To reduce the de-mapping complexity of cross-constellation QAM in G.HN standard, we first introduce a novel concept of contribution weight to evaluate the contribution of given constellation to the log-likelihood ratio. With its help, a low complexity de-mapping algorithm within shrunk search range is proposed. And then, some boundary conditions are analyzed and solved. Furthermore, an adaptive configuration of search range is also presented with the assistance of channel estimation. The simulation result validates that the proposed algorithm downsizes the search range without any sacrifice of performance. 2 048-QAM and 512-QAM shrink their average of searching constellations to 1.9％ and 5.7％ respectively of the whole constellations. Therefore, the proposed algorithm makes good tradeoffs between performance and complexity, and is expected to have important engineering value and widely application foreground.