CROSS-BAND ANTENNA HEALTH INFERENCE FOR SMARTPHONE REFURBISHMENT: A MULTI-DEVICE, CRYPTOGRAPHICALLY-ATTESTED 90- SECOND RECEPTION TEST

When purchasing a pre-owned smartphone, physical defects like cracked glass are obvious, but damaged
internal antennas remain invisible to the consumer. Reception failures following handover are significant drivers
of resale disputes; however, current on-device “RF tests” conflate hardware quality with local environmental
attenuation. We present Cross-Band Antenna Health Inference (CBAHI), a 90-second on-device test that
differentiates hardware deficits from environmental factors on stock Android. CBAHI captures concurrent
reception data across WiFi (2.4/5/6 GHz), Cellular (LTE, 5G NR), GNSS, and Bluetooth-LE. Our core
contribution is a leave-one-out studentised residual test with an exact t-distribution null, allowing for calibrated
p-values. We evaluated CBAHI on 420 physical devices tested in the Retronics refurbishment laboratory. The
corpus spans 120 healthy calibration units and 300 labelled test units across Samsung, Google, and Xiaomi
OEMs, incorporating documented water damage, post-screen-replacement, post-back-glass-replacement, and
drop-history cohorts. Zero false-positive hardware-concern flags were observed in the cross-environment stress
test (60 runs across environment shifts), while the full labelled-cohort evaluation at η = 0.01 yielded one false
positive out of 120 healthy devices (specificity 99.2%). Each result is signed by a hardware-attested
AndroidKeyStore key bound to a buyer-supplied nonce plus the device’s IMEI hash, mitigating replay, twindevice substitution, pre-caching, and band-uniform selective-environment manipulation. The system achieves a
weighted sensitivity of 87.1% for documented water and mechanical damage (C1–C3), supported by a
differentially-private federated protocol with cohort-rolling keys. We further characterise the system through a
per-band Bayesian posterior baseline, a software-perturbation grid that exercises the graceful-degradation
behaviour of the leave-one-out test under partial radio loss, and a cross-environment validation yielding zero
false-positive hardware-concern flags across a > 13 dB cellular shift, a > 7 dB-Hz GPS L1 shift, and an
opposite-direction Bluetooth shift. We are explicit about what we do not claim: per-antenna isolation (closed at
the OEM combiner firmware), graded NFC sensitivity (binary functional check only), and reliability in
pathological RF environments. A flagged band is a hint to inspect, not a verdict.