Support in DeepMIMO v4
Hi,
In the code, the deepmimo version supported seems to be v3 only, as loading it loads parameters.mat file and not parameters.json file.
If so, then how did you generate patches for pretraining on data that is compatible with deepmimo v4 only, like new york https://www.deepmimo.net/scenarios/v4/city_0_newyork_3p5_lwm? I tried using the data under the legacy part (deepmimo v3 https://www.deepmimo.net/scenarios/v3/city_series/new-york) but i keep getting for most cities the error: "ValueError: The rows are not selected properly. The available rows in this scenario are in [0, 43]."
Hi @tomerraviv95 ,
Thank you for your feedback.
You are right! LWM 1.1 was pre-trained on densified DeepMIMO v3 scenarios, where the grid spacing was reduced from 2.5 m to 1 m to increase the number of samples (which is why the code expects more rows than in the standard v3 scenarios). These specific densified v3 scenarios are not publicly available on the website, although the corresponding v4 versions are.
For your use case, we recommend using our LWM competition package:
https://huggingface.co/wi-lab/lwm-competition-2025
This refactored LWM 1.1 package, originally built for our ongoing competition (https://lwm-wireless.net/challenge), is fully based on DeepMIMO v4. It allows you to access and use over 160 DeepMIMO v4 scenarios (including the ones used for LWM 1.1) directly through our APIs, with no additional coding or manual data downloading and organization required.
Please let us know if you have any questions.