Noise & Hardware Simulation

---

Q-RLSTC includes a full noise simulation stack with no classical RLSTC equivalent. All components are in the quantum/ package.

Backend Factory

Defined in backends.py:

def get_backend(mode: str, noise_model_name: str = None):
    if mode == "ideal":
        return AerSimulator()
    elif mode == "noisy_sim":
        noise_model = get_noise_model(noise_model_name)
        return AerSimulator(noise_model=noise_model)

Available Noise Profiles

Name	1Q Error	2Q Error	Readout	T₁	T₂	Target
Ideal	0	0	0	∞	∞	Algorithmic debugging
Simple	0.1%	1.0%	2.0%	—	—	Quick noise impact checks
Eagle	0.05%	0.8%	—	300μs	150μs	IBM Eagle 127-qubit emulation
Heron	0.02%	0.2%	—	400μs	200μs	IBM Heron next-gen emulation

Estimated Circuit Fidelity

For the 5-qubit VQ-DQN (depth ~11, 8 CNOTs):

Backend	Estimated Fidelity
Ideal	100%
Simple	~90%
Eagle	~85%
Heron	~95%

Readout Error Mitigation

Defined in mitigation.py:

class ReadoutMitigator:
    def calibrate(self, backend, shots=8192):
        """Build calibration matrix by measuring all basis states."""
        # Run 2ⁿ calibration circuits
        # Build M[i,j] = P(measure i | prepared j)
        self.calibration_matrix = M

    def apply(self, counts):
        """Correct raw counts via matrix pseudo-inverse."""
        # Solve linear system → corrected probabilities
        # Clip negative values, re-normalise
        return corrected_counts

Mitigation Pipeline

Raw counts → Calibration matrix (pseudo-inverse) → Clip negatives → Re-normalise → Corrected counts

Falls back to pass-through if no calibration has been performed.

Configuration

@dataclass
class NoiseConfig:
    use_noise: bool = False
    noise_model: str = "depolarizing"     # or "thermal", "ibm_fake"
    use_mitigation: bool = True
    calibration_shots: int = 8192

Noise Impact on Training

Effect	Mechanism	Mitigation
Q-value variance	Shot noise in expectations	More shots (512→4096 for evaluation)
Gradient noise	SPSA gradient estimate corrupted	Decaying perturbation cₖ
Readout bias	Systematic measurement errors	Calibration matrix
Decoherence	T₁/T₂ decay during circuit	Shallow depth (≤11 layers)

---

Next: Experimental Design →