{"id":3290,"date":"2025-09-12T00:55:03","date_gmt":"2025-09-12T00:55:03","guid":{"rendered":"https:\/\/172-234-197-23.ip.linodeusercontent.com\/?p=3290"},"modified":"2025-09-12T06:12:25","modified_gmt":"2025-09-12T06:12:25","slug":"neural-beamforming-with-scene-priors-a-minimal-reproducible-benchmark-with-calibration","status":"publish","type":"post","link":"https:\/\/172-234-197-23.ip.linodeusercontent.com\/?p=3290","title":{"rendered":"Neural Beamforming with Scene Priors: A Minimal, Reproducible Benchmark with Calibration"},"content":{"rendered":"\n<figure class=\"wp-block-embed is-type-wp-embed is-provider-spectrcyde wp-block-embed-spectrcyde\"><div class=\"wp-block-embed__wrapper\">\n<blockquote class=\"wp-embedded-content\" data-secret=\"MtKpujD99T\"><a href=\"https:\/\/172-234-197-23.ip.linodeusercontent.com\/?page_id=3281\">Neural Beamforming with Scene Priors:A Minimal, Reproducible Benchmark withCalibration<\/a><\/blockquote><iframe class=\"wp-embedded-content\" sandbox=\"allow-scripts\" security=\"restricted\" style=\"position: absolute; visibility: hidden;\" title=\"&#8220;Neural Beamforming with Scene Priors:A Minimal, Reproducible Benchmark withCalibration&#8221; &#8212; Spectrcyde\" src=\"https:\/\/172-234-197-23.ip.linodeusercontent.com\/?page_id=3281&#038;embed=true#?secret=yiHtGQWwzd#?secret=MtKpujD99T\" data-secret=\"MtKpujD99T\" width=\"600\" height=\"338\" frameborder=\"0\" marginwidth=\"0\" marginheight=\"0\" scrolling=\"no\"><\/iframe>\n<\/div><\/figure>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h1 class=\"wp-block-heading\"><\/h1>\n\n\n\n<p>A Minimal, Reproducible Benchmark with Calibration<\/p>\n\n\n\n<p><strong>Benjamin J. Gilbert<\/strong><br>Spectrcyde RF Quantum SCYTHE, College of the Mainland<br><a href=\"mailto:bgilbert2@com.edu\">bgilbert2@com.edu<\/a><br>ORCID: [0000-0002-1234-5678]<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">Abstract<\/h2>\n\n\n\n<p>We introduce a <strong>lightweight benchmark<\/strong> for RF beam selection that fuses <strong>scene priors<\/strong> with channel state information (CSI). A compact neural head runs on top of a simulated environment and auto-generates results (accuracy, Succ@\u00b11, ECE, Brier score) across discrete beam sweeps (8\/12\/16 beams).<\/p>\n\n\n\n<p>The framework ensures <strong>reviewer-safe reproducibility<\/strong>: one command yields figures, tables, and calibration analysis. The approach leverages <strong>NeRF-derived scene priors<\/strong> combined with WiFi CSI to predict optimal beamforming angles, with <strong>temperature scaling<\/strong> applied for calibrated uncertainty estimates.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">I. Introduction<\/h2>\n\n\n\n<p>Traditional RF beamforming depends on <strong>geometric propagation models<\/strong> or <strong>hand-engineered signal processing<\/strong>. These methods lack robustness under real-world blockage and multipath.<\/p>\n\n\n\n<p>Recent advances in <strong>Neural Radiance Fields (NeRF)<\/strong> show that compact scene encodings can act as priors for geometry and material layout. This paper integrates <strong>NeRF scene features<\/strong> with <strong>CSI snapshots<\/strong> in a neural predictor, targeting beam direction selection.<\/p>\n\n\n\n<p>Our contribution is <strong>minimal but reproducible<\/strong>:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>A supervised neural beamforming module that fuses scene priors with CSI.<\/li>\n\n\n\n<li>Automatic evaluation with accuracy and calibration metrics.<\/li>\n\n\n\n<li>End-to-end reproducibility via scripted builds.<\/li>\n<\/ul>\n\n\n\n<p>The emphasis is not field-grade deployment but <strong>clear reproducibility<\/strong>, calibration reporting, and lightweight runtime \u2014 the Guangdong philosophy of <em>engineer small, prove reproducible, then scale<\/em>.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">II. Related Work<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Neural Scene Priors<\/strong>: NeRF [1] encodes geometry and materials; we exploit compact features as side information to regularize beam selection.<\/li>\n\n\n\n<li><strong>Learning-Based Beam Selection<\/strong>: mmWave studies show learning policies outperform exhaustive search, especially under blockage [2].<\/li>\n\n\n\n<li><strong>Propagation Studies<\/strong>: Classical models justify high-frequency links but underscore the need for aggressive beam steering [3].<\/li>\n\n\n\n<li><strong>Calibration<\/strong>: Probabilistic models require confidence control; temperature scaling [4], ECE [5], and Brier [6] are standard.<\/li>\n<\/ul>\n\n\n\n<p>Our approach differs by combining <strong>scene priors + CSI<\/strong> in a lightweight neural net and reporting <strong>both accuracy and calibration<\/strong>, yielding <strong>Pareto tradeoffs across beam count, accuracy, and runtime<\/strong>.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">III. Method<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">A. Neural Architecture<\/h3>\n\n\n\n<p>We implement <strong>RFBeamformingNN<\/strong>, which maps concatenated scene+CSI features to beam logits. Input vector:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>NeRF Scene Features (50 dims)<\/strong>: Depth, material, geometry encodings.<\/li>\n\n\n\n<li><strong>WiFi CSI (40 dims)<\/strong>: Channel state vectors with injected noise.<\/li>\n\n\n\n<li><strong>RF Environment (20 dims)<\/strong>: Signal strength, interference, temporal dynamics.<\/li>\n<\/ul>\n\n\n\n<p>Backbone: fully connected layers (128 \u2192 64 neurons), with <strong>batch norm + dropout<\/strong>. Training uses <strong>cross-entropy loss<\/strong> against simulated optimal beams. For calibration, <strong>Platt\/temperature scaling<\/strong> adjusts probabilities.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">B. Evaluation Metrics<\/h3>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Exact Accuracy<\/strong>: % of perfect beam index predictions.<\/li>\n\n\n\n<li><strong>Succ@\u00b11<\/strong>: Success if |predicted \u2013 optimal| \u2264 1 index. Bin widths: 45\u00b0 (8 beams), 30\u00b0 (12 beams), 22.5\u00b0 (16 beams).<\/li>\n\n\n\n<li><strong>ECE (Expected Calibration Error)<\/strong>: 15-bin histogram, measures miscalibration.<\/li>\n\n\n\n<li><strong>Brier Score<\/strong>: Quadratic scoring rule for probabilistic accuracy.<\/li>\n<\/ul>\n\n\n\n<p>Calibration procedure: Fit temperature <strong>T &gt; 0<\/strong> on a held-out set by minimizing negative log-likelihood. Report pre\/post ECE and Brier on test set.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">Guangdong Framing<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Minimal build<\/strong>: one command reproduces all results.<\/li>\n\n\n\n<li><strong>Compact network<\/strong>: small enough for embedded RF controllers.<\/li>\n\n\n\n<li><strong>Full calibration reporting<\/strong>: reviewer-safe, no hidden heuristics.<\/li>\n\n\n\n<li><strong>Practical scope<\/strong>: synthetic benchmark only, but structured for rapid extension into mmWave\/WiFi-6 testbeds.<\/li>\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p>\u2699\ufe0f <strong>Summary:<\/strong> A <strong>neural beamforming benchmark<\/strong> that unifies NeRF scene priors with CSI. Lightweight, reproducible, calibrated. Demonstrates the Guangdong ethos: <em>small kit, clear metrics, reproducible pipeline, deployment mindset<\/em>.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p>Do you want me to also <strong>recast the evaluation metrics into LaTeX tabular form<\/strong> (like you\u2019d drop straight into IEEE style), or keep them in narrative form for now?<\/p>\n","protected":false},"excerpt":{"rendered":"<p>A Minimal, Reproducible Benchmark with Calibration Benjamin J. GilbertSpectrcyde RF Quantum SCYTHE, College of the Mainlandbgilbert2@com.eduORCID: [0000-0002-1234-5678] Abstract We introduce a lightweight benchmark for RF beam selection that fuses scene priors with channel state information (CSI). A compact neural head runs on top of a simulated environment and auto-generates results (accuracy, Succ@\u00b11, ECE, Brier score)&hellip;&nbsp;<a href=\"https:\/\/172-234-197-23.ip.linodeusercontent.com\/?p=3290\" rel=\"bookmark\"><span class=\"screen-reader-text\">Neural Beamforming with Scene Priors: A Minimal, Reproducible Benchmark with Calibration<\/span><\/a><\/p>\n","protected":false},"author":1,"featured_media":3292,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"neve_meta_sidebar":"","neve_meta_container":"","neve_meta_enable_content_width":"","neve_meta_content_width":0,"neve_meta_title_alignment":"","neve_meta_author_avatar":"","neve_post_elements_order":"","neve_meta_disable_header":"","neve_meta_disable_footer":"","neve_meta_disable_title":"","footnotes":""},"categories":[10],"tags":[],"class_list":["post-3290","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-signal_scythe"],"_links":{"self":[{"href":"https:\/\/172-234-197-23.ip.linodeusercontent.com\/index.php?rest_route=\/wp\/v2\/posts\/3290","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/172-234-197-23.ip.linodeusercontent.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/172-234-197-23.ip.linodeusercontent.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/172-234-197-23.ip.linodeusercontent.com\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/172-234-197-23.ip.linodeusercontent.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=3290"}],"version-history":[{"count":3,"href":"https:\/\/172-234-197-23.ip.linodeusercontent.com\/index.php?rest_route=\/wp\/v2\/posts\/3290\/revisions"}],"predecessor-version":[{"id":3295,"href":"https:\/\/172-234-197-23.ip.linodeusercontent.com\/index.php?rest_route=\/wp\/v2\/posts\/3290\/revisions\/3295"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/172-234-197-23.ip.linodeusercontent.com\/index.php?rest_route=\/wp\/v2\/media\/3292"}],"wp:attachment":[{"href":"https:\/\/172-234-197-23.ip.linodeusercontent.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=3290"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/172-234-197-23.ip.linodeusercontent.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=3290"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/172-234-197-23.ip.linodeusercontent.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=3290"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}