Skip to main content

05Experience

Royal Military College of Canada

Machine Learning Researcher

Developed and evaluated more than five PyTorch CNN architectures for classifying five RF-signal categories from a corpus of approximately 150,000 samples.

May 2025 – Sep 2025Kingston, Ontario, Canada · On-site
  • Python
  • PyTorch
  • CNNs
  • NumPy
  • Signal processing
  • RTL-SDR
  • Embedded inference
  1. CaptureRTL-SDR captures
  2. Dataset150K samples, 5 classes
  3. PrepareAugmentation + normalization
  4. Model5+ PyTorch CNNs
  5. EvaluateUnseen + noisy-condition tests

Public-scope workflow+1 more

Contributions

  • Built and compared more than five convolutional neural-network architectures.
  • Worked with a roughly 150,000-sample, five-class RF dataset.
  • Used RTL-SDR-oriented signal-processing and ML workflows.
  • Evaluated held-out and unseen performance and noisy-condition behavior.
  • Improved noisy-condition accuracy by approximately 10%.
  • Contributed to an embedded inference workflow operating at roughly 2 Hz.

Context

The work emphasized noisy-condition robustness and its connection to an embedded inference pipeline, rather than optimizing a single clean-data score.

In one line

Developed and compared PyTorch CNNs for five-class RF-signal classification on approximately 150,000 samples, including noisy-condition and embedded tests.

Measured outcomes

  • ≈90%

    Average accuracy on unseen real-world signals

    Research-role result reported in the current Master and MLE/SWE resumes and the LinkedIn record.

    Approximate; no public dataset supports reproduction.

  • ≈10%

    Improvement in noisy-band generalization

    Achieved through data augmentation, learning-rate decay, frequency-offset sampling, and spectrogram normalization.

    The source does not state relative versus percentage-point lift; it is reported as written.

  • 150K / 5

    Sample RF dataset and signal classes

    Dataset scale as reported in the current resume.

  • ≈2 Hz

    Embedded inference cadence

    Refresh rate supported by the embedded inference pipeline.

    An inference cadence, not a latency percentile.

Tools & stack

  • Python
  • PyTorch
  • CNNs
  • NumPy
  • Signal processing
  • RTL-SDR
  • Embedded inference

Scope of this page

This page includes only the high-level project scale and approximate outcomes already present in the current public resumes and LinkedIn profile. Code, raw data, signal details, and operational context are not public, and the specific signal classes are not named.

  • Every figure keeps its approximation mark, as the source states it.
  • Not production-deployed; no operational or state-of-the-art claim is made.