It has long been the trend that modern wireless networks have been getting denser as time goes on; where, by the term dense, we refer to the gap between network nodes (the ‘points of presence’ of the network) and the average area intended to be covered by the single node. Till now, this has primarily been driven by the increasing demands for bandwidth for users, exponentially increasing numbers of users and the deployment of networks in higher and higher frequencies. The last point is important because scattering increases (and hence, the effective radius decreases) as the network node uses a higher frequency.
This trend was predicted well in advance by (Gupta, et al., 2000), (Gastpar, et al., 2002)and others; that the capacity of the different types of wireless networks (cellular, relay, local area, mesh, etc.) network increases O(N^(1/2)); where N is the number of network nodes or points of presence in a given area. In recent times, one of the driving factors is also the miniaturization of digital and RF hardware (which allows lower footprint deployments) and the development of new types of architectures, such as the centralized ‘cloud RAN’ for cellular networks.
It is common knowledge in the world that Wireless Networks(cellular, local area or combined cellular and local access) are interference limited and, as the network shrinks, co-channel interference shall become a tighter and tighter constraint. Another constraint that is rapidly developing, on the other hand, is the total overall power consumption of the network. Whereas, in earlier networks, power constraints were only due to the need to manage interference and similar issues, now it has become a limitation in its own right.
On the other hand, technology has developed rapidly and the deployment of distributed antenna systems (DAS), multiple-antenna systems (MIMO, MISO, etc.) is widespread. Fast synchronous communications between network nodes make real-time coordination possible, which has given rise to the new technique of Coordinated Multipoint (Comp). All of these are powerful weapons to help carve the network out into finer and finer slices, so as to make space for an ever-increasing number of network users.
We now come to the specific problem description.
Our problem is equally valid for both cellular and WLAN networks, both of which are rapidly evolving in terms of MIMO capabilities and the need for throughput. In the following discussion, we use network nodes to refer to both eNodeB/DAS transceivers for cellular networks and Access Points or equivalent transceivers for WLAN networks. The ‘main’ node refers to the macro-cell or main AP. The auxiliary node refers to the hotspot Access Points or the small-cells. The term UE is used for cell-phone user equipment and STA for WLAN user equipment. However, it is well known that with the advent of smart-phones, it is quite a common feature to find the same device acting as both.
In Figure 1, we see a next generation cellular network, comprising of a macro-cell and two small-cells covering a given area. The macro-cell transmits to two User Equipment (UE) at a relatively larger distance, whereas the small-cells cater, at a lower power, to relatively close-by UEs. The specific problem to be solved is macro to small-cell interference on the downlink, which is a well-known problem. We assume that the macro is a high power transmitter with multiple transmitting elements; the UEs and the small-cells are dual-stream MIMO capable
Figure 1 5G network with Active Interference Mgmt. CRAN
In the example in Figure 2, we have an in-office managed WLAN network, comprising of 802.11ac Access Points. The Access Points are divided into low-power hot-spots, intended to provide high speed access for specific areas, as well as one or more centralized AP, whose purpose is to provide coverage to the entire floor. The Access Points share the same frequency and are connected via a central server, which handles issues of authentication (Passpoint based), discovery and admission control (ANQP/GAS based); to that, we add the function of interference management using the CIMSS.
Figure 2 Indoor WLAN enterprise network with Interference Mitigation
As can be seen in the descriptions in the previous section, we have introduced a new node, the CIMSS. The function of the CIMSS is to coordinate cross-network-node interference measurement and mitigation. In other words, it tracks the feedback reports (forwarded by the network nodes) between the different UEs in the area and the corresponding network nodes that they can see, along with the channel state information. As we shall see below, advanced forms of Interference Management require a fair amount of information to be processed in relatively short term, and combined scheduling decisions for multiple network nodes. Hence, a centralized processing point for a given cluster of nodes is more convenient. The CIMSS will play a role in other functions such as Cellular-Wi-Fi offload, WLAN network node discovery, etc.
In the rest of this article, we shall consider the different modes of interference management available to CIMSS and the consequent challenges of scheduling and power control.
We, in this paper, are primarily focusing on co-channel interference; indeed, as network reuse distances are shrinking and the selectivity and other characteristics of solid-state RF chips improve, CCI is the dominant factor in interference at the cell edge. In this section, we consider, at a very high level, the top most methods for CCI reduction in modern networks. We classify them into ‘passive’ and ‘active’ modes. The distinction shall be explained below.
Power scaling simply refers to the setting of power levels for each network node, based on the SNR required by the UEs connected to it, as well as the interference that it is generating on nearby UEs connected to other network nodes
The simplest approach (which we refer to as D-SON) is to allow appropriate power setting by individual network nodes, based on measured interference and bandwidth requirements. The crucial difference is the feedback mechanism, which has to be built into the system; especially for an asymmetric radio interface (FDD). While a simple improvement on static power setting, it is a challenging problem in global optimization; however, even a sub-optimal solution can bring enormous improvements to network performance. In recent times, we have seen attempts to solve the power scaling problem using the mean-field approach, which can be tailored to multiple network criteria. Distributed power control using local gradient search methods (Saha, 2009)as well as the mean-field method (Nourian, 2013) have been demonstrated by HSC in the laboratory.
A modern variation of this is ICIC and its successors (eICIC and feICIC) (Chrysovalantis Kosta, 2013). In these methods, network nodes coordinate with each other, so that individual high-power/macro nodes switch off /reduce power in specific time-slots and frequencies, allowing close by low-power nodes an opportunity to transmit. This is more selective than the network wide power control method and can significantly reduce interference at the cost of some idling of network resources.
While the above mechanisms work in all cases, more sophisticated methods can be used in multiple-antenna cases, by adjusting the precoding methodology. Precoding is an existing feature of MISO and MIMO transmission, where the transmit data is pre-multiplied by a precoding matrix and subsequently, after being received at the receiver is suitably post-multiplied as part of the receive process. By adjusting the pre-coding (and corresponding post-coding), the capacity of the multiple-antenna channel can be maximized.
Passive modes of interference management are those which do not consume additional network resources. The original method was to simply reduce transmit powers network wide, to a point where the interference was tolerable; obviously, this also leads to reduced capacity over the network.
The precoding in pure MIMO transmission only considers the transmit channel between the transmitting network node and the UE. However, it can be modified to take into account the presence of other UEs nearby and adjusting accordingly.
In orthogonal coding or zero-forcing coding, we have a transmitter, with its connected “target” UEs and at least one “co-resident” UE, which is nearby, but is receiving transmissions from some other network node. The transmitter adjusts the precoding matrix so as to ensure that the interference at the co-resident UE is zero; it does this by forcing the precoding matrix to be in the null-space of the matrix representing the channel between itself and the co-resident UE. The method can be extended to multiple co-resident UEs, if the number of antennae available in the transmitter are large enough. The method works well only if the co-resident and target UEs have orthogonal channel matrices. In a practical realization, the transmitter has multiple UEs available for transmission and chooses the one whose channel matrix for that transmission opportunity is the most orthogonal to that of the co-resident UEs.
Interference Alignment is a new concept which attempts to massage the transmission so that all causal interference is mapped to a ‘wastebasket’; a sub-space of the entire signal space, which can be ignored by the receiver. A good introduction is provided in (S.Jafar, 2008), with a MIMO extension being given in (Ignacio Santamaria, 2010). Interference Alignment has been shown to be capable of providing half the bandwidth for each user in a K-user environment, but this has to be extended over multiple timeslots/resource opportunities, in order to find the appropriate vector mapping.
In contrast to Passive methods, we have Active Interference Management, where network nodes actually use additional power and possibly other networks to actively battle interference. These methods have been made possible due to research in modern MIMO systems and precoding techniques.
Interference pre-subtraction is a technique commonly used in single transmitter, multiple receiver multi-user MIMO scenarios (MU-MIMO) and is based on Costas seminal work on Dirty Paper Coding. In single user MIMO, the transmission is coded in such a way that the signal for the nth receiver is contaminated only by the signals from the 1st through (n-1) th signals; the first receiver, hence, gets an interference free signal. There are many techniques for interference pre-subtraction; for example, the block diagonalization approach by (Caire, 2003), (Yu, 2004). The residual interference in the nth signal can then be further modified using some version of dirty paper coding; ranging from the simple Tomlinson Harashima Precoding method to more sophisticated solutions like Lattice coding.
Active Null Forming was proposed by us as a modification to the Interference pre-subtraction approach described above. Rather than use additional power to transmit to two receivers, it creates a ‘nulling signaling’ to the second receiver, so as to minimize the energy of the signal that the second receiver actually receives. We call this a ‘nulling’ signal. In (Saha, 2016) we proposed an algorithm to construct the second signal subject to constraints in terms of the maximum additional power that could be used for this purpose.
Curiously, whereas at one time reducing interference globally meant reducing power globally, with the new interference management techniques available to us, this is not true. We can indeed “spend” a little extra in power, in order to achieve reduced interference. We can illustrate it by a simple strawman case below, using a canonical example of a two node scenario, which is the template for all the discussion on interference management. For the problem described in 1.2.1, we can break up the network so as to have two of these for the macro-cell and each of the small-cells. A similar equivalence exists for the indoor WLAN scenario in 1.2.2
Figure 3 Two Node Reduction
In the Figure 3, we have two network nodes and two UEs/STAs, with a common shared channel i.e. single frequency and bandwidth. Each network node has Nt antennae and each UE has Nrantennae for reception, Nr < Nt. Without loss of generality, we have that the ith network node is transmitting to the ith UE. The target SINR for each UE is given by si, i=1,2. The channel matrix between the ith transmitter and the jth receiver is given by . If each network node is transmitting with the transmit power Pi and there is a general zero-mean noise term of variance n, the SINR at the ith node is given by
Let us assume a situation where, at the current value of , UE 2 is not meeting their SINR target i.e Pi
We have the following courses of action available to us.
We now consider the task of the CIMSS. It has to determine the optimal operating configuration on a TTI by TTI basis so as to minimize interference and maximize network capacity, subject to the equally important task of limiting power consumption.
Transmit scheduling function of the CIMSS requires it to make the following choices
We can further simplify the problem by making the assumption of primary node autonomy. In other words, the primary node autonomously determines the UEs it needs to transmit to in the current transmission opportunity and announces this decision to the CIMSS as a fait accompli. It also signals the corresponding CSI to the CIMSS. Hence the task of the CIMSS is to determine the corresponding UEs in the auxiliary nodes which are to be simultaneously targeted.
We further allow the primary node to choose the number of UEs to transmit to and the corresponding precoding vectors. If the primary node chooses to transmit to more than one UE, we assume that it will use a version of broadcast transmission with interference pre-subtraction, so as to achieve optimal SINR for its targeted UEs. The baseline power required for this is provided by the node as well.
In our formulation, the CIMSS receives CSI information from multiple UEs; the UEs connected to the primary node as well as the UEs connected to the small-cell and has to determine the optimal set of UEs to be transmitted to in the next TTI. To achieve this, we need the following determinations
We can develop the outlines of a simple CIMSS scheduling algorithm as follows:
Making the tradeoff between ANF and eICIC and related power reduction methods is discussed in detail in (Saha, 2015).
As is obvious from the above, the ability to receive feedback from multiple UEs regarding multiple nearby network nodes is the crucial element. While MU-MIMO has recently entered the standards, it is mostly focused on a single network-node to multiple UE scenario.
The primary problem is for a UE/STA to be able to measure the channel matrix from multiple network nodes; in order for this to happen, the network nodes must transmit pilots or schedule sounding channels which are readable separately; either by separating them in time, or by using some orthogonal tone arrangement. A simple arrangement (which at least works in our case) is to reserve one of the blocks for the primary. The rest of the blocks are reserved for the ‘local’ auxiliary network node.
An alternate arrangement is to simply combine the central network node and the local network node as one ‘virtual’ network node of double the number of transmitting elements, for the purpose of transmitting the pilot/sounding PDU/training field. The receiver would measure one single combined channel matrix, which would be relayed via its attached network node to the CIMSS, which would be able to separate it out into the contributions from each channel. This requires coordination between the network nodes in terms of the selected pilot, but otherwise reduces the complexity of the system.
In this article, we have presented a new network architecture for cooperative interference mitigation and capacity enhancement for dense networks. We have also shown that there are multiple methods available today, which can be used within this architecture to improve the performance of dense coordinated networks. These methods are suitable for different forms of current and upcoming networks; small-cell networks, LTE-WLAN het-nets, etc. which are increasingly suffering from the problem of network density. Rather than focusing purely on older techniques of power scaling, we use newly developed methods for optimal broadcast MIMO and related technologies to intelligently reduce interference within a single-frequency network. We demonstrate how a very common use-case of small-cell coexistence can take advantage of this service. We also demonstrate the necessary feedback mechanisms to make this technology possible.
3GPP R1-105622 // Discussion of TD eICIC schemes.
3GPP R1-114424 // On the need for signaling enhancements for feICIC.
Caire Giuseppe, and Shlomo Shamai On the achievable throughput of a multiantenna Gaussian broadcast channel [Journal] // Information Theory, IEEE Transactions on. - 2003. - 7 : Vol. 49. - pp. 1691-1706.
Chrysovalantis Kosta Bernard Hunt, Atta UI Quddus, Rahim Tafazolli On Interference Avoidance Through Inter-Cell Interference Coordination (ICIC) Based on OFDMA Mobile Systems [Conference]. - [s.l.] : IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 15, NO. 3, 2013.
David L´opez-P´erez ˙Ismail G¨uvenc, Guillaume de la Roche, Marios Kountouris, Tony Q.S. Quek, Jie Zhang, Enhanced Inter-Cell Interference Coordination Challenges in Heterogeneous Networks [Conference].
Gastpar M. and Vetterli M On the capacity of wireless networks: the relay case [Conference]. - [s.l.] : IEEE, 2002. - 0743-166X.
Gupta P. and Kumar P.R The capacity of wireless networks [Journal] // Information Theory, IEEE Transactions on. - 2000. - pp. 388-404.
Ignacio Santamaria Oscar Gonzalez, Robert W. Heath Jr., Steven W. Peters Maximum Sum-Rate Interference Alignment Algorithms for MIMO Channels [Conference] // GlobeCOM. - Miami, FL : [s.n.], 2010. - pp. 1--6.
Nourian Mojtaba and Caines, Peter E and Malhame, Roland P and Huang, Minyi Nash, social and centralized solutions to consensus problems via mean field control theory [Journal] // Automatic Control, IEEE Transactions On. - [s.l.] : IEEE, 2013. - 3 : Vol. 58. - pp. 639--653.
S.Jafar V. Cadambe Interference alignment and degrees of freedom of the k-user interference channel [Journal] // Information Theory, IEEE Transactions on. - [s.l.] : IEEE, 2008. - 8 : Vol. 54. - pp. 3425--3451.
Saha Abheek A Distributed Power Management Algorithm for a Self-optimizing WiFi Network. [Conference] // 9th Advanced International Conference on Telecommunications. - [s.l.] : AICT, 2009. - pp. 23-28.
Saha Abheek A Novel Technique for Cross-Cell Interference Mitigation using Joint Encoding [Online] // www.hsc.com. - October 2015.
Saha Abheek Active Null Forming: A new technique for interference mitigation in multi-user MIMO channels [Conference] // AICT, Proceedings of. - Valencia : [s.n.], 2016.
Yu Wei, and John M. Cioffi. Sum capacity of Gaussian vector broadcast channels [Journal] // Information Theory, IEEE Transactions on. - [s.l.] : IEEE, 2004. - 9 : Vol. 50. - pp. 1875--1892.