Modern radar systems operate in increasingly crowded and contested electromagnetic environments. Multiple emitters can produce pulse streams at the same time, creating a complex signal classification challenge: how do we separate those pulses accurately, identify which signals belong together and support faster recognition of potential threats?

Working with QinetiQ, OQC has developed a hybrid quantum-classical AI approach to radar pulse deinterleaving. The work explores how quantum techniques can support higher-confidence clustering in some of the most challenging cases, where signals are ambiguous and conventional approaches may struggle.

The result is a hybrid pipeline that combines the strengths of classical machine learning with a quantum-enhanced model. The classical AI model processes the majority of signals, while uncertain pulse pairs are routed to a quantum stage, the Guided Quantum Compressor (GQC) for additional analysis.

This approach reflects a practical route to quantum utility: applying quantum where it can add value, while continuing to use classical computing where it performs well.

The Challenge: Separating Signals From Multiple Emitters

When a radar receives pulses from multiple emitters, those pulses must be clustered so that each group corresponds to a single source. Accurate clustering is critical as misclassified signals can lead to missed threats, delayed responses or false alarms.

Radar pulse deinterleaving is especially challenging because the number of emitters may be unknown, signals may overlap and the electromagnetic environment may change rapidly. In these conditions, confidence matters.

To address this, the algorithm treats each pulse as a node in a graph. It samples pairs of pulses, engineers new features based on the differences between each pulse’s characteristics and determines whether the two pulses belong to the same emitter.

If the model answers yes with high confidence, an edge is created between the two pulse nodes. The final clusters are then identified by finding the connected components of the graph, with each cluster representing a likely emitter.

A Hybrid Quantum-Classical Pipeline

The approach uses a classical machine learning model as the main pipeline. This model handles most pulse pair classifications efficiently with a mixture-of-experts framework. A router decides which classifier to use, classical or quantum, and routes the data accordingly to its decision.

The GQC acts as a quantum expert within the workflow. Its role is not to replace the classical model, but to support it in the hardest cases. This hybrid design allows the system to focus quantum resources where they are most useful.

The approach builds on anomaly detection techniques OQC has previously explored in financial services. While radar pulse deinterleaving is a very different application, the underlying problem has a familiar structure: identifying unusual, ambiguous, or high-risk patterns in complex data.

Measuring Performance

The results were generated using the OQC Toshiko simulator and evaluated against three clustering metrics: completeness, homogeneity and V-measure. We used what is considered a scan dataset in the Turing Deinterleaving Challenge.

Completeness measures whether pulses from the same emitter are grouped together.

Homogeneity looks at the quality of each cluster, measuring whether the signals within it belong to the same emitter.

V-measure combines completeness and homogeneity into a single score, providing a balanced view of clustering performance.

The hybrid quantum-classical pipeline outperformed both the Alan Turing Institute benchmark and an equivalent purely classical version of the algorithm. The results showed a small trade-off: homogeneity reduced slightly, meaning some clusters were marginally less pure, but completeness improved more significantly. Overall, this led to a higher V-measure.

In practical terms, the quantum-enhanced approach grouped pulses from the same emitter more consistently, improving overall radar pulse deinterleaving performance.

Why It Matters

This work demonstrates how quantum techniques can be integrated into practical AI workflows today. Rather than positioning quantum as a wholesale replacement for classical computing, it shows the value of a targeted hybrid approach: classical systems handle what they do best, while quantum techniques are applied to the hardest and most ambiguous parts of the problem.

For defence and security applications, improved signal classification can support faster, more reliable identification of potential threats in complex electromagnetic environments.

For quantum computing more broadly, this is an important step toward operationally relevant use cases. It shows how quantum-enhanced machine learning can be tested, benchmarked, and applied to real-world challenges using OQC’s platform.

At OQC, we are focused on making quantum accessible, usable, and valuable. Collaborations like this help move quantum from possibility to practical exploration, giving organisations a clearer path to understanding where quantum can make a measurable difference.