On the Benefits of Multi-hop Communication for Indoor 60 GHz Wireless Networks

Due to the emerging applications and the need for higher capacity, current sub-6 GHz wireless spectrum is not enough to cope with the recent demands for high data rate applications ranging from augmented/virtual reality to wireless HD video streaming. Millimeter wave (mmWave) bands – between 30 GHz to 300 GHz – have been contemplated as a solution to mitigate the existing spectrum scarcity in the sub-6 GHz. However, enabling mmWave wireless systems in general requires properly dealing with the channel impairments and propagation characteristics of the high frequency bands. In particular, due to the high propagation losses, mmWave provides a limited range of communications. Also, high penetration, reflection and diffraction losses reduce the available diversity and limit non-line-of-sight (NLOS) communications. Due to the small wavelength of mmWave, most objects in the propagation environment lead to blocking and reflection as opposed to scattering and diffraction as in the sub-6 GHz bands.

To experimentally investigate the effect of human blockage on mmWave links, we conducted a set of measurements in the 30 GHz band with a stationary transmitter and a mobile receiver that moves away from the transmitter with the speed of 1 m/s [1]. From the results shown in Fig. 1, we observe that during human blockage intervals (denoted by “HB”), the received signal strength falls to almost zero. This is consistent with the measurement results in [2] that suggest human body can increase the path loss by more than 20 dB.

Currently, there is arguably an adequate understanding of physical layer issues and signal characterization [3, 4] as well as MAC layer designs for mmWave systems [5]. By contrast, the upper layers of the protocol stack are still largely unexplored and the existing protocols are not tailored for the mmWave bands. Multi-hop networking can be a viable solution to mitigate the issue of limited range of communications as well as the sensitivity to blockage. Notwithstanding the benefits, multi-hop communication adds additional overhead by requiring exchange of metadata (e.g., route discovery messages), which is more challenging due to directional communication in mmWave bands. In addition, it is not clear that multi-hop networking satisfies the latency requirements due to extra processing, queuing, and channel access latency at relay nodes. For instance, the IEEE 802.11ay use-case document specifies the latency and jitter requirements of less than 5ms for 8K UHD wireless transfer at smart homes as well as for augmented reality/virtual reality headsets and other high-end wearable devices [6].

In this paper, we investigate the performance of multi-hop mmWave communication for indoor 60 GHz applications and demonstrate the benefits of multi-hop routing protocols for blockage-prone and range-limited mmWave links. Due to the nature of use-cases for indoor mmWave systems (e.g., real-time high data rate applications), a blocked link should quickly be detected and replaced by an alternative link. As such, we propose a hop-by-hop multi-path routing protocol that is efficient and fast in switching to a reserved ready-to-use path towards the destination. We implement the proposed protocol on top of the 802.11ad PHY and MAC specification, and investigate its performance under different scenarios.

There are two main categories of related works: (i) On-demand routing protocols that include the basic Ad-hoc On-Demand Distance Vector (AODV), multi-path AODV, and directional multi-hop protocols, and (ii) methods to combat blockage in mmWave bands.

We consider a mmWave network that consists of several STAs such that the intermediate STAs are able to relay data traffic from the originating STA to the destination STA (depending on the connectivity and links configurations between STAs). Each directional link is associated with a cost metric that captures the quality of that link. Cost of a route is defined as the summation over the cost of individual constituent links of the route. The goal is to find one primary and one backup link from the originating STA to neighbor STAs to achieve a hop-by-hop multi-path routing. In hop-by-hop multi-path routing, the routing table at each STA contains two next-hops per destination (not for the entire route). If the link to the primary next-hop is blocked, the STA switches to its backup next-hop. In [25], hop-by-hop routing is proposed for the Internet data packets that can be split between multiple next hop switches. This in turn improves the efficiency and robustness. In this paper, hop-by-hop routing is considered for mmWave systems in order to mitigate blockage issues. In traditional multi-path AODV protocols (e.g., [8, 10]), when the primary path from the source to destination is unavailable, the source node switches to the next alternate end-to-end path. In contrast, the hop-by-hop multi-path routing adds a local repair capability to each node (source and intermediate) by switching to an alternative link for the blocked link, without the need for an entire route change by the source. This may result in a transient non-optimal end-to-end route, which can be fixed in the next round of routing tables reset.

Route Discovery Messages: In order to disseminate the routing control messages across the network, either omni-directional transmissions or directional signals can be used. In the former, due to the short-range nature of mmWave signals, omni-directional signaling may not reach to the neighbor nodes. In the latter, directional transmissions incur high signaling overhead to disseminate control messages since the same message needs to be transmitted across different sectors to cover all the neighbor nodes. In this paper, we assume that the STAs have performed the sector sweep (SSW) procedure with the 1-hop neighbors. In order to establish a multi-hop route, the originating STA sends directional RREQ frames to its neighbor STAs. Each 1-hop neighbor receives the RREQ frame and updates its reverse route to the originating STA. Each neighbor STA then forwards the RREQ to its 1-hop neighbor as well, excluding the transmitter STA from which the RREQ was received. As the forwarding continues, intermediate STAs may receive duplicate RREQ from other STAs. For example, consider a network with three STAs as shown in Fig. 2. In this example, STA B receives two RREQ frames: one from the originating STA S and one through STA A.

By receiving RREQ messages, the best RREQ and second best RREQ frames (in terms of cost) determine the next-hop and backup next-hop node to the originating STA in the routing table of the intermediate STAs. In the example above, STA B sets STA A as the backup next-hop to reach to STA S, assuming that the link metric from S to B is better than sum of the link metrics S to A and A to B. In order to reduce the routing overhead, the intermediate STA picks the best received RREQ, and forwards it to its neighbor STAs. To avoid looping, the STA records the forwarding action in its Forwarding Table.

Destination STA receives potentially several RREQ messages, and sends an RREP frame to the same STA from which an RREQ was received. Each intermediate STA that receives an RREP message, updates its routing table to the destination STA. If the intermediate STA receives more than one RREP, it picks the best RREP frame and forwards it to its 1-hop neighbor STAs. Intermediate STA records the forwarding operation in its Forwarding Table. Similar to RREQ, each RREP frame and its duplicate versions determine the next-hop and backup next-hop. The process of forwarding RREP continues until the RREP message is received at the originating STA. Originating STA potentially receives more than one RREP message. It picks the best and second best RREP message and records them as the next-hop and backup next-hop to reach to the destination STA. Each STA proactively makes sure its routing table entries are up-to-date and two next-hop options are reachable.

To demonstrate this process, let us consider the network topology depicted in Fig. 3(a) where STA S is the source, STA D is the destination, and STAs A and B serve as relay nodes. Weight of each edge determines the cost of that directional link. In this case, the routing table at STA B has the primary next-hop of D to reach STA D (with the cost of 3), while STA A is listed as the backup next-hop for STA B to reach STA D with the cost of 5. The routing table at STA B is shown in Table I. Now, assume that the direct link from B to D gets blocked as shown in Fig. 3(b). In this case, based on the routing table, STA B switches to its backup next-hop (STA A) that is ready to be deployed. The updated network topology is shown in Fig. 3(c). After the local repair step at STA B, it transmits an error message to its neighbors so that they can correct their routes to converge to a global optimal route. Temporarily switching to a backup link at STA B is faster than multi-path AODV that would typically require switching to the alternative end-to-end route S$\rightarrow$A$\rightarrow$D [10].

Route Discovery Triggers: We use three mechanisms to detect that a neighbor STA is no longer available. First, similar to the AODV protocol, we use HELLO messages such that if a neighbor STA misses HELLO transmissions, that neighbor STA is assumed to be not available. Second, if the number of transmissions for a frame exceeds a threshold, the link is assumed to be broken. This threshold is set to 7 for short frames and 4 for long frames. Third, we implement a trigger based on the modulation and coding scheme (MCS) index such that if the MCS index drops below a specified value¹¹1We assume that an adaptive rate controller such as ARF is deployed., we consider that link not useful, and route discovery process (i.e., sending RREQ) is initiated. This in turn, can guarantee a minimum throughput at the destination STA.

In this section, we present simulation results to demonstrate the benefits of multi-hop communications for indoor 60 GHz systems such as AR/VR and 8K UHD wireless streaming. The main contribution of this paper is to investigate the performance of multi-hop mmWave for indoor scenarios, rather than improving the performance of AODV-type protocols.

In this paper, we investigated the benefits of network layer solutions and on-demand routing protocol with backup routes to ensure reliable and robust mmWave communication under severe conditions (blockage and NLOS). Our hop-by-hop multi-path routing protocol establishes one primary and one reserved link per destination such that once the primary link is blocked, the backup link is ready to be deployed. To verify the performance of our protocol, we conducted system-level simulations based on the IEEE 802.11ad PHY and MAC specifications. Our simulations confirm the validity of our approach to sustain high throughput and low latency performance under blockage and NLOS scenarios.