020 performance requirements specified in ITU-R M.2410 for low latency with high speed mobility will be met anytime soon (by either 3GPP Release 16 or IMT 2020 compliant specifications).
Editor’s Note: In ITU-R M.2410, the minimum requirements for IMT 2020 (“5G”) user plane latency are: 4 ms for eMBB (enhanced mobile broadband) and 1 ms for URLLC (ultra high reliability, ultra low latency communications).
IMT 2020 is expected to be approved by ITU-R SG D after their November 23-24,2020 meeting, which is one week after the ITU-R WP 5D approval at their November 17-19, 2020 meeting.
There are three different “5G Radios” being progressed as IMT 2020 RIT/SRIT submissions: 3GPP, DECT/ETSI, and Nufront. The TSDSI’s (India) submission adds Low Mobility Large Cell (LMLC) to 3GPP’s “5G NR.”
The fundamental reason why we do not experience high data rates using 4G-LTE lies in the signal format. That did not change much with 3GPP’s “5G NR,” which is the leading candidate IMT 2020 Radio Interface Technology (RIT). Please refer to Editor’s Note above.
In coherent detection, a local carrier mixes with the received radio frequency (RF) signal to generate a product term. As a result, the received RF signal can be frequency translated and demodulated. When using coherent detection, we need to estimate the channel (frequency band). The amount of overhead strongly depends on the channel variations. That is, the faster we are moving, the higher the overhead. Therefore, the only way to obtain higher data rates in these circumstances is to increase the allocated bandwidth (e.g. with carrier aggregation ) for a particular connection, which is obviously a non-scalable solution.
Coherent Communications, CSI, and OFDM Explained:
A coherent receiver creates a replica of the transmitted carrier, as perfectly synchronized (using the same frequency and the same phase) as possible. Combining coherent detection with the received signal, the baseband data is recovered with additive noise being the only impairment.
However, the propagation channel usually introduces some additional negative effects that distorts the amplitude and phase of the received signal (when compared to the transmitted signal). Hence, the need to estimate the channel characteristics and remove the total distortion. In wireless communications, channel state information (CSI) refers to known channel properties of a communication link, i.e. the channel characteristics. CSI needs to be estimated at the receiver and is usually quantized and sent back to the transmitter.
Orthogonal frequency-division multiplexing (OFDM) is a method of digital signal modulation in which a single data stream is split across several separate narrowband channels at different frequencies to reduce interference and crosstalk. Modern communications systems using OFDM carefully design reference signals to be able to estimate the CSI as accurately as possible. That requires pilot signals in the composite Physical layer frame (in addition to the digital information being transmitted) in order to estimate the CSI. The frequency of those reference signals and the corresponding amount of overhead depends on the characteristics of the channel that we would like to estimate from some (hopefully) reduced number of samples.
Wireless communications were not always based on coherent detection. At the time of the initial amplitude modulation (AM) and frequency modulation (FM), the receivers obtained an estimate of the transmitted data by detecting the amplitude or frequency variations of the received signal without creating a local replica of the carrier. But their performance was very limited. Indeed, coherent receivers were a break-through to achieve high quality communications.
Other Methods of Signal Detection:
More recently, there are two popular ways of non-coherently detecting the transmitted data correctly at the receiver.
One way is to perform energy or frequency detection in a similar way to the initial AM and FM receivers.
In differential encoding, we encode the information in the phase shifts (or phase differences) of the transmitted carrier. Then, the absolute phase is not important, but just its transitions from one symbol to the other. The differential receivers are much simpler than the coherent ones, but their performance is worse since noise is increased in the detection process.
Communications systems that prioritize simple and inexpensive receivers, such as Bluetooth , use non-coherent receivers. Also, differential encoding is an added feature in some standards, such as Digital Audio Broadcasting (DAB). The latter was one of the first, if not the first standard, to use OFDM in wireless communications. It increases the robustness to mitigate phase distortions, caused by the propagation channel for mobile, portable or fixed receivers.
However, the vast majority of contemporary wireless communications systems use coherent detection. That is true for 4G-LTE and “5G NR.”
Combining non-coherent communications with massive MIMO:
Massive MIMO (multiple-input, multiple-output) groups together antennas at the transmitter and receiver to provide better throughput and better spectrum efficiency. When massive MIMO is used, obtaining and sharing CSI threatened to become a bottleneck, because of the large number of channels that need to be estimated because there are a very large number of antennas.
A Universidad Carlos III de Madrid research group started looking at a combination of massive MIMO with non-coherent receivers as a possible solution for good quality (user experience) high speed mobile communications. It is an interesting combination. The improvement of performance brought by the excess of antennas may counteract the fundamental performance loss of non-coherent schemes (usually a 3 dB signal-to-noise ratio loss).
Indeed, our research showed that if we take into account the overhead caused by CSI estimation in coherent schemes, we have shown several cases in which non-coherent massive MIMO performs better than its coherent counterpart. There are even cases where coherent schemes do not work at all, at least with the overheads considered by 4G-LTE and 5G (IMT 2020) standards. Yet non-coherent detection usually works well under those conditions. These latter cases are most prevalent in high-mobility environments.
Editor’s Note: In ITU-R M.2410, high speed vehicular communications (120 km/hr to 500 km/hr) is mainly envisioned for high speed trains. No “dead zones” are permitted as the “minimum” mobility interruption time is 0 ms!
When to use non-coherent massive MIMO?
Clearly in those situations where coherent schemes work well with a reasonable pilot signal overhead, we do not need to search for alternatives. However, there are other scenarios of interest where non-coherent schemes may substitute or complement the coherent ones. These are cases when the propagation channel is very frequency selective and/or very time-varying. In these situations, estimating the CSI is very costly in terms of resources that need to be used as pilots for the estimation. Alternatives that do not require channel estimation are often more efficient.
An interesting combination of non-coherent and coherent data streams is presented in reference , where the non-coherent stream is used at the same time to transmit data and to estimate the CSI for the coherent stream. This is an example of how coherent and non-coherent approaches are complementary and the best combination can be chosen depending on the scenario. Such a hybrid scheme is depicted in the figure below.
Figure 1. Suitability of coherent (C), non-coherent (NC) and hybrid schemes (from reference )
What about Millimeter Waves and Beam Steering?
The advantage of millimeter waves (very high frequencies) is the spectrum availability and high speeds. The disadvantages are short distances and line of sight communications required.
Compensating for the overhead by adding more bandwidth, may be a viable solution. However, the high propagation loss that characterizes these millimeter wave high frequency bands creates the need for highly directive antennas. Such antennas would need to create narrow beams and then steer them towards the user’s position. This is easy when the user equipment is fixed or slowly moving, but doing it in a high speed environment is a real challenge.
Note that the beam searching and tracking systems that are proposed in today’s wireless communications standards, won’t work in high speed mobile communications when the User Endpoint (UE) has moved to the coverage of another base station at the time the steering beams are aligned! There is certainly a lot of research to be done here.
In summary, the combination of non-coherent techniques with massive MIMO does not present any additional problems when they are carried out in millimeter wave frequencies. For example, reference  shows how a non-coherent scheme can be combined with beamforming, provided the beamforming is performed by a beam tracking procedure. However, the problem of how to achieve fast beam alignment remains to be solved.
Non-coherent massive MIMO makes sense in wireless communications systems that need to have very low complexity or that need to work in scenarios with high mobility. Its advantage is that it makes possible communications in places or circumstances where the classical coherent communications fail. However, this scheme will not perform as well as coherent schemes under normal conditions.
Most probably, non-coherent massive MIMO will be used in the future as a complement to well-understood and (usually) well-performing coherent systems. This will happen when there are clear market opportunities for high mobility, high speed, low latency use cases and applications.
 ITU report: “Setting the scene for 5G: opportunities and challenges”, 2018, https://www.itu.int/en/ITU-D/Documents/ITU_5G_REPORT-2018.pdf
 F. Kaltenberger et al., “Broadband wireless channel measurements for high speed trains,” 2015 IEEE International Conference on Communications (ICC), London, 2015, pp. 2620-2625, doi: 10.1109/ICC.2015.7248720.
 L. Lampe, R. Schober and M. Jain, “Noncoherent sequence detection receiver for Bluetooth systems,” in IEEE Journal on Selected Areas in Communications, vol. 23, no. 9, pp. 1718-1727, Sept. 2005, doi: 10.1109/JSAC.2005.853791.
 ETSI ETS 300 401, “Radio broadcasting systems; DAB to mobile, portable and fixed receivers,” 1997.
 M Lopez-Morales, K Chen Hu, A Garcia Armada, “Differential Data-aided Channel Estimation for Up-link Massive SIMO-OFDM”, IEEE Open Journal of the Communications Society -> in press.
 K Chen Hu, L Yong, A Garcia Armada, “Non-Coherent Massive MIMO-OFDM Down-Link based on Differential Modulation”, IEEE Trans. on Vehicular Technology -> in press.
About Ana García Armada, PhD: