Radio Signal Intelligence
From zero to training convolutional neural networks that classify radio signals - then quantum computing and QCNNs, transformer and self-supervised RF-ML, and finally quantum RF sensing - the one quantum technology that touches radio today.
How this course works
- One spine per phase. Each phase has a single primary resource you work through in order. Everything else is a reference - open it when you are stuck, not by default.
- The exercises are the course. Reading and watching are warm-ups. You have learned a week's material when its exercises run and you can explain the results.
- Phase gates. Each phase ends with a self-test. Score yourself honestly; 80 percent is a pass. Below that, spend a catch-up week before moving on.
- Keep a lab notebook. Make one git repository on day one. Every exercise, experiment, and note lives there. By the end it is your portfolio.
- Timebox, do not perfect. If a week runs long, ship what you have and move forward. The capstones are where depth pays off.
Progress checkboxes are saved in this browser only - they will not sync between your phone and laptop. Your git lab notebook is the real record.
The eight phases
-
Phase 1 · Weeks 1-4
Foundations
Linear algebra, calculus and probability for ML, Python and NumPy fluency, and digital signal processing: I/Q data, FFTs, and modulation.
0 of 4 weeks complete -
Phase 2 · Weeks 5-12
Machine Learning to CNNs
Classical ML fundamentals, neural networks from scratch, PyTorch, then convolutional networks and the craft of training them well.
0 of 8 weeks complete -
Phase 3 · Weeks 13-15
Capstone A: Modulation Classifier
Build a CNN that classifies radio modulation schemes on the RadioML benchmark, reproducing and then beating the founding papers.
0 of 3 weeks complete -
Phase 4 · Weeks 16-21
Quantum Computing
Qubits, entanglement, circuits, and the canonical algorithms - hands-on in Qiskit, including runs on real IBM hardware.
0 of 6 weeks complete -
Phase 5 · Weeks 22-25
Quantum Machine Learning
Variational circuits, data encoding, quantum kernels, barren plateaus - and the QCNN architecture, implemented in PennyLane.
0 of 4 weeks complete -
Phase 6 · Weeks 26-28
Capstone B: Quantum Classifier
A hybrid quantum-classical and QCNN classifier on compressed radio signal features, benchmarked honestly against your Capstone A model.
0 of 3 weeks complete -
Phase 7 · Weeks 29-34
Advanced Classical RF-ML
Transformers from scratch and for signals, self-supervised pretraining on unlabeled spectrum, real captures with your own SDR, and a live over-the-air classifier.
0 of 6 weeks complete -
Phase 8 · Weeks 35-38
Quantum RF Sensing
Rydberg-atom receivers simulated with research tools, the quantum limits of measurement, and real bench hardware: an atomic clock, a coherent SDR array, a DIY quantum magnetometer.
0 of 4 weeks complete
What to buy (and what not to)
Almost everything in this course is free. These are the only purchases worth making, in priority order:
| Item | Used in | Priority |
|---|---|---|
| Hands-On Machine Learning with Scikit-Learn, Keras & TensorFlow, 3rd ed. - Aurélien Géron | Phase 2 spine (chapters 1-4, 10-14) | Buy now |
| Machine Learning with Quantum Computers, 2nd ed. - Schuld & Petruccione (Springer) | Phase 5 spine. This is the current edition of the book listed as "Supervised Learning with Quantum Computers" - buy the 2nd edition, not the 2018 one. | Buy before week 22 |
| Quantum Computation and Quantum Information - Nielsen & Chuang | Phase 4-5 reference. Never read cover to cover here; used for chapter look-ups. | Optional |
| RTL-SDR Blog v4 dongle with antenna kit (about 40 USD) | Phase 1 week 4, Capstone A stretch, and required for Phase 7 week 32 (capturing real signals) | Buy by week 32; earlier is more fun |
| Surplus rubidium frequency standard, e.g. FE-5680A (roughly 150-250 USD plus a small power supply) | Phase 8 week 37 - a real quantum device (Rb-87 hyperfine clock) on your bench, disciplining your SDR | Buy before week 37 |
| KrakenSDR 5-channel coherent receiver (about 750 USD) or DIY proton magnetometer parts (about 150 USD) | Phase 8 week 38 - pick one: phase-coherent direction finding, or building an actual quantum spin sensor | Pick one before week 38 |
| ADALM-PlutoSDR (about 230 USD) or HackRF One (about 330 USD) | Phase 7 - transmit-capable SDR for generating your own test signals into a cable or dummy load | Optional |
| Colab Pro or a cloud GPU (about 10 USD per month) | Phases 2-3, only if training feels slow on free tiers or your own machine | Only if needed |
- fast.ai and the Coursera ML Specialization - excellent, but redundant with the Géron plus CS231n spine. Use fast.ai instead of Phase 2 weeks 5-9 if you prefer top-down learning; do not do both.
- Wittek, "Quantum Machine Learning" (2014) - predates the variational-circuit era that QCNNs live in. Dropped.
- Microsoft Quantum Katas - good drills, but Q# while this course standardizes on Python. Optional extra practice in Phase 4.
- deeplearningbook.org - kept as a theory reference for Phase 2, not a read-through.
- Most items you listed as books are actually research papers - they are scheduled as readings in Phases 5 and 6, where you will have the background to enjoy them.
Classical CNNs for signal classification are mature, deployed technology. QCNNs are research: today they run on simulators or a few dozen noisy qubits, on heavily compressed inputs, and quantum advantage for this task is an open question. By Capstone B you will be reading current papers rather than catching up on decades - that is the honest meaning of "hero" here, and it is a genuinely good place to be.
Weekly rhythm that works
- Two study blocks (about 2 hours each): the week's reading or lectures.
- Two build blocks (about 2-3 hours each): the exercises, in your lab notebook repo.
- One review block (about 1 hour): write a half-page summary of what you learned, in your own words. This is the highest-leverage hour of the week.
Start with Phase 1: Foundations.