LITF-PA-2026-017 · AgTech / Animal Health

System and Method for Non-Contact Continuous Vital Sign Monitoring and Early Disease Outbreak Detection in Commercial Livestock Facilities Using Ceiling-Mounted Millimeter-Wave FMCW Radar Arrays, Per-Animal Micro-Doppler Decomposition, and Edge-Deployed Epidemiological Anomaly Networks

Defensive Prior Art Disclosure
Ceiling-mounted millimeter-wave radar sensor pods in a dairy barn monitoring cattle vital signs with signal visualization overlay
⚖️ Prior Art Notice: This document is published as defensive prior art under 35 U.S.C. § 102(a)(1). The inventions described herein are dedicated to the public domain as of the publication date above. This disclosure is intended to prevent the patenting of these concepts by any party.

Abstract

Disclosed is a system and method for continuously monitoring the vital signs of individual animals in commercial livestock facilities using ceiling-mounted frequency-modulated continuous-wave (FMCW) millimeter-wave radar arrays operating in the 60 GHz or 77 GHz bands. Each radar unit simultaneously tracks multiple animals within its field of view using range-angle-Doppler processing, extracts per-animal micro-Doppler signatures encoding chest wall displacement (respiration, typical 12-30 breaths/min for cattle) and cardiac-induced surface vibration (heart rate, typical 48-84 bpm for cattle), and classifies individual health states using an edge-deployed temporal convolutional network. The system detects respiratory rate elevation, heart rate variability depression, and locomotion gait anomalies 24-72 hours before clinical signs of bovine respiratory disease (BRD), mastitis, and lameness become visible. A facility-level epidemiological anomaly propagation module models spatial clustering of vital sign deviations across the herd, distinguishing isolated individual illness from emerging outbreak events requiring immediate veterinary intervention.

Field of the Invention

This invention relates to precision livestock farming, specifically to non-contact radio-frequency sensing of animal vital signs combined with edge-deployed machine learning for individual animal health monitoring and herd-level disease outbreak early warning in commercial cattle, swine, and poultry operations.

Background

Livestock disease causes estimated annual losses of $300 billion globally (Global Burden of Animal Diseases, 2024), with bovine respiratory disease (BRD) alone accounting for approximately $900 million per year in the U.S. cattle industry through mortality, treatment costs, and reduced performance (USDA APHIS). BRD is the leading cause of death in feedlot cattle, affecting 15-45% of animals within 45 days of feedlot arrival, yet clinical diagnosis relies on visual observation of symptoms (nasal discharge, depression, labored breathing) that manifest only after the disease has progressed significantly.

Current approaches to livestock health monitoring fall into several categories:

  • Visual observation by pen riders: Trained personnel on horseback or foot evaluate animals 1-2 times daily. Sensitivity for BRD detection is 56-62% (White and Renter, 2009), meaning nearly half of sick animals are missed. Cost: $0.50-1.50 per head per day in labor. Subjective, weather-dependent, and impossible at night.
  • Rectal temperature monitoring: The gold standard for fever detection requires physical restraint in a chute, generating stress that itself elevates temperature by 0.3-0.8°C (cortisol response). Practical only as a one-time check at processing, not continuous monitoring.
  • Rumen bolus sensors: Swallowed transmitters (e.g., smaXtec) measure rumen temperature and activity. Cost: $80-120 per animal. Provide continuous temperature but no respiratory or cardiac data. Require individual oral administration. Not recoverable at slaughter in many jurisdictions. Battery life: 4-6 months.
  • Ear tag accelerometers: Products like CowManager and Allflex SenseTime measure rumination time, activity level, and ear temperature. Cost: $60-90 per tag. Detect feeding behavior changes but lack the physiological resolution to measure respiration rate or heart rate. Ear temperature correlates poorly with core temperature (r = 0.3-0.5).
  • Thermal imaging cameras: Infrared cameras pointed at pen areas can detect elevated surface temperature. Limited by ambient temperature variation, solar radiation, and wet/dirty coats. Cannot extract heart rate or respiration rate. Schaefer et al., 2012 showed sensitivity of 73% for BRD detection via infrared thermography, but specificity was only 64%.

Millimeter-wave radar for human vital sign monitoring has been validated extensively. Zhang et al. (2024) demonstrated contactless heart rate and respiration monitoring of macaques using 60 GHz FMCW radar with mean absolute error of 1.2 bpm for heart rate and 0.8 breaths/min for respiration rate. WO2015174879A1 describes a general-purpose mm-wave radar vital signs apparatus but targets human subjects and single-target scenarios. US20090189740A1 describes contactless vital signs monitoring of animals but uses ultrawideband radar without micro-Doppler decomposition for per-animal separation in multi-animal environments and without epidemiological outbreak modeling. US11317608B2 (Texas Instruments) covers general mmWave vital signs processing but does not address livestock-specific applications, multi-animal tracking, or disease prediction.

The gap in the art is a complete, facility-deployable system that: (a) uses ceiling-mounted FMCW mmWave radar to simultaneously monitor vital signs of multiple animals without any contact sensors or per-animal hardware, (b) decomposes overlapping micro-Doppler signatures to extract per-animal respiration and heart rate in dense environments (feedlot pens, dairy barns, swine finishers), (c) runs all inference at the edge with no cloud dependency, (d) applies livestock-specific physiological models calibrated to species and breed norms, and (e) performs herd-level epidemiological anomaly propagation analysis to distinguish individual illness from outbreak events.

Detailed Description

1. Radar Hardware Architecture

Each sensor unit is a ceiling-mounted or overhead-beam-mounted FMCW radar module operating in one of two bands:

  • 60 GHz (V-band): Using a 4-TX, 4-RX MIMO antenna array (e.g., based on the Infineon BGT60TR24B transceiver, unit cost $18). Bandwidth: 4 GHz (57-61 GHz). Range resolution: 3.75 cm. Maximum unambiguous range: 12 m. This band is unlicensed globally under FCC Part 15.255 and ETSI EN 302 567.
  • 77 GHz (W-band): Using a Texas Instruments IWR6843 or equivalent single-chip radar SoC with 3-TX, 4-RX MIMO antenna array. Bandwidth: 4 GHz (76-81 GHz). Range resolution: 3.75 cm. Integrated ARM Cortex-R4F DSP + C674x DSP for on-chip FFT processing. Unit cost: $22 (chip) + $35 (PCB, enclosure, PoE interface) = $57 per unit.

Each unit is enclosed in an IP66-rated polycarbonate housing with an RF-transparent radome window (PTFE or Rexolite lens). Mounting height: 3-5 m above the animal plane. Field of view: ±60° azimuth, ±30° elevation per unit. Effective coverage area per unit: approximately 50-80 m² at 4 m mounting height, sufficient for a standard 12-head dairy freestall pen or a 20 m section of feedlot bunk line.

Facility-wide coverage uses overlapping radar units spaced at 8-12 m intervals along overhead beams. A typical 1,000-head feedlot requires 40-60 radar units. Total hardware cost: $2,280-3,420 ($2.28-3.42 per head), amortized over the sensor's projected 7-year operational life.

2. FMCW Signal Processing and Multi-Animal Separation

The radar transmits linear frequency chirps (chirp duration: 60 μs, chirp slope: 66.7 MHz/μs for 4 GHz bandwidth, chirp repetition rate: 5 kHz). For each chirp, the received IF (intermediate frequency) signal is digitized at 10 MSPS with 12-bit resolution.

Processing proceeds in three stages:

Stage 1: Range-Doppler-Angle Cube. A 2D FFT (range FFT across ADC samples within each chirp, Doppler FFT across consecutive chirps in a frame of 128 chirps = 25.6 ms frame time) produces a range-Doppler map. The MIMO virtual array (12 virtual elements for IWR6843) enables digital beamforming via a third FFT (angle FFT) or Capon/MUSIC super-resolution algorithms, yielding a 3D range-angle-Doppler cube updated at 39 frames/second.

Stage 2: Multi-Animal Detection and Tracking. Constant false alarm rate (CFAR) detection on the range-angle map identifies discrete scattering centers corresponding to individual animal torsos. An extended Kalman filter (EKF) tracker maintains persistent tracks for each animal, associating detections across frames using a global nearest-neighbor assignment with gating based on predicted position, velocity, and radar cross-section (RCS). Typical cattle RCS at 77 GHz: 5-15 dBsm depending on aspect angle. The tracker maintains identity through occlusions (animal behind another) using motion prediction and re-association logic with a 30-second track coast timeout.

Stage 3: Per-Animal Micro-Doppler Extraction. For each tracked animal, the system extracts the phase history of the dominant range-angle bin corresponding to the animal's torso centroid. The raw phase signal φ(t) encodes chest wall displacement d(t) as φ(t) = 4πd(t)/λ, where λ = 3.9 mm at 77 GHz. A displacement of 5 mm (typical bovine thoracic excursion during respiration) produces a phase shift of 16.1 radians, providing excellent SNR. The phase signal is unwrapped and bandpass-filtered into two channels: respiration band (0.15-0.8 Hz, corresponding to 9-48 breaths/min) and cardiac band (0.7-2.5 Hz, corresponding to 42-150 bpm). Heart rate extraction uses an arctangent demodulation algorithm followed by autocorrelation peak detection with harmonic rejection to suppress respiration harmonics that fall within the cardiac band.

3. Per-Animal Health State Classification

A lightweight temporal convolutional network (TCN) processes a rolling 60-minute window of per-animal vital sign features. The input feature vector (computed every 30 seconds) contains 14 elements:

  • Respiratory features (5): mean respiration rate, respiration rate standard deviation, peak inspiratory velocity (from Doppler), respiratory effort index (ratio of inspiratory to expiratory duration), and apnea event count (inter-breath interval > 2× mean).
  • Cardiac features (4): mean heart rate, heart rate variability (SDNN over 5-minute windows), low-frequency/high-frequency HRV ratio (LF/HF, a sympathovagal balance indicator), and cardiac rhythm regularity index.
  • Locomotion features (3): total distance traveled per hour, lying bout duration (time stationary with low centroid height), and gait symmetry index (computed from Doppler spectrogram periodicity during walking episodes).
  • Context features (2): time of day (sine/cosine encoded), ambient temperature from a co-located BME280 sensor.

The TCN architecture comprises 4 dilated causal convolutional blocks (dilation factors 1, 2, 4, 8) with 32 filters each, followed by a global average pooling layer and a 5-class softmax output: healthy, pre-clinical respiratory (elevated resp. rate, normal temp proxy), pre-clinical systemic (HRV depression, locomotion change), clinical illness (multiple vital sign deviations), and post-treatment recovery. Model size: 62 KB (INT8 quantized). Inference time: 8 ms on the IWR6843's C674x DSP.

Training data is bootstrapped from existing labeled datasets where BRD-positive animals were identified by clinical diagnosis and paired with retrospective sensor data. Transfer learning from human vital sign models provides initialization for the cardiac and respiratory feature extractors.

4. Breed and Species Calibration

Baseline vital sign ranges vary significantly across livestock species and breeds. The system maintains a calibration lookup table:

Species / BreedResting Resp. Rate (breaths/min)Resting Heart Rate (bpm)Thoracic Displacement (mm)
Beef cattle (Angus adult)15-3048-844-8
Dairy cattle (Holstein adult)18-2848-845-10
Swine (finisher, 100 kg)15-2570-1202-5
Sheep (adult ewe)16-3460-1202-4
Poultry (broiler)15-30200-4000.3-0.8

For poultry, the reduced thoracic displacement (sub-millimeter) pushes phase SNR limits. The system addresses this by using the 60 GHz band (λ = 5 mm, yielding 0.38 radians of phase shift for 0.3 mm displacement) with coherent integration over 10-second windows, achieving sufficient SNR for flock-level respiratory rate estimation (not individual bird identification).

Calibration tables are selectable via facility configuration. An auto-calibration mode observes the herd for 72 hours after installation, fits Gaussian mixture models to the population's baseline vital sign distributions, and automatically adjusts detection thresholds to facility-specific norms.

5. Epidemiological Anomaly Propagation Module

Individual animal anomalies are common (stress, minor injury, estrus). The critical capability is distinguishing isolated events from spatiotemporally correlated anomaly clusters that indicate contagious disease spread.

The propagation module constructs a dynamic spatial graph where nodes represent individual animals and edges represent proximity events (two animals within 2 m for > 5 minutes, as measured by the radar tracking system). Edge weights encode cumulative contact duration over rolling 7-day windows. When multiple animals within a connected subgraph transition from "healthy" to "pre-clinical" states within a 48-hour window, the system computes a propagation score using a discrete-time susceptible-infected (SI) model fitted to the observed transition rate. If the fitted basic reproduction number R₀ exceeds 1.0 with 90% confidence (bootstrapped from the observed contact graph and transition times), the system escalates from individual animal alerts to a herd-level outbreak warning.

This outbreak warning includes: the estimated index case (animal with the earliest onset), the predicted attack rate under the current contact pattern, a spatial heatmap of infection probability per animal based on graph distance from confirmed cases, and a recommended isolation zone (connected subgraph components within 2 hops of confirmed cases).

6. Edge Computing and Data Architecture

All signal processing, tracking, and health classification run on the radar SoC's embedded DSP (for per-unit processing) or on a facility-edge server (NVIDIA Jetson Orin Nano, $199, for facility-wide aggregation and epidemiological modeling). No cloud connectivity is required for real-time operation. A facility with 60 radar units generates approximately 2.4 MB/min of compressed telemetry (per-animal vital signs, track positions, health classifications) stored locally on a 1 TB NVMe drive with 290-day retention at full capacity.

Alert delivery uses local protocols: MQTT messages to facility management software, SMS/email via optional cellular gateway, and integration with existing herd management systems (e.g., Dairy Comp 305, Valley Vet) via a REST API exposing JSON endpoints for per-animal health scores, herd-level outbreak status, and historical vital sign time series.

7. Figures Description

  • Figure 1: Facility layout showing ceiling-mounted radar unit placement in a 1,000-head feedlot with overlapping coverage zones, edge server location, and alert delivery pathways.
  • Figure 2: FMCW signal processing pipeline from chirp generation through range-Doppler-angle cube to per-animal micro-Doppler phase extraction, showing respiration and cardiac band separation.
  • Figure 3: Example 60-minute vital sign time series for a healthy animal and a pre-clinical BRD case, showing elevated respiration rate and depressed HRV 36 hours before clinical diagnosis.
  • Figure 4: Epidemiological propagation graph showing animal contact network, anomaly onset times, and fitted SI model with estimated R₀ for a simulated BRD outbreak event.

Claims

  1. A system for non-contact continuous vital sign monitoring of livestock, comprising: one or more ceiling-mounted or overhead-mounted FMCW millimeter-wave radar units operating in the 60 GHz or 77 GHz band; a multi-animal detection and tracking module that separates radar returns from individual animals within a shared enclosure using range-angle-Doppler processing and extended Kalman filter tracking; and a per-animal micro-Doppler extraction module that derives respiration rate and heart rate from the phase history of each tracked animal's dominant scattering center.
  2. The system of claim 1, wherein per-animal chest wall displacement is extracted via arctangent demodulation of the radar phase signal and bandpass-filtered into a respiration channel (0.15-0.8 Hz) and a cardiac channel (0.7-2.5 Hz) with harmonic rejection to prevent respiration harmonics from contaminating cardiac rate estimation.
  3. The system of claim 1, further comprising an edge-deployed temporal convolutional network that classifies each tracked animal into health states including healthy, pre-clinical respiratory, pre-clinical systemic, clinical illness, and post-treatment recovery, based on a rolling window of respiratory, cardiac, locomotion, and environmental features.
  4. The system of claim 1, further comprising a breed and species calibration module that adjusts vital sign baseline ranges and anomaly detection thresholds according to species-specific and breed-specific physiological norms, with an auto-calibration mode that fits Gaussian mixture models to the herd's observed baseline distributions over an initial observation period.
  5. The system of claim 1, further comprising an epidemiological anomaly propagation module that constructs a dynamic spatial contact graph from radar-measured animal proximity events, fits a discrete-time susceptible-infected model to observed health state transitions within the graph, and generates herd-level outbreak warnings when the estimated basic reproduction number R₀ exceeds a configurable threshold.
  6. A method for early detection of disease outbreaks in commercial livestock facilities, comprising: continuously transmitting FMCW millimeter-wave chirps from ceiling-mounted radar arrays; processing received signals to generate range-angle-Doppler cubes and tracking individual animals via persistent tracker associations; extracting per-animal respiration rate and heart rate from micro-Doppler phase signatures; classifying per-animal health states using an edge-deployed neural network; constructing a dynamic contact graph from radar-measured proximity events; and escalating to herd-level outbreak alerts when spatiotemporally clustered health state transitions indicate contagious disease propagation.
  7. The method of claim 6, wherein the system detects pre-clinical signs of bovine respiratory disease, mastitis, or lameness 24-72 hours before clinical symptoms become visible, by identifying sustained elevation in respiration rate, depression of heart rate variability, or gait symmetry index degradation relative to the animal's own rolling baseline and breed-specific population norms.
  8. The method of claim 6, further comprising generating a recommended isolation zone by identifying connected subgraph components within a configurable graph-hop distance of animals classified as pre-clinical or clinical, and transmitting the isolation recommendation with spatial coordinates to facility management systems.
  9. The system of claim 1, wherein all signal processing, animal tracking, health classification, and epidemiological modeling are performed on edge computing hardware co-located at the facility with no cloud connectivity required, and wherein compressed telemetry is stored locally with configurable retention periods.
  10. The system of claim 1, wherein the MIMO virtual antenna array enables angular resolution sufficient to separate animals standing within 1.5 meters of each other using super-resolution algorithms (Capon beamforming or MUSIC), and wherein the tracker maintains animal identity through brief occlusion events using motion prediction and re-association with a configurable coast timeout.

Prior Art References

  1. Global Burden of Animal Diseases (2024) — Quantifying cost of disease in livestock: $300B+ annual global losses
  2. Comprehensive Review: Bovine Respiratory Disease (2025) — Epidemiology, diagnostic challenges, $900M annual U.S. cost
  3. BRD prediction technology comparison (2025) — Pen rider sensitivity 56-62% for BRD detection
  4. Wearable wireless biosensor technology for monitoring cattle (2021) — Review of ear tags, boluses, accelerometers
  5. IR thermography for livestock health (2023) — 73% sensitivity, 64% specificity for BRD
  6. Zhang et al. (2024) — Contactless vital signs monitoring in macaques using mm-wave FMCW radar, MAE 1.2 bpm HR / 0.8 breaths/min RR
  7. Method for distinguishing humans and animals in vital signs monitoring using IR-UWB radar (2019)
  8. WO2015174879A1 — Mm-wave radar vital signs detection apparatus (human-focused, single-target)
  9. US20090189740A1 — Contactless vital signs monitoring of animals (UWB, no micro-Doppler decomposition, no outbreak modeling)
  10. US11317608B2 — Texas Instruments mmWave vital signs processing (general, no livestock application)
  11. Texas Instruments IWR6843 — 76-81 GHz FMCW radar SoC with integrated DSP
  12. smaXtec — Rumen bolus sensor for cattle temperature monitoring ($80-120/head)
  13. TensorFlow Lite for Microcontrollers — On-device ML inference runtime
  14. Application of precision livestock farming (2025) — Challenges and opportunities review