The evolution of wide-area cellular networks has been a tug of war between connecting more users and providing more services on one side and managing costs on the other side. One of the most important factors on the side of users and services is the degrees of freedom of the network, i.e., the maximum simultaneous streams of information transmitted or received in the network. Cellular systems have witnessed a steady increase in the network degrees of freedom from spatial reuse, new allocation of spectrum, and the use of multiple antennas otherwise known as MIMO. The other side is cost effectiveness. To make the wide-area network economically viable, a critical hallmark of cellular architecture is the strategically centralized management of complexity. A macro cellular base station is designed to serve many users and thereby justifies the relatively high capital and operating costs. Small cells are another approach to increase the network degrees of freedom; however, their high operating expenses (e.g., backhaul, site acquisition/rent and equipment maintenance) remain an obstacle to wide market adoption.
MIMO has been a key enabling technology in the current LTE cellular systems, as it enables the possibility for simultaneous transmission of multiple streams (to single or multiple users), while most of the technological complexity can be integrated in base station and user equipment processing. More recently, the concept of Massive MIMO was introduced, in which multiple simultaneous users are to be served by a base station with a large number of antenna elements. In the original Massive MIMO architecture, the many antenna elements are not required to be coherent at radio frequency (RF). The beamforming coefficients (which adjust the RF signal magnitudes and phases at the antenna elements) are derived from channel estimation to the target users, which includes the radio chains. Typically, this approach uses channel reciprocity in time division duplex (TDD) systems.
In the original Massive MIMO architecture, a full transceiver radio chain is used behind each antenna element, such that the system has access to the individual digitized baseband signal associated with any given antenna element. We refer to such architecture as a Digital Massive MIMO system. There are two principal sources of limitations in Digital Massive MIMO: hardware complexity and processing errors. The first comes from the large number of transceiver radio chains. While in the past digital circuits have enjoyed large integration and cost reduction by way of Moore’s law, the analog components in the RF chains, e.g., amplifiers, filters, have followed a much slower evolution towards cost and energy efficiency. Furthermore, Moore’s law has slowed dramatically, and many are predicting its death (see, e.g., https://www.technologyreview.com/s/601441/moores-law-is-dead-now-what/). With no signs of analog circuit breakthroughs and a sick Moore’s Law at best, the hope of making Digital Massive MIMO cost efficient is on thin ice. In addition, Digital Massive MIMO is prone to suffer from demanding thermal management and sizable weight.
The second source of complexity in Digital Massive MIMO comes from its reliance on massive channel estimation. In order for the base station to focus the RF energy to particular users and avoid others with RF nulls, the base station needs to know the wireless channel condition from each antenna element to all users with high precision, as it needs to adjust the transmission signal at each antenna element accordingly. The traditional technique of having the mobile user measure and feed back channel information from each of the base station antenna elements becomes impractical due to large overhead, since the transmission of each measurement feedback requires dedicated system bandwidth. Instead of feedback, Digital Massive MIMO relies on reciprocal channel estimation in a TDD system. In TDD, the downlink and the uplink transmission between the base station and the mobile users take place at the same frequency but over different time slots. When a mobile user transmits on the uplink, each of the antenna elements at the base station may simultaneously estimates its respective channel from the mobile from the uplink reference signal. When uplink and downlink happen at the same frequency, their channel conditions can be taken as the same, i.e., reciprocal, due to the underlying physics of wireless signal propagation. Therefore, the one-to-many uplink simultaneous measurements become the many-to-one downlink channel estimations.
A direct consequence of relying on channel reciprocity without RF coherent radios is unpredictable channel estimation errors. Specifically, while the over-the-air uplink and downlink channels can be taken as reciprocal, the transmit and receive radio paths are not reciprocal in general. The transmit power amplifier phase and magnitude characteristics typically exhibit a strong dependency on temperature, which in turn depends on the downlink traffic level. It is challenging to match the phase and magnitude characteristics in the transmit and receive radio paths over a wide range of operating conditions to allow for accurate end-to-end channel estimation. Though not an inherent technical limitation, another practical corollary of reciprocity-based channel estimation is the difficulty in applying this technique in frequency division duplex (FDD) systems, where uplink and downlink transmission take place at different frequencies and channel reciprocity does not hold. While many new deployments are in TDD bands, over 85% of the currently in-use cellular spectrum is FDD. Furthermore, the same limitation applies to systems with channel aggregation in downlink without the corresponding uplink.
In contrast to embedding the phase and magnitude characteristics associated with the different antenna elements as part of the channel estimation process, we consider an architecture where the antenna elements are RF-coherent. In the context of Massive MIMO systems, RF coherency refers to all transmit antenna elements having matching phase and magnitude characteristics at the operating RF frequency in the downlink, and likewise matching received paths in the uplink (though the downlink and uplink paths are not necessarily matched). When the antenna elements are RF-coherent, they may be placed close together, typically fractions of wavelengths apart, to form a phased array.
Phased arrays have been used extensively in radar systems, e.g., in detecting angles-of-arrival of signals of interest. Traditionally, RF coherency is achieved by employing a distribution network known as corporate feed, which comprises matching transmission paths with stable RF property over temperature and time. Due to the tight tolerances required, corporate feeds are typically costly and it is often necessary to perform regular offline calibrations. As a result, while phased arrays are common in, e.g., aviation and military systems, they are deemed cost-prohibitive for commercial telecommunication networks, especially since a corporate feed would be needed for each generated beam.
Recently, we at Blue Danube introduced techniques to achieve RF coherency using a custom RF integrated circuit (RFIC) and a new array architecture. Effectively, self-calibrating mechanisms are implemented using standard RFIC technology. In our approach, high-cost corporate feeds are no longer necessary, and this enables us to produce a cost-effective RF-coherent phased array. In the next series of articles, we consider the implications of RF coherency in Massive MIMO systems, and how a RF-coherent Massive MIMO antenna array effectively takes advantage of its spatial degrees of freedom while avoiding undue system complexity.