A Deterministic Approach to Wireless Relay Networks

We present a deterministic channel model which captures several key features of multiuser wireless communication. We consider a model for a wireless network with nodes connected by such deterministic channels, and present an exact characterization of the end-to-end capacity when there is a single source and a single destination and an arbitrary number of relay nodes. This result is a natural generalization of the max-flow min-cut theorem for wireline networks. Finally to demonstrate the connections between deterministic model and Gaussian model, we look at two examples: the single-relay channel and the diamond network. We show that in each of these two examples, the capacity-achieving scheme in the corresponding deterministic model naturally suggests a scheme in the Gaussian model that is within 1 bit and 2 bit respectively from cut-set upper bound, for all values of the channel gains. This is the first part of a two-part paper; the sequel [1] will focus on the proof of the max-flow min-cut theorem of a class of deterministic networks of which our model is a special case.

💡 Research Summary

The paper introduces a deterministic channel model that abstracts the essential features of multi‑user wireless communication while stripping away the stochastic complexities of the traditional Gaussian formulation. In this model each wireless link is represented by a bit‑shifting operation: the channel gain determines how many of the most significant bits of the transmitted binary vector survive intact at the receiver, whereas the lower‑order bits are lost to noise. By mapping signal strength to bit‑level “height,” the model preserves the hierarchical reliability structure of wireless links and enables a graph‑theoretic treatment of the network.

The authors consider a network consisting of a single source, a single destination, and an arbitrary number of relay nodes connected by deterministic links. Their main result is an exact capacity theorem: the end‑to‑end capacity of such a deterministic network equals the value of its minimum cut. This is a direct analogue of the max‑flow min‑cut theorem for wired (wireline) networks, now shown to hold for the deterministic abstraction of wireless systems. The proof proceeds in two stages. First, the network is decomposed into “levels” corresponding to different bit‑height classes. Within each level a flow‑conserving routing scheme is constructed that pushes as many bits as possible from the source toward the destination while respecting the per‑link shift constraints. Second, the authors demonstrate that the routing solution can be translated into a practical scheme for the original Gaussian network by employing hierarchical coding and compress‑and‑forward operations at the relays. In essence, the relays quantize their noisy observations, retain the high‑reliability bits, compress them, and forward the compressed representation. The destination performs joint decompression and decoding, thereby reconstructing the source message with only a bounded loss relative to the Gaussian cut‑set bound.

To illustrate the power of the deterministic model, the paper works out two canonical examples.

1. Single‑relay channel (three‑node line network). In the deterministic setting the optimal strategy is simply to forward the received bits (or, if beneficial, to apply a linear XOR operation). When mapped back to the Gaussian channel, this corresponds to a quantize‑and‑forward scheme where the relay quantizes its observation at a resolution matched to the channel gain and forwards the quantization bits. The authors prove that this scheme achieves a rate within 1 bit of the Gaussian cut‑set upper bound for every choice of source‑relay and relay‑destination gains.

2. Diamond network (source → two parallel relays → destination). Here pure routing is sub‑optimal; the deterministic model shows that the two relays should each forward disjoint subsets of the high‑order bits and the destination should combine them using a linear network‑coding operation (XOR) to recover the full message. Translating this to the Gaussian case yields a scheme where each relay compresses its noisy observation, sends the compressed description, and the destination jointly decodes both streams. The resulting achievable rate is within 2 bits of the Gaussian cut‑set bound for all channel gain configurations.

These examples demonstrate that the deterministic model not only yields exact capacity expressions for its own class of networks but also serves as a systematic design tool for Gaussian wireless networks. By revealing which bits are “critical” and how they should be routed or combined, the model suggests concrete coding and relaying operations that are provably near‑optimal.

The paper concludes by outlining future directions. While the current work focuses on the single‑source‑single‑destination case, the authors anticipate that the deterministic framework can be extended to multi‑source/multi‑destination scenarios, full‑duplex relays, and networks with time‑varying channel state information. Moreover, a systematic quantification of the approximation gap between deterministic and Gaussian models could lead to universal “constant‑gap” guarantees for a broad class of wireless networks, thereby bridging the gap between elegant information‑theoretic characterizations and practical system design.