{"paper":{"title":"Analog RF Computing: A New Paradigm for Energy-Efficient Edge AI Over MU-MIMO Systems","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"In MU-MIMO systems a base station broadcasts weight-encoded RF waveforms so clients perform neural-network matrix-vector multiplications with passive mixers, cutting client energy use by nearly two orders of magnitude.","cross_cats":["cs.AI","cs.ET","cs.IT","cs.LG","math.IT"],"primary_cat":"eess.SP","authors_text":"Vincent W.S. Wong, Wentao Yu","submitted_at":"2026-05-14T03:50:27Z","abstract_excerpt":"Modern edge devices increasingly rely on neural networks for intelligent applications. However, conventional digital computing-based edge inference requires substantial memory and energy consumption. In analog radio frequency (RF) computing, a base station (BS) encodes the weights of the neural networks and broadcasts the RF waveforms to the clients. Each client reuses its passive mixer to multiply the received weight-encoded waveform with a locally generated input-encoded waveform. This enables wireless receivers to perform the matrix-vector multiplications (MVMs) that account for most of the"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Simulations under 3GPP specifications show that analog RF computing can significantly reduce client-side energy consumption by nearly two orders of magnitude compared to digital computing, while mixed-precision inference requires even lower energy consumption than uniform-precision inference.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The derived tractable models for analog MVM accuracy and energy consumption accurately represent real-world passive mixer behavior, wireless channel effects, and hardware impairments without significant unmodeled errors.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Analog RF computing performs neural network matrix-vector multiplications via RF waveform mixing at clients in MU-MIMO systems, reducing energy consumption by nearly two orders of magnitude compared to digital computing.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"In MU-MIMO systems a base station broadcasts weight-encoded RF waveforms so clients perform neural-network matrix-vector multiplications with passive mixers, cutting client energy use by nearly two orders of magnitude.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"df4b2c1672677db6521f5b3421eda909fe02e726e8be9e7bec06b09435847cc5"},"source":{"id":"2605.14331","kind":"arxiv","version":1},"verdict":{"id":"f1a7042c-9436-46f4-a01a-8a67ee57bcd1","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T02:17:26.563206Z","strongest_claim":"Simulations under 3GPP specifications show that analog RF computing can significantly reduce client-side energy consumption by nearly two orders of magnitude compared to digital computing, while mixed-precision inference requires even lower energy consumption than uniform-precision inference.","one_line_summary":"Analog RF computing performs neural network matrix-vector multiplications via RF waveform mixing at clients in MU-MIMO systems, reducing energy consumption by nearly two orders of magnitude compared to digital computing.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The derived tractable models for analog MVM accuracy and energy consumption accurately represent real-world passive mixer behavior, wireless channel effects, and hardware impairments without significant unmodeled errors.","pith_extraction_headline":"In MU-MIMO systems a base station broadcasts weight-encoded RF waveforms so clients perform neural-network matrix-vector multiplications with passive mixers, cutting client energy use by nearly two orders of magnitude."},"references":{"count":34,"sample":[{"doi":"","year":2022,"title":"Edge artificial intelligence for 6G: Vision, enabling technologies, and applications,","work_id":"ab89921c-7ad4-4e34-b666-31b216200c0c","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2014,"title":"Computing’s energy problem (and what we can do about it),","work_id":"15b01c7e-abe2-4a47-b250-c0593bf97c91","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":1985,"title":"Deep reinforcement learning for task offloading in mobile edge computing systems,","work_id":"f69630c6-287a-4d54-9936-33bb99f65f2e","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2017,"title":"Joint optimal pricing and task scheduling in mobile cloud computing systems,","work_id":"95c5a984-4d8c-4da0-ab7f-4c2570ab24f7","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2011,"title":"Razavi,RF Microelectronics, 2nd ed","work_id":"7f35736f-5147-4f1b-9afc-eaeb529a002a","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":34,"snapshot_sha256":"76f6ba2efe69359523c2494a2d274f9e35451838a3377ddbc7a010c1dad49061","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}