IR-UWB Ranging Protocol and Model

To obtain IR-UWB ranges, vehicles need to perform a ranging protocol which may be challenging in VANETs. We have identified the following problems:

• One-way ranging protocol is not suitable as vehicles might not be perfectly synchro-nized due to many reasons (e.g., GNSS-denied environments, insufficient millisecond accuracy provided by Network Time Protocol (NTP)).

• Multiple-way ranging protocols must mitigate clock frequency offset-induced range errors and minimize the number of exchanged ranging frames. Numerous variants are detailed and benchmarked in [123] including: two-way ranging [124], symmetri-cal double-sided two-way ranging [125], asymmetrisymmetri-cal doubled-sided two-way rang-ing [126], double two-way rangrang-ing [127], and burst-mode symmetrical double-sided

Beacon Period Data Period Superframe N Superframe N+1 Superframe N-1

Beacon 5 ms

Contention Access Period (CAP) 9 slots × 5 ms = 45 ms

Contention Free Period (CFP) 30 slots × 5 ms = 150 ms Guaranteed Time Slot (GTS)

5 ms

… …

Figure 5.2: Beacon-aided TDMA MAC SF format supporting the localization functionality (SF duration of 200 ms).

two-way ranging [128]. Besides clock drift and clock offset issues, it is indeed im-portant to shorten ranging transactions (and thus reduce acquisition latency, for instance through ranging data aggregation and broadcast), which may cause mea-surement biases in high mobility contexts (resulting from a lack of spatial coherence as vehicles are moving, between the moments when the first transaction is initiated and the moment when it ends).

• Each vehicle performs ranging with multiple neighbors requiring careful and efficient scheduling to avoid packet collision.

So as to support the previous ranging transactions (initially, not in the vehicular domain), the standard IEEE 802.15.4 beacon-enabled time division multiple access (TDMA) MAC superframe (SF) has been initially modified, as depicted in Figure 5.2. Note that several variants, directly inheriting from the latter MAC structure, have been proposed, leading to different trade-offs in terms of ranging accuracy versus acquisition latency (e.g., [129–

132]). In our specific vehicular context, we assume that a vehicle (e.g., temporarily self-elected as local coordinator, if no other coordinator is already detected as active in the area) periodically transmits beacons to synchronize the vehicles in the vicinity in order to indicate the beginning of the SF and allocate time slot (TSs) for ranging. Paired vehicles demand the coordinator for ranging TSs in the contention access period (CAP) and are allocated guarantee time slots (GTSs) in the contention free period (CFP).

Besides, we use a three-way ranging procedure to compensate for the asynchronous vehicles’ clocks (i.e., clock drifts and offsets), thus requiring at least 3 adjacent GTS to complete a ranging transaction between two given nodes in the most basic allocation schemes (i.e, with no data aggregation and broadcast). Generally speaking, for a N -node VANET, each vehicle needs 3(N −1) GTSs (star configuration) for one estimates

with respect to its one-hop neighbors and the full VANET would require accordingly 3N(N −1) GTSs (mesh configuration). This situation may lead to an extremely long SF (or alternatively to multiple SFs) to complete the ranging procedures, which is harmful to CLoc under high mobility, as already highlighted (i.e., resulting in biased and/or severely asynchronous range measurements, low-rate CLoc, etc.). Thus, we assume that a classical aggregate and broadcast (A-B) scheme is applied to minimize the amount of overhead or the number of required GTSs to perform all the possible pairwise measurements in a mesh configuration, similarly to [129–131]. Specifically, such A-B scheme enables to share time resource in such a way that each node initiates specific ranging transactions with all the other nodes, and each transmitted packets can play multiple roles e.g., either a request or a response or even a drift correction packet, depending on the receiving neighbor status and on the current step in the three-way ranging protocol [129,130]. Quantitatively, under full connectivity 3N GTSs are needed to guarantee ranging transactions between any pair of nodes, instead of 3N(N−1) GTSs. Figure 5.3 illustrates an example of A-B scheme in a SF for 3 vehicles. The extension to more numerous vehicles is straightforward. Although the IR-UWB penetration is out the scope of this study, we hint in Figure 5.3 the fact that several TSs after the first and the second transmission rounds of all vehicles should be reserved for new vehicles to join. Finally, when the ranging/SF is completed, each vehicle is aware of the full distance matrix where dbj→i and dbi→j are different estimates produced by vehicles i and j, respectively of the relative distance between them. So different schemes can be applied to obtain the refined range ¯dj→i (by vehicle i) by either averaging 1/2(dbj→i+dbi→j) or considering only the latest estimate between them or the nearest estimate based on innovation monitoring to reject outliers². These measurement redundancies may also be beneficial in case some transactions are incomplete (due to the loss of at least one packet over the three required ones), and thus, related range estimates are missing.

Thus far, through a cooperative ranging protocol (e.g., based on the TOF estimation of transmitted packets involved in multiple-way handshake transactions), vehicleiat time

2Performing marginal innovation monitoring in a tracking filter at the system level (i.e., while integrating multiple links and thus, multiple range measurements with respect to neighbors) can indeed be used to detect link-wise inconsistent measurements and hence, discard outliers.

vehicle 3

coarse distance estimated by vehicle i

corrected distance estimated by vehicle i corrected distance estimated by vehicle i and broadcast to any vehicles in piconet GTS Slot

Figure 5.3: Example of the A-B protocol scheme in a SF for ranging within a VANET of 3 vehicles.

ti,k estimates the V2V distancedbj→i,k to node j,j∈ S_→i,k in positionxj,ki:

dbj→i,k =kx_i,k−x_j,kik+nj→i,k, (5.1)

where ranging noise n_j→i,k ∼ N(0, σ²_UWB) with σ_UWBthe ranging standard deviation. At the protocol level at least since the clock drift compensation mechanisms are expected to remove measurement biases so that noise is assumed to be zero-mean in first approximation (at least in LOS). Accordingly, the standard deviation accounts for both the arrival time uncertainty of each unitary packet involved in a ranging transaction and the residual noise resulting from clock drift compensation mechanisms (i.e., after combining several of these times of arrival).