Skip to content

Broadband (Coherent) Scenario – Clarification on Training Setup and Dataset Size #12

Description

@Neeraja-pk

I am currently implementing SubspaceNet for the broadband coherent scenario, and I am attempting to reproduce the performance reported in Table III of the paper. However, I am facing difficulties in reproducing the reported loss/performance values and would appreciate clarification regarding the training configuration used.

I generated datasets:
Generic_DataSet_Broadband_coherent_45000_M=2_N=8_T=200_SNR=10_eta=0_sv_noise_var0_bias=0 and SubspaceNet_DataSet_Broadband_coherent_45000_M=2_N=8_T=200_SNR=10_eta=0_sv_noise_var0_bias=0.

However, I observe relatively high loss (~0.56 (rad) at epoch 30), which does not seem to align with the performance trends in Table III. Hyperparameters used are as follows:
simulation_parameters = (
TrainingParams()
.set_batch_size(2048)
.set_epochs(80)
# .set_model(system_model=samples_model, tau=model.tau, diff_method=model.diff_method)
.set_model(model=model_config)
.set_optimizer(optimizer="Adam", learning_rate=0.001, weight_decay=1e-9)
.set_training_dataset(train_dataset)
.set_schedular(step_size=80, gamma=0.2)
.set_criterion()
)

I would like to confirm whether: My dataset size may be insufficient, or Broadband configuration requires different hyperparameters.
Any clarification regarding the exact training setup used for the broadband coherent scenario would be greatly appreciated.
Thank you in advance for your support and guidance.

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions