Skip to content

benchmarks_reference.md

Latest commit

 

History

History
63 lines (47 loc) · 3.23 KB

benchmarks_reference.md

File metadata and controls

63 lines (47 loc) · 3.23 KB
Raw

CORDIS Benchmark Registry Reference

Generated automatically from cordis.algorithms.benchmarks.BENCHMARK_REGISTRY. Regenerate with: ./scripts/list_benchmarks.sh

Each benchmark pairs a communication beamformer (varied across rows) with a power-allocation strategy (Phase II of CORDIS-Split, or a uniform fixed ratio ρ). All benchmarks share the same interface and return a BenchmarkResult for the simulation runner.

Note on LR-MMSE benchmarks (paper revision).

lr_mmse_split (LR-MMSE Phase-I + optimal P-Split PA) is numerically identical to CORDIS-Split (which is precisely LR-MMSE BF + optimal PSR by construction). To avoid duplicating the CORDIS-Split curve in figures, the default benchmark surface (cordis_vs_benchmarks, all_algorithms) now uses lr_mmse_fixed (PSR ρ=0.5, no PA optimisation) as the LR-MMSE representative. The lr_mmse_split entry is retained for debugging / legacy access but is not in any default spec set.

The two extra variants lr_mmse_fixed_p020 (ρ=0.2, sensing-biased) and lr_mmse_fixed_p080 (ρ=0.8, comm-biased) support the paper's benefit of optimal PSR figure via the psr_baselines spec set (cordis.experiments.specs.psr_baselines).

Quick usage

from cordis.algorithms.benchmarks import (
    BENCHMARK_REGISTRY, run_benchmark, run_all_benchmarks,
)

# Single benchmark
res = run_benchmark(
    "mrt_split", topo, cfg, est, sensing_stats,
    association, sigma_n_sq, Pmax,
)
# res.W_tx, res.phase_i, res.split_res, res.psr

# Whole sweep on one scenario
all_res = run_all_benchmarks(
    topo, cfg, est, sensing_stats, association,
    sigma_n_sq, Pmax, skip_failures=True,
)

PA optimised via P-Split (Phase II of CORDIS-Split)

Each AP's power-splitting ratio ρ_a is chosen by the projected-gradient / CVXPY Phase II solver to satisfy the per-user SINR target γ while maximising the sensing utility. These variants isolate the choice of communication beamformer.

Name Comm BF ρ Description
lr_mmse_split lr_mmse — LR-MMSE + P-Split PA (≡ CORDIS-Split)
mrt_split mrt — Local MRT + P-Split PA
rzf_split rzf — Local RZF + P-Split PA
zf_split zf — Local ZF + P-Split PA
global_zf_split global_zf — Global ZF + P-Split PA
global_mrt_split global_mrt — Global MRT + P-Split PA

PA fixed (no Phase II optimisation)

All APs use a uniform ρ (no PA optimisation). These variants isolate the value of the PA optimisation itself by holding the beamformer fixed and skipping Phase II.

Name Comm BF ρ Description
lr_mmse_fixed lr_mmse 0.50 LR-MMSE + equal ρ=0.5 (no PA opt)
lr_mmse_fixed_p020 lr_mmse 0.20 LR-MMSE + equal ρ=0.2 (sensing-biased)
lr_mmse_fixed_p080 lr_mmse 0.80 LR-MMSE + equal ρ=0.8 (comm-biased)
mrt_fixed mrt 0.50 Local MRT + equal ρ=0.5
rzf_fixed rzf 0.50 Local RZF + equal ρ=0.5

Adding a benchmark

Add a new entry to BENCHMARK_REGISTRY in cordis/algorithms/benchmarks.py. The tuple is (description, comm_bf_method, pa_optimized, fixed_psr). Then run ./scripts/list_benchmarks.sh to refresh this file.