Implementation of Multipath and Multiple Description Coding in OLSR

In this paper we discussed the application and the implementation of multipath routing and multiple description coding (MDC) extension of OLSR, called MP-OLSR. It is based on the link state algorithm and employs periodic exchange of messages to maintain topology information of the networks. In the mean time, it updates the routing table in an on-demand scheme and forwards the packets in multiple paths which have been determined at the source. If a link failure is detected, the algorithm recovers the route automatically. Concerning the instability of the wireless networks, the multiple description coding is used to improve reliability of the network transmission, and several methods are proposed to allocate the redundancy in different paths. The simulation in NS2 shows that the new protocol can effectively improve the performance of the networks. The implementation of MP-OLSR is also proposed in the end.

💡 Research Summary

The paper presents MP‑OLSR, an extension of the Optimized Link State Routing (OLSR) protocol that integrates multipath routing and Multiple Description Coding (MDC) to improve reliability and performance in mobile ad‑hoc wireless networks. MP‑OLSR retains OLSR’s periodic HELLO and TC message exchange for topology discovery, but departs from the classic approach of continuously maintaining a full routing table. Instead, routes are computed on‑demand when a data flow needs to be sent. The core of the multipath computation is a “Multipath Dijkstra Algorithm”. After the first shortest path to a destination is found, the algorithm increases the cost of the arcs belonging to that path (using a function fₚ) and the cost of arcs leading to vertices of that path (using a function fₑ). By adjusting fₚ and fₑ, the protocol can generate link‑disjoint or node‑disjoint paths, or a mixture when strict disjointness is impossible in sparse topologies.

Routing is performed using a semi‑source‑routing scheme: the complete hop list of each selected path is inserted into the packet header. Intermediate nodes simply verify that the next hop indicated in the header is a current neighbor and forward accordingly. This gives the source fine‑grained control over traffic distribution across multiple paths, but it also creates two failure scenarios: (1) the source’s topology view may be stale, causing it to select a non‑existent link, and (2) the topology may change while a packet is in transit, breaking the pre‑computed path. To mitigate these problems, MP‑OLSR introduces a route‑recovery mechanism. When a node discovers that the next hop listed in the header is no longer a neighbor, it locally recomputes a fresh shortest path to the destination and forwards the packet along this new route. The recovery process is further accelerated by exploiting link‑layer feedback from the IEEE 802.11 MAC (MAC RET events), which informs the routing layer immediately when a transmission fails.

The second major contribution is the integration of MDC to increase robustness against packet loss. The authors adopt the Mojette transform, a discrete, exactly invertible version of the Radon transform that requires only addition operations. Given an original data block I, the transform produces N independent descriptions (P₁…P_N). Any subset of at least M descriptions (0 < M ≤ N) suffices to reconstruct I perfectly. In the implementation, the descriptions are sent over distinct paths discovered by the multipath algorithm. To avoid excessive traffic overhead, the authors propose two practical measures: (a) buffering a group of original packets before applying the transform, thus reducing the number of MDC operations, and (b) using a systematic construction where the first M descriptions are exact copies of the original packets, guaranteeing zero decoding cost when no loss occurs.

Redundancy allocation is made adaptive to network conditions. At the physical layer, link quality (e.g., Bit Error Rate) can be propagated to influence the cost functions of the Multipath Dijkstra Algorithm. At the network layer, each node monitors the length of its FIFO buffer; the longest buffer along a path determines how much extra redundancy should be assigned to that path. An alternative, application‑layer approach maximizes the expected quality E