IN Brief:
- The US Army’s 501st Military Intelligence Brigade reported field experiments conducted over eight months at Camp Humphreys and Baegiri Airfield.
- Project MANOWAR uses distributed battery-powered acoustic nodes, printed housings, locally developed classification software and mesh communications.
- Reported confidence exceeded 90% for individual analysis windows in September trials, not as an independently validated overall drone detection rate.
The US Army has reported results from Project MANOWAR, a distributed acoustic sensor network designed to detect and classify small drones flying in difficult terrain. US military intelligence personnel and Army researchers have spent eight months developing and testing battery-powered nodes with 3D-printed housings and onboard machine-learning software. The latest account, published on 13 November, describes trials at Camp Humphreys and Baegiri Airfield in South Korea.
The 532nd Military Intelligence Battalion, part of the 501st Military Intelligence Brigade, led the work with the Army Research Laboratory, the 2nd Infantry Division, industry participants and Republic of Korea Army personnel. MANOWAR stands for Mobile Acoustic Network for Operational Warning and Aerial Reconnaissance. Two-week field experiments explored how an acoustic sensing layer might supplement existing counter-drone defences.
Small drones flying close to buildings or vegetation are difficult to track consistently with sensors that depend on a clear line of sight. MANOWAR uses sound as an additional observation source, listening for patterns associated with propulsion systems while radar, optical and other technologies provide complementary information. Acoustic detection can contribute warnings without emitting a tracking signal, although the useful range and quality of a reading depend on background noise, weather and terrain.
Each sensor uses a 3D-printed chassis, local processing and battery power, with a machine-learning model developed by military intelligence personnel and researchers in the Army Research Laboratory’s Mechanical Sensing Branch. Printing a housing can simplify prototype changes and the local replacement of certain parts, but the Army has not published unit costs, production yields or a validated durability specification. The project remains an experimental sensor architecture rather than an approved mass-produced item.
Sound classification requires a model to discriminate between drone propellers and numerous other acoustic sources. Captured audio is analysed over time windows, with the software evaluating patterns rather than relying on a single loudness threshold. Recordings from multiple targets, distances and noise environments can improve the model, but testing must also establish how it performs on unfamiliar aircraft and against sound reflections that complicate source identification.
Between trial events in May and September, the developers used recordings from their own sensors and labels created by military intelligence soldiers to train a smaller onboard model. Processing at the node can reduce the need to send continuous raw audio over a radio link, although the Army has not quantified the resulting bandwidth or energy savings. Edge processing also creates trade-offs among model complexity, computational load and the ability to distinguish similar signals.
By September, the Army reported more than 90% window-level confidence against low-altitude threats, alongside improved classification and localisation. That describes a reported confidence value associated with individual analysis windows, not the proportion of all drones detected, an independent false-alarm rate or weapon effectiveness. Public results do not give test sample sizes, detection ranges or an externally verified performance comparison.
Sensors are linked by what the Army describes as a self-healing mesh, in which battery-powered nodes can relay reports from other nodes. Multiple communication paths may allow information to reach a local display even if one connection is lost. Coverage and resilience will nevertheless depend on node spacing, obstructions, radio conditions, remaining battery energy and the network configuration; a mesh does not remove those constraints.
During the field experiments, operators received sensor reports through the mesh network on a separate local display. Passing detections to another interface is an integration step because warnings must be available to personnel and other defence systems in a usable form. The Army has not said that MANOWAR directly controls any interceptor, jammer or gun, and no engagement outcome follows from the reported acoustic classification results.
Locating a sound source is more demanding than merely identifying that it resembles a drone. A distributed network can potentially compare the timing or other characteristics of recordings from different known sensor positions, but reflections and uncertain sensor placement complicate the calculation. The Army says localisation improved during the trials without specifying the method, distance error or operational detection envelope. Those limits matter before the system can be incorporated into wider warning networks.
An inexpensive, replaceable sensing layer could cover more dispersed locations than a small number of higher-cost sensors, provided the network can be installed and maintained at the required density. Scale would introduce demands for battery resupply, sensor calibration, software configuration, repair and security updates. The project has not reached a stage at which an accepted production baseline, military logistics package or contracted unit cost has been published.
Testing with US and South Korean personnel also exposed technical and procedural integration gaps, according to the brigade. Trials at an allied airfield can check whether local reporting reaches operators within an agreed workflow, although the Army has not published a complete interoperability specification. No joint manufacturing contract or procurement programme with the Republic of Korea has been announced.
The 532nd Military Intelligence Battalion wants to move MANOWAR into scaled manufacture and ultimately a formal Army programme of record. Those are stated intentions: approval, procurement funding and a production contract remain outstanding. Broader trials would need to measure performance across aircraft types, terrain, distances and noise conditions while assessing cyber resilience and sustainment as well as the acoustic models.
MANOWAR’s current achievement is a set of field experiments combining printed hardware, locally developed classification software and mesh-linked reports. The reported increase in confidence between trial events offers a measurable development claim, but not an independently established detection-success rate. A formal acquisition decision and repeatable operational testing would determine whether this experimental network becomes a widely fielded capability.


