TRMMA: Trajectory Recovery & Time-Reversal Access

• TRMMA is a context-dependent framework that includes efficient sparse trajectory recovery with map matching on road networks as well as interpretations in time-reversal communications.
• It employs a two-stage approach where an initial map matching algorithm (MMA) uses candidate classification followed by dual-transformer encoding to accurately recover trajectories.
• In communication applications, TRMMA is associated with protocols like TRMAC and iterative TRDMA that improve throughput and mitigate interference in multi-hop underwater and UWB networks.

Searching arXiv for the exact acronym and nearest variants to ground the article. arxiv_search query="^^^^^^^^5TRMMA OR TRMAC OR TRDMA OR \5^^^^^^^^"trajectory recovery\"^^^^^^^^5 \5^^^^^^^^"map matching\"^^^^^^^^5 \5^^^^^^^^"Time Reversal\"^^^^^^^^5 \5^^^^^^^^"Multiple Access\"^^^^^^^^5 \5^^^^^^^^"Mobility\"" max_results=^^^^5 \5TRMMA OR TRMAC OR TRDMA OR \5^^^^ TRMMA is not a standardized acronym across the papers considered here. In the most direct usage, it denotes an efficient method for sparse trajectory recovery on road networks, with MMA as its map-matching first stage (&&&^^^^5TRMMA OR TRMAC OR TRDMA OR \5^^^^&&&). In other papers, the same string is used only by interpretation: as equivalent to the active time-reversal MAC protocol TRMAC in multi-hop underwater acoustic networks (&&&^^^^5 \5^^^^&&&), as a shorthand for time-reversal multiple access and mobility via iterative TRDMA precoding (&&&^^^^5 \5^^^^&&&), or as a label for model-free algorithms built around Timed Reward Machines and related planner-facing reachability metrics in reinforcement learning (&&&^^^^5 \5^^^^&&&). The term is therefore best understood as context-dependent rather than canonical.

^^^^5 \5^^^^. Terminological status and scope

Within the supplied literature, only the road-network paper explicitly presents TRMMA as a named method; there, MMA serves as the first step of TRMMA, and the overall objective is “accurate trajectory recovery and map matching” for sparse GPS trajectories on road networks (&&&^^^^5TRMMA OR TRMAC OR TRDMA OR \5^^^^&&&). By contrast, several other papers do not introduce TRMMA as an official acronym, but the surrounding notes map it to nearby constructs: TRMAC in underwater acoustic MAC design, iterative TRDMA in time-reversal multiple access and mobility, and TRM with monotonic improvement assurance or Timed Reward Machine algorithms in reinforcement learning (&&&^^^^5 \5^^^^&&&, &&&^^^^5 \5^^^^&&&, Li et al., 28 Dec 2025, &&&^^^^5 \5^^^^&&&).

A further source of ambiguity is that neighboring acronyms are distinct methods. “TRAMA” is the correct acronym for “TRajectory-class-Aware Multi-Agent reinforcement learning,” and the paper states that if “TRMMA” appears elsewhere, it is a typo or alternate naming not used in that work (Na et al., 3 Mar 2025). Likewise, “TRiMM” denotes “Transformer-Based Rich Motion Matching” for digital humans rather than TRMMA (&&&^^^^5 \5TRMMA OR TRMAC OR TRDMA OR \5^^^^&&&).

Usage in the supplied papers	Domain	Canonical paper
TRMMA with MMA as first step	Sparse trajectory recovery and map matching	(&&&^^^^5TRMMA OR TRMAC OR TRDMA OR \5^^^^&&&)
TRMMA treated as equivalent to TRMAC	Multi-hop underwater acoustic MAC	(&&&^^^^5 \5^^^^&&&)
TRMMA mapped to iterative TRDMA for multiple access and mobility	UWB time-reversal communications	(&&&^^^^5 \5^^^^&&&)
TRMMA interpreted as TRM-based RL algorithms or guarantees	Reinforcement learning and planning	(&&&^^^^5 \5^^^^&&&, Li et al., 28 Dec 2025, &&&^^^^5 \5^^^^6&&&)

^^^^5 \5^^^^. TRMMA as sparse trajectory recovery on road networks

In its direct and explicit form, TRMMA addresses two coupled problems on a directed road network PRESERVED_PLACEHOLDER_^^^^5TRMMA OR TRMAC OR TRDMA OR \5^^^^: map matching, which aligns sparse GPS points to road segments, and trajectory recovery, which infers missing points so that a sparse trajectory can be converted into a high-sampling trajectory on the road network (&&&^^^^5TRMMA OR TRMAC OR TRDMA OR \5^^^^&&&). A sparse GPS trajectory is written as PRESERVED_PLACEHOLDER_^^^^5 \5^^^^, while recovered map-matched points are represented as tuples PRESERVED_PLACEHOLDER_^^^^5 \5^^^^, where PRESERVED_PLACEHOLDER_^^^^5 \5^^^^ is a road segment and PRESERVED_PLACEHOLDER_^^^^5 \5^^^^ is the position ratio along that segment.

The first stage, MMA, formulates map matching as classification over a small candidate segment set rather than over the entire road network. For each GPS point $p_i$, the candidate set $S_i$ is the top-$k_c$ nearest segments ranked by perpendicular distance, with $k_c=10$ in the reported implementation. MMA combines a Node^^^^5 \5^^^^Vec-initialized segment embedding, four cosine directional features, a transformer over normalized GPS inputs, and an attention-based fusion between point and candidate-segment representations. Training uses binary cross-entropy over candidates in $S_i$. This local-candidate formulation is a central design choice because it reduces the output space while preserving the relevant alternatives (&&&^^^^5TRMMA OR TRMAC OR TRDMA OR \5^^^^&&&).

TRMMA then uses the route returned by MMA and restricts recovery to segments on that route. Its encoder is a dual-transformer encoding process: one transformer encodes the observed sparse trajectory, and a second encodes the matched route. A cross-attention fusion produces segment-level representations PRESERVED_PLACEHOLDER_^^^^5 \5TRMMA OR TRMAC OR TRDMA OR \5^^^^, which initialize a decoder. For each interval between consecutive observed points, the decoder predicts the number of missing points from the target sampling interval, then performs constrained segment classification over the sub-route suffix and position-ratio regression for each missing point. The overall loss combines segment classification and ratio regression, written as PRESERVED_PLACEHOLDER_^^^^5 \5 \5^^^^ (&&&^^^^5TRMMA OR TRMAC OR TRDMA OR \5^^^^&&&).

The reported empirical evaluation uses four large real-world datasets: Porto with ^^^^5 \5^^^^,^^^^5TRMMA OR TRMAC OR TRDMA OR \5 \5 \5^^^^,^^^^5 \5 \5^^^^7 trajectories, Xi’an with ^^^^5 \5^^^^,^^^^5 \5 \5^^^^6,95^^^^5TRMMA OR TRMAC OR TRDMA OR \5^^^^ trajectories, Beijing with ^^^^5 \5^^^^,^^^^5 \5^^^^76,^^^^5TRMMA OR TRMAC OR TRDMA OR \5^^^^97 trajectories, and Chengdu with ^^^^5 \5^^^^,^^^^5 \5^^^^8^^^^5 \5^^^^,^^^^5 \5 \5 \5^^^^ trajectories. MMA consistently attains the strongest map-matching quality among the reported methods, with F^^^^5 \5^^^^/Jaccard values of 9^^^^5 \5^^^^.^^^^5 \55^^^^/9^^^^5 \5^^^^.5^^^^5 \5^^^^ on Porto, 97.^^^^5 \5^^^^6/95.97 on Xi’an, 8^^^^5TRMMA OR TRMAC OR TRDMA OR \5^^^^.9^^^^5 \5^^^^/75.^^^^5 \5^^^^8 on Beijing, and 96.5^^^^5 \5^^^^/9^^^^5 \5^^^^.9^^^^5 \5^^^^ on Chengdu. TRMMA also leads the trajectory-recovery benchmarks, reporting F^^^^5 \5^^^^/Accuracy/MAE of 75.87/57.8^^^^5 \5^^^^/8^^^^5 \5^^^^.^^^^5 \5TRMMA OR TRMAC OR TRDMA OR \5^^^^ m on Porto, 9^^^^5TRMMA OR TRMAC OR TRDMA OR \5^^^^.^^^^5 \5 \5^^^^/78.95/68.^^^^5 \5^^^^ m on Xi’an, 6^^^^5 \5^^^^.6^^^^5 \5^^^^/5^^^^5 \5^^^^.7^^^^5 \5^^^^/^^^^5 \5 \5 \5^^^^.^^^^5 \5^^^^ m on Beijing, and 89.^^^^5 \5^^^^9/75.^^^^5 \5^^^^8/75.^^^^5 \5^^^^ m on Chengdu (&&&^^^^5TRMMA OR TRMAC OR TRDMA OR \5^^^^&&&).

The efficiency claims are equally specific. MMA requires ^^^^5TRMMA OR TRMAC OR TRDMA OR \5^^^^.^^^^5 \5^^^^9 s, ^^^^5TRMMA OR TRMAC OR TRDMA OR \5^^^^.^^^^5 \5^^^^6 s, ^^^^5TRMMA OR TRMAC OR TRDMA OR \5^^^^.^^^^5 \5 \5^^^^ s, and ^^^^5TRMMA OR TRMAC OR TRDMA OR \5^^^^.^^^^5 \5 \5^^^^ s per ^^^^5 \5^^^^,^^^^5TRMMA OR TRMAC OR TRDMA OR \5TRMMA OR TRMAC OR TRDMA OR \5TRMMA OR TRMAC OR TRDMA OR \5^^^^ trajectories on Porto, Xi’an, Beijing, and Chengdu, respectively. TRMMA requires ^^^^5TRMMA OR TRMAC OR TRDMA OR \5^^^^.88 s, ^^^^5TRMMA OR TRMAC OR TRDMA OR \5^^^^.8^^^^5TRMMA OR TRMAC OR TRDMA OR \5^^^^ s, ^^^^5TRMMA OR TRMAC OR TRDMA OR \5^^^^.59 s, and ^^^^5TRMMA OR TRMAC OR TRDMA OR \5^^^^.85 s per ^^^^5 \5^^^^,^^^^5TRMMA OR TRMAC OR TRDMA OR \5TRMMA OR TRMAC OR TRDMA OR \5TRMMA OR TRMAC OR TRDMA OR \5^^^^ trajectory recoveries on the same datasets. The paper attributes these gains to the fact that TRMMA focuses on the segments in the route returned by MMA and therefore avoids evaluation of all road segments (&&&^^^^5TRMMA OR TRMAC OR TRDMA OR \5^^^^&&&).

^^^^5 \5^^^^. Time-reversal communication interpretations

A second family of usages connects TRMMA to time-reversal-based multiple access. In “Time Reversal based MAC for Multi-Hop Underwater Acoustic Networks,” the relevant protocol is explicitly named TRMAC, and the accompanying note states that TRMMA should be treated as equivalent to TRMAC in the paper (&&&^^^^5 \5^^^^&&&). TRMAC is an active time-reversal-based medium access control protocol for multi-hop underwater acoustic networks, designed to exploit underwater multipath, spatial focusing, and temporal focusing. The paper states that active TR spatially focuses the signals to the location of the intended receiver, thereby diminishing interference among parallel links, and that the protocol derives a threshold of the link cross-correlation to resolve collisions caused by high cross-correlation between realistic links. Simulations in OPNET and BELLHOP show significantly improved throughput, decreased delay, and reduced data drop ratio (&&&^^^^5 \5^^^^&&&).

A related but distinct interpretation appears in “Time Reversal for Multiple Access and Mobility: Algorithmic Design and Experimental Results,” where TRMMA is mapped to the paper’s iterative Time-Reversal Division Multiple Access (TRDMA) framework (&&&^^^^5 \5^^^^&&&). Here the core contribution is an iterative TRDMA (ITRDMA) precoder that suppresses inter-symbol interference and inter-user interference while preserving low receiver complexity. The experiments use a ^^^^5 \5^^^^ GHz carrier, ^^^^5 \5TRMMA OR TRMAC OR TRDMA OR \5TRMMA OR TRMAC OR TRDMA OR \5^^^^ MHz bandwidth, and an 8-antenna transmitter in a reverberation chamber. The paper reports up to 6.5 dB SIR improvement after ^^^^5 \5TRMMA OR TRMAC OR TRDMA OR \5TRMMA OR TRMAC OR TRDMA OR \5^^^^ iterations at PRESERVED_PLACEHOLDER_^^^^5 \5 \5^^^^ MHz and studies robustness under receiver mobility, including a ^^^^5 \5TRMMA OR TRMAC OR TRDMA OR \5^^^^ cm linear bench experiment and speed thresholds tied to channel-estimation interval PRESERVED_PLACEHOLDER_^^^^5 \5 \5^^^^ &&&^^^^5 \55^^^^&&&)?]